3 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
7 * Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
8 * Copyright (c) 2010 Janne Grunau <janne-libav@jannau.net>
10 * This file is part of FFmpeg.
12 * FFmpeg is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public
14 * License as published by the Free Software Foundation; either
15 * version 2.1 of the License, or (at your option) any later version.
17 * FFmpeg is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with FFmpeg; if not, write to the Free Software
24 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
30 * @author Oded Shimon ( ods15 ods15 dyndns org )
31 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
38 * N (code in SoC repo) gain control
40 * Y window shapes - standard
41 * N window shapes - Low Delay
42 * Y filterbank - standard
43 * N (code in SoC repo) filterbank - Scalable Sample Rate
44 * Y Temporal Noise Shaping
45 * Y Long Term Prediction
48 * Y frequency domain prediction
49 * Y Perceptual Noise Substitution
51 * N Scalable Inverse AAC Quantization
52 * N Frequency Selective Switch
54 * Y quantization & coding - AAC
55 * N quantization & coding - TwinVQ
56 * N quantization & coding - BSAC
57 * N AAC Error Resilience tools
58 * N Error Resilience payload syntax
59 * N Error Protection tool
61 * N Silence Compression
64 * N Structured Audio tools
65 * N Structured Audio Sample Bank Format
67 * N Harmonic and Individual Lines plus Noise
68 * N Text-To-Speech Interface
69 * Y Spectral Band Replication
70 * Y (not in this code) Layer-1
71 * Y (not in this code) Layer-2
72 * Y (not in this code) Layer-3
73 * N SinuSoidal Coding (Transient, Sinusoid, Noise)
75 * N Direct Stream Transfer
77 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
78 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
82 #include "libavutil/float_dsp.h"
83 #include "libavutil/opt.h"
89 #include "fmtconvert.h"
96 #include "aacdectab.h"
97 #include "cbrt_tablegen.h"
100 #include "mpeg4audio.h"
101 #include "aacadtsdec.h"
102 #include "libavutil/intfloat.h"
110 # include "arm/aac.h"
113 static VLC vlc_scalefactors;
114 static VLC vlc_spectral[11];
116 #define overread_err "Input buffer exhausted before END element found\n"
118 static int count_channels(uint8_t (*layout)[3], int tags)
121 for (i = 0; i < tags; i++) {
122 int syn_ele = layout[i][0];
123 int pos = layout[i][2];
124 sum += (1 + (syn_ele == TYPE_CPE)) *
125 (pos != AAC_CHANNEL_OFF && pos != AAC_CHANNEL_CC);
131 * Check for the channel element in the current channel position configuration.
132 * If it exists, make sure the appropriate element is allocated and map the
133 * channel order to match the internal FFmpeg channel layout.
135 * @param che_pos current channel position configuration
136 * @param type channel element type
137 * @param id channel element id
138 * @param channels count of the number of channels in the configuration
140 * @return Returns error status. 0 - OK, !0 - error
142 static av_cold int che_configure(AACContext *ac,
143 enum ChannelPosition che_pos,
144 int type, int id, int *channels)
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 if (*channels >= MAX_CHANNELS - (type == TYPE_CPE || (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1))) {
154 av_log(ac->avctx, AV_LOG_ERROR, "Too many channels\n");
155 return AVERROR_INVALIDDATA;
157 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[0];
158 if (type == TYPE_CPE ||
159 (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1)) {
160 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[1];
164 if (ac->che[type][id])
165 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
166 av_freep(&ac->che[type][id]);
171 static int frame_configure_elements(AVCodecContext *avctx)
173 AACContext *ac = avctx->priv_data;
174 int type, id, ch, ret;
176 /* set channel pointers to internal buffers by default */
177 for (type = 0; type < 4; type++) {
178 for (id = 0; id < MAX_ELEM_ID; id++) {
179 ChannelElement *che = ac->che[type][id];
181 che->ch[0].ret = che->ch[0].ret_buf;
182 che->ch[1].ret = che->ch[1].ret_buf;
187 /* get output buffer */
188 ac->frame.nb_samples = 2048;
189 if ((ret = ff_get_buffer(avctx, &ac->frame)) < 0) {
