3 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
7 * Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
8 * Copyright (c) 2010 Janne Grunau <janne-libav@jannau.net>
10 * This file is part of Libav.
12 * Libav is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public
14 * License as published by the Free Software Foundation; either
15 * version 2.1 of the License, or (at your option) any later version.
17 * Libav is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with Libav; if not, write to the Free Software
24 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
30 * @author Oded Shimon ( ods15 ods15 dyndns org )
31 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
38 * N (code in SoC repo) gain control
40 * Y window shapes - standard
41 * N window shapes - Low Delay
42 * Y filterbank - standard
43 * N (code in SoC repo) filterbank - Scalable Sample Rate
44 * Y Temporal Noise Shaping
45 * Y Long Term Prediction
48 * Y frequency domain prediction
49 * Y Perceptual Noise Substitution
51 * N Scalable Inverse AAC Quantization
52 * N Frequency Selective Switch
54 * Y quantization & coding - AAC
55 * N quantization & coding - TwinVQ
56 * N quantization & coding - BSAC
57 * N AAC Error Resilience tools
58 * N Error Resilience payload syntax
59 * N Error Protection tool
61 * N Silence Compression
64 * N Structured Audio tools
65 * N Structured Audio Sample Bank Format
67 * N Harmonic and Individual Lines plus Noise
68 * N Text-To-Speech Interface
69 * Y Spectral Band Replication
70 * Y (not in this code) Layer-1
71 * Y (not in this code) Layer-2
72 * Y (not in this code) Layer-3
73 * N SinuSoidal Coding (Transient, Sinusoid, Noise)
75 * N Direct Stream Transfer
77 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
78 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
82 #include "libavutil/float_dsp.h"
87 #include "fmtconvert.h"
94 #include "aacdectab.h"
95 #include "cbrt_tablegen.h"
98 #include "mpeg4audio.h"
99 #include "aacadtsdec.h"
100 #include "libavutil/intfloat.h"
108 # include "arm/aac.h"
111 static VLC vlc_scalefactors;
112 static VLC vlc_spectral[11];
114 static const char overread_err[] = "Input buffer exhausted before END element found\n";
116 static int count_channels(uint8_t (*layout)[3], int tags)
119 for (i = 0; i < tags; i++) {
120 int syn_ele = layout[i][0];
121 int pos = layout[i][2];
122 sum += (1 + (syn_ele == TYPE_CPE)) *
123 (pos != AAC_CHANNEL_OFF && pos != AAC_CHANNEL_CC);
129 * Check for the channel element in the current channel position configuration.
130 * If it exists, make sure the appropriate element is allocated and map the
131 * channel order to match the internal Libav channel layout.
133 * @param che_pos current channel position configuration
134 * @param type channel element type
135 * @param id channel element id
136 * @param channels count of the number of channels in the configuration
138 * @return Returns error status. 0 - OK, !0 - error
140 static av_cold int che_configure(AACContext *ac,
141 enum ChannelPosition che_pos,
142 int type, int id, int *channels)
144 if (*channels >= MAX_CHANNELS)
145 return AVERROR_INVALIDDATA;
147 if (!ac->che[type][id]) {
148 if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
149 return AVERROR(ENOMEM);
150 ff_aac_sbr_ctx_init(ac, &ac->che[type][id]->sbr);
152 if (type != TYPE_CCE) {
153 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[0];
154 if (type == TYPE_CPE ||
155 (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1)) {
156 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[1];
160 if (ac->che[type][id])
161 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
162 av_freep(&ac->che[type][id]);
167 static int frame_configure_elements(AVCodecContext *avctx)
169 AACContext *ac = avctx->priv_data;
170 int type, id, ch, ret;
172 /* set channel pointers to internal buffers by default */
173 for (type = 0; type < 4; type++) {
174 for (id = 0; id < MAX_ELEM_ID; id++) {
175 ChannelElement *che = ac->che[type][id];
177 che->ch[0].ret = che->ch[0].ret_buf;
178 che->ch[1].ret = che->ch[1].ret_buf;
183 /* get output buffer */
184 av_frame_unref(ac->frame);
185 ac->frame->nb_samples = 2048;
186 if ((ret = ff_get_buffer(avctx, ac->frame, 0)) < 0) {
187 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
191 /* map output channel pointers to AVFrame data */
192 for (ch = 0; ch < avctx->channels; ch++) {
193 if (ac->output_element[ch])
194 ac->output_element[ch]->ret = (float *)ac->frame->extended_data[ch];
200 struct elem_to_channel {
201 uint64_t av_position;
204 uint8_t aac_position;
207 static int assign_pair(struct elem_to_channel e2c_vec[MAX_ELEM_ID],
208 uint8_t (*layout_map)[3], int offset, uint64_t left,
209 uint64_t right, int pos)
211 if (layout_map[offset][0] == TYPE_CPE) {
212 e2c_vec[offset] = (struct elem_to_channel) {
213 .av_position = left | right,
215 .elem_id = layout_map[offset][1],
220 e2c_vec[offset] = (struct elem_to_channel) {
223 .elem_id = layout_map[offset][1],
226 e2c_vec[offset + 1] = (struct elem_to_channel) {
227 .av_position = right,
229 .elem_id = layout_map[offset + 1][1],
236 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos,
239 int num_pos_channels = 0;
243 for (i = *current; i < tags; i++) {
244 if (layout_map[i][2] != pos)
246 if (layout_map[i][0] == TYPE_CPE) {
248 if (pos == AAC_CHANNEL_FRONT && !first_cpe) {
254 num_pos_channels += 2;
262 ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
265 return num_pos_channels;
268 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
270 int i, n, total_non_cc_elements;
271 struct elem_to_channel e2c_vec[4 * MAX_ELEM_ID] = { { 0 } };
272 int num_front_channels, num_side_channels, num_back_channels;
275 if (FF_ARRAY_ELEMS(e2c_vec) < tags)
280 count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
281 if (num_front_channels < 0)
284 count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
285 if (num_side_channels < 0)
288 count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
289 if (num_back_channels < 0)
293 if (num_front_channels & 1) {
294 e2c_vec[i] = (struct elem_to_channel) {
295 .av_position = AV_CH_FRONT_CENTER,
297 .elem_id = layout_map[i][1],
298 .aac_position = AAC_CHANNEL_FRONT
301 num_front_channels--;
303 if (num_front_channels >= 4) {
304 i += assign_pair(e2c_vec, layout_map, i,
305 AV_CH_FRONT_LEFT_OF_CENTER,
306 AV_CH_FRONT_RIGHT_OF_CENTER,
308 num_front_channels -= 2;
310 if (num_front_channels >= 2) {
311 i += assign_pair(e2c_vec, layout_map, i,
315 num_front_channels -= 2;
317 while (num_front_channels >= 2) {
318 i += assign_pair(e2c_vec, layout_map, i,
322 num_front_channels -= 2;
325 if (num_side_channels >= 2) {
326 i += assign_pair(e2c_vec, layout_map, i,
330 num_side_channels -= 2;
332 while (num_side_channels >= 2) {
333 i += assign_pair(e2c_vec, layout_map, i,
337 num_side_channels -= 2;
340 while (num_back_channels >= 4) {
341 i += assign_pair(e2c_vec, layout_map, i,
345 num_back_channels -= 2;
347 if (num_back_channels >= 2) {
348 i += assign_pair(e2c_vec, layout_map, i,
352 num_back_channels -= 2;
354 if (num_back_channels) {
355 e2c_vec[i] = (struct elem_to_channel) {
356 .av_position = AV_CH_BACK_CENTER,
358 .elem_id = layout_map[i][1],
359 .aac_position = AAC_CHANNEL_BACK
365 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
366 e2c_vec[i] = (struct elem_to_channel) {
367 .av_position = AV_CH_LOW_FREQUENCY,
369 .elem_id = layout_map[i][1],
370 .aac_position = AAC_CHANNEL_LFE
374 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
375 e2c_vec[i] = (struct elem_to_channel) {
376 .av_position = UINT64_MAX,
378 .elem_id = layout_map[i][1],
379 .aac_position = AAC_CHANNEL_LFE
384 // Must choose a stable sort
385 total_non_cc_elements = n = i;
388 for (i = 1; i < n; i++)
389 if (e2c_vec[i - 1].av_position > e2c_vec[i].av_position) {
390 FFSWAP(struct elem_to_channel, e2c_vec[i - 1], e2c_vec[i]);
397 for (i = 0; i < total_non_cc_elements; i++) {
398 layout_map[i][0] = e2c_vec[i].syn_ele;
399 layout_map[i][1] = e2c_vec[i].elem_id;
400 layout_map[i][2] = e2c_vec[i].aac_position;
401 if (e2c_vec[i].av_position != UINT64_MAX) {
402 layout |= e2c_vec[i].av_position;
410 * Save current output configuration if and only if it has been locked.
