3 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
7 * Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
8 * Copyright (c) 2010 Janne Grunau <janne-ffmpeg@jannau.net>
10 * This file is part of FFmpeg.
12 * FFmpeg is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public
14 * License as published by the Free Software Foundation; either
15 * version 2.1 of the License, or (at your option) any later version.
17 * FFmpeg is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with FFmpeg; if not, write to the Free Software
24 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
30 * @author Oded Shimon ( ods15 ods15 dyndns org )
31 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
38 * N (code in SoC repo) gain control
40 * Y window shapes - standard
41 * N window shapes - Low Delay
42 * Y filterbank - standard
43 * N (code in SoC repo) filterbank - Scalable Sample Rate
44 * Y Temporal Noise Shaping
45 * Y Long Term Prediction
48 * Y frequency domain prediction
49 * Y Perceptual Noise Substitution
51 * N Scalable Inverse AAC Quantization
52 * N Frequency Selective Switch
54 * Y quantization & coding - AAC
55 * N quantization & coding - TwinVQ
56 * N quantization & coding - BSAC
57 * N AAC Error Resilience tools
58 * N Error Resilience payload syntax
59 * N Error Protection tool
61 * N Silence Compression
64 * N Structured Audio tools
65 * N Structured Audio Sample Bank Format
67 * N Harmonic and Individual Lines plus Noise
68 * N Text-To-Speech Interface
69 * Y Spectral Band Replication
70 * Y (not in this code) Layer-1
71 * Y (not in this code) Layer-2
72 * Y (not in this code) Layer-3
73 * N SinuSoidal Coding (Transient, Sinusoid, Noise)
75 * N Direct Stream Transfer
77 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
78 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
88 #include "fmtconvert.h"
95 #include "aacdectab.h"
96 #include "cbrt_tablegen.h"
99 #include "mpeg4audio.h"
100 #include "aacadtsdec.h"
108 # include "arm/aac.h"
116 static VLC vlc_scalefactors;
117 static VLC vlc_spectral[11];
119 static const char overread_err[] = "Input buffer exhausted before END element found\n";
121 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
123 // For PCE based channel configurations map the channels solely based on tags.
124 if (!ac->m4ac.chan_config) {
125 return ac->tag_che_map[type][elem_id];
127 // For indexed channel configurations map the channels solely based on position.
128 switch (ac->m4ac.chan_config) {
130 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
132 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
135 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
136 instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
137 encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
138 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
140 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
143 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
145 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
148 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
150 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
154 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
156 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
157 } else if (ac->m4ac.chan_config == 2) {
161 if (!ac->tags_mapped && type == TYPE_SCE) {
163 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
171 * Check for the channel element in the current channel position configuration.
172 * If it exists, make sure the appropriate element is allocated and map the
173 * channel order to match the internal FFmpeg channel layout.
175 * @param che_pos current channel position configuration
176 * @param type channel element type
177 * @param id channel element id
178 * @param channels count of the number of channels in the configuration
180 * @return Returns error status. 0 - OK, !0 - error
182 static av_cold int che_configure(AACContext *ac,
183 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
187 if (che_pos[type][id]) {
188 if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
189 return AVERROR(ENOMEM);
190 ff_aac_sbr_ctx_init(&ac->che[type][id]->sbr);
191 if (type != TYPE_CCE) {
192 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
193 if (type == TYPE_CPE ||
194 (type == TYPE_SCE && ac->m4ac.ps == 1)) {
195 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
199 if (ac->che[type][id])
200 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
201 av_freep(&ac->che[type][id]);
207 * Configure output channel order based on the current program configuration element.
209 * @param che_pos current channel position configuration
210 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
212 * @return Returns error status. 0 - OK, !0 - error
214 static av_cold int output_configure(AACContext *ac,
215 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
216 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
217 int channel_config, enum OCStatus oc_type)
219 AVCodecContext *avctx = ac->avctx;
220 int i, type, channels = 0, ret;
222 if (new_che_pos != che_pos)
223 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
225 if (channel_config) {
226 for (i = 0; i < tags_per_config[channel_config]; i++) {
227 if ((ret = che_configure(ac, che_pos,
228 aac_channel_layout_map[channel_config - 1][i][0],
229 aac_channel_layout_map[channel_config - 1][i][1],
234 memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
236 avctx->channel_layout = aac_channel_layout[channel_config - 1];
238 /* Allocate or free elements depending on if they are in the
239 * current program configuration.
241 * Set up default 1:1 output mapping.
243 * For a 5.1 stream the output order will be:
244 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
247 for (i = 0; i < MAX_ELEM_ID; i++) {
248 for (type = 0; type < 4; type++) {
249 if ((ret = che_configure(ac, che_pos, type, i, &channels)))
254 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
256 avctx->channel_layout = 0;
259 avctx->channels = channels;
261 ac->output_configured = oc_type;
267 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
269 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
270 * @param sce_map mono (Single Channel Element) map
271 * @param type speaker type/position for these channels
273 static void decode_channel_map(enum ChannelPosition *cpe_map,
274 enum ChannelPosition *sce_map,
275 enum ChannelPosition type,
276 GetBitContext *gb, int n)
279 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
280 map[get_bits(gb, 4)] = type;
285 * Decode program configuration element; reference: table 4.2.
287 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
289 * @return Returns error status. 0 - OK, !0 - error
291 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
292 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
295 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
298 skip_bits(gb, 2); // object_type
300 sampling_index = get_bits(gb, 4);
301 if (m4ac->sampling_index != sampling_index)
302 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");
304 num_front = get_bits(gb, 4);
305 num_side = get_bits(gb, 4);
306 num_back = get_bits(gb, 4);
307 num_lfe = get_bits(gb, 2);
308 num_assoc_data = get_bits(gb, 3);
309 num_cc = get_bits(gb, 4);
312 skip_bits(gb, 4); // mono_mixdown_tag
314 skip_bits(gb, 4); // stereo_mixdown_tag
317 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
319 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
320 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
321 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
322 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
324 skip_bits_long(gb, 4 * num_assoc_data);
326 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
330 /* comment field, first byte is length */
331 comment_len = get_bits(gb, 8) * 8;
332 if (get_bits_left(gb) < comment_len) {
333 av_log(avctx, AV_LOG_ERROR, overread_err);
336 skip_bits_long(gb, comment_len);
341 * Set up channel positions based on a default channel configuration
342 * as specified in table 1.17.
344 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
346 * @return Returns error status. 0 - OK, !0 - error
348 static av_cold int set_default_channel_config(AVCodecContext *avctx,
349 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
352 if (channel_config < 1 || channel_config > 7) {
353 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
358 /* default channel configurations:
360 * 1ch : front center (mono)
361 * 2ch : L + R (stereo)
362 * 3ch : front center + L + R
363 * 4ch : front center + L + R + back center
364 * 5ch : front center + L + R + back stereo
365 * 6ch : front center + L + R + back stereo + LFE
366 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
369 if (channel_config != 2)
370 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
371 if (channel_config > 1)
372 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
373 if (channel_config == 4)
374 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
375 if (channel_config > 4)
376 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
377 = AAC_CHANNEL_BACK; // back stereo
378 if (channel_config > 5)
379 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
380 if (channel_config == 7)
381 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
387 * Decode GA "General Audio" specific configuration; reference: table 4.1.
