3 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 * @author Oded Shimon ( ods15 ods15 dyndns org )
27 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
34 * N (code in SoC repo) gain control
36 * Y window shapes - standard
37 * N window shapes - Low Delay
38 * Y filterbank - standard
39 * N (code in SoC repo) filterbank - Scalable Sample Rate
40 * Y Temporal Noise Shaping
41 * N (code in SoC repo) Long Term Prediction
44 * Y frequency domain prediction
45 * Y Perceptual Noise Substitution
47 * N Scalable Inverse AAC Quantization
48 * N Frequency Selective Switch
50 * Y quantization & coding - AAC
51 * N quantization & coding - TwinVQ
52 * N quantization & coding - BSAC
53 * N AAC Error Resilience tools
54 * N Error Resilience payload syntax
55 * N Error Protection tool
57 * N Silence Compression
60 * N Structured Audio tools
61 * N Structured Audio Sample Bank Format
63 * N Harmonic and Individual Lines plus Noise
64 * N Text-To-Speech Interface
65 * Y Spectral Band Replication
66 * Y (not in this code) Layer-1
67 * Y (not in this code) Layer-2
68 * Y (not in this code) Layer-3
69 * N SinuSoidal Coding (Transient, Sinusoid, Noise)
70 * N (planned) Parametric Stereo
71 * N Direct Stream Transfer
73 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
74 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
88 #include "aacdectab.h"
89 #include "cbrt_tablegen.h"
92 #include "mpeg4audio.h"
93 #include "aac_parser.h"
101 # include "arm/aac.h"
109 static VLC vlc_scalefactors;
110 static VLC vlc_spectral[11];
112 static const char overread_err[] = "Input buffer exhausted before END element found\n";
114 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
116 if (ac->tag_che_map[type][elem_id]) {
117 return ac->tag_che_map[type][elem_id];
119 if (ac->tags_mapped >= tags_per_config[ac->m4ac.chan_config]) {
122 switch (ac->m4ac.chan_config) {
124 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
126 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
129 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
130 instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have
131 encountered such a stream, transfer the LFE[0] element to SCE[1] */
132 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
134 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
137 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
139 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
142 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
144 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
148 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
150 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
151 } else if (ac->m4ac.chan_config == 2) {
155 if (!ac->tags_mapped && type == TYPE_SCE) {
157 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
165 * Check for the channel element in the current channel position configuration.
166 * If it exists, make sure the appropriate element is allocated and map the
167 * channel order to match the internal FFmpeg channel layout.
169 * @param che_pos current channel position configuration
170 * @param type channel element type
171 * @param id channel element id
172 * @param channels count of the number of channels in the configuration
174 * @return Returns error status. 0 - OK, !0 - error
176 static av_cold int che_configure(AACContext *ac,
177 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
181 if (che_pos[type][id]) {
182 if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
183 return AVERROR(ENOMEM);
184 ff_aac_sbr_ctx_init(&ac->che[type][id]->sbr);
185 if (type != TYPE_CCE) {
186 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
187 if (type == TYPE_CPE) {
188 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
192 if (ac->che[type][id])
193 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
194 av_freep(&ac->che[type][id]);
200 * Configure output channel order based on the current program configuration element.
202 * @param che_pos current channel position configuration
203 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
205 * @return Returns error status. 0 - OK, !0 - error
207 static av_cold int output_configure(AACContext *ac,
208 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
209 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
210 int channel_config, enum OCStatus oc_type)
212 AVCodecContext *avctx = ac->avccontext;
213 int i, type, channels = 0, ret;
215 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
217 if (channel_config) {
218 for (i = 0; i < tags_per_config[channel_config]; i++) {
219 if ((ret = che_configure(ac, che_pos,
220 aac_channel_layout_map[channel_config - 1][i][0],
221 aac_channel_layout_map[channel_config - 1][i][1],
226 memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
229 avctx->channel_layout = aac_channel_layout[channel_config - 1];
231 /* Allocate or free elements depending on if they are in the
232 * current program configuration.
234 * Set up default 1:1 output mapping.
236 * For a 5.1 stream the output order will be:
237 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
240 for (i = 0; i < MAX_ELEM_ID; i++) {
241 for (type = 0; type < 4; type++) {
242 if ((ret = che_configure(ac, che_pos, type, i, &channels)))
247 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
248 ac->tags_mapped = 4 * MAX_ELEM_ID;
250 avctx->channel_layout = 0;
253 avctx->channels = channels;
255 ac->output_configured = oc_type;
261 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
263 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
264 * @param sce_map mono (Single Channel Element) map
265 * @param type speaker type/position for these channels
267 static void decode_channel_map(enum ChannelPosition *cpe_map,
268 enum ChannelPosition *sce_map,
269 enum ChannelPosition type,
270 GetBitContext *gb, int n)
273 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
274 map[get_bits(gb, 4)] = type;
279 * Decode program configuration element; reference: table 4.2.
281 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
283 * @return Returns error status. 0 - OK, !0 - error
285 static int decode_pce(AACContext *ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
288 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
291 skip_bits(gb, 2); // object_type
293 sampling_index = get_bits(gb, 4);
294 if (ac->m4ac.sampling_index != sampling_index)
295 av_log(ac->avccontext, AV_LOG_WARNING, "Sample rate index in program config element does not match the sample rate index configured by the container.\n");
297 num_front = get_bits(gb, 4);
298 num_side = get_bits(gb, 4);
299 num_back = get_bits(gb, 4);
300 num_lfe = get_bits(gb, 2);
301 num_assoc_data = get_bits(gb, 3);
302 num_cc = get_bits(gb, 4);
305 skip_bits(gb, 4); // mono_mixdown_tag
307 skip_bits(gb, 4); // stereo_mixdown_tag
310 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
312 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
313 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
314 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
315 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
317 skip_bits_long(gb, 4 * num_assoc_data);
319 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
323 /* comment field, first byte is length */
324 comment_len = get_bits(gb, 8) * 8;
325 if (get_bits_left(gb) < comment_len) {
326 av_log(ac->avccontext, AV_LOG_ERROR, overread_err);
329 skip_bits_long(gb, comment_len);
334 * Set up channel positions based on a default channel configuration
335 * as specified in table 1.17.