412 static void push_output_configuration(AACContext *ac) {
413 if (ac->oc[1].status == OC_LOCKED) {
414 ac->oc[0] = ac->oc[1];
416 ac->oc[1].status = OC_NONE;
420 * Restore the previous output configuration if and only if the current
421 * configuration is unlocked.
423 static void pop_output_configuration(AACContext *ac) {
424 if (ac->oc[1].status != OC_LOCKED && ac->oc[0].status != OC_NONE) {
425 ac->oc[1] = ac->oc[0];
426 ac->avctx->channels = ac->oc[1].channels;
427 ac->avctx->channel_layout = ac->oc[1].channel_layout;
432 * Configure output channel order based on the current program
433 * configuration element.
435 * @return Returns error status. 0 - OK, !0 - error
437 static int output_configure(AACContext *ac,
438 uint8_t layout_map[MAX_ELEM_ID * 4][3], int tags,
439 enum OCStatus oc_type, int get_new_frame)
441 AVCodecContext *avctx = ac->avctx;
442 int i, channels = 0, ret;
445 if (ac->oc[1].layout_map != layout_map) {
446 memcpy(ac->oc[1].layout_map, layout_map, tags * sizeof(layout_map[0]));
447 ac->oc[1].layout_map_tags = tags;
450 // Try to sniff a reasonable channel order, otherwise output the
451 // channels in the order the PCE declared them.
452 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
453 layout = sniff_channel_order(layout_map, tags);
454 for (i = 0; i < tags; i++) {
455 int type = layout_map[i][0];
456 int id = layout_map[i][1];
457 int position = layout_map[i][2];
458 // Allocate or free elements depending on if they are in the
459 // current program configuration.
460 ret = che_configure(ac, position, type, id, &channels);
464 if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
465 if (layout == AV_CH_FRONT_CENTER) {
466 layout = AV_CH_FRONT_LEFT|AV_CH_FRONT_RIGHT;
472 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
473 avctx->channel_layout = ac->oc[1].channel_layout = layout;
474 avctx->channels = ac->oc[1].channels = channels;
475 ac->oc[1].status = oc_type;
478 if ((ret = frame_configure_elements(ac->avctx)) < 0)
486 * Set up channel positions based on a default channel configuration
487 * as specified in table 1.17.
489 * @return Returns error status. 0 - OK, !0 - error
491 static int set_default_channel_config(AVCodecContext *avctx,
492 uint8_t (*layout_map)[3],
496 if (channel_config < 1 || channel_config > 7) {
497 av_log(avctx, AV_LOG_ERROR,
498 "invalid default channel configuration (%d)\n",
500 return AVERROR_INVALIDDATA;
502 *tags = tags_per_config[channel_config];
503 memcpy(layout_map, aac_channel_layout_map[channel_config - 1],
504 *tags * sizeof(*layout_map));
508 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
510 /* For PCE based channel configurations map the channels solely based
512 if (!ac->oc[1].m4ac.chan_config) {
513 return ac->tag_che_map[type][elem_id];
515 // Allow single CPE stereo files to be signalled with mono configuration.
516 if (!ac->tags_mapped && type == TYPE_CPE &&
517 ac->oc[1].m4ac.chan_config == 1) {
518 uint8_t layout_map[MAX_ELEM_ID*4][3];
520 push_output_configuration(ac);
522 if (set_default_channel_config(ac->avctx, layout_map,
523 &layout_map_tags, 2) < 0)
525 if (output_configure(ac, layout_map, layout_map_tags,
526 OC_TRIAL_FRAME, 1) < 0)
529 ac->oc[1].m4ac.chan_config = 2;
530 ac->oc[1].m4ac.ps = 0;
533 if (!ac->tags_mapped && type == TYPE_SCE &&
534 ac->oc[1].m4ac.chan_config == 2) {
535 uint8_t layout_map[MAX_ELEM_ID * 4][3];
537 push_output_configuration(ac);
539 if (set_default_channel_config(ac->avctx, layout_map,
540 &layout_map_tags, 1) < 0)
542 if (output_configure(ac, layout_map, layout_map_tags,
543 OC_TRIAL_FRAME, 1) < 0)
546 ac->oc[1].m4ac.chan_config = 1;
547 if (ac->oc[1].m4ac.sbr)
548 ac->oc[1].m4ac.ps = -1;
550 /* For indexed channel configurations map the channels solely based
552 switch (ac->oc[1].m4ac.chan_config) {
554 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
556 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
559 /* Some streams incorrectly code 5.1 audio as
560 * SCE[0] CPE[0] CPE[1] SCE[1]
562 * SCE[0] CPE[0] CPE[1] LFE[0].
563 * If we seem to have encountered such a stream, transfer
564 * the LFE[0] element to the SCE[1]'s mapping */
565 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
567 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
570 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
572 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
575 if (ac->tags_mapped == 2 &&
576 ac->oc[1].m4ac.chan_config == 4 &&
579 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
583 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) &&
586 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
587 } else if (ac->oc[1].m4ac.chan_config == 2) {
591 if (!ac->tags_mapped && type == TYPE_SCE) {
593 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
601 * Decode an array of 4 bit element IDs, optionally interleaved with a
602 * stereo/mono switching bit.
604 * @param type speaker type/position for these channels
606 static void decode_channel_map(uint8_t layout_map[][3],
607 enum ChannelPosition type,
608 GetBitContext *gb, int n)
611 enum RawDataBlockType syn_ele;
613 case AAC_CHANNEL_FRONT:
614 case AAC_CHANNEL_BACK:
615 case AAC_CHANNEL_SIDE:
616 syn_ele = get_bits1(gb);
622 case AAC_CHANNEL_LFE:
626 layout_map[0][0] = syn_ele;
627 layout_map[0][1] = get_bits(gb, 4);
628 layout_map[0][2] = type;
634 * Decode program configuration element; reference: table 4.2.
636 * @return Returns error status. 0 - OK, !0 - error
638 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
639 uint8_t (*layout_map)[3],
642 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc;
647 skip_bits(gb, 2); // object_type
649 sampling_index = get_bits(gb, 4);
650 if (m4ac->sampling_index != sampling_index)
651 av_log(avctx, AV_LOG_WARNING,
652 "Sample rate index in program config element does not "
653 "match the sample rate index configured by the container.\n");
655 num_front = get_bits(gb, 4);
656 num_side = get_bits(gb, 4);
657 num_back = get_bits(gb, 4);
658 num_lfe = get_bits(gb, 2);
659 num_assoc_data = get_bits(gb, 3);
660 num_cc = get_bits(gb, 4);
663 skip_bits(gb, 4); // mono_mixdown_tag
665 skip_bits(gb, 4); // stereo_mixdown_tag
668 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
670 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
672 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
674 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
676 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
679 skip_bits_long(gb, 4 * num_assoc_data);
681 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
686 /* comment field, first byte is length */
687 comment_len = get_bits(gb, 8) * 8;
688 if (get_bits_left(gb) < comment_len) {
689 av_log(avctx, AV_LOG_ERROR, overread_err);
690 return AVERROR_INVALIDDATA;
692 skip_bits_long(gb, comment_len);
697 * Decode GA "General Audio" specific configuration; reference: table 4.1.
699 * @param ac pointer to AACContext, may be null
700 * @param avctx pointer to AVCCodecContext, used for logging
702 * @return Returns error status. 0 - OK, !0 - error
704 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
706 MPEG4AudioConfig *m4ac,
709 int extension_flag, ret;
710 uint8_t layout_map[MAX_ELEM_ID*4][3];
713 if (get_bits1(gb)) { // frameLengthFlag
714 avpriv_request_sample(avctx, "960/120 MDCT window");
715 return AVERROR_PATCHWELCOME;
718 if (get_bits1(gb)) // dependsOnCoreCoder
719 skip_bits(gb, 14); // coreCoderDelay
720 extension_flag = get_bits1(gb);
722 if (m4ac->object_type == AOT_AAC_SCALABLE ||
723 m4ac->object_type == AOT_ER_AAC_SCALABLE)
724 skip_bits(gb, 3); // layerNr
726 if (channel_config == 0) {
727 skip_bits(gb, 4); // element_instance_tag
728 tags = decode_pce(avctx, m4ac, layout_map, gb);
732 if ((ret = set_default_channel_config(avctx, layout_map,
733 &tags, channel_config)))
737 if (count_channels(layout_map, tags) > 1) {
739 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
742 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
745 if (extension_flag) {
746 switch (m4ac->object_type) {
748 skip_bits(gb, 5); // numOfSubFrame
749 skip_bits(gb, 11); // layer_length
753 case AOT_ER_AAC_SCALABLE:
755 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
756 * aacScalefactorDataResilienceFlag
757 * aacSpectralDataResilienceFlag
761 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
767 * Decode audio specific configuration; reference: table 1.13.