389 * @param ac pointer to AACContext, may be null
390 * @param avctx pointer to AVCCodecContext, used for logging
392 * @return Returns error status. 0 - OK, !0 - error
394 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
396 MPEG4AudioConfig *m4ac,
399 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
400 int extension_flag, ret;
402 if (get_bits1(gb)) { // frameLengthFlag
403 av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
407 if (get_bits1(gb)) // dependsOnCoreCoder
408 skip_bits(gb, 14); // coreCoderDelay
409 extension_flag = get_bits1(gb);
411 if (m4ac->object_type == AOT_AAC_SCALABLE ||
412 m4ac->object_type == AOT_ER_AAC_SCALABLE)
413 skip_bits(gb, 3); // layerNr
415 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
416 if (channel_config == 0) {
417 skip_bits(gb, 4); // element_instance_tag
418 if ((ret = decode_pce(avctx, m4ac, new_che_pos, gb)))
421 if ((ret = set_default_channel_config(avctx, new_che_pos, channel_config)))
424 if (ac && (ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
427 if (extension_flag) {
428 switch (m4ac->object_type) {
430 skip_bits(gb, 5); // numOfSubFrame
431 skip_bits(gb, 11); // layer_length
435 case AOT_ER_AAC_SCALABLE:
437 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
438 * aacScalefactorDataResilienceFlag
439 * aacSpectralDataResilienceFlag
443 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
449 * Decode audio specific configuration; reference: table 1.13.
451 * @param ac pointer to AACContext, may be null
452 * @param avctx pointer to AVCCodecContext, used for logging
453 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
454 * @param data pointer to AVCodecContext extradata
455 * @param data_size size of AVCCodecContext extradata
457 * @return Returns error status or number of consumed bits. <0 - error
459 static int decode_audio_specific_config(AACContext *ac,
460 AVCodecContext *avctx,
461 MPEG4AudioConfig *m4ac,
462 const uint8_t *data, int data_size)
467 init_get_bits(&gb, data, data_size * 8);
469 if ((i = ff_mpeg4audio_get_config(m4ac, data, data_size)) < 0)
471 if (m4ac->sampling_index > 12) {
472 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
475 if (m4ac->sbr == 1 && m4ac->ps == -1)
478 skip_bits_long(&gb, i);
480 switch (m4ac->object_type) {
484 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
488 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
489 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
493 return get_bits_count(&gb);
497 * linear congruential pseudorandom number generator
499 * @param previous_val pointer to the current state of the generator
501 * @return Returns a 32-bit pseudorandom integer
503 static av_always_inline int lcg_random(int previous_val)
505 return previous_val * 1664525 + 1013904223;
508 static av_always_inline void reset_predict_state(PredictorState *ps)
518 static void reset_all_predictors(PredictorState *ps)
521 for (i = 0; i < MAX_PREDICTORS; i++)
522 reset_predict_state(&ps[i]);
525 static void reset_predictor_group(PredictorState *ps, int group_num)
528 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
529 reset_predict_state(&ps[i]);
532 #define AAC_INIT_VLC_STATIC(num, size) \
533 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
534 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
535 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
538 static av_cold int aac_decode_init(AVCodecContext *avctx)
540 AACContext *ac = avctx->priv_data;
543 ac->m4ac.sample_rate = avctx->sample_rate;
545 if (avctx->extradata_size > 0) {
546 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
548 avctx->extradata_size) < 0)
552 /* ffdshow custom code */
553 #if CONFIG_AUDIO_FLOAT
554 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
556 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
559 AAC_INIT_VLC_STATIC( 0, 304);
560 AAC_INIT_VLC_STATIC( 1, 270);
561 AAC_INIT_VLC_STATIC( 2, 550);
562 AAC_INIT_VLC_STATIC( 3, 300);
563 AAC_INIT_VLC_STATIC( 4, 328);
564 AAC_INIT_VLC_STATIC( 5, 294);
565 AAC_INIT_VLC_STATIC( 6, 306);
566 AAC_INIT_VLC_STATIC( 7, 268);
567 AAC_INIT_VLC_STATIC( 8, 510);
568 AAC_INIT_VLC_STATIC( 9, 366);
569 AAC_INIT_VLC_STATIC(10, 462);
573 dsputil_init(&ac->dsp, avctx);
574 ff_fmt_convert_init(&ac->fmt_conv, avctx);
576 ac->random_state = 0x1f2e3d4c;
578 // -1024 - Compensate wrong IMDCT method.
579 // 60 - Required to scale values to the correct range [-32768,32767]
580 // for float to int16 conversion. (1 << (60 / 4)) == 32768
581 ac->sf_scale = 1. / -1024.;
586 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
587 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
588 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
591 ff_mdct_init(&ac->mdct, 11, 1, 1.0);
592 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
593 ff_mdct_init(&ac->mdct_ltp, 11, 0, 1.0);
594 // window initialization
595 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
596 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
597 ff_init_ff_sine_windows(10);
598 ff_init_ff_sine_windows( 7);
606 * Skip data_stream_element; reference: table 4.10.
608 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
610 int byte_align = get_bits1(gb);
611 int count = get_bits(gb, 8);
613 count += get_bits(gb, 8);
617 if (get_bits_left(gb) < 8 * count) {
618 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
621 skip_bits_long(gb, 8 * count);
625 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
630 ics->predictor_reset_group = get_bits(gb, 5);
631 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
632 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
636 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
637 ics->prediction_used[sfb] = get_bits1(gb);
643 * Decode Long Term Prediction data; reference: table 4.xx.
645 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
646 GetBitContext *gb, uint8_t max_sfb)
650 ltp->lag = get_bits(gb, 11);
651 ltp->coef = ltp_coef[get_bits(gb, 3)] * ac->sf_scale;
652 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
653 ltp->used[sfb] = get_bits1(gb);
657 * Decode Individual Channel Stream info; reference: table 4.6.