190 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
194 /* map output channel pointers to AVFrame data */
195 for (ch = 0; ch < avctx->channels; ch++) {
196 if (ac->output_element[ch])
197 ac->output_element[ch]->ret = (float *)ac->frame.extended_data[ch];
203 struct elem_to_channel {
204 uint64_t av_position;
207 uint8_t aac_position;
210 static int assign_pair(struct elem_to_channel e2c_vec[MAX_ELEM_ID],
211 uint8_t (*layout_map)[3], int offset, uint64_t left,
212 uint64_t right, int pos)
214 if (layout_map[offset][0] == TYPE_CPE) {
215 e2c_vec[offset] = (struct elem_to_channel) {
216 .av_position = left | right, .syn_ele = TYPE_CPE,
217 .elem_id = layout_map[offset ][1], .aac_position = pos };
220 e2c_vec[offset] = (struct elem_to_channel) {
221 .av_position = left, .syn_ele = TYPE_SCE,
222 .elem_id = layout_map[offset ][1], .aac_position = pos };
223 e2c_vec[offset + 1] = (struct elem_to_channel) {
224 .av_position = right, .syn_ele = TYPE_SCE,
225 .elem_id = layout_map[offset + 1][1], .aac_position = pos };
230 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos, int *current) {
231 int num_pos_channels = 0;
235 for (i = *current; i < tags; i++) {
236 if (layout_map[i][2] != pos)
238 if (layout_map[i][0] == TYPE_CPE) {
240 if (pos == AAC_CHANNEL_FRONT && !first_cpe) {
246 num_pos_channels += 2;
254 ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
257 return num_pos_channels;
260 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
262 int i, n, total_non_cc_elements;
263 struct elem_to_channel e2c_vec[4*MAX_ELEM_ID] = {{ 0 }};
264 int num_front_channels, num_side_channels, num_back_channels;
267 if (FF_ARRAY_ELEMS(e2c_vec) < tags)
272 count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
273 if (num_front_channels < 0)
276 count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
277 if (num_side_channels < 0)
280 count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
281 if (num_back_channels < 0)
285 if (num_front_channels & 1) {
286 e2c_vec[i] = (struct elem_to_channel) {
287 .av_position = AV_CH_FRONT_CENTER, .syn_ele = TYPE_SCE,
288 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_FRONT };
290 num_front_channels--;
292 if (num_front_channels >= 4) {
293 i += assign_pair(e2c_vec, layout_map, i,
294 AV_CH_FRONT_LEFT_OF_CENTER,
295 AV_CH_FRONT_RIGHT_OF_CENTER,
297 num_front_channels -= 2;
299 if (num_front_channels >= 2) {
300 i += assign_pair(e2c_vec, layout_map, i,
304 num_front_channels -= 2;
306 while (num_front_channels >= 2) {
307 i += assign_pair(e2c_vec, layout_map, i,
311 num_front_channels -= 2;
314 if (num_side_channels >= 2) {
315 i += assign_pair(e2c_vec, layout_map, i,
319 num_side_channels -= 2;
321 while (num_side_channels >= 2) {
322 i += assign_pair(e2c_vec, layout_map, i,
326 num_side_channels -= 2;
329 while (num_back_channels >= 4) {
330 i += assign_pair(e2c_vec, layout_map, i,
334 num_back_channels -= 2;
336 if (num_back_channels >= 2) {
337 i += assign_pair(e2c_vec, layout_map, i,
341 num_back_channels -= 2;
343 if (num_back_channels) {
344 e2c_vec[i] = (struct elem_to_channel) {
345 .av_position = AV_CH_BACK_CENTER, .syn_ele = TYPE_SCE,
346 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_BACK };
351 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
352 e2c_vec[i] = (struct elem_to_channel) {
353 .av_position = AV_CH_LOW_FREQUENCY, .syn_ele = TYPE_LFE,
354 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_LFE };
357 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
358 e2c_vec[i] = (struct elem_to_channel) {
359 .av_position = UINT64_MAX, .syn_ele = TYPE_LFE,
360 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_LFE };
364 // Must choose a stable sort
365 total_non_cc_elements = n = i;
368 for (i = 1; i < n; i++) {
369 if (e2c_vec[i-1].av_position > e2c_vec[i].av_position) {
370 FFSWAP(struct elem_to_channel, e2c_vec[i-1], e2c_vec[i]);
378 for (i = 0; i < total_non_cc_elements; i++) {
379 layout_map[i][0] = e2c_vec[i].syn_ele;
380 layout_map[i][1] = e2c_vec[i].elem_id;
381 layout_map[i][2] = e2c_vec[i].aac_position;
382 if (e2c_vec[i].av_position != UINT64_MAX) {
383 layout |= e2c_vec[i].av_position;
391 * Save current output configuration if and only if it has been locked.