769 * @param ac pointer to AACContext, may be null
770 * @param avctx pointer to AVCCodecContext, used for logging
771 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
772 * @param data pointer to buffer holding an audio specific config
773 * @param bit_size size of audio specific config or data in bits
774 * @param sync_extension look for an appended sync extension
776 * @return Returns error status or number of consumed bits. <0 - error
778 static int decode_audio_specific_config(AACContext *ac,
779 AVCodecContext *avctx,
780 MPEG4AudioConfig *m4ac,
781 const uint8_t *data, int bit_size,
787 av_dlog(avctx, "extradata size %d\n", avctx->extradata_size);
788 for (i = 0; i < avctx->extradata_size; i++)
789 av_dlog(avctx, "%02x ", avctx->extradata[i]);
790 av_dlog(avctx, "\n");
792 init_get_bits(&gb, data, bit_size);
794 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size,
795 sync_extension)) < 0)
796 return AVERROR_INVALIDDATA;
797 if (m4ac->sampling_index > 12) {
798 av_log(avctx, AV_LOG_ERROR,
799 "invalid sampling rate index %d\n",
800 m4ac->sampling_index);
801 return AVERROR_INVALIDDATA;
804 skip_bits_long(&gb, i);
806 switch (m4ac->object_type) {
810 if ((ret = decode_ga_specific_config(ac, avctx, &gb,
811 m4ac, m4ac->chan_config)) < 0)
815 av_log(avctx, AV_LOG_ERROR,
816 "Audio object type %s%d is not supported.\n",
817 m4ac->sbr == 1 ? "SBR+" : "",
819 return AVERROR(ENOSYS);
823 "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
824 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
825 m4ac->sample_rate, m4ac->sbr,
828 return get_bits_count(&gb);
832 * linear congruential pseudorandom number generator
834 * @param previous_val pointer to the current state of the generator
836 * @return Returns a 32-bit pseudorandom integer
838 static av_always_inline int lcg_random(int previous_val)
840 union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
844 static av_always_inline void reset_predict_state(PredictorState *ps)
854 static void reset_all_predictors(PredictorState *ps)
857 for (i = 0; i < MAX_PREDICTORS; i++)
858 reset_predict_state(&ps[i]);
861 static int sample_rate_idx (int rate)
863 if (92017 <= rate) return 0;
864 else if (75132 <= rate) return 1;
865 else if (55426 <= rate) return 2;
866 else if (46009 <= rate) return 3;
867 else if (37566 <= rate) return 4;
868 else if (27713 <= rate) return 5;
869 else if (23004 <= rate) return 6;
870 else if (18783 <= rate) return 7;
871 else if (13856 <= rate) return 8;
872 else if (11502 <= rate) return 9;
873 else if (9391 <= rate) return 10;
877 static void reset_predictor_group(PredictorState *ps, int group_num)
880 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
881 reset_predict_state(&ps[i]);
884 #define AAC_INIT_VLC_STATIC(num, size) \
885 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
886 ff_aac_spectral_bits[num], sizeof(ff_aac_spectral_bits[num][0]), \
887 sizeof(ff_aac_spectral_bits[num][0]), \
888 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), \
889 sizeof(ff_aac_spectral_codes[num][0]), \
892 static av_cold int aac_decode_init(AVCodecContext *avctx)
894 AACContext *ac = avctx->priv_data;
898 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
900 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
902 if (avctx->extradata_size > 0) {
903 if ((ret = decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
905 avctx->extradata_size * 8,
910 uint8_t layout_map[MAX_ELEM_ID*4][3];
913 sr = sample_rate_idx(avctx->sample_rate);
914 ac->oc[1].m4ac.sampling_index = sr;
915 ac->oc[1].m4ac.channels = avctx->channels;
916 ac->oc[1].m4ac.sbr = -1;
917 ac->oc[1].m4ac.ps = -1;
919 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
920 if (ff_mpeg4audio_channels[i] == avctx->channels)
922 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
925 ac->oc[1].m4ac.chan_config = i;
927 if (ac->oc[1].m4ac.chan_config) {
928 int ret = set_default_channel_config(avctx, layout_map,
929 &layout_map_tags, ac->oc[1].m4ac.chan_config);
931 output_configure(ac, layout_map, layout_map_tags,
933 else if (avctx->err_recognition & AV_EF_EXPLODE)
934 return AVERROR_INVALIDDATA;
938 AAC_INIT_VLC_STATIC( 0, 304);
939 AAC_INIT_VLC_STATIC( 1, 270);
940 AAC_INIT_VLC_STATIC( 2, 550);
941 AAC_INIT_VLC_STATIC( 3, 300);
942 AAC_INIT_VLC_STATIC( 4, 328);
943 AAC_INIT_VLC_STATIC( 5, 294);
944 AAC_INIT_VLC_STATIC( 6, 306);
945 AAC_INIT_VLC_STATIC( 7, 268);
946 AAC_INIT_VLC_STATIC( 8, 510);
947 AAC_INIT_VLC_STATIC( 9, 366);
948 AAC_INIT_VLC_STATIC(10, 462);
952 ff_fmt_convert_init(&ac->fmt_conv, avctx);
953 avpriv_float_dsp_init(&ac->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
955 ac->random_state = 0x1f2e3d4c;
959 INIT_VLC_STATIC(&vlc_scalefactors, 7,
960 FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
961 ff_aac_scalefactor_bits,
962 sizeof(ff_aac_scalefactor_bits[0]),
963 sizeof(ff_aac_scalefactor_bits[0]),
964 ff_aac_scalefactor_code,
965 sizeof(ff_aac_scalefactor_code[0]),
966 sizeof(ff_aac_scalefactor_code[0]),
969 ff_mdct_init(&ac->mdct, 11, 1, 1.0 / (32768.0 * 1024.0));
970 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0 / (32768.0 * 128.0));
971 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0 * 32768.0);
972 // window initialization
973 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
974 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
975 ff_init_ff_sine_windows(10);
976 ff_init_ff_sine_windows( 7);
984 * Skip data_stream_element; reference: table 4.10.
986 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
988 int byte_align = get_bits1(gb);
989 int count = get_bits(gb, 8);
991 count += get_bits(gb, 8);
995 if (get_bits_left(gb) < 8 * count) {
996 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
997 return AVERROR_INVALIDDATA;
999 skip_bits_long(gb, 8 * count);
1003 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
1007 if (get_bits1(gb)) {
1008 ics->predictor_reset_group = get_bits(gb, 5);
1009 if (ics->predictor_reset_group == 0 ||
1010 ics->predictor_reset_group > 30) {
1011 av_log(ac->avctx, AV_LOG_ERROR,
1012 "Invalid Predictor Reset Group.\n");
1013 return AVERROR_INVALIDDATA;
1016 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
1017 ics->prediction_used[sfb] = get_bits1(gb);
1023 * Decode Long Term Prediction data; reference: table 4.xx.
1025 static void decode_ltp(LongTermPrediction *ltp,
1026 GetBitContext *gb, uint8_t max_sfb)
1030 ltp->lag = get_bits(gb, 11);
1031 ltp->coef = ltp_coef[get_bits(gb, 3)];
1032 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1033 ltp->used[sfb] = get_bits1(gb);
1037 * Decode Individual Channel Stream info; reference: table 4.6.
1039 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
1042 if (get_bits1(gb)) {
1043 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1044 return AVERROR_INVALIDDATA;
1046 ics->window_sequence[1] = ics->window_sequence[0];
1047 ics->window_sequence[0] = get_bits(gb, 2);
1048 ics->use_kb_window[1] = ics->use_kb_window[0];
1049 ics->use_kb_window[0] = get_bits1(gb);
1050 ics->num_window_groups = 1;
1051 ics->group_len[0] = 1;
1052 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1054 ics->max_sfb = get_bits(gb, 4);
1055 for (i = 0; i < 7; i++) {
1056 if (get_bits1(gb)) {
1057 ics->group_len[ics->num_window_groups - 1]++;
1059 ics->num_window_groups++;
1060 ics->group_len[ics->num_window_groups - 1] = 1;
1063 ics->num_windows = 8;
1064 ics->swb_offset = ff_swb_offset_128[ac->oc[1].m4ac.sampling_index];
1065 ics->num_swb = ff_aac_num_swb_128[ac->oc[1].m4ac.sampling_index];
1066 ics->tns_max_bands = ff_tns_max_bands_128[ac->oc[1].m4ac.sampling_index];
1067 ics->predictor_present = 0;
1069 ics->max_sfb = get_bits(gb, 6);
1070 ics->num_windows = 1;
1071 ics->swb_offset = ff_swb_offset_1024[ac->oc[1].m4ac.sampling_index];
1072 ics->num_swb = ff_aac_num_swb_1024[ac->oc[1].m4ac.sampling_index];
1073 ics->tns_max_bands = ff_tns_max_bands_1024[ac->oc[1].m4ac.sampling_index];
1074 ics->predictor_present = get_bits1(gb);
1075 ics->predictor_reset_group = 0;
1076 if (ics->predictor_present) {
1077 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1078 if (decode_prediction(ac, ics, gb)) {
1079 return AVERROR_INVALIDDATA;
1081 } else if (ac->oc[1].m4ac.object_type == AOT_AAC_LC) {
1082 av_log(ac->avctx, AV_LOG_ERROR,
1083 "Prediction is not allowed in AAC-LC.\n");
1084 return AVERROR_INVALIDDATA;
1086 if ((ics->ltp.present = get_bits(gb, 1)))
1087 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1092 if (ics->max_sfb > ics->num_swb) {
1093 av_log(ac->avctx, AV_LOG_ERROR,
1094 "Number of scalefactor bands in group (%d) "
1095 "exceeds limit (%d).\n",
1096 ics->max_sfb, ics->num_swb);
1097 return AVERROR_INVALIDDATA;
1104 * Decode band types (section_data payload); reference: table 4.46.