659 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
661 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
662 GetBitContext *gb, int common_window)
665 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
666 memset(ics, 0, sizeof(IndividualChannelStream));
669 ics->window_sequence[1] = ics->window_sequence[0];
670 ics->window_sequence[0] = get_bits(gb, 2);
671 ics->use_kb_window[1] = ics->use_kb_window[0];
672 ics->use_kb_window[0] = get_bits1(gb);
673 ics->num_window_groups = 1;
674 ics->group_len[0] = 1;
675 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
677 ics->max_sfb = get_bits(gb, 4);
678 for (i = 0; i < 7; i++) {
680 ics->group_len[ics->num_window_groups - 1]++;
682 ics->num_window_groups++;
683 ics->group_len[ics->num_window_groups - 1] = 1;
686 ics->num_windows = 8;
687 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
688 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
689 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
690 ics->predictor_present = 0;
692 ics->max_sfb = get_bits(gb, 6);
693 ics->num_windows = 1;
694 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
695 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
696 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
697 ics->predictor_present = get_bits1(gb);
698 ics->predictor_reset_group = 0;
699 if (ics->predictor_present) {
700 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
701 if (decode_prediction(ac, ics, gb)) {
702 memset(ics, 0, sizeof(IndividualChannelStream));
705 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
706 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
707 memset(ics, 0, sizeof(IndividualChannelStream));
710 if ((ics->ltp.present = get_bits(gb, 1)))
711 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
716 if (ics->max_sfb > ics->num_swb) {
717 av_log(ac->avctx, AV_LOG_ERROR,
718 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
719 ics->max_sfb, ics->num_swb);
720 memset(ics, 0, sizeof(IndividualChannelStream));
728 * Decode band types (section_data payload); reference: table 4.46.
730 * @param band_type array of the used band type
731 * @param band_type_run_end array of the last scalefactor band of a band type run
733 * @return Returns error status. 0 - OK, !0 - error
735 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
736 int band_type_run_end[120], GetBitContext *gb,
737 IndividualChannelStream *ics)
740 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
741 for (g = 0; g < ics->num_window_groups; g++) {
743 while (k < ics->max_sfb) {
744 uint8_t sect_end = k;
746 int sect_band_type = get_bits(gb, 4);
747 if (sect_band_type == 12) {
748 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
751 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
752 sect_end += sect_len_incr;
753 sect_end += sect_len_incr;
754 if (get_bits_left(gb) < 0) {
755 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
758 if (sect_end > ics->max_sfb) {
759 av_log(ac->avctx, AV_LOG_ERROR,
760 "Number of bands (%d) exceeds limit (%d).\n",
761 sect_end, ics->max_sfb);
764 for (; k < sect_end; k++) {
765 band_type [idx] = sect_band_type;
766 band_type_run_end[idx++] = sect_end;
774 * Decode scalefactors; reference: table 4.47.
776 * @param global_gain first scalefactor value as scalefactors are differentially coded
777 * @param band_type array of the used band type
778 * @param band_type_run_end array of the last scalefactor band of a band type run
779 * @param sf array of scalefactors or intensity stereo positions
781 * @return Returns error status. 0 - OK, !0 - error
783 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
784 unsigned int global_gain,
785 IndividualChannelStream *ics,
786 enum BandType band_type[120],
787 int band_type_run_end[120])
789 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
791 int offset[3] = { global_gain, global_gain - 90, 100 };
793 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
794 for (g = 0; g < ics->num_window_groups; g++) {
795 for (i = 0; i < ics->max_sfb;) {
796 int run_end = band_type_run_end[idx];
797 if (band_type[idx] == ZERO_BT) {
798 for (; i < run_end; i++, idx++)
800 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
801 for (; i < run_end; i++, idx++) {
802 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
803 if (offset[2] > 255U) {
804 av_log(ac->avctx, AV_LOG_ERROR,
805 "%s (%d) out of range.\n", sf_str[2], offset[2]);
808 sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
810 } else if (band_type[idx] == NOISE_BT) {
811 for (; i < run_end; i++, idx++) {
812 if (noise_flag-- > 0)
813 offset[1] += get_bits(gb, 9) - 256;
815 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
816 if (offset[1] > 255U) {
817 av_log(ac->avctx, AV_LOG_ERROR,
818 "%s (%d) out of range.\n", sf_str[1], offset[1]);
821 sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
824 for (; i < run_end; i++, idx++) {
825 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
826 if (offset[0] > 255U) {
827 av_log(ac->avctx, AV_LOG_ERROR,
828 "%s (%d) out of range.\n", sf_str[0], offset[0]);
831 sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
840 * Decode pulse data; reference: table 4.7.
842 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
843 const uint16_t *swb_offset, int num_swb)
846 pulse->num_pulse = get_bits(gb, 2) + 1;
847 pulse_swb = get_bits(gb, 6);
848 if (pulse_swb >= num_swb)
850 pulse->pos[0] = swb_offset[pulse_swb];
851 pulse->pos[0] += get_bits(gb, 5);
852 if (pulse->pos[0] > 1023)
854 pulse->amp[0] = get_bits(gb, 4);
855 for (i = 1; i < pulse->num_pulse; i++) {
856 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
857 if (pulse->pos[i] > 1023)
859 pulse->amp[i] = get_bits(gb, 4);
865 * Decode Temporal Noise Shaping data; reference: table 4.48.
867 * @return Returns error status. 0 - OK, !0 - error
869 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
870 GetBitContext *gb, const IndividualChannelStream *ics)
872 int w, filt, i, coef_len, coef_res, coef_compress;
873 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
874 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
875 for (w = 0; w < ics->num_windows; w++) {
876 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
877 coef_res = get_bits1(gb);
879 for (filt = 0; filt < tns->n_filt[w]; filt++) {
881 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
883 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
884 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
885 tns->order[w][filt], tns_max_order);
886 tns->order[w][filt] = 0;
889 if (tns->order[w][filt]) {
890 tns->direction[w][filt] = get_bits1(gb);
891 coef_compress = get_bits1(gb);
892 coef_len = coef_res + 3 - coef_compress;
893 tmp2_idx = 2 * coef_compress + coef_res;
895 for (i = 0; i < tns->order[w][filt]; i++)
896 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
905 * Decode Mid/Side data; reference: table 4.54.
907 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
908 * [1] mask is decoded from bitstream; [2] mask is all 1s;
909 * [3] reserved for scalable AAC
911 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
915 if (ms_present == 1) {
916 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
917 cpe->ms_mask[idx] = get_bits1(gb);
918 } else if (ms_present == 2) {
919 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
924 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
928 *dst++ = v[idx & 15] * s;
929 *dst++ = v[idx>>4 & 15] * s;
935 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
939 *dst++ = v[idx & 3] * s;
940 *dst++ = v[idx>>2 & 3] * s;
941 *dst++ = v[idx>>4 & 3] * s;
942 *dst++ = v[idx>>6 & 3] * s;
948 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
949 unsigned sign, const float *scale)
951 union float754 s0, s1;
953 s0.f = s1.f = *scale;
954 s0.i ^= sign >> 1 << 31;
957 *dst++ = v[idx & 15] * s0.f;
958 *dst++ = v[idx>>4 & 15] * s1.f;
965 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
966 unsigned sign, const float *scale)
968 unsigned nz = idx >> 12;
969 union float754 s = { .f = *scale };
972 t.i = s.i ^ (sign & 1<<31);
973 *dst++ = v[idx & 3] * t.f;
975 sign <<= nz & 1; nz >>= 1;
976 t.i = s.i ^ (sign & 1<<31);
977 *dst++ = v[idx>>2 & 3] * t.f;
979 sign <<= nz & 1; nz >>= 1;
980 t.i = s.i ^ (sign & 1<<31);
981 *dst++ = v[idx>>4 & 3] * t.f;
983 sign <<= nz & 1; nz >>= 1;
984 t.i = s.i ^ (sign & 1<<31);
985 *dst++ = v[idx>>6 & 3] * t.f;
992 * Decode spectral data; reference: table 4.50.