337 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
339 * @return Returns error status. 0 - OK, !0 - error
341 static av_cold int set_default_channel_config(AACContext *ac,
342 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
345 if (channel_config < 1 || channel_config > 7) {
346 av_log(ac->avccontext, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
351 /* default channel configurations:
353 * 1ch : front center (mono)
354 * 2ch : L + R (stereo)
355 * 3ch : front center + L + R
356 * 4ch : front center + L + R + back center
357 * 5ch : front center + L + R + back stereo
358 * 6ch : front center + L + R + back stereo + LFE
359 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
362 if (channel_config != 2)
363 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
364 if (channel_config > 1)
365 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
366 if (channel_config == 4)
367 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
368 if (channel_config > 4)
369 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
370 = AAC_CHANNEL_BACK; // back stereo
371 if (channel_config > 5)
372 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
373 if (channel_config == 7)
374 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
380 * Decode GA "General Audio" specific configuration; reference: table 4.1.
382 * @return Returns error status. 0 - OK, !0 - error
384 static int decode_ga_specific_config(AACContext *ac, GetBitContext *gb,
387 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
388 int extension_flag, ret;
390 if (get_bits1(gb)) { // frameLengthFlag
391 av_log_missing_feature(ac->avccontext, "960/120 MDCT window is", 1);
395 if (get_bits1(gb)) // dependsOnCoreCoder
396 skip_bits(gb, 14); // coreCoderDelay
397 extension_flag = get_bits1(gb);
399 if (ac->m4ac.object_type == AOT_AAC_SCALABLE ||
400 ac->m4ac.object_type == AOT_ER_AAC_SCALABLE)
401 skip_bits(gb, 3); // layerNr
403 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
404 if (channel_config == 0) {
405 skip_bits(gb, 4); // element_instance_tag
406 if ((ret = decode_pce(ac, new_che_pos, gb)))
409 if ((ret = set_default_channel_config(ac, new_che_pos, channel_config)))
412 if ((ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
415 if (extension_flag) {
416 switch (ac->m4ac.object_type) {
418 skip_bits(gb, 5); // numOfSubFrame
419 skip_bits(gb, 11); // layer_length
423 case AOT_ER_AAC_SCALABLE:
425 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
426 * aacScalefactorDataResilienceFlag
427 * aacSpectralDataResilienceFlag
431 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
437 * Decode audio specific configuration; reference: table 1.13.
439 * @param data pointer to AVCodecContext extradata
440 * @param data_size size of AVCCodecContext extradata
442 * @return Returns error status. 0 - OK, !0 - error
444 static int decode_audio_specific_config(AACContext *ac, void *data,
450 init_get_bits(&gb, data, data_size * 8);
452 if ((i = ff_mpeg4audio_get_config(&ac->m4ac, data, data_size)) < 0)
454 if (ac->m4ac.sampling_index > 12) {
455 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
459 skip_bits_long(&gb, i);
461 switch (ac->m4ac.object_type) {
464 if (decode_ga_specific_config(ac, &gb, ac->m4ac.chan_config))
468 av_log(ac->avccontext, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
469 ac->m4ac.sbr == 1? "SBR+" : "", ac->m4ac.object_type);
476 * linear congruential pseudorandom number generator
478 * @param previous_val pointer to the current state of the generator
480 * @return Returns a 32-bit pseudorandom integer
482 static av_always_inline int lcg_random(int previous_val)
484 return previous_val * 1664525 + 1013904223;
487 static av_always_inline void reset_predict_state(PredictorState *ps)
497 static void reset_all_predictors(PredictorState *ps)
500 for (i = 0; i < MAX_PREDICTORS; i++)
501 reset_predict_state(&ps[i]);
504 static void reset_predictor_group(PredictorState *ps, int group_num)
507 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
508 reset_predict_state(&ps[i]);
511 static av_cold int aac_decode_init(AVCodecContext *avccontext)
513 AACContext *ac = avccontext->priv_data;
516 ac->avccontext = avccontext;
517 ac->m4ac.sample_rate = avccontext->sample_rate;
519 if (avccontext->extradata_size > 0) {
520 if (decode_audio_specific_config(ac, avccontext->extradata, avccontext->extradata_size))
524 avccontext->sample_fmt = SAMPLE_FMT_S16;
526 AAC_INIT_VLC_STATIC( 0, 304);
527 AAC_INIT_VLC_STATIC( 1, 270);
528 AAC_INIT_VLC_STATIC( 2, 550);
529 AAC_INIT_VLC_STATIC( 3, 300);
530 AAC_INIT_VLC_STATIC( 4, 328);
531 AAC_INIT_VLC_STATIC( 5, 294);
532 AAC_INIT_VLC_STATIC( 6, 306);
533 AAC_INIT_VLC_STATIC( 7, 268);
534 AAC_INIT_VLC_STATIC( 8, 510);
535 AAC_INIT_VLC_STATIC( 9, 366);
536 AAC_INIT_VLC_STATIC(10, 462);
540 dsputil_init(&ac->dsp, avccontext);
542 ac->random_state = 0x1f2e3d4c;
544 // -1024 - Compensate wrong IMDCT method.
545 // 32768 - Required to scale values to the correct range for the bias method
546 // for float to int16 conversion.