1106 * @param band_type array of the used band type
1107 * @param band_type_run_end array of the last scalefactor band of a band type run
1109 * @return Returns error status. 0 - OK, !0 - error
1111 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1112 int band_type_run_end[120], GetBitContext *gb,
1113 IndividualChannelStream *ics)
1116 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1117 for (g = 0; g < ics->num_window_groups; g++) {
1119 while (k < ics->max_sfb) {
1120 uint8_t sect_end = k;
1122 int sect_band_type = get_bits(gb, 4);
1123 if (sect_band_type == 12) {
1124 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1125 return AVERROR_INVALIDDATA;
1128 sect_len_incr = get_bits(gb, bits);
1129 sect_end += sect_len_incr;
1130 if (get_bits_left(gb) < 0) {
1131 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
1132 return AVERROR_INVALIDDATA;
1134 if (sect_end > ics->max_sfb) {
1135 av_log(ac->avctx, AV_LOG_ERROR,
1136 "Number of bands (%d) exceeds limit (%d).\n",
1137 sect_end, ics->max_sfb);
1138 return AVERROR_INVALIDDATA;
1140 } while (sect_len_incr == (1 << bits) - 1);
1141 for (; k < sect_end; k++) {
1142 band_type [idx] = sect_band_type;
1143 band_type_run_end[idx++] = sect_end;
1151 * Decode scalefactors; reference: table 4.47.
1153 * @param global_gain first scalefactor value as scalefactors are differentially coded
1154 * @param band_type array of the used band type
1155 * @param band_type_run_end array of the last scalefactor band of a band type run
1156 * @param sf array of scalefactors or intensity stereo positions
1158 * @return Returns error status. 0 - OK, !0 - error
1160 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1161 unsigned int global_gain,
1162 IndividualChannelStream *ics,
1163 enum BandType band_type[120],
1164 int band_type_run_end[120])
1167 int offset[3] = { global_gain, global_gain - 90, 0 };
1170 for (g = 0; g < ics->num_window_groups; g++) {
1171 for (i = 0; i < ics->max_sfb;) {
1172 int run_end = band_type_run_end[idx];
1173 if (band_type[idx] == ZERO_BT) {
1174 for (; i < run_end; i++, idx++)
1176 } else if ((band_type[idx] == INTENSITY_BT) ||
1177 (band_type[idx] == INTENSITY_BT2)) {
1178 for (; i < run_end; i++, idx++) {
1179 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1180 clipped_offset = av_clip(offset[2], -155, 100);
1181 if (offset[2] != clipped_offset) {
1182 avpriv_request_sample(ac->avctx,
1183 "If you heard an audible artifact, there may be a bug in the decoder. "
1184 "Clipped intensity stereo position (%d -> %d)",
1185 offset[2], clipped_offset);
1187 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1189 } else if (band_type[idx] == NOISE_BT) {
1190 for (; i < run_end; i++, idx++) {
1191 if (noise_flag-- > 0)
1192 offset[1] += get_bits(gb, 9) - 256;
1194 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1195 clipped_offset = av_clip(offset[1], -100, 155);
1196 if (offset[1] != clipped_offset) {
1197 avpriv_request_sample(ac->avctx,
1198 "If you heard an audible artifact, there may be a bug in the decoder. "
1199 "Clipped noise gain (%d -> %d)",
1200 offset[1], clipped_offset);
1202 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1205 for (; i < run_end; i++, idx++) {
1206 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1207 if (offset[0] > 255U) {
1208 av_log(ac->avctx, AV_LOG_ERROR,
1209 "Scalefactor (%d) out of range.\n", offset[0]);
1210 return AVERROR_INVALIDDATA;
1212 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1221 * Decode pulse data; reference: table 4.7.
1223 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1224 const uint16_t *swb_offset, int num_swb)
1227 pulse->num_pulse = get_bits(gb, 2) + 1;
1228 pulse_swb = get_bits(gb, 6);
1229 if (pulse_swb >= num_swb)
1231 pulse->pos[0] = swb_offset[pulse_swb];
1232 pulse->pos[0] += get_bits(gb, 5);
1233 if (pulse->pos[0] > 1023)
1235 pulse->amp[0] = get_bits(gb, 4);
1236 for (i = 1; i < pulse->num_pulse; i++) {
1237 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1238 if (pulse->pos[i] > 1023)
1240 pulse->amp[i] = get_bits(gb, 4);
1246 * Decode Temporal Noise Shaping data; reference: table 4.48.
1248 * @return Returns error status. 0 - OK, !0 - error
1250 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1251 GetBitContext *gb, const IndividualChannelStream *ics)
1253 int w, filt, i, coef_len, coef_res, coef_compress;
1254 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1255 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1256 for (w = 0; w < ics->num_windows; w++) {
1257 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1258 coef_res = get_bits1(gb);
1260 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1262 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1264 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1265 av_log(ac->avctx, AV_LOG_ERROR,
1266 "TNS filter order %d is greater than maximum %d.\n",
1267 tns->order[w][filt], tns_max_order);
1268 tns->order[w][filt] = 0;
1269 return AVERROR_INVALIDDATA;
1271 if (tns->order[w][filt]) {
1272 tns->direction[w][filt] = get_bits1(gb);
1273 coef_compress = get_bits1(gb);
1274 coef_len = coef_res + 3 - coef_compress;
1275 tmp2_idx = 2 * coef_compress + coef_res;
1277 for (i = 0; i < tns->order[w][filt]; i++)
1278 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1287 * Decode Mid/Side data; reference: table 4.54.
1289 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1290 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1291 * [3] reserved for scalable AAC
1293 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1297 if (ms_present == 1) {
1299 idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb;
1301 cpe->ms_mask[idx] = get_bits1(gb);
1302 } else if (ms_present == 2) {
1303 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
1308 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1312 *dst++ = v[idx & 15] * s;
1313 *dst++ = v[idx>>4 & 15] * s;
1319 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1323 *dst++ = v[idx & 3] * s;
1324 *dst++ = v[idx>>2 & 3] * s;
1325 *dst++ = v[idx>>4 & 3] * s;
1326 *dst++ = v[idx>>6 & 3] * s;
1332 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1333 unsigned sign, const float *scale)
1335 union av_intfloat32 s0, s1;
1337 s0.f = s1.f = *scale;
1338 s0.i ^= sign >> 1 << 31;
1341 *dst++ = v[idx & 15] * s0.f;
1342 *dst++ = v[idx>>4 & 15] * s1.f;
1349 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1350 unsigned sign, const float *scale)
1352 unsigned nz = idx >> 12;
1353 union av_intfloat32 s = { .f = *scale };
1354 union av_intfloat32 t;
1356 t.i = s.i ^ (sign & 1U<<31);
1357 *dst++ = v[idx & 3] * t.f;
1359 sign <<= nz & 1; nz >>= 1;
1360 t.i = s.i ^ (sign & 1U<<31);
1361 *dst++ = v[idx>>2 & 3] * t.f;
1363 sign <<= nz & 1; nz >>= 1;
1364 t.i = s.i ^ (sign & 1U<<31);
1365 *dst++ = v[idx>>4 & 3] * t.f;
1368 t.i = s.i ^ (sign & 1U<<31);
1369 *dst++ = v[idx>>6 & 3] * t.f;
1376 * Decode spectral data; reference: table 4.50.
1377 * Dequantize and scale spectral data; reference: 4.6.3.3.