993 * Dequantize and scale spectral data; reference: 4.6.3.3.
995 * @param coef array of dequantized, scaled spectral data
996 * @param sf array of scalefactors or intensity stereo positions
997 * @param pulse_present set if pulses are present
998 * @param pulse pointer to pulse data struct
999 * @param band_type array of the used band type
1001 * @return Returns error status. 0 - OK, !0 - error
1003 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1004 GetBitContext *gb, const float sf[120],
1005 int pulse_present, const Pulse *pulse,
1006 const IndividualChannelStream *ics,
1007 enum BandType band_type[120])
1009 int i, k, g, idx = 0;
1010 const int c = 1024 / ics->num_windows;
1011 const uint16_t *offsets = ics->swb_offset;
1012 float *coef_base = coef;
1014 for (g = 0; g < ics->num_windows; g++)
1015 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1017 for (g = 0; g < ics->num_window_groups; g++) {
1018 unsigned g_len = ics->group_len[g];
1020 for (i = 0; i < ics->max_sfb; i++, idx++) {
1021 const unsigned cbt_m1 = band_type[idx] - 1;
1022 float *cfo = coef + offsets[i];
1023 int off_len = offsets[i + 1] - offsets[i];
1026 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1027 for (group = 0; group < g_len; group++, cfo+=128) {
1028 memset(cfo, 0, off_len * sizeof(float));
1030 } else if (cbt_m1 == NOISE_BT - 1) {
1031 for (group = 0; group < g_len; group++, cfo+=128) {
1035 for (k = 0; k < off_len; k++) {
1036 ac->random_state = lcg_random(ac->random_state);
1037 cfo[k] = ac->random_state;
1040 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1041 scale = sf[idx] / sqrtf(band_energy);
1042 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1045 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1046 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1047 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1048 OPEN_READER(re, gb);
1050 switch (cbt_m1 >> 1) {
1052 for (group = 0; group < g_len; group++, cfo+=128) {
1060 UPDATE_CACHE(re, gb);
1061 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1062 cb_idx = cb_vector_idx[code];
1063 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1069 for (group = 0; group < g_len; group++, cfo+=128) {
1079 UPDATE_CACHE(re, gb);
1080 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1081 cb_idx = cb_vector_idx[code];
1082 nnz = cb_idx >> 8 & 15;
1083 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1084 LAST_SKIP_BITS(re, gb, nnz);
1085 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1091 for (group = 0; group < g_len; group++, cfo+=128) {
1099 UPDATE_CACHE(re, gb);
1100 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1101 cb_idx = cb_vector_idx[code];
1102 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1109 for (group = 0; group < g_len; group++, cfo+=128) {
1119 UPDATE_CACHE(re, gb);
1120 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1121 cb_idx = cb_vector_idx[code];
1122 nnz = cb_idx >> 8 & 15;
1123 sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1124 LAST_SKIP_BITS(re, gb, nnz);
1125 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1131 for (group = 0; group < g_len; group++, cfo+=128) {
1133 uint32_t *icf = (uint32_t *) cf;
1143 UPDATE_CACHE(re, gb);
1144 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1152 cb_idx = cb_vector_idx[code];
1155 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1156 LAST_SKIP_BITS(re, gb, nnz);
1158 for (j = 0; j < 2; j++) {
1162 /* The total length of escape_sequence must be < 22 bits according
1163 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1164 UPDATE_CACHE(re, gb);
1165 b = GET_CACHE(re, gb);
1166 b = 31 - av_log2(~b);
1169 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1173 SKIP_BITS(re, gb, b + 1);
1175 n = (1 << b) + SHOW_UBITS(re, gb, b);
1176 LAST_SKIP_BITS(re, gb, b);
1177 *icf++ = cbrt_tab[n] | (bits & 1<<31);
1180 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1181 *icf++ = (bits & 1<<31) | v;
1188 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1192 CLOSE_READER(re, gb);
1198 if (pulse_present) {
1200 for (i = 0; i < pulse->num_pulse; i++) {
1201 float co = coef_base[ pulse->pos[i] ];
1202 while (offsets[idx + 1] <= pulse->pos[i])
1204 if (band_type[idx] != NOISE_BT && sf[idx]) {
1205 float ico = -pulse->amp[i];
1208 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1210 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1217 static av_always_inline float flt16_round(float pf)
1221 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1225 static av_always_inline float flt16_even(float pf)
1229 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1233 static av_always_inline float flt16_trunc(float pf)
1237 pun.i &= 0xFFFF0000U;
1241 static av_always_inline void predict(PredictorState *ps, float *coef,
1242 float sf_scale, float inv_sf_scale,
1245 const float a = 0.953125; // 61.0 / 64
1246 const float alpha = 0.90625; // 29.0 / 32
1250 float r0 = ps->r0, r1 = ps->r1;
1251 float cor0 = ps->cor0, cor1 = ps->cor1;
1252 float var0 = ps->var0, var1 = ps->var1;
1254 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1255 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1257 pv = flt16_round(k1 * r0 + k2 * r1);
1259 *coef += pv * sf_scale;
1261 e0 = *coef * inv_sf_scale;
1264 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1265 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1266 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1267 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1269 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1270 ps->r0 = flt16_trunc(a * e0);
1274 * Apply AAC-Main style frequency domain prediction.
1276 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1279 float sf_scale = ac->sf_scale, inv_sf_scale = 1 / ac->sf_scale;
1281 if (!sce->ics.predictor_initialized) {
1282 reset_all_predictors(sce->predictor_state);
1283 sce->ics.predictor_initialized = 1;
1286 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1287 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1288 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1289 predict(&sce->predictor_state[k], &sce->coeffs[k],
1290 sf_scale, inv_sf_scale,
1291 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1294 if (sce->ics.predictor_reset_group)
1295 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1297 reset_all_predictors(sce->predictor_state);
1301 * Decode an individual_channel_stream payload; reference: table 4.44.
1303 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1304 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1306 * @return Returns error status. 0 - OK, !0 - error
1308 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1309 GetBitContext *gb, int common_window, int scale_flag)
1312 TemporalNoiseShaping *tns = &sce->tns;
1313 IndividualChannelStream *ics = &sce->ics;
1314 float *out = sce->coeffs;
1315 int global_gain, pulse_present = 0;
1317 /* This assignment is to silence a GCC warning about the variable being used
1318 * uninitialized when in fact it always is.