548 if (ac->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
549 ac->add_bias = 385.0f;
550 ac->sf_scale = 1. / (-1024. * 32768.);
554 ac->sf_scale = 1. / -1024.;
558 #if !CONFIG_HARDCODED_TABLES
559 for (i = 0; i < 428; i++)
560 ff_aac_pow2sf_tab[i] = pow(2, (i - 200) / 4.);
561 #endif /* CONFIG_HARDCODED_TABLES */
563 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
564 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
565 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
568 ff_mdct_init(&ac->mdct, 11, 1, 1.0);
569 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
570 // window initialization
571 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
572 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
573 ff_init_ff_sine_windows(10);
574 ff_init_ff_sine_windows( 7);
582 * Skip data_stream_element; reference: table 4.10.
584 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
586 int byte_align = get_bits1(gb);
587 int count = get_bits(gb, 8);
589 count += get_bits(gb, 8);
593 if (get_bits_left(gb) < 8 * count) {
594 av_log(ac->avccontext, AV_LOG_ERROR, overread_err);
597 skip_bits_long(gb, 8 * count);
601 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
606 ics->predictor_reset_group = get_bits(gb, 5);
607 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
608 av_log(ac->avccontext, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
612 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
613 ics->prediction_used[sfb] = get_bits1(gb);
619 * Decode Individual Channel Stream info; reference: table 4.6.
621 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
623 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
624 GetBitContext *gb, int common_window)
627 av_log(ac->avccontext, AV_LOG_ERROR, "Reserved bit set.\n");
628 memset(ics, 0, sizeof(IndividualChannelStream));
631 ics->window_sequence[1] = ics->window_sequence[0];
632 ics->window_sequence[0] = get_bits(gb, 2);
633 ics->use_kb_window[1] = ics->use_kb_window[0];
634 ics->use_kb_window[0] = get_bits1(gb);
635 ics->num_window_groups = 1;
636 ics->group_len[0] = 1;
637 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
639 ics->max_sfb = get_bits(gb, 4);
640 for (i = 0; i < 7; i++) {
642 ics->group_len[ics->num_window_groups - 1]++;
644 ics->num_window_groups++;
645 ics->group_len[ics->num_window_groups - 1] = 1;
648 ics->num_windows = 8;
649 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
650 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
651 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
652 ics->predictor_present = 0;
654 ics->max_sfb = get_bits(gb, 6);
655 ics->num_windows = 1;
656 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
657 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
658 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
659 ics->predictor_present = get_bits1(gb);
660 ics->predictor_reset_group = 0;
661 if (ics->predictor_present) {
662 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
663 if (decode_prediction(ac, ics, gb)) {
664 memset(ics, 0, sizeof(IndividualChannelStream));
667 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
668 av_log(ac->avccontext, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
669 memset(ics, 0, sizeof(IndividualChannelStream));
672 av_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
673 memset(ics, 0, sizeof(IndividualChannelStream));
679 if (ics->max_sfb > ics->num_swb) {
680 av_log(ac->avccontext, AV_LOG_ERROR,
681 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
682 ics->max_sfb, ics->num_swb);
683 memset(ics, 0, sizeof(IndividualChannelStream));
691 * Decode band types (section_data payload); reference: table 4.46.
693 * @param band_type array of the used band type
694 * @param band_type_run_end array of the last scalefactor band of a band type run
696 * @return Returns error status. 0 - OK, !0 - error
698 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
699 int band_type_run_end[120], GetBitContext *gb,
700 IndividualChannelStream *ics)
703 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
704 for (g = 0; g < ics->num_window_groups; g++) {
706 while (k < ics->max_sfb) {
707 uint8_t sect_end = k;
709 int sect_band_type = get_bits(gb, 4);
710 if (sect_band_type == 12) {
711 av_log(ac->avccontext, AV_LOG_ERROR, "invalid band type\n");
714 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
715 sect_end += sect_len_incr;
716 sect_end += sect_len_incr;
717 if (get_bits_left(gb) < 0) {
718 av_log(ac->avccontext, AV_LOG_ERROR, overread_err);
721 if (sect_end > ics->max_sfb) {
722 av_log(ac->avccontext, AV_LOG_ERROR,
723 "Number of bands (%d) exceeds limit (%d).\n",
724 sect_end, ics->max_sfb);
727 for (; k < sect_end; k++) {
728 band_type [idx] = sect_band_type;
729 band_type_run_end[idx++] = sect_end;
737 * Decode scalefactors; reference: table 4.47.
739 * @param global_gain first scalefactor value as scalefactors are differentially coded
740 * @param band_type array of the used band type
741 * @param band_type_run_end array of the last scalefactor band of a band type run
742 * @param sf array of scalefactors or intensity stereo positions
744 * @return Returns error status. 0 - OK, !0 - error
746 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
747 unsigned int global_gain,
748 IndividualChannelStream *ics,
749 enum BandType band_type[120],
750 int band_type_run_end[120])
752 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
754 int offset[3] = { global_gain, global_gain - 90, 100 };
756 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
757 for (g = 0; g < ics->num_window_groups; g++) {
758 for (i = 0; i < ics->max_sfb;) {
759 int run_end = band_type_run_end[idx];
760 if (band_type[idx] == ZERO_BT) {
761 for (; i < run_end; i++, idx++)
763 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
764 for (; i < run_end; i++, idx++) {
765 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
766 if (offset[2] > 255U) {
767 av_log(ac->avccontext, AV_LOG_ERROR,
768 "%s (%d) out of range.\n", sf_str[2], offset[2]);
771 sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
773 } else if (band_type[idx] == NOISE_BT) {
774 for (; i < run_end; i++, idx++) {
775 if (noise_flag-- > 0)
776 offset[1] += get_bits(gb, 9) - 256;
778 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
779 if (offset[1] > 255U) {
780 av_log(ac->avccontext, AV_LOG_ERROR,
781 "%s (%d) out of range.\n", sf_str[1], offset[1]);
784 sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
787 for (; i < run_end; i++, idx++) {
788 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
789 if (offset[0] > 255U) {
790 av_log(ac->avccontext, AV_LOG_ERROR,
791 "%s (%d) out of range.\n", sf_str[0], offset[0]);
794 sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
803 * Decode pulse data; reference: table 4.7.