1379 * @param coef array of dequantized, scaled spectral data
1380 * @param sf array of scalefactors or intensity stereo positions
1381 * @param pulse_present set if pulses are present
1382 * @param pulse pointer to pulse data struct
1383 * @param band_type array of the used band type
1385 * @return Returns error status. 0 - OK, !0 - error
1387 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1388 GetBitContext *gb, const float sf[120],
1389 int pulse_present, const Pulse *pulse,
1390 const IndividualChannelStream *ics,
1391 enum BandType band_type[120])
1393 int i, k, g, idx = 0;
1394 const int c = 1024 / ics->num_windows;
1395 const uint16_t *offsets = ics->swb_offset;
1396 float *coef_base = coef;
1398 for (g = 0; g < ics->num_windows; g++)
1399 memset(coef + g * 128 + offsets[ics->max_sfb], 0,
1400 sizeof(float) * (c - offsets[ics->max_sfb]));
1402 for (g = 0; g < ics->num_window_groups; g++) {
1403 unsigned g_len = ics->group_len[g];
1405 for (i = 0; i < ics->max_sfb; i++, idx++) {
1406 const unsigned cbt_m1 = band_type[idx] - 1;
1407 float *cfo = coef + offsets[i];
1408 int off_len = offsets[i + 1] - offsets[i];
1411 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1412 for (group = 0; group < g_len; group++, cfo+=128) {
1413 memset(cfo, 0, off_len * sizeof(float));
1415 } else if (cbt_m1 == NOISE_BT - 1) {
1416 for (group = 0; group < g_len; group++, cfo+=128) {
1420 for (k = 0; k < off_len; k++) {
1421 ac->random_state = lcg_random(ac->random_state);
1422 cfo[k] = ac->random_state;
1425 band_energy = ac->fdsp.scalarproduct_float(cfo, cfo, off_len);
1426 scale = sf[idx] / sqrtf(band_energy);
1427 ac->fdsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1430 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1431 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1432 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1433 OPEN_READER(re, gb);
1435 switch (cbt_m1 >> 1) {
1437 for (group = 0; group < g_len; group++, cfo+=128) {
1445 UPDATE_CACHE(re, gb);
1446 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1447 cb_idx = cb_vector_idx[code];
1448 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1454 for (group = 0; group < g_len; group++, cfo+=128) {
1464 UPDATE_CACHE(re, gb);
1465 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1466 cb_idx = cb_vector_idx[code];
1467 nnz = cb_idx >> 8 & 15;
1468 bits = nnz ? GET_CACHE(re, gb) : 0;
1469 LAST_SKIP_BITS(re, gb, nnz);
1470 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1476 for (group = 0; group < g_len; group++, cfo+=128) {
1484 UPDATE_CACHE(re, gb);
1485 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1486 cb_idx = cb_vector_idx[code];
1487 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1494 for (group = 0; group < g_len; group++, cfo+=128) {
1504 UPDATE_CACHE(re, gb);
1505 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1506 cb_idx = cb_vector_idx[code];
1507 nnz = cb_idx >> 8 & 15;
1508 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1509 LAST_SKIP_BITS(re, gb, nnz);
1510 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1516 for (group = 0; group < g_len; group++, cfo+=128) {
1518 uint32_t *icf = (uint32_t *) cf;
1528 UPDATE_CACHE(re, gb);
1529 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1537 cb_idx = cb_vector_idx[code];
1540 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1541 LAST_SKIP_BITS(re, gb, nnz);
1543 for (j = 0; j < 2; j++) {
1547 /* The total length of escape_sequence must be < 22 bits according
1548 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1549 UPDATE_CACHE(re, gb);
1550 b = GET_CACHE(re, gb);
1551 b = 31 - av_log2(~b);
1554 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1555 return AVERROR_INVALIDDATA;
1558 SKIP_BITS(re, gb, b + 1);
1560 n = (1 << b) + SHOW_UBITS(re, gb, b);
1561 LAST_SKIP_BITS(re, gb, b);
1562 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1565 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1566 *icf++ = (bits & 1U<<31) | v;
1573 ac->fdsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1577 CLOSE_READER(re, gb);
1583 if (pulse_present) {
1585 for (i = 0; i < pulse->num_pulse; i++) {
1586 float co = coef_base[ pulse->pos[i] ];
1587 while (offsets[idx + 1] <= pulse->pos[i])
1589 if (band_type[idx] != NOISE_BT && sf[idx]) {
1590 float ico = -pulse->amp[i];
1593 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1595 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1602 static av_always_inline float flt16_round(float pf)
1604 union av_intfloat32 tmp;
1606 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1610 static av_always_inline float flt16_even(float pf)
1612 union av_intfloat32 tmp;
1614 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1618 static av_always_inline float flt16_trunc(float pf)
1620 union av_intfloat32 pun;
1622 pun.i &= 0xFFFF0000U;
1626 static av_always_inline void predict(PredictorState *ps, float *coef,
1629 const float a = 0.953125; // 61.0 / 64
1630 const float alpha = 0.90625; // 29.0 / 32
1634 float r0 = ps->r0, r1 = ps->r1;
1635 float cor0 = ps->cor0, cor1 = ps->cor1;
1636 float var0 = ps->var0, var1 = ps->var1;
1638 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1639 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1641 pv = flt16_round(k1 * r0 + k2 * r1);
1648 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1649 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1650 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1651 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1653 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1654 ps->r0 = flt16_trunc(a * e0);
1658 * Apply AAC-Main style frequency domain prediction.
1660 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1664 if (!sce->ics.predictor_initialized) {
1665 reset_all_predictors(sce->predictor_state);
1666 sce->ics.predictor_initialized = 1;
1669 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1671 sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index];
1673 for (k = sce->ics.swb_offset[sfb];
1674 k < sce->ics.swb_offset[sfb + 1];
1676 predict(&sce->predictor_state[k], &sce->coeffs[k],
1677 sce->ics.predictor_present &&
1678 sce->ics.prediction_used[sfb]);
1681 if (sce->ics.predictor_reset_group)
1682 reset_predictor_group(sce->predictor_state,
1683 sce->ics.predictor_reset_group);
1685 reset_all_predictors(sce->predictor_state);
1689 * Decode an individual_channel_stream payload; reference: table 4.44.
1691 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1692 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1694 * @return Returns error status. 0 - OK, !0 - error
1696 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1697 GetBitContext *gb, int common_window, int scale_flag)
1700 TemporalNoiseShaping *tns = &sce->tns;
1701 IndividualChannelStream *ics = &sce->ics;
1702 float *out = sce->coeffs;
1703 int global_gain, pulse_present = 0;
1706 /* This assignment is to silence a GCC warning about the variable being used
1707 * uninitialized when in fact it always is.
1709 pulse.num_pulse = 0;
1711 global_gain = get_bits(gb, 8);
1713 if (!common_window && !scale_flag) {
1714 if (decode_ics_info(ac, ics, gb) < 0)
1715 return AVERROR_INVALIDDATA;
1718 if ((ret = decode_band_types(ac, sce->band_type,
1719 sce->band_type_run_end, gb, ics)) < 0)
1721 if ((ret = decode_scalefactors(ac, sce->sf, gb, global_gain, ics,
1722 sce->band_type, sce->band_type_run_end)) < 0)
1727 if ((pulse_present = get_bits1(gb))) {
1728 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1729 av_log(ac->avctx, AV_LOG_ERROR,
1730 "Pulse tool not allowed in eight short sequence.\n");
1731 return AVERROR_INVALIDDATA;
1733 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1734 av_log(ac->avctx, AV_LOG_ERROR,
1735 "Pulse data corrupt or invalid.\n");
1736 return AVERROR_INVALIDDATA;
1739 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1740 return AVERROR_INVALIDDATA;
1741 if (get_bits1(gb)) {
1742 avpriv_request_sample(ac->avctx, "SSR");
1743 return AVERROR_PATCHWELCOME;
1747 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present,
1748 &pulse, ics, sce->band_type) < 0)
1749 return AVERROR_INVALIDDATA;
1751 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1752 apply_prediction(ac, sce);
1758 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1760 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1762 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1763 float *ch0 = cpe->ch[0].coeffs;
1764 float *ch1 = cpe->ch[1].coeffs;
1765 int g, i, group, idx = 0;
1766 const uint16_t *offsets = ics->swb_offset;
1767 for (g = 0; g < ics->num_window_groups; g++) {
1768 for (i = 0; i < ics->max_sfb; i++, idx++) {
1769 if (cpe->ms_mask[idx] &&
1770 cpe->ch[0].band_type[idx] < NOISE_BT &&
1771 cpe->ch[1].band_type[idx] < NOISE_BT) {
1772 for (group = 0; group < ics->group_len[g]; group++) {
1773 ac->fdsp.butterflies_float(ch0 + group * 128 + offsets[i],
1774 ch1 + group * 128 + offsets[i],
1775 offsets[i+1] - offsets[i]);
1779 ch0 += ics->group_len[g] * 128;
1780 ch1 += ics->group_len[g] * 128;
1785 * intensity stereo decoding; reference: 4.6.8.2.3
1787 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1788 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1789 * [3] reserved for scalable AAC
1791 static void apply_intensity_stereo(AACContext *ac,
1792 ChannelElement *cpe, int ms_present)
1794 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1795 SingleChannelElement *sce1 = &cpe->ch[1];
1796 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1797 const uint16_t *offsets = ics->swb_offset;
1798 int g, group, i, idx = 0;
1801 for (g = 0; g < ics->num_window_groups; g++) {
1802 for (i = 0; i < ics->max_sfb;) {
1803 if (sce1->band_type[idx] == INTENSITY_BT ||
1804 sce1->band_type[idx] == INTENSITY_BT2) {
1805 const int bt_run_end = sce1->band_type_run_end[idx];
1806 for (; i < bt_run_end; i++, idx++) {
1807 c = -1 + 2 * (sce1->band_type[idx] - 14);
1809 c *= 1 - 2 * cpe->ms_mask[idx];
1810 scale = c * sce1->sf[idx];
1811 for (group = 0; group < ics->group_len[g]; group++)
1812 ac->fdsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1813 coef0 + group * 128 + offsets[i],
1815 offsets[i + 1] - offsets[i]);
1818 int bt_run_end = sce1->band_type_run_end[idx];
1819 idx += bt_run_end - i;
1823 coef0 += ics->group_len[g] * 128;
1824 coef1 += ics->group_len[g] * 128;
1829 * Decode a channel_pair_element; reference: table 4.4.