1320 pulse.num_pulse = 0;
1322 global_gain = get_bits(gb, 8);
1324 if (!common_window && !scale_flag) {
1325 if (decode_ics_info(ac, ics, gb, 0) < 0)
1329 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1331 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1336 if ((pulse_present = get_bits1(gb))) {
1337 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1338 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1341 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1342 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1346 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1348 if (get_bits1(gb)) {
1349 av_log_missing_feature(ac->avctx, "SSR", 1);
1354 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1357 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1358 apply_prediction(ac, sce);
1364 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1366 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1368 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1369 float *ch0 = cpe->ch[0].coeffs;
1370 float *ch1 = cpe->ch[1].coeffs;
1371 int g, i, group, idx = 0;
1372 const uint16_t *offsets = ics->swb_offset;
1373 for (g = 0; g < ics->num_window_groups; g++) {
1374 for (i = 0; i < ics->max_sfb; i++, idx++) {
1375 if (cpe->ms_mask[idx] &&
1376 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1377 for (group = 0; group < ics->group_len[g]; group++) {
1378 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1379 ch1 + group * 128 + offsets[i],
1380 offsets[i+1] - offsets[i]);
1384 ch0 += ics->group_len[g] * 128;
1385 ch1 += ics->group_len[g] * 128;
1390 * intensity stereo decoding; reference: 4.6.8.2.3
1392 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1393 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1394 * [3] reserved for scalable AAC
1396 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1398 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1399 SingleChannelElement *sce1 = &cpe->ch[1];
1400 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1401 const uint16_t *offsets = ics->swb_offset;
1402 int g, group, i, idx = 0;
1405 for (g = 0; g < ics->num_window_groups; g++) {
1406 for (i = 0; i < ics->max_sfb;) {
1407 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1408 const int bt_run_end = sce1->band_type_run_end[idx];
1409 for (; i < bt_run_end; i++, idx++) {
1410 c = -1 + 2 * (sce1->band_type[idx] - 14);
1412 c *= 1 - 2 * cpe->ms_mask[idx];
1413 scale = c * sce1->sf[idx];
1414 for (group = 0; group < ics->group_len[g]; group++)
1415 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1416 coef0 + group * 128 + offsets[i],
1418 offsets[i + 1] - offsets[i]);
1421 int bt_run_end = sce1->band_type_run_end[idx];
1422 idx += bt_run_end - i;
1426 coef0 += ics->group_len[g] * 128;
1427 coef1 += ics->group_len[g] * 128;
1432 * Decode a channel_pair_element; reference: table 4.4.
1434 * @return Returns error status. 0 - OK, !0 - error
1436 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1438 int i, ret, common_window, ms_present = 0;
1440 common_window = get_bits1(gb);
1441 if (common_window) {
1442 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1444 i = cpe->ch[1].ics.use_kb_window[0];
1445 cpe->ch[1].ics = cpe->ch[0].ics;
1446 cpe->ch[1].ics.use_kb_window[1] = i;
1447 if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1448 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1449 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1450 ms_present = get_bits(gb, 2);
1451 if (ms_present == 3) {
1452 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1454 } else if (ms_present)
1455 decode_mid_side_stereo(cpe, gb, ms_present);
1457 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1459 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1462 if (common_window) {
1464 apply_mid_side_stereo(ac, cpe);
1465 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1466 apply_prediction(ac, &cpe->ch[0]);
1467 apply_prediction(ac, &cpe->ch[1]);
1471 apply_intensity_stereo(ac, cpe, ms_present);
1475 static const float cce_scale[] = {
1476 1.09050773266525765921, //2^(1/8)
1477 1.18920711500272106672, //2^(1/4)
1483 * Decode coupling_channel_element; reference: table 4.8.
1485 * @return Returns error status. 0 - OK, !0 - error
1487 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1493 SingleChannelElement *sce = &che->ch[0];
1494 ChannelCoupling *coup = &che->coup;
1496 coup->coupling_point = 2 * get_bits1(gb);
1497 coup->num_coupled = get_bits(gb, 3);
1498 for (c = 0; c <= coup->num_coupled; c++) {
1500 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1501 coup->id_select[c] = get_bits(gb, 4);
1502 if (coup->type[c] == TYPE_CPE) {
1503 coup->ch_select[c] = get_bits(gb, 2);
1504 if (coup->ch_select[c] == 3)
1507 coup->ch_select[c] = 2;
1509 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1511 sign = get_bits(gb, 1);
1512 scale = cce_scale[get_bits(gb, 2)];
1514 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1517 for (c = 0; c < num_gain; c++) {
1521 float gain_cache = 1.;
1523 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1524 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1525 gain_cache = powf(scale, -gain);
1527 if (coup->coupling_point == AFTER_IMDCT) {
1528 coup->gain[c][0] = gain_cache;
1530 for (g = 0; g < sce->ics.num_window_groups; g++) {
1531 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1532 if (sce->band_type[idx] != ZERO_BT) {
1534 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1542 gain_cache = powf(scale, -t) * s;
1545 coup->gain[c][idx] = gain_cache;
1555 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1557 * @return Returns number of bytes consumed.
1559 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1563 int num_excl_chan = 0;
1566 for (i = 0; i < 7; i++)
1567 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1568 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1570 return num_excl_chan / 7;
1574 * Decode dynamic range information; reference: table 4.52.
1576 * @param cnt length of TYPE_FIL syntactic element in bytes
1578 * @return Returns number of bytes consumed.
1580 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1581 GetBitContext *gb, int cnt)
1584 int drc_num_bands = 1;
1587 /* pce_tag_present? */
1588 if (get_bits1(gb)) {
1589 che_drc->pce_instance_tag = get_bits(gb, 4);
1590 skip_bits(gb, 4); // tag_reserved_bits
1594 /* excluded_chns_present? */
1595 if (get_bits1(gb)) {
1596 n += decode_drc_channel_exclusions(che_drc, gb);
1599 /* drc_bands_present? */
1600 if (get_bits1(gb)) {
1601 che_drc->band_incr = get_bits(gb, 4);
1602 che_drc->interpolation_scheme = get_bits(gb, 4);
1604 drc_num_bands += che_drc->band_incr;
1605 for (i = 0; i < drc_num_bands; i++) {
1606 che_drc->band_top[i] = get_bits(gb, 8);
1611 /* prog_ref_level_present? */
1612 if (get_bits1(gb)) {
1613 che_drc->prog_ref_level = get_bits(gb, 7);
1614 skip_bits1(gb); // prog_ref_level_reserved_bits
1618 for (i = 0; i < drc_num_bands; i++) {
1619 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1620 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1628 * Decode extension data (incomplete); reference: table 4.51.