805 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
806 const uint16_t *swb_offset, int num_swb)
809 pulse->num_pulse = get_bits(gb, 2) + 1;
810 pulse_swb = get_bits(gb, 6);
811 if (pulse_swb >= num_swb)
813 pulse->pos[0] = swb_offset[pulse_swb];
814 pulse->pos[0] += get_bits(gb, 5);
815 if (pulse->pos[0] > 1023)
817 pulse->amp[0] = get_bits(gb, 4);
818 for (i = 1; i < pulse->num_pulse; i++) {
819 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
820 if (pulse->pos[i] > 1023)
822 pulse->amp[i] = get_bits(gb, 4);
828 * Decode Temporal Noise Shaping data; reference: table 4.48.
830 * @return Returns error status. 0 - OK, !0 - error
832 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
833 GetBitContext *gb, const IndividualChannelStream *ics)
835 int w, filt, i, coef_len, coef_res, coef_compress;
836 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
837 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
838 for (w = 0; w < ics->num_windows; w++) {
839 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
840 coef_res = get_bits1(gb);
842 for (filt = 0; filt < tns->n_filt[w]; filt++) {
844 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
846 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
847 av_log(ac->avccontext, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
848 tns->order[w][filt], tns_max_order);
849 tns->order[w][filt] = 0;
852 if (tns->order[w][filt]) {
853 tns->direction[w][filt] = get_bits1(gb);
854 coef_compress = get_bits1(gb);
855 coef_len = coef_res + 3 - coef_compress;
856 tmp2_idx = 2 * coef_compress + coef_res;
858 for (i = 0; i < tns->order[w][filt]; i++)
859 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
868 * Decode Mid/Side data; reference: table 4.54.
870 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
871 * [1] mask is decoded from bitstream; [2] mask is all 1s;
872 * [3] reserved for scalable AAC
874 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
878 if (ms_present == 1) {
879 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
880 cpe->ms_mask[idx] = get_bits1(gb);
881 } else if (ms_present == 2) {
882 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
887 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
891 *dst++ = v[idx & 15] * s;
892 *dst++ = v[idx>>4 & 15] * s;
898 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
902 *dst++ = v[idx & 3] * s;
903 *dst++ = v[idx>>2 & 3] * s;
904 *dst++ = v[idx>>4 & 3] * s;
905 *dst++ = v[idx>>6 & 3] * s;
911 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
912 unsigned sign, const float *scale)
914 union float754 s0, s1;
916 s0.f = s1.f = *scale;
917 s0.i ^= sign >> 1 << 31;
920 *dst++ = v[idx & 15] * s0.f;
921 *dst++ = v[idx>>4 & 15] * s1.f;
928 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
929 unsigned sign, const float *scale)
931 unsigned nz = idx >> 12;
932 union float754 s = { .f = *scale };
935 t.i = s.i ^ (sign & 1<<31);
936 *dst++ = v[idx & 3] * t.f;
938 sign <<= nz & 1; nz >>= 1;
939 t.i = s.i ^ (sign & 1<<31);
940 *dst++ = v[idx>>2 & 3] * t.f;
942 sign <<= nz & 1; nz >>= 1;
943 t.i = s.i ^ (sign & 1<<31);
944 *dst++ = v[idx>>4 & 3] * t.f;
946 sign <<= nz & 1; nz >>= 1;
947 t.i = s.i ^ (sign & 1<<31);
948 *dst++ = v[idx>>6 & 3] * t.f;
955 * Decode spectral data; reference: table 4.50.
956 * Dequantize and scale spectral data; reference: 4.6.3.3.
958 * @param coef array of dequantized, scaled spectral data
959 * @param sf array of scalefactors or intensity stereo positions
960 * @param pulse_present set if pulses are present
961 * @param pulse pointer to pulse data struct
962 * @param band_type array of the used band type
964 * @return Returns error status. 0 - OK, !0 - error
966 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
967 GetBitContext *gb, const float sf[120],
968 int pulse_present, const Pulse *pulse,
969 const IndividualChannelStream *ics,
970 enum BandType band_type[120])
972 int i, k, g, idx = 0;
973 const int c = 1024 / ics->num_windows;
974 const uint16_t *offsets = ics->swb_offset;
975 float *coef_base = coef;
978 for (g = 0; g < ics->num_windows; g++)
979 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
981 for (g = 0; g < ics->num_window_groups; g++) {
982 unsigned g_len = ics->group_len[g];
984 for (i = 0; i < ics->max_sfb; i++, idx++) {
985 const unsigned cbt_m1 = band_type[idx] - 1;
986 float *cfo = coef + offsets[i];
987 int off_len = offsets[i + 1] - offsets[i];
990 if (cbt_m1 >= INTENSITY_BT2 - 1) {
991 for (group = 0; group < g_len; group++, cfo+=128) {
992 memset(cfo, 0, off_len * sizeof(float));
994 } else if (cbt_m1 == NOISE_BT - 1) {
995 for (group = 0; group < g_len; group++, cfo+=128) {
999 for (k = 0; k < off_len; k++) {
1000 ac->random_state = lcg_random(ac->random_state);
1001 cfo[k] = ac->random_state;
1004 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1005 scale = sf[idx] / sqrtf(band_energy);
1006 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1009 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1010 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1011 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1012 const int cb_size = ff_aac_spectral_sizes[cbt_m1];
1013 OPEN_READER(re, gb);
1015 switch (cbt_m1 >> 1) {
1017 for (group = 0; group < g_len; group++, cfo+=128) {
1025 UPDATE_CACHE(re, gb);
1026 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1028 if (code >= cb_size) {