1831 * @return Returns error status. 0 - OK, !0 - error
1833 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1835 int i, ret, common_window, ms_present = 0;
1837 common_window = get_bits1(gb);
1838 if (common_window) {
1839 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1840 return AVERROR_INVALIDDATA;
1841 i = cpe->ch[1].ics.use_kb_window[0];
1842 cpe->ch[1].ics = cpe->ch[0].ics;
1843 cpe->ch[1].ics.use_kb_window[1] = i;
1844 if (cpe->ch[1].ics.predictor_present &&
1845 (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
1846 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1847 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1848 ms_present = get_bits(gb, 2);
1849 if (ms_present == 3) {
1850 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1851 return AVERROR_INVALIDDATA;
1852 } else if (ms_present)
1853 decode_mid_side_stereo(cpe, gb, ms_present);
1855 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1857 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1860 if (common_window) {
1862 apply_mid_side_stereo(ac, cpe);
1863 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1864 apply_prediction(ac, &cpe->ch[0]);
1865 apply_prediction(ac, &cpe->ch[1]);
1869 apply_intensity_stereo(ac, cpe, ms_present);
1873 static const float cce_scale[] = {
1874 1.09050773266525765921, //2^(1/8)
1875 1.18920711500272106672, //2^(1/4)
1881 * Decode coupling_channel_element; reference: table 4.8.
1883 * @return Returns error status. 0 - OK, !0 - error
1885 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1891 SingleChannelElement *sce = &che->ch[0];
1892 ChannelCoupling *coup = &che->coup;
1894 coup->coupling_point = 2 * get_bits1(gb);
1895 coup->num_coupled = get_bits(gb, 3);
1896 for (c = 0; c <= coup->num_coupled; c++) {
1898 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1899 coup->id_select[c] = get_bits(gb, 4);
1900 if (coup->type[c] == TYPE_CPE) {
1901 coup->ch_select[c] = get_bits(gb, 2);
1902 if (coup->ch_select[c] == 3)
1905 coup->ch_select[c] = 2;
1907 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1909 sign = get_bits(gb, 1);
1910 scale = cce_scale[get_bits(gb, 2)];
1912 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1915 for (c = 0; c < num_gain; c++) {
1919 float gain_cache = 1.0;
1921 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1922 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1923 gain_cache = powf(scale, -gain);
1925 if (coup->coupling_point == AFTER_IMDCT) {
1926 coup->gain[c][0] = gain_cache;
1928 for (g = 0; g < sce->ics.num_window_groups; g++) {
1929 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1930 if (sce->band_type[idx] != ZERO_BT) {
1932 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1940 gain_cache = powf(scale, -t) * s;
1943 coup->gain[c][idx] = gain_cache;
1953 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1955 * @return Returns number of bytes consumed.
1957 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1961 int num_excl_chan = 0;
1964 for (i = 0; i < 7; i++)
1965 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1966 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1968 return num_excl_chan / 7;
1972 * Decode dynamic range information; reference: table 4.52.
1974 * @return Returns number of bytes consumed.
1976 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1980 int drc_num_bands = 1;
1983 /* pce_tag_present? */
1984 if (get_bits1(gb)) {
1985 che_drc->pce_instance_tag = get_bits(gb, 4);
1986 skip_bits(gb, 4); // tag_reserved_bits
1990 /* excluded_chns_present? */
1991 if (get_bits1(gb)) {
1992 n += decode_drc_channel_exclusions(che_drc, gb);
1995 /* drc_bands_present? */
1996 if (get_bits1(gb)) {
1997 che_drc->band_incr = get_bits(gb, 4);
1998 che_drc->interpolation_scheme = get_bits(gb, 4);
2000 drc_num_bands += che_drc->band_incr;
2001 for (i = 0; i < drc_num_bands; i++) {
2002 che_drc->band_top[i] = get_bits(gb, 8);
2007 /* prog_ref_level_present? */
2008 if (get_bits1(gb)) {
2009 che_drc->prog_ref_level = get_bits(gb, 7);
2010 skip_bits1(gb); // prog_ref_level_reserved_bits
2014 for (i = 0; i < drc_num_bands; i++) {
2015 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
2016 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
2024 * Decode extension data (incomplete); reference: table 4.51.
2026 * @param cnt length of TYPE_FIL syntactic element in bytes
2028 * @return Returns number of bytes consumed
2030 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2031 ChannelElement *che, enum RawDataBlockType elem_type)
2035 switch (get_bits(gb, 4)) { // extension type
2036 case EXT_SBR_DATA_CRC:
2040 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2042 } else if (!ac->oc[1].m4ac.sbr) {
2043 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2044 skip_bits_long(gb, 8 * cnt - 4);
2046 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2047 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2048 skip_bits_long(gb, 8 * cnt - 4);
2050 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2051 ac->oc[1].m4ac.sbr = 1;
2052 ac->oc[1].m4ac.ps = 1;
2053 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2054 ac->oc[1].status, 1);
2056 ac->oc[1].m4ac.sbr = 1;
2058 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2060 case EXT_DYNAMIC_RANGE:
2061 res = decode_dynamic_range(&ac->che_drc, gb);
2065 case EXT_DATA_ELEMENT:
2067 skip_bits_long(gb, 8 * cnt - 4);
2074 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2076 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2077 * @param coef spectral coefficients
2079 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2080 IndividualChannelStream *ics, int decode)
2082 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2084 int bottom, top, order, start, end, size, inc;
2085 float lpc[TNS_MAX_ORDER];
2086 float tmp[TNS_MAX_ORDER + 1];
2088 for (w = 0; w < ics->num_windows; w++) {
2089 bottom = ics->num_swb;
2090 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2092 bottom = FFMAX(0, top - tns->length[w][filt]);
2093 order = tns->order[w][filt];
2098 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2100 start = ics->swb_offset[FFMIN(bottom, mmm)];
2101 end = ics->swb_offset[FFMIN( top, mmm)];
2102 if ((size = end - start) <= 0)
2104 if (tns->direction[w][filt]) {
2114 for (m = 0; m < size; m++, start += inc)
2115 for (i = 1; i <= FFMIN(m, order); i++)
2116 coef[start] -= coef[start - i * inc] * lpc[i - 1];
2119 for (m = 0; m < size; m++, start += inc) {
2120 tmp[0] = coef[start];
2121 for (i = 1; i <= FFMIN(m, order); i++)
2122 coef[start] += tmp[i] * lpc[i - 1];
2123 for (i = order; i > 0; i--)
2124 tmp[i] = tmp[i - 1];
2132 * Apply windowing and MDCT to obtain the spectral
2133 * coefficient from the predicted sample by LTP.
2135 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2136 float *in, IndividualChannelStream *ics)
2138 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2139 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2140 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2141 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2143 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2144 ac->fdsp.vector_fmul(in, in, lwindow_prev, 1024);
2146 memset(in, 0, 448 * sizeof(float));
2147 ac->fdsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2149 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2150 ac->fdsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2152 ac->fdsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2153 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2155 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2159 * Apply the long term prediction
2161 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2163 const LongTermPrediction *ltp = &sce->ics.ltp;
2164 const uint16_t *offsets = sce->ics.swb_offset;
2167 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2168 float *predTime = sce->ret;
2169 float *predFreq = ac->buf_mdct;
2170 int16_t num_samples = 2048;
2172 if (ltp->lag < 1024)
2173 num_samples = ltp->lag + 1024;
2174 for (i = 0; i < num_samples; i++)
2175 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2176 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2178 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2180 if (sce->tns.present)
2181 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2183 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2185 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2186 sce->coeffs[i] += predFreq[i];
2191 * Update the LTP buffer for next frame
2193 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2195 IndividualChannelStream *ics = &sce->ics;
2196 float *saved = sce->saved;
2197 float *saved_ltp = sce->coeffs;
2198 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2199 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2202 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2203 memcpy(saved_ltp, saved, 512 * sizeof(float));
2204 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2205 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2206 for (i = 0; i < 64; i++)
2207 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2208 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2209 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2210 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2211 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2212 for (i = 0; i < 64; i++)
2213 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2214 } else { // LONG_STOP or ONLY_LONG
2215 ac->fdsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2216 for (i = 0; i < 512; i++)
2217 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2220 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2221 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2222 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2226 * Conduct IMDCT and windowing.