1630 * @param cnt length of TYPE_FIL syntactic element in bytes
1632 * @return Returns number of bytes consumed
1634 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1635 ChannelElement *che, enum RawDataBlockType elem_type)
1639 switch (get_bits(gb, 4)) { // extension type
1640 case EXT_SBR_DATA_CRC:
1644 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1646 } else if (!ac->m4ac.sbr) {
1647 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1648 skip_bits_long(gb, 8 * cnt - 4);
1650 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1651 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1652 skip_bits_long(gb, 8 * cnt - 4);
1654 } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1657 output_configure(ac, ac->che_pos, ac->che_pos, ac->m4ac.chan_config, ac->output_configured);
1661 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1663 case EXT_DYNAMIC_RANGE:
1664 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1668 case EXT_DATA_ELEMENT:
1670 skip_bits_long(gb, 8 * cnt - 4);
1677 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1679 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1680 * @param coef spectral coefficients
1682 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1683 IndividualChannelStream *ics, int decode)
1685 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1687 int bottom, top, order, start, end, size, inc;
1688 float lpc[TNS_MAX_ORDER];
1689 float tmp[TNS_MAX_ORDER];
1691 for (w = 0; w < ics->num_windows; w++) {
1692 bottom = ics->num_swb;
1693 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1695 bottom = FFMAX(0, top - tns->length[w][filt]);
1696 order = tns->order[w][filt];
1701 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1703 start = ics->swb_offset[FFMIN(bottom, mmm)];
1704 end = ics->swb_offset[FFMIN( top, mmm)];
1705 if ((size = end - start) <= 0)
1707 if (tns->direction[w][filt]) {
1717 for (m = 0; m < size; m++, start += inc)
1718 for (i = 1; i <= FFMIN(m, order); i++)
1719 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1722 for (m = 0; m < size; m++, start += inc) {
1723 tmp[0] = coef[start];
1724 for (i = 1; i <= FFMIN(m, order); i++)
1725 coef[start] += tmp[i] * lpc[i - 1];
1726 for (i = order; i > 0; i--)
1727 tmp[i] = tmp[i - 1];
1735 * Apply windowing and MDCT to obtain the spectral
1736 * coefficient from the predicted sample by LTP.
1738 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
1739 float *in, IndividualChannelStream *ics)
1741 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1742 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1743 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1744 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1746 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
1747 ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
1749 memset(in, 0, 448 * sizeof(float));
1750 ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
1751 memcpy(in + 576, in + 576, 448 * sizeof(float));
1753 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
1754 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
1756 memcpy(in + 1024, in + 1024, 448 * sizeof(float));
1757 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
1758 memset(in + 1024 + 576, 0, 448 * sizeof(float));
1760 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
1764 * Apply the long term prediction
1766 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
1768 const LongTermPrediction *ltp = &sce->ics.ltp;
1769 const uint16_t *offsets = sce->ics.swb_offset;
1772 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1773 float *predTime = sce->ret;
1774 float *predFreq = ac->buf_mdct;
1775 int16_t num_samples = 2048;
1777 if (ltp->lag < 1024)
1778 num_samples = ltp->lag + 1024;
1779 for (i = 0; i < num_samples; i++)
1780 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
1781 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
1783 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
1785 if (sce->tns.present)
1786 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
1788 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
1790 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
1791 sce->coeffs[i] += predFreq[i];
1796 * Update the LTP buffer for next frame
1798 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
1800 IndividualChannelStream *ics = &sce->ics;
1801 float *saved = sce->saved;
1802 float *saved_ltp = sce->coeffs;
1803 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1804 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1807 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1808 memcpy(saved_ltp, saved, 512 * sizeof(float));
1809 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1810 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1811 for (i = 0; i < 64; i++)
1812 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1813 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1814 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
1815 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1816 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1817 for (i = 0; i < 64; i++)
1818 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1819 } else { // LONG_STOP or ONLY_LONG
1820 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
1821 for (i = 0; i < 512; i++)
1822 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
1825 memcpy(sce->ltp_state, &sce->ltp_state[1024], 1024 * sizeof(int16_t));
1826 ac->fmt_conv.float_to_int16(&(sce->ltp_state[1024]), sce->ret, 1024);
1827 ac->fmt_conv.float_to_int16(&(sce->ltp_state[2048]), saved_ltp, 1024);
1831 * Conduct IMDCT and windowing.
1833 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1835 IndividualChannelStream *ics = &sce->ics;
1836 float *in = sce->coeffs;
1837 float *out = sce->ret;
1838 float *saved = sce->saved;
1839 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1840 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1841 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1842 float *buf = ac->buf_mdct;
1843 float *temp = ac->temp;
1847 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1848 for (i = 0; i < 1024; i += 128)
1849 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
1851 ac->mdct.imdct_half(&ac->mdct, buf, in);
1853 /* window overlapping
1854 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1855 * and long to short transitions are considered to be short to short
1856 * transitions. This leaves just two cases (long to long and short to short)
1857 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1859 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1860 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1861 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
1863 memcpy( out, saved, 448 * sizeof(float));
1865 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1866 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
1867 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
1868 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
1869 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
1870 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
1871 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1873 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
1874 memcpy( out + 576, buf + 64, 448 * sizeof(float));
1879 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1880 memcpy( saved, temp + 64, 64 * sizeof(float));
1881 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
1882 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
1883 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
1884 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1885 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1886 memcpy( saved, buf + 512, 448 * sizeof(float));
1887 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1888 } else { // LONG_STOP or ONLY_LONG
1889 memcpy( saved, buf + 512, 512 * sizeof(float));
1894 * Apply dependent channel coupling (applied before IMDCT).
1896 * @param index index into coupling gain array
1898 static void apply_dependent_coupling(AACContext *ac,
1899 SingleChannelElement *target,
1900 ChannelElement *cce, int index)
1902 IndividualChannelStream *ics = &cce->ch[0].ics;
1903 const uint16_t *offsets = ics->swb_offset;
1904 float *dest = target->coeffs;
1905 const float *src = cce->ch[0].coeffs;
1906 int g, i, group, k, idx = 0;
1907 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1908 av_log(ac->avctx, AV_LOG_ERROR,
1909 "Dependent coupling is not supported together with LTP\n");
1912 for (g = 0; g < ics->num_window_groups; g++) {
1913 for (i = 0; i < ics->max_sfb; i++, idx++) {
1914 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1915 const float gain = cce->coup.gain[index][idx];
1916 for (group = 0; group < ics->group_len[g]; group++) {
1917 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1919 dest[group * 128 + k] += gain * src[group * 128 + k];
1924 dest += ics->group_len[g] * 128;
1925 src += ics->group_len[g] * 128;
1930 * Apply independent channel coupling (applied after IMDCT).
1932 * @param index index into coupling gain array
1934 static void apply_independent_coupling(AACContext *ac,
1935 SingleChannelElement *target,
1936 ChannelElement *cce, int index)
1939 const float gain = cce->coup.gain[index][0];
1940 const float *src = cce->ch[0].ret;
1941 float *dest = target->ret;
1942 const int len = 1024 << (ac->m4ac.sbr == 1);
1944 for (i = 0; i < len; i++)
1945 dest[i] += gain * src[i];
1949 * channel coupling transformation interface
1951 * @param apply_coupling_method pointer to (in)dependent coupling function
1953 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1954 enum RawDataBlockType type, int elem_id,
1955 enum CouplingPoint coupling_point,
1956 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1960 for (i = 0; i < MAX_ELEM_ID; i++) {
1961 ChannelElement *cce = ac->che[TYPE_CCE][i];
1964 if (cce && cce->coup.coupling_point == coupling_point) {
1965 ChannelCoupling *coup = &cce->coup;
1967 for (c = 0; c <= coup->num_coupled; c++) {
1968 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1969 if (coup->ch_select[c] != 1) {
1970 apply_coupling_method(ac, &cc->ch[0], cce, index);
1971 if (coup->ch_select[c] != 0)
1974 if (coup->ch_select[c] != 2)
1975 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1977 index += 1 + (coup->ch_select[c] == 3);
1984 * Convert spectral data to float samples, applying all supported tools as appropriate.