1030 goto err_cb_overflow;
1033 cb_idx = cb_vector_idx[code];
1034 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1040 for (group = 0; group < g_len; group++, cfo+=128) {
1050 UPDATE_CACHE(re, gb);
1051 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1053 if (code >= cb_size) {
1055 goto err_cb_overflow;
1058 #if MIN_CACHE_BITS < 20
1059 UPDATE_CACHE(re, gb);
1061 cb_idx = cb_vector_idx[code];
1062 nnz = cb_idx >> 8 & 15;
1063 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1064 LAST_SKIP_BITS(re, gb, nnz);
1065 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1071 for (group = 0; group < g_len; group++, cfo+=128) {
1079 UPDATE_CACHE(re, gb);
1080 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1082 if (code >= cb_size) {
1084 goto err_cb_overflow;
1087 cb_idx = cb_vector_idx[code];
1088 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1095 for (group = 0; group < g_len; group++, cfo+=128) {
1105 UPDATE_CACHE(re, gb);
1106 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1108 if (code >= cb_size) {
1110 goto err_cb_overflow;
1113 cb_idx = cb_vector_idx[code];
1114 nnz = cb_idx >> 8 & 15;
1115 sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1116 LAST_SKIP_BITS(re, gb, nnz);
1117 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1123 for (group = 0; group < g_len; group++, cfo+=128) {
1125 uint32_t *icf = (uint32_t *) cf;
1135 UPDATE_CACHE(re, gb);
1136 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1144 if (code >= cb_size) {
1146 goto err_cb_overflow;
1149 cb_idx = cb_vector_idx[code];
1152 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1153 LAST_SKIP_BITS(re, gb, nnz);
1155 for (j = 0; j < 2; j++) {
1159 /* The total length of escape_sequence must be < 22 bits according
1160 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1161 UPDATE_CACHE(re, gb);
1162 b = GET_CACHE(re, gb);
1163 b = 31 - av_log2(~b);
1166 av_log(ac->avccontext, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1170 #if MIN_CACHE_BITS < 21
1171 LAST_SKIP_BITS(re, gb, b + 1);
1172 UPDATE_CACHE(re, gb);
1174 SKIP_BITS(re, gb, b + 1);
1177 n = (1 << b) + SHOW_UBITS(re, gb, b);
1178 LAST_SKIP_BITS(re, gb, b);
1179 *icf++ = cbrt_tab[n] | (bits & 1<<31);
1182 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1183 *icf++ = (bits & 1<<31) | v;
1190 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1194 CLOSE_READER(re, gb);
1200 if (pulse_present) {
1202 for (i = 0; i < pulse->num_pulse; i++) {
1203 float co = coef_base[ pulse->pos[i] ];
1204 while (offsets[idx + 1] <= pulse->pos[i])
1206 if (band_type[idx] != NOISE_BT && sf[idx]) {
1207 float ico = -pulse->amp[i];
1210 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1212 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1219 av_log(ac->avccontext, AV_LOG_ERROR,
1220 "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
1221 band_type[idx], err_idx, ff_aac_spectral_sizes[band_type[idx]]);
1225 static av_always_inline float flt16_round(float pf)
1229 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1233 static av_always_inline float flt16_even(float pf)
1237 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1241 static av_always_inline float flt16_trunc(float pf)
1245 pun.i &= 0xFFFF0000U;
1249 static av_always_inline void predict(AACContext *ac, PredictorState *ps, float *coef,
1252 const float a = 0.953125; // 61.0 / 64
1253 const float alpha = 0.90625; // 29.0 / 32
1258 k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0;
1259 k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0;
1261 pv = flt16_round(k1 * ps->r0 + k2 * ps->r1);
1263 *coef += pv * ac->sf_scale;
1265 e0 = *coef / ac->sf_scale;
1266 e1 = e0 - k1 * ps->r0;
1268 ps->cor1 = flt16_trunc(alpha * ps->cor1 + ps->r1 * e1);
1269 ps->var1 = flt16_trunc(alpha * ps->var1 + 0.5 * (ps->r1 * ps->r1 + e1 * e1));
1270 ps->cor0 = flt16_trunc(alpha * ps->cor0 + ps->r0 * e0);
1271 ps->var0 = flt16_trunc(alpha * ps->var0 + 0.5 * (ps->r0 * ps->r0 + e0 * e0));
1273 ps->r1 = flt16_trunc(a * (ps->r0 - k1 * e0));
1274 ps->r0 = flt16_trunc(a * e0);
1278 * Apply AAC-Main style frequency domain prediction.
1280 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1284 if (!sce->ics.predictor_initialized) {
1285 reset_all_predictors(sce->predictor_state);
1286 sce->ics.predictor_initialized = 1;
1289 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1290 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1291 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1292 predict(ac, &sce->predictor_state[k], &sce->coeffs[k],
1293 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1296 if (sce->ics.predictor_reset_group)
1297 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1299 reset_all_predictors(sce->predictor_state);
1303 * Decode an individual_channel_stream payload; reference: table 4.44.
1305 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1306 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1308 * @return Returns error status. 0 - OK, !0 - error
1310 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1311 GetBitContext *gb, int common_window, int scale_flag)
1314 TemporalNoiseShaping *tns = &sce->tns;
1315 IndividualChannelStream *ics = &sce->ics;
1316 float *out = sce->coeffs;
1317 int global_gain, pulse_present = 0;
1319 /* This assignment is to silence a GCC warning about the variable being used
1320 * uninitialized when in fact it always is.