393 static void push_output_configuration(AACContext *ac) {
394 if (ac->oc[1].status == OC_LOCKED) {
395 ac->oc[0] = ac->oc[1];
397 ac->oc[1].status = OC_NONE;
401 * Restore the previous output configuration if and only if the current
402 * configuration is unlocked.
404 static void pop_output_configuration(AACContext *ac) {
405 if (ac->oc[1].status != OC_LOCKED && ac->oc[0].status != OC_NONE) {
406 ac->oc[1] = ac->oc[0];
407 ac->avctx->channels = ac->oc[1].channels;
408 ac->avctx->channel_layout = ac->oc[1].channel_layout;
413 * Configure output channel order based on the current program configuration element.
415 * @return Returns error status. 0 - OK, !0 - error
417 static int output_configure(AACContext *ac,
418 uint8_t layout_map[MAX_ELEM_ID*4][3], int tags,
419 enum OCStatus oc_type, int get_new_frame)
421 AVCodecContext *avctx = ac->avctx;
422 int i, channels = 0, ret;
425 if (ac->oc[1].layout_map != layout_map) {
426 memcpy(ac->oc[1].layout_map, layout_map, tags * sizeof(layout_map[0]));
427 ac->oc[1].layout_map_tags = tags;
430 // Try to sniff a reasonable channel order, otherwise output the
431 // channels in the order the PCE declared them.
432 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
433 layout = sniff_channel_order(layout_map, tags);
434 for (i = 0; i < tags; i++) {
435 int type = layout_map[i][0];
436 int id = layout_map[i][1];
437 int position = layout_map[i][2];
438 // Allocate or free elements depending on if they are in the
439 // current program configuration.
440 ret = che_configure(ac, position, type, id, &channels);
444 if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
445 if (layout == AV_CH_FRONT_CENTER) {
446 layout = AV_CH_FRONT_LEFT|AV_CH_FRONT_RIGHT;
452 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
453 if (layout) avctx->channel_layout = layout;
454 ac->oc[1].channel_layout = layout;
455 avctx->channels = ac->oc[1].channels = channels;
456 ac->oc[1].status = oc_type;
459 if ((ret = frame_configure_elements(ac->avctx)) < 0)
466 static void flush(AVCodecContext *avctx)
468 AACContext *ac= avctx->priv_data;
471 for (type = 3; type >= 0; type--) {
472 for (i = 0; i < MAX_ELEM_ID; i++) {
473 ChannelElement *che = ac->che[type][i];
475 for (j = 0; j <= 1; j++) {
476 memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
484 * Set up channel positions based on a default channel configuration
485 * as specified in table 1.17.
487 * @return Returns error status. 0 - OK, !0 - error
489 static int set_default_channel_config(AVCodecContext *avctx,
490 uint8_t (*layout_map)[3],
494 if (channel_config < 1 || channel_config > 7) {
495 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
499 *tags = tags_per_config[channel_config];
500 memcpy(layout_map, aac_channel_layout_map[channel_config-1], *tags * sizeof(*layout_map));
504 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
506 // For PCE based channel configurations map the channels solely based on tags.
507 if (!ac->oc[1].m4ac.chan_config) {
508 return ac->tag_che_map[type][elem_id];
510 // Allow single CPE stereo files to be signalled with mono configuration.
511 if (!ac->tags_mapped && type == TYPE_CPE && ac->oc[1].m4ac.chan_config == 1) {
512 uint8_t layout_map[MAX_ELEM_ID*4][3];
514 push_output_configuration(ac);
516 av_log(ac->avctx, AV_LOG_DEBUG, "mono with CPE\n");
518 if (set_default_channel_config(ac->avctx, layout_map, &layout_map_tags,
521 if (output_configure(ac, layout_map, layout_map_tags,
522 OC_TRIAL_FRAME, 1) < 0)
525 ac->oc[1].m4ac.chan_config = 2;
526 ac->oc[1].m4ac.ps = 0;
529 if (!ac->tags_mapped && type == TYPE_SCE && ac->oc[1].m4ac.chan_config == 2) {
530 uint8_t layout_map[MAX_ELEM_ID*4][3];
532 push_output_configuration(ac);
534 av_log(ac->avctx, AV_LOG_DEBUG, "stereo with SCE\n");
536 if (set_default_channel_config(ac->avctx, layout_map, &layout_map_tags,
539 if (output_configure(ac, layout_map, layout_map_tags,
540 OC_TRIAL_FRAME, 1) < 0)
543 ac->oc[1].m4ac.chan_config = 1;
544 if (ac->oc[1].m4ac.sbr)
545 ac->oc[1].m4ac.ps = -1;
547 // For indexed channel configurations map the channels solely based on position.