1986 static void spectral_to_sample(AACContext *ac)
1989 for (type = 3; type >= 0; type--) {
1990 for (i = 0; i < MAX_ELEM_ID; i++) {
1991 ChannelElement *che = ac->che[type][i];
1993 if (type <= TYPE_CPE)
1994 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1995 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1996 if (che->ch[0].ics.predictor_present) {
1997 if (che->ch[0].ics.ltp.present)
1998 apply_ltp(ac, &che->ch[0]);
1999 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2000 apply_ltp(ac, &che->ch[1]);
2003 if (che->ch[0].tns.present)
2004 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2005 if (che->ch[1].tns.present)
2006 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2007 if (type <= TYPE_CPE)
2008 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2009 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2010 imdct_and_windowing(ac, &che->ch[0]);
2011 if (ac->m4ac.object_type == AOT_AAC_LTP)
2012 update_ltp(ac, &che->ch[0]);
2013 if (type == TYPE_CPE) {
2014 imdct_and_windowing(ac, &che->ch[1]);
2015 if (ac->m4ac.object_type == AOT_AAC_LTP)
2016 update_ltp(ac, &che->ch[1]);
2018 if (ac->m4ac.sbr > 0) {
2019 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2022 if (type <= TYPE_CCE)
2023 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2029 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2032 AACADTSHeaderInfo hdr_info;
2034 size = ff_aac_parse_header(gb, &hdr_info);
2036 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
2037 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2038 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2039 ac->m4ac.chan_config = hdr_info.chan_config;
2040 if (set_default_channel_config(ac->avctx, new_che_pos, hdr_info.chan_config))
2042 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
2044 } else if (ac->output_configured != OC_LOCKED) {
2045 ac->output_configured = OC_NONE;
2047 if (ac->output_configured != OC_LOCKED) {
2051 ac->m4ac.sample_rate = hdr_info.sample_rate;
2052 ac->m4ac.sampling_index = hdr_info.sampling_index;
2053 ac->m4ac.object_type = hdr_info.object_type;
2054 if (!ac->avctx->sample_rate)
2055 ac->avctx->sample_rate = hdr_info.sample_rate;
2056 if (hdr_info.num_aac_frames == 1) {
2057 if (!hdr_info.crc_absent)
2060 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2067 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2068 int *data_size, GetBitContext *gb)
2070 AACContext *ac = avctx->priv_data;
2071 ChannelElement *che = NULL, *che_prev = NULL;
2072 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2073 int err, elem_id, data_size_tmp;
2074 int samples = 0, multiplier;
2076 if (show_bits(gb, 12) == 0xfff) {
2077 if (parse_adts_frame_header(ac, gb) < 0) {
2078 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2081 if (ac->m4ac.sampling_index > 12) {
2082 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2087 ac->tags_mapped = 0;
2089 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2090 elem_id = get_bits(gb, 4);
2092 if (elem_type < TYPE_DSE) {
2093 if (!(che=get_che(ac, elem_type, elem_id))) {
2094 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2095 elem_type, elem_id);
2101 switch (elem_type) {
2104 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2108 err = decode_cpe(ac, gb, che);
2112 err = decode_cce(ac, gb, che);
2116 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2120 err = skip_data_stream_element(ac, gb);
2124 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2125 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2126 if ((err = decode_pce(avctx, &ac->m4ac, new_che_pos, gb)))
2128 if (ac->output_configured > OC_TRIAL_PCE)
2129 av_log(avctx, AV_LOG_ERROR,
2130 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2132 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2138 elem_id += get_bits(gb, 8) - 1;
2139 if (get_bits_left(gb) < 8 * elem_id) {
2140 av_log(avctx, AV_LOG_ERROR, overread_err);
2144 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2145 err = 0; /* FIXME */
2149 err = -1; /* should not happen, but keeps compiler happy */
2154 elem_type_prev = elem_type;
2159 if (get_bits_left(gb) < 3) {
2160 av_log(avctx, AV_LOG_ERROR, overread_err);
2165 spectral_to_sample(ac);
2167 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2168 samples <<= multiplier;
2169 if (ac->output_configured < OC_LOCKED) {
2170 avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2171 avctx->frame_size = samples;
2174 /* ffdshow custom code */
2175 #if CONFIG_AUDIO_FLOAT
2176 data_size_tmp = samples * avctx->channels * sizeof(float);
2178 data_size_tmp = samples * avctx->channels * sizeof(int16_t);
2180 if (*data_size < data_size_tmp) {
2181 av_log(avctx, AV_LOG_ERROR,
2182 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2183 *data_size, data_size_tmp);
2186 *data_size = data_size_tmp;
2189 /* ffdshow custom code */
2190 #if CONFIG_AUDIO_FLOAT
2191 float_interleave(data, (const float **)ac->output_data, samples, avctx->channels);
2193 ac->fmt_conv.float_to_int16_interleave(data, (const float **)ac->output_data, samples, avctx->channels);
2197 if (ac->output_configured)
2198 ac->output_configured = OC_LOCKED;
2203 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2204 int *data_size, AVPacket *avpkt)
2206 const uint8_t *buf = avpkt->data;
2207 int buf_size = avpkt->size;
2213 init_get_bits(&gb, buf, buf_size * 8);
2215 if ((err = aac_decode_frame_int(avctx, data, data_size, &gb)) < 0)
2218 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2219 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2220 if (buf[buf_offset])
2223 return buf_size > buf_offset ? buf_consumed : buf_size;
2226 static av_cold int aac_decode_close(AVCodecContext *avctx)
2228 AACContext *ac = avctx->priv_data;
2231 for (i = 0; i < MAX_ELEM_ID; i++) {
2232 for (type = 0; type < 4; type++) {
2233 if (ac->che[type][i])
2234 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2235 av_freep(&ac->che[type][i]);
2239 ff_mdct_end(&ac->mdct);
2240 ff_mdct_end(&ac->mdct_small);
2241 ff_mdct_end(&ac->mdct_ltp);
2246 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2248 struct LATMContext {
2249 AACContext aac_ctx; ///< containing AACContext
2250 int initialized; ///< initilized after a valid extradata was seen
2253 int audio_mux_version_A; ///< LATM syntax version
2254 int frame_length_type; ///< 0/1 variable/fixed frame length
2255 int frame_length; ///< frame length for fixed frame length
2258 static inline uint32_t latm_get_value(GetBitContext *b)