1322 pulse.num_pulse = 0;
1324 global_gain = get_bits(gb, 8);
1326 if (!common_window && !scale_flag) {
1327 if (decode_ics_info(ac, ics, gb, 0) < 0)
1331 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1333 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1338 if ((pulse_present = get_bits1(gb))) {
1339 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1340 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1343 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1344 av_log(ac->avccontext, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1348 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1350 if (get_bits1(gb)) {
1351 av_log_missing_feature(ac->avccontext, "SSR", 1);
1356 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1359 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1360 apply_prediction(ac, sce);
1366 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1368 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1370 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1371 float *ch0 = cpe->ch[0].coeffs;
1372 float *ch1 = cpe->ch[1].coeffs;
1373 int g, i, group, idx = 0;
1374 const uint16_t *offsets = ics->swb_offset;
1375 for (g = 0; g < ics->num_window_groups; g++) {
1376 for (i = 0; i < ics->max_sfb; i++, idx++) {
1377 if (cpe->ms_mask[idx] &&
1378 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1379 for (group = 0; group < ics->group_len[g]; group++) {
1380 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1381 ch1 + group * 128 + offsets[i],
1382 offsets[i+1] - offsets[i]);
1386 ch0 += ics->group_len[g] * 128;
1387 ch1 += ics->group_len[g] * 128;
1392 * intensity stereo decoding; reference: 4.6.8.2.3
1394 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1395 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1396 * [3] reserved for scalable AAC
1398 static void apply_intensity_stereo(ChannelElement *cpe, int ms_present)
1400 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1401 SingleChannelElement *sce1 = &cpe->ch[1];
1402 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1403 const uint16_t *offsets = ics->swb_offset;
1404 int g, group, i, k, idx = 0;
1407 for (g = 0; g < ics->num_window_groups; g++) {
1408 for (i = 0; i < ics->max_sfb;) {
1409 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1410 const int bt_run_end = sce1->band_type_run_end[idx];
1411 for (; i < bt_run_end; i++, idx++) {
1412 c = -1 + 2 * (sce1->band_type[idx] - 14);
1414 c *= 1 - 2 * cpe->ms_mask[idx];
1415 scale = c * sce1->sf[idx];
1416 for (group = 0; group < ics->group_len[g]; group++)
1417 for (k = offsets[i]; k < offsets[i + 1]; k++)
1418 coef1[group * 128 + k] = scale * coef0[group * 128 + k];
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 * @param elem_id Identifies the instance of a syntax element.
1436 * @return Returns error status. 0 - OK, !0 - error
1438 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1440 int i, ret, common_window, ms_present = 0;
1442 common_window = get_bits1(gb);
1443 if (common_window) {
1444 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1446 i = cpe->ch[1].ics.use_kb_window[0];
1447 cpe->ch[1].ics = cpe->ch[0].ics;
1448 cpe->ch[1].ics.use_kb_window[1] = i;
1449 ms_present = get_bits(gb, 2);
1450 if (ms_present == 3) {
1451 av_log(ac->avccontext, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1453 } else if (ms_present)
1454 decode_mid_side_stereo(cpe, gb, ms_present);
1456 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1458 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1461 if (common_window) {
1463 apply_mid_side_stereo(ac, cpe);
1464 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1465 apply_prediction(ac, &cpe->ch[0]);
1466 apply_prediction(ac, &cpe->ch[1]);
1470 apply_intensity_stereo(cpe, ms_present);
1475 * Decode coupling_channel_element; reference: table 4.8.
1477 * @param elem_id Identifies the instance of a syntax element.
1479 * @return Returns error status. 0 - OK, !0 - error
1481 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1487 SingleChannelElement *sce = &che->ch[0];
1488 ChannelCoupling *coup = &che->coup;
1490 coup->coupling_point = 2 * get_bits1(gb);
1491 coup->num_coupled = get_bits(gb, 3);
1492 for (c = 0; c <= coup->num_coupled; c++) {
1494 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1495 coup->id_select[c] = get_bits(gb, 4);
1496 if (coup->type[c] == TYPE_CPE) {
1497 coup->ch_select[c] = get_bits(gb, 2);
1498 if (coup->ch_select[c] == 3)
1501 coup->ch_select[c] = 2;
1503 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1505 sign = get_bits(gb, 1);
1506 scale = pow(2., pow(2., (int)get_bits(gb, 2) - 3));
1508 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1511 for (c = 0; c < num_gain; c++) {
1515 float gain_cache = 1.;
1517 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1518 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1519 gain_cache = pow(scale, -gain);
1521 if (coup->coupling_point == AFTER_IMDCT) {
1522 coup->gain[c][0] = gain_cache;
1524 for (g = 0; g < sce->ics.num_window_groups; g++) {
1525 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1526 if (sce->band_type[idx] != ZERO_BT) {
1528 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1536 gain_cache = pow(scale, -t) * s;
1539 coup->gain[c][idx] = gain_cache;
1549 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1551 * @return Returns number of bytes consumed.
1553 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1557 int num_excl_chan = 0;
1560 for (i = 0; i < 7; i++)
1561 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1562 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1564 return num_excl_chan / 7;
1568 * Decode dynamic range information; reference: table 4.52.
1570 * @param cnt length of TYPE_FIL syntactic element in bytes
1572 * @return Returns number of bytes consumed.
1574 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1575 GetBitContext *gb, int cnt)
1578 int drc_num_bands = 1;
1581 /* pce_tag_present? */
1582 if (get_bits1(gb)) {
1583 che_drc->pce_instance_tag = get_bits(gb, 4);
1584 skip_bits(gb, 4); // tag_reserved_bits
1588 /* excluded_chns_present? */
1589 if (get_bits1(gb)) {
1590 n += decode_drc_channel_exclusions(che_drc, gb);
1593 /* drc_bands_present? */
1594 if (get_bits1(gb)) {
1595 che_drc->band_incr = get_bits(gb, 4);
1596 che_drc->interpolation_scheme = get_bits(gb, 4);
1598 drc_num_bands += che_drc->band_incr;
1599 for (i = 0; i < drc_num_bands; i++) {
1600 che_drc->band_top[i] = get_bits(gb, 8);
1605 /* prog_ref_level_present? */
1606 if (get_bits1(gb)) {
1607 che_drc->prog_ref_level = get_bits(gb, 7);
1608 skip_bits1(gb); // prog_ref_level_reserved_bits
1612 for (i = 0; i < drc_num_bands; i++) {
1613 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1614 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1622 * Decode extension data (incomplete); reference: table 4.51.