2228 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2230 IndividualChannelStream *ics = &sce->ics;
2231 float *in = sce->coeffs;
2232 float *out = sce->ret;
2233 float *saved = sce->saved;
2234 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2235 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2236 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2237 float *buf = ac->buf_mdct;
2238 float *temp = ac->temp;
2242 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2243 for (i = 0; i < 1024; i += 128)
2244 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2246 ac->mdct.imdct_half(&ac->mdct, buf, in);
2248 /* window overlapping
2249 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2250 * and long to short transitions are considered to be short to short
2251 * transitions. This leaves just two cases (long to long and short to short)
2252 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2254 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2255 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2256 ac->fdsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2258 memcpy( out, saved, 448 * sizeof(float));
2260 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2261 ac->fdsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2262 ac->fdsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2263 ac->fdsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2264 ac->fdsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2265 ac->fdsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2266 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2268 ac->fdsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2269 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2274 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2275 memcpy( saved, temp + 64, 64 * sizeof(float));
2276 ac->fdsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2277 ac->fdsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2278 ac->fdsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2279 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2280 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2281 memcpy( saved, buf + 512, 448 * sizeof(float));
2282 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2283 } else { // LONG_STOP or ONLY_LONG
2284 memcpy( saved, buf + 512, 512 * sizeof(float));
2289 * Apply dependent channel coupling (applied before IMDCT).
2291 * @param index index into coupling gain array
2293 static void apply_dependent_coupling(AACContext *ac,
2294 SingleChannelElement *target,
2295 ChannelElement *cce, int index)
2297 IndividualChannelStream *ics = &cce->ch[0].ics;
2298 const uint16_t *offsets = ics->swb_offset;
2299 float *dest = target->coeffs;
2300 const float *src = cce->ch[0].coeffs;
2301 int g, i, group, k, idx = 0;
2302 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2303 av_log(ac->avctx, AV_LOG_ERROR,
2304 "Dependent coupling is not supported together with LTP\n");
2307 for (g = 0; g < ics->num_window_groups; g++) {
2308 for (i = 0; i < ics->max_sfb; i++, idx++) {
2309 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2310 const float gain = cce->coup.gain[index][idx];
2311 for (group = 0; group < ics->group_len[g]; group++) {
2312 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2314 dest[group * 128 + k] += gain * src[group * 128 + k];
2319 dest += ics->group_len[g] * 128;
2320 src += ics->group_len[g] * 128;
2325 * Apply independent channel coupling (applied after IMDCT).
2327 * @param index index into coupling gain array
2329 static void apply_independent_coupling(AACContext *ac,
2330 SingleChannelElement *target,
2331 ChannelElement *cce, int index)
2334 const float gain = cce->coup.gain[index][0];
2335 const float *src = cce->ch[0].ret;
2336 float *dest = target->ret;
2337 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2339 for (i = 0; i < len; i++)
2340 dest[i] += gain * src[i];
2344 * channel coupling transformation interface
2346 * @param apply_coupling_method pointer to (in)dependent coupling function
2348 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2349 enum RawDataBlockType type, int elem_id,
2350 enum CouplingPoint coupling_point,
2351 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2355 for (i = 0; i < MAX_ELEM_ID; i++) {
2356 ChannelElement *cce = ac->che[TYPE_CCE][i];
2359 if (cce && cce->coup.coupling_point == coupling_point) {
2360 ChannelCoupling *coup = &cce->coup;
2362 for (c = 0; c <= coup->num_coupled; c++) {
2363 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2364 if (coup->ch_select[c] != 1) {
2365 apply_coupling_method(ac, &cc->ch[0], cce, index);
2366 if (coup->ch_select[c] != 0)
2369 if (coup->ch_select[c] != 2)
2370 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2372 index += 1 + (coup->ch_select[c] == 3);
2379 * Convert spectral data to float samples, applying all supported tools as appropriate.
2381 static void spectral_to_sample(AACContext *ac)
2384 for (type = 3; type >= 0; type--) {
2385 for (i = 0; i < MAX_ELEM_ID; i++) {
2386 ChannelElement *che = ac->che[type][i];
2388 if (type <= TYPE_CPE)
2389 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2390 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2391 if (che->ch[0].ics.predictor_present) {
2392 if (che->ch[0].ics.ltp.present)
2393 apply_ltp(ac, &che->ch[0]);
2394 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2395 apply_ltp(ac, &che->ch[1]);
2398 if (che->ch[0].tns.present)
2399 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2400 if (che->ch[1].tns.present)
2401 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2402 if (type <= TYPE_CPE)
2403 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2404 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2405 imdct_and_windowing(ac, &che->ch[0]);
2406 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2407 update_ltp(ac, &che->ch[0]);
2408 if (type == TYPE_CPE) {
2409 imdct_and_windowing(ac, &che->ch[1]);
2410 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2411 update_ltp(ac, &che->ch[1]);
2413 if (ac->oc[1].m4ac.sbr > 0) {
2414 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2417 if (type <= TYPE_CCE)
2418 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2424 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2427 AACADTSHeaderInfo hdr_info;
2428 uint8_t layout_map[MAX_ELEM_ID*4][3];
2429 int layout_map_tags;
2431 size = avpriv_aac_parse_header(gb, &hdr_info);
2433 if (hdr_info.num_aac_frames != 1) {
2434 avpriv_report_missing_feature(ac->avctx,
2435 "More than one AAC RDB per ADTS frame");
2436 return AVERROR_PATCHWELCOME;
2438 push_output_configuration(ac);
2439 if (hdr_info.chan_config) {
2440 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2441 if (set_default_channel_config(ac->avctx, layout_map,
2442 &layout_map_tags, hdr_info.chan_config))
2444 if (output_configure(ac, layout_map, layout_map_tags,
2445 FFMAX(ac->oc[1].status, OC_TRIAL_FRAME), 0))
2448 ac->oc[1].m4ac.chan_config = 0;
2450 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2451 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2452 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2453 if (ac->oc[0].status != OC_LOCKED ||
2454 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2455 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2456 ac->oc[1].m4ac.sbr = -1;
2457 ac->oc[1].m4ac.ps = -1;
2459 if (!hdr_info.crc_absent)
2465 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2466 int *got_frame_ptr, GetBitContext *gb)
2468 AACContext *ac = avctx->priv_data;
2469 ChannelElement *che = NULL, *che_prev = NULL;
2470 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2472 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2476 if (show_bits(gb, 12) == 0xfff) {
2477 if (parse_adts_frame_header(ac, gb) < 0) {
2478 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2482 if (ac->oc[1].m4ac.sampling_index > 12) {
2483 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2489 if (frame_configure_elements(avctx) < 0) {
2494 ac->tags_mapped = 0;
2496 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2497 elem_id = get_bits(gb, 4);
2499 if (elem_type < TYPE_DSE) {
2500 if (!(che=get_che(ac, elem_type, elem_id))) {
2501 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2502 elem_type, elem_id);
2509 switch (elem_type) {
2512 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2517 err = decode_cpe(ac, gb, che);
2522 err = decode_cce(ac, gb, che);
2526 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2531 err = skip_data_stream_element(ac, gb);
2535 uint8_t layout_map[MAX_ELEM_ID*4][3];
2537 push_output_configuration(ac);
2538 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
2544 av_log(avctx, AV_LOG_ERROR,
2545 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2546 pop_output_configuration(ac);
2548 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
2556 elem_id += get_bits(gb, 8) - 1;
2557 if (get_bits_left(gb) < 8 * elem_id) {
2558 av_log(avctx, AV_LOG_ERROR, overread_err);
2563 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2564 err = 0; /* FIXME */
2568 err = -1; /* should not happen, but keeps compiler happy */
2573 elem_type_prev = elem_type;
2578 if (get_bits_left(gb) < 3) {
2579 av_log(avctx, AV_LOG_ERROR, overread_err);
2585 spectral_to_sample(ac);
2587 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
2588 samples <<= multiplier;
2591 ac->frame->nb_samples = samples;
2592 *got_frame_ptr = !!samples;