2260 int length = get_bits(b, 2);
2262 return get_bits_long(b, (length+1)*8);
2265 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2268 AVCodecContext *avctx = latmctx->aac_ctx.avctx;
2269 int config_start_bit = get_bits_count(gb);
2270 int bits_consumed, esize;
2272 if (config_start_bit % 8) {
2273 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2274 "config not byte aligned.\n", 1);
2275 return AVERROR_INVALIDDATA;
2278 decode_audio_specific_config(&latmctx->aac_ctx, avctx,
2279 &latmctx->aac_ctx.m4ac,
2280 gb->buffer + (config_start_bit / 8),
2281 get_bits_left(gb) / 8);
2283 if (bits_consumed < 0)
2284 return AVERROR_INVALIDDATA;
2286 esize = (bits_consumed+7) / 8;
2288 if (avctx->extradata_size <= esize) {
2289 av_free(avctx->extradata);
2290 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2291 if (!avctx->extradata)
2292 return AVERROR(ENOMEM);
2295 avctx->extradata_size = esize;
2296 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2297 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2299 skip_bits_long(gb, bits_consumed);
2302 return bits_consumed;
2305 static int read_stream_mux_config(struct LATMContext *latmctx,
2308 int ret, audio_mux_version = get_bits(gb, 1);
2310 latmctx->audio_mux_version_A = 0;
2311 if (audio_mux_version)
2312 latmctx->audio_mux_version_A = get_bits(gb, 1);
2314 if (!latmctx->audio_mux_version_A) {
2316 if (audio_mux_version)
2317 latm_get_value(gb); // taraFullness
2319 skip_bits(gb, 1); // allStreamSameTimeFraming
2320 skip_bits(gb, 6); // numSubFrames
2322 if (get_bits(gb, 4)) { // numPrograms
2323 av_log_missing_feature(latmctx->aac_ctx.avctx,
2324 "multiple programs are not supported\n", 1);
2325 return AVERROR_PATCHWELCOME;
2328 // for each program (which there is only on in DVB)
2330 // for each layer (which there is only on in DVB)
2331 if (get_bits(gb, 3)) { // numLayer
2332 av_log_missing_feature(latmctx->aac_ctx.avctx,
2333 "multiple layers are not supported\n", 1);
2334 return AVERROR_PATCHWELCOME;
2337 // for all but first stream: use_same_config = get_bits(gb, 1);
2338 if (!audio_mux_version) {
2339 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2342 int ascLen = latm_get_value(gb);
2343 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2346 skip_bits_long(gb, ascLen);
2349 latmctx->frame_length_type = get_bits(gb, 3);
2350 switch (latmctx->frame_length_type) {
2352 skip_bits(gb, 8); // latmBufferFullness
2355 latmctx->frame_length = get_bits(gb, 9);
2360 skip_bits(gb, 6); // CELP frame length table index
2364 skip_bits(gb, 1); // HVXC frame length table index
2368 if (get_bits(gb, 1)) { // other data
2369 if (audio_mux_version) {
2370 latm_get_value(gb); // other_data_bits
2374 esc = get_bits(gb, 1);
2380 if (get_bits(gb, 1)) // crc present
2381 skip_bits(gb, 8); // config_crc
2387 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2391 if (ctx->frame_length_type == 0) {
2392 int mux_slot_length = 0;
2394 tmp = get_bits(gb, 8);
2395 mux_slot_length += tmp;
2396 } while (tmp == 255);
2397 return mux_slot_length;
2398 } else if (ctx->frame_length_type == 1) {
2399 return ctx->frame_length;
2400 } else if (ctx->frame_length_type == 3 ||
2401 ctx->frame_length_type == 5 ||
2402 ctx->frame_length_type == 7) {
2403 skip_bits(gb, 2); // mux_slot_length_coded
2408 static int read_audio_mux_element(struct LATMContext *latmctx,
2412 uint8_t use_same_mux = get_bits(gb, 1);
2413 if (!use_same_mux) {
2414 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2416 } else if (!latmctx->aac_ctx.avctx->extradata) {
2417 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2418 "no decoder config found\n");
2419 return AVERROR(EAGAIN);
2421 if (latmctx->audio_mux_version_A == 0) {
2422 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2423 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2424 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2425 return AVERROR_INVALIDDATA;
2426 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2427 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2428 "frame length mismatch %d << %d\n",
2429 mux_slot_length_bytes * 8, get_bits_left(gb));
2430 return AVERROR_INVALIDDATA;
2437 static int latm_decode_frame(AVCodecContext *avctx, void *out, int *out_size,
2440 struct LATMContext *latmctx = avctx->priv_data;
2444 if (avpkt->size == 0)
2447 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2449 // check for LOAS sync word
2450 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2451 return AVERROR_INVALIDDATA;
2453 muxlength = get_bits(&gb, 13) + 3;
2454 // not enough data, the parser should have sorted this
2455 if (muxlength > avpkt->size)
2456 return AVERROR_INVALIDDATA;
2458 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2461 if (!latmctx->initialized) {
2462 if (!avctx->extradata) {
2466 if ((err = aac_decode_init(avctx)) < 0)
2468 latmctx->initialized = 1;
2472 if (show_bits(&gb, 12) == 0xfff) {
2473 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2474 "ADTS header detected, probably as result of configuration "
2476 return AVERROR_INVALIDDATA;
2479 if ((err = aac_decode_frame_int(avctx, out, out_size, &gb)) < 0)
2485 av_cold static int latm_decode_init(AVCodecContext *avctx)
2487 struct LATMContext *latmctx = avctx->priv_data;
2490 ret = aac_decode_init(avctx);
2492 if (avctx->extradata_size > 0) {
2493 latmctx->initialized = !ret;
2495 latmctx->initialized = 0;
2502 AVCodec ff_aac_decoder = {
2511 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2512 .sample_fmts = (const enum AVSampleFormat[]) {
2513 #if CONFIG_AUDIO_FLOAT
2514 AV_SAMPLE_FMT_FLT,AV_SAMPLE_FMT_NONE
2516 AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE
2519 .channel_layouts = aac_channel_layout,
2523 Note: This decoder filter is intended to decode LATM streams transferred
2524 in MPEG transport streams which only contain one program.
2525 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2527 AVCodec ff_aac_latm_decoder = {
2529 .type = AVMEDIA_TYPE_AUDIO,
2530 .id = CODEC_ID_AAC_LATM,
2531 .priv_data_size = sizeof(struct LATMContext),
2532 .init = latm_decode_init,
2533 .close = aac_decode_close,
2534 .decode = latm_decode_frame,
2535 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2536 .sample_fmts = (const enum AVSampleFormat[]) {
2537 #if CONFIG_AUDIO_FLOAT
2538 AV_SAMPLE_FMT_FLT,AV_SAMPLE_FMT_NONE
2540 AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE
2543 .channel_layouts = aac_channel_layout,