1624 * @param cnt length of TYPE_FIL syntactic element in bytes
1626 * @return Returns number of bytes consumed
1628 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1629 ChannelElement *che, enum RawDataBlockType elem_type)
1633 switch (get_bits(gb, 4)) { // extension type
1634 case EXT_SBR_DATA_CRC:
1638 av_log(ac->avccontext, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1640 } else if (!ac->m4ac.sbr) {
1641 av_log(ac->avccontext, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1642 skip_bits_long(gb, 8 * cnt - 4);
1644 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1645 av_log(ac->avccontext, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1646 skip_bits_long(gb, 8 * cnt - 4);
1651 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1653 case EXT_DYNAMIC_RANGE:
1654 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1658 case EXT_DATA_ELEMENT:
1660 skip_bits_long(gb, 8 * cnt - 4);
1667 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1669 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1670 * @param coef spectral coefficients
1672 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1673 IndividualChannelStream *ics, int decode)
1675 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1677 int bottom, top, order, start, end, size, inc;
1678 float lpc[TNS_MAX_ORDER];
1680 for (w = 0; w < ics->num_windows; w++) {
1681 bottom = ics->num_swb;
1682 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1684 bottom = FFMAX(0, top - tns->length[w][filt]);
1685 order = tns->order[w][filt];
1690 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1692 start = ics->swb_offset[FFMIN(bottom, mmm)];
1693 end = ics->swb_offset[FFMIN( top, mmm)];
1694 if ((size = end - start) <= 0)
1696 if (tns->direction[w][filt]) {
1705 for (m = 0; m < size; m++, start += inc)
1706 for (i = 1; i <= FFMIN(m, order); i++)
1707 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1713 * Conduct IMDCT and windowing.
1715 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce, float bias)
1717 IndividualChannelStream *ics = &sce->ics;
1718 float *in = sce->coeffs;
1719 float *out = sce->ret;
1720 float *saved = sce->saved;
1721 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1722 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1723 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1724 float *buf = ac->buf_mdct;
1725 float *temp = ac->temp;
1729 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1730 if (ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE)
1731 av_log(ac->avccontext, AV_LOG_WARNING,
1732 "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1733 "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
1734 for (i = 0; i < 1024; i += 128)
1735 ff_imdct_half(&ac->mdct_small, buf + i, in + i);
1737 ff_imdct_half(&ac->mdct, buf, in);
1739 /* window overlapping
1740 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1741 * and long to short transitions are considered to be short to short
1742 * transitions. This leaves just two cases (long to long and short to short)
1743 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1745 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1746 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1747 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, bias, 512);
1749 for (i = 0; i < 448; i++)
1750 out[i] = saved[i] + bias;
1752 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1753 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, bias, 64);
1754 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, bias, 64);
1755 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, bias, 64);
1756 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, bias, 64);
1757 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, bias, 64);
1758 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1760 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, bias, 64);
1761 for (i = 576; i < 1024; i++)
1762 out[i] = buf[i-512] + bias;
1767 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1768 for (i = 0; i < 64; i++)
1769 saved[i] = temp[64 + i] - bias;
1770 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1771 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1772 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1773 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1774 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1775 memcpy( saved, buf + 512, 448 * sizeof(float));
1776 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1777 } else { // LONG_STOP or ONLY_LONG
1778 memcpy( saved, buf + 512, 512 * sizeof(float));
1783 * Apply dependent channel coupling (applied before IMDCT).
1785 * @param index index into coupling gain array
1787 static void apply_dependent_coupling(AACContext *ac,
1788 SingleChannelElement *target,
1789 ChannelElement *cce, int index)
1791 IndividualChannelStream *ics = &cce->ch[0].ics;
1792 const uint16_t *offsets = ics->swb_offset;
1793 float *dest = target->coeffs;
1794 const float *src = cce->ch[0].coeffs;
1795 int g, i, group, k, idx = 0;
1796 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1797 av_log(ac->avccontext, AV_LOG_ERROR,
1798 "Dependent coupling is not supported together with LTP\n");
1801 for (g = 0; g < ics->num_window_groups; g++) {
1802 for (i = 0; i < ics->max_sfb; i++, idx++) {
1803 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1804 const float gain = cce->coup.gain[index][idx];
1805 for (group = 0; group < ics->group_len[g]; group++) {
1806 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1808 dest[group * 128 + k] += gain * src[group * 128 + k];
1813 dest += ics->group_len[g] * 128;
1814 src += ics->group_len[g] * 128;
1819 * Apply independent channel coupling (applied after IMDCT).
1821 * @param index index into coupling gain array
1823 static void apply_independent_coupling(AACContext *ac,
1824 SingleChannelElement *target,
1825 ChannelElement *cce, int index)
1828 const float gain = cce->coup.gain[index][0];
1829 const float bias = ac->add_bias;
1830 const float *src = cce->ch[0].ret;
1831 float *dest = target->ret;
1832 const int len = 1024 << (ac->m4ac.sbr == 1);
1834 for (i = 0; i < len; i++)
1835 dest[i] += gain * (src[i] - bias);
1839 * channel coupling transformation interface
1841 * @param index index into coupling gain array
1842 * @param apply_coupling_method pointer to (in)dependent coupling function
1844 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1845 enum RawDataBlockType type, int elem_id,
1846 enum CouplingPoint coupling_point,
1847 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1851 for (i = 0; i < MAX_ELEM_ID; i++) {
1852 ChannelElement *cce = ac->che[TYPE_CCE][i];
1855 if (cce && cce->coup.coupling_point == coupling_point) {
1856 ChannelCoupling *coup = &cce->coup;
1858 for (c = 0; c <= coup->num_coupled; c++) {
1859 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1860 if (coup->ch_select[c] != 1) {
1861 apply_coupling_method(ac, &cc->ch[0], cce, index);
1862 if (coup->ch_select[c] != 0)
1865 if (coup->ch_select[c] != 2)
1866 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1868 index += 1 + (coup->ch_select[c] == 3);
1875 * Convert spectral data to float samples, applying all supported tools as appropriate.