548 switch (ac->oc[1].m4ac.chan_config) {
550 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
552 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
555 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
556 instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
557 encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
558 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
560 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
563 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
565 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
568 if (ac->tags_mapped == 2 && ac->oc[1].m4ac.chan_config == 4 && type == TYPE_SCE) {
570 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
574 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) && type == TYPE_CPE) {
576 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
577 } else if (ac->oc[1].m4ac.chan_config == 2) {
581 if (!ac->tags_mapped && type == TYPE_SCE) {
583 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
591 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
593 * @param type speaker type/position for these channels
595 static void decode_channel_map(uint8_t layout_map[][3],
596 enum ChannelPosition type,
597 GetBitContext *gb, int n)
600 enum RawDataBlockType syn_ele;
602 case AAC_CHANNEL_FRONT:
603 case AAC_CHANNEL_BACK:
604 case AAC_CHANNEL_SIDE:
605 syn_ele = get_bits1(gb);
611 case AAC_CHANNEL_LFE:
617 layout_map[0][0] = syn_ele;
618 layout_map[0][1] = get_bits(gb, 4);
619 layout_map[0][2] = type;
625 * Decode program configuration element; reference: table 4.2.
627 * @return Returns error status. 0 - OK, !0 - error
629 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
630 uint8_t (*layout_map)[3],
633 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
637 skip_bits(gb, 2); // object_type
639 sampling_index = get_bits(gb, 4);
640 if (m4ac->sampling_index != sampling_index)
641 av_log(avctx, AV_LOG_WARNING, "Sample rate index in program config element does not match the sample rate index configured by the container.\n");
643 num_front = get_bits(gb, 4);
644 num_side = get_bits(gb, 4);
645 num_back = get_bits(gb, 4);
646 num_lfe = get_bits(gb, 2);
647 num_assoc_data = get_bits(gb, 3);
648 num_cc = get_bits(gb, 4);
651 skip_bits(gb, 4); // mono_mixdown_tag
653 skip_bits(gb, 4); // stereo_mixdown_tag
656 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
658 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
659 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
662 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
664 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
666 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
668 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
671 skip_bits_long(gb, 4 * num_assoc_data);
673 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
678 /* comment field, first byte is length */
679 comment_len = get_bits(gb, 8) * 8;
680 if (get_bits_left(gb) < comment_len) {
681 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
684 skip_bits_long(gb, comment_len);
689 * Decode GA "General Audio" specific configuration; reference: table 4.1.
691 * @param ac pointer to AACContext, may be null
692 * @param avctx pointer to AVCCodecContext, used for logging
694 * @return Returns error status. 0 - OK, !0 - error
696 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
698 MPEG4AudioConfig *m4ac,
701 int extension_flag, ret;
702 uint8_t layout_map[MAX_ELEM_ID*4][3];
705 if (get_bits1(gb)) { // frameLengthFlag
706 av_log_missing_feature(avctx, "960/120 MDCT window", 1);
707 return AVERROR_PATCHWELCOME;
710 if (get_bits1(gb)) // dependsOnCoreCoder
711 skip_bits(gb, 14); // coreCoderDelay
712 extension_flag = get_bits1(gb);
714 if (m4ac->object_type == AOT_AAC_SCALABLE ||
715 m4ac->object_type == AOT_ER_AAC_SCALABLE)
716 skip_bits(gb, 3); // layerNr
718 if (channel_config == 0) {
719 skip_bits(gb, 4); // element_instance_tag
720 tags = decode_pce(avctx, m4ac, layout_map, gb);
724 if ((ret = set_default_channel_config(avctx, layout_map, &tags, channel_config)))
728 if (count_channels(layout_map, tags) > 1) {
730 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
733 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
736 if (extension_flag) {
737 switch (m4ac->object_type) {
739 skip_bits(gb, 5); // numOfSubFrame
740 skip_bits(gb, 11); // layer_length
744 case AOT_ER_AAC_SCALABLE:
746 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
747 * aacScalefactorDataResilienceFlag
748 * aacSpectralDataResilienceFlag
752 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
758 * Decode audio specific configuration; reference: table 1.13.