2594 if (ac->oc[1].status && audio_found) {
2595 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
2596 avctx->frame_size = samples;
2597 ac->oc[1].status = OC_LOCKED;
2602 pop_output_configuration(ac);
2606 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2607 int *got_frame_ptr, AVPacket *avpkt)
2609 AACContext *ac = avctx->priv_data;
2610 const uint8_t *buf = avpkt->data;
2611 int buf_size = avpkt->size;
2616 int new_extradata_size;
2617 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2618 AV_PKT_DATA_NEW_EXTRADATA,
2619 &new_extradata_size);
2621 if (new_extradata) {
2622 av_free(avctx->extradata);
2623 avctx->extradata = av_mallocz(new_extradata_size +
2624 FF_INPUT_BUFFER_PADDING_SIZE);
2625 if (!avctx->extradata)
2626 return AVERROR(ENOMEM);
2627 avctx->extradata_size = new_extradata_size;
2628 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2629 push_output_configuration(ac);
2630 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
2632 avctx->extradata_size*8, 1) < 0) {
2633 pop_output_configuration(ac);
2634 return AVERROR_INVALIDDATA;
2638 init_get_bits(&gb, buf, buf_size * 8);
2640 if ((err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb)) < 0)
2643 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2644 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2645 if (buf[buf_offset])
2648 return buf_size > buf_offset ? buf_consumed : buf_size;
2651 static av_cold int aac_decode_close(AVCodecContext *avctx)
2653 AACContext *ac = avctx->priv_data;
2656 for (i = 0; i < MAX_ELEM_ID; i++) {
2657 for (type = 0; type < 4; type++) {
2658 if (ac->che[type][i])
2659 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2660 av_freep(&ac->che[type][i]);
2664 ff_mdct_end(&ac->mdct);
2665 ff_mdct_end(&ac->mdct_small);
2666 ff_mdct_end(&ac->mdct_ltp);
2671 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2673 struct LATMContext {
2674 AACContext aac_ctx; ///< containing AACContext
2675 int initialized; ///< initilized after a valid extradata was seen
2678 int audio_mux_version_A; ///< LATM syntax version
2679 int frame_length_type; ///< 0/1 variable/fixed frame length
2680 int frame_length; ///< frame length for fixed frame length
2683 static inline uint32_t latm_get_value(GetBitContext *b)
2685 int length = get_bits(b, 2);
2687 return get_bits_long(b, (length+1)*8);
2690 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2691 GetBitContext *gb, int asclen)
2693 AACContext *ac = &latmctx->aac_ctx;
2694 AVCodecContext *avctx = ac->avctx;
2695 MPEG4AudioConfig m4ac = { 0 };
2696 int config_start_bit = get_bits_count(gb);
2697 int sync_extension = 0;
2698 int bits_consumed, esize;
2702 asclen = FFMIN(asclen, get_bits_left(gb));
2704 asclen = get_bits_left(gb);
2706 if (config_start_bit % 8) {
2707 avpriv_request_sample(latmctx->aac_ctx.avctx,
2708 "Non-byte-aligned audio-specific config");
2709 return AVERROR_PATCHWELCOME;
2712 return AVERROR_INVALIDDATA;
2713 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
2714 gb->buffer + (config_start_bit / 8),
2715 asclen, sync_extension);
2717 if (bits_consumed < 0)
2718 return AVERROR_INVALIDDATA;
2720 if (ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
2721 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
2723 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
2724 latmctx->initialized = 0;
2726 esize = (bits_consumed+7) / 8;
2728 if (avctx->extradata_size < esize) {
2729 av_free(avctx->extradata);
2730 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2731 if (!avctx->extradata)
2732 return AVERROR(ENOMEM);
2735 avctx->extradata_size = esize;
2736 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2737 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2739 skip_bits_long(gb, bits_consumed);
2741 return bits_consumed;
2744 static int read_stream_mux_config(struct LATMContext *latmctx,
2747 int ret, audio_mux_version = get_bits(gb, 1);
2749 latmctx->audio_mux_version_A = 0;
2750 if (audio_mux_version)
2751 latmctx->audio_mux_version_A = get_bits(gb, 1);
2753 if (!latmctx->audio_mux_version_A) {
2755 if (audio_mux_version)
2756 latm_get_value(gb); // taraFullness
2758 skip_bits(gb, 1); // allStreamSameTimeFraming
2759 skip_bits(gb, 6); // numSubFrames
2761 if (get_bits(gb, 4)) { // numPrograms
2762 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple programs");
2763 return AVERROR_PATCHWELCOME;
2766 // for each program (which there is only on in DVB)
2768 // for each layer (which there is only on in DVB)
2769 if (get_bits(gb, 3)) { // numLayer
2770 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple layers");
2771 return AVERROR_PATCHWELCOME;
2774 // for all but first stream: use_same_config = get_bits(gb, 1);
2775 if (!audio_mux_version) {
2776 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
2779 int ascLen = latm_get_value(gb);
2780 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
2783 skip_bits_long(gb, ascLen);
2786 latmctx->frame_length_type = get_bits(gb, 3);
2787 switch (latmctx->frame_length_type) {
2789 skip_bits(gb, 8); // latmBufferFullness
2792 latmctx->frame_length = get_bits(gb, 9);
2797 skip_bits(gb, 6); // CELP frame length table index
2801 skip_bits(gb, 1); // HVXC frame length table index
2805 if (get_bits(gb, 1)) { // other data
2806 if (audio_mux_version) {
2807 latm_get_value(gb); // other_data_bits
2811 esc = get_bits(gb, 1);
2817 if (get_bits(gb, 1)) // crc present
2818 skip_bits(gb, 8); // config_crc
2824 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2828 if (ctx->frame_length_type == 0) {
2829 int mux_slot_length = 0;
2831 tmp = get_bits(gb, 8);
2832 mux_slot_length += tmp;
2833 } while (tmp == 255);
2834 return mux_slot_length;
2835 } else if (ctx->frame_length_type == 1) {
2836 return ctx->frame_length;
2837 } else if (ctx->frame_length_type == 3 ||
2838 ctx->frame_length_type == 5 ||
2839 ctx->frame_length_type == 7) {
2840 skip_bits(gb, 2); // mux_slot_length_coded
2845 static int read_audio_mux_element(struct LATMContext *latmctx,
2849 uint8_t use_same_mux = get_bits(gb, 1);
2850 if (!use_same_mux) {
2851 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2853 } else if (!latmctx->aac_ctx.avctx->extradata) {
2854 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2855 "no decoder config found\n");
2856 return AVERROR(EAGAIN);
2858 if (latmctx->audio_mux_version_A == 0) {
2859 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2860 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2861 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2862 return AVERROR_INVALIDDATA;
2863 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2864 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2865 "frame length mismatch %d << %d\n",
2866 mux_slot_length_bytes * 8, get_bits_left(gb));
2867 return AVERROR_INVALIDDATA;
2874 static int latm_decode_frame(AVCodecContext *avctx, void *out,
2875 int *got_frame_ptr, AVPacket *avpkt)
2877 struct LATMContext *latmctx = avctx->priv_data;
2881 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2883 // check for LOAS sync word
2884 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2885 return AVERROR_INVALIDDATA;
2887 muxlength = get_bits(&gb, 13) + 3;
2888 // not enough data, the parser should have sorted this
2889 if (muxlength > avpkt->size)
2890 return AVERROR_INVALIDDATA;
2892 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2895 if (!latmctx->initialized) {
2896 if (!avctx->extradata) {
2900 push_output_configuration(&latmctx->aac_ctx);
2901 if ((err = decode_audio_specific_config(
2902 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
2903 avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
2904 pop_output_configuration(&latmctx->aac_ctx);
2907 latmctx->initialized = 1;
2911 if (show_bits(&gb, 12) == 0xfff) {
2912 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2913 "ADTS header detected, probably as result of configuration "
2915 return AVERROR_INVALIDDATA;
2918 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb)) < 0)
2924 static av_cold int latm_decode_init(AVCodecContext *avctx)
2926 struct LATMContext *latmctx = avctx->priv_data;
2927 int ret = aac_decode_init(avctx);
2929 if (avctx->extradata_size > 0)
2930 latmctx->initialized = !ret;
2936 AVCodec ff_aac_decoder = {
2938 .type = AVMEDIA_TYPE_AUDIO,
2939 .id = AV_CODEC_ID_AAC,
2940 .priv_data_size = sizeof(AACContext),
2941 .init = aac_decode_init,
2942 .close = aac_decode_close,
2943 .decode = aac_decode_frame,
2944 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
2945 .sample_fmts = (const enum AVSampleFormat[]) {
2946 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
2948 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2949 .channel_layouts = aac_channel_layout,
2953 Note: This decoder filter is intended to decode LATM streams transferred
2954 in MPEG transport streams which only contain one program.
2955 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2957 AVCodec ff_aac_latm_decoder = {
2959 .type = AVMEDIA_TYPE_AUDIO,
2960 .id = AV_CODEC_ID_AAC_LATM,
2961 .priv_data_size = sizeof(struct LATMContext),
2962 .init = latm_decode_init,
2963 .close = aac_decode_close,
2964 .decode = latm_decode_frame,
2965 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
2966 .sample_fmts = (const enum AVSampleFormat[]) {
2967 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
2969 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2970 .channel_layouts = aac_channel_layout,