1877 static void spectral_to_sample(AACContext *ac)
1880 float imdct_bias = (ac->m4ac.sbr <= 0) ? ac->add_bias : 0.0f;
1881 for (type = 3; type >= 0; type--) {
1882 for (i = 0; i < MAX_ELEM_ID; i++) {
1883 ChannelElement *che = ac->che[type][i];
1885 if (type <= TYPE_CPE)
1886 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1887 if (che->ch[0].tns.present)
1888 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1889 if (che->ch[1].tns.present)
1890 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
1891 if (type <= TYPE_CPE)
1892 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
1893 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
1894 imdct_and_windowing(ac, &che->ch[0], imdct_bias);
1895 if (type == TYPE_CPE) {
1896 imdct_and_windowing(ac, &che->ch[1], imdct_bias);
1898 if (ac->m4ac.sbr > 0) {
1899 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
1902 if (type <= TYPE_CCE)
1903 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
1909 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
1912 AACADTSHeaderInfo hdr_info;
1914 size = ff_aac_parse_header(gb, &hdr_info);
1916 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
1917 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1918 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1919 ac->m4ac.chan_config = hdr_info.chan_config;
1920 if (set_default_channel_config(ac, new_che_pos, hdr_info.chan_config))
1922 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
1924 } else if (ac->output_configured != OC_LOCKED) {
1925 ac->output_configured = OC_NONE;
1927 if (ac->output_configured != OC_LOCKED)
1929 ac->m4ac.sample_rate = hdr_info.sample_rate;
1930 ac->m4ac.sampling_index = hdr_info.sampling_index;
1931 ac->m4ac.object_type = hdr_info.object_type;
1932 if (!ac->avccontext->sample_rate)
1933 ac->avccontext->sample_rate = hdr_info.sample_rate;
1934 if (hdr_info.num_aac_frames == 1) {
1935 if (!hdr_info.crc_absent)
1938 av_log_missing_feature(ac->avccontext, "More than one AAC RDB per ADTS frame is", 0);
1945 static int aac_decode_frame(AVCodecContext *avccontext, void *data,
1946 int *data_size, AVPacket *avpkt)
1948 const uint8_t *buf = avpkt->data;
1949 int buf_size = avpkt->size;
1950 AACContext *ac = avccontext->priv_data;
1951 ChannelElement *che = NULL, *che_prev = NULL;
1953 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
1954 int err, elem_id, data_size_tmp;
1956 int samples = 1024, multiplier;
1959 init_get_bits(&gb, buf, buf_size * 8);
1961 if (show_bits(&gb, 12) == 0xfff) {
1962 if (parse_adts_frame_header(ac, &gb) < 0) {
1963 av_log(avccontext, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
1966 if (ac->m4ac.sampling_index > 12) {
1967 av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
1973 while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
1974 elem_id = get_bits(&gb, 4);
1976 if (elem_type < TYPE_DSE && !(che=get_che(ac, elem_type, elem_id))) {
1977 av_log(ac->avccontext, AV_LOG_ERROR, "channel element %d.%d is not allocated\n", elem_type, elem_id);
1981 switch (elem_type) {
1984 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1988 err = decode_cpe(ac, &gb, che);
1992 err = decode_cce(ac, &gb, che);
1996 err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
2000 err = skip_data_stream_element(ac, &gb);
2004 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2005 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2006 if ((err = decode_pce(ac, new_che_pos, &gb)))
2008 if (ac->output_configured > OC_TRIAL_PCE)
2009 av_log(avccontext, AV_LOG_ERROR,
2010 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2012 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2018 elem_id += get_bits(&gb, 8) - 1;
2019 if (get_bits_left(&gb) < 8 * elem_id) {
2020 av_log(avccontext, AV_LOG_ERROR, overread_err);
2024 elem_id -= decode_extension_payload(ac, &gb, elem_id, che_prev, elem_type_prev);
2025 err = 0; /* FIXME */
2029 err = -1; /* should not happen, but keeps compiler happy */
2034 elem_type_prev = elem_type;
2039 if (get_bits_left(&gb) < 3) {
2040 av_log(avccontext, AV_LOG_ERROR, overread_err);
2045 spectral_to_sample(ac);
2047 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2048 samples <<= multiplier;
2049 if (ac->output_configured < OC_LOCKED) {
2050 avccontext->sample_rate = ac->m4ac.sample_rate << multiplier;
2051 avccontext->frame_size = samples;
2054 data_size_tmp = samples * avccontext->channels * sizeof(int16_t);
2055 if (*data_size < data_size_tmp) {
2056 av_log(avccontext, AV_LOG_ERROR,
2057 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2058 *data_size, data_size_tmp);
2061 *data_size = data_size_tmp;
2063 ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, samples, avccontext->channels);
2065 if (ac->output_configured)
2066 ac->output_configured = OC_LOCKED;
2068 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2069 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2070 if (buf[buf_offset])
2073 return buf_size > buf_offset ? buf_consumed : buf_size;
2076 static av_cold int aac_decode_close(AVCodecContext *avccontext)
2078 AACContext *ac = avccontext->priv_data;
2081 for (i = 0; i < MAX_ELEM_ID; i++) {
2082 for (type = 0; type < 4; type++) {
2083 if (ac->che[type][i])
2084 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2085 av_freep(&ac->che[type][i]);
2089 ff_mdct_end(&ac->mdct);
2090 ff_mdct_end(&ac->mdct_small);
2094 AVCodec aac_decoder = {
2103 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2104 .sample_fmts = (const enum SampleFormat[]) {
2105 SAMPLE_FMT_S16,SAMPLE_FMT_NONE
2107 .channel_layouts = aac_channel_layout,