760 * @param ac pointer to AACContext, may be null
761 * @param avctx pointer to AVCCodecContext, used for logging
762 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
763 * @param data pointer to buffer holding an audio specific config
764 * @param bit_size size of audio specific config or data in bits
765 * @param sync_extension look for an appended sync extension
767 * @return Returns error status or number of consumed bits. <0 - error
769 static int decode_audio_specific_config(AACContext *ac,
770 AVCodecContext *avctx,
771 MPEG4AudioConfig *m4ac,
772 const uint8_t *data, int bit_size,
778 av_dlog(avctx, "audio specific config size %d\n", bit_size >> 3);
779 for (i = 0; i < bit_size >> 3; i++)
780 av_dlog(avctx, "%02x ", data[i]);
781 av_dlog(avctx, "\n");
783 init_get_bits(&gb, data, bit_size);
785 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size, sync_extension)) < 0)
787 if (m4ac->sampling_index > 12) {
788 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
792 skip_bits_long(&gb, i);
794 switch (m4ac->object_type) {
798 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
802 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
803 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
807 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
808 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
809 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
811 return get_bits_count(&gb);
815 * linear congruential pseudorandom number generator
817 * @param previous_val pointer to the current state of the generator
819 * @return Returns a 32-bit pseudorandom integer
821 static av_always_inline int lcg_random(unsigned previous_val)
823 union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
827 static av_always_inline void reset_predict_state(PredictorState *ps)
837 static void reset_all_predictors(PredictorState *ps)
840 for (i = 0; i < MAX_PREDICTORS; i++)
841 reset_predict_state(&ps[i]);
844 static int sample_rate_idx (int rate)
846 if (92017 <= rate) return 0;
847 else if (75132 <= rate) return 1;
848 else if (55426 <= rate) return 2;
849 else if (46009 <= rate) return 3;
850 else if (37566 <= rate) return 4;
851 else if (27713 <= rate) return 5;
852 else if (23004 <= rate) return 6;
853 else if (18783 <= rate) return 7;
854 else if (13856 <= rate) return 8;
855 else if (11502 <= rate) return 9;
856 else if (9391 <= rate) return 10;
860 static void reset_predictor_group(PredictorState *ps, int group_num)
863 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
864 reset_predict_state(&ps[i]);
867 #define AAC_INIT_VLC_STATIC(num, size) \
868 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
869 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
870 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
873 static av_cold int aac_decode_init(AVCodecContext *avctx)
875 AACContext *ac = avctx->priv_data;
878 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
880 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
882 if (avctx->extradata_size > 0) {
883 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
885 avctx->extradata_size*8, 1) < 0)
889 uint8_t layout_map[MAX_ELEM_ID*4][3];
892 sr = sample_rate_idx(avctx->sample_rate);
893 ac->oc[1].m4ac.sampling_index = sr;
894 ac->oc[1].m4ac.channels = avctx->channels;
895 ac->oc[1].m4ac.sbr = -1;
896 ac->oc[1].m4ac.ps = -1;
898 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
899 if (ff_mpeg4audio_channels[i] == avctx->channels)
901 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
904 ac->oc[1].m4ac.chan_config = i;
906 if (ac->oc[1].m4ac.chan_config) {
907 int ret = set_default_channel_config(avctx, layout_map,
908 &layout_map_tags, ac->oc[1].m4ac.chan_config);
910 output_configure(ac, layout_map, layout_map_tags,
912 else if (avctx->err_recognition & AV_EF_EXPLODE)
913 return AVERROR_INVALIDDATA;
917 AAC_INIT_VLC_STATIC( 0, 304);
918 AAC_INIT_VLC_STATIC( 1, 270);
919 AAC_INIT_VLC_STATIC( 2, 550);
920 AAC_INIT_VLC_STATIC( 3, 300);
921 AAC_INIT_VLC_STATIC( 4, 328);
922 AAC_INIT_VLC_STATIC( 5, 294);
923 AAC_INIT_VLC_STATIC( 6, 306);
924 AAC_INIT_VLC_STATIC( 7, 268);
925 AAC_INIT_VLC_STATIC( 8, 510);
926 AAC_INIT_VLC_STATIC( 9, 366);
927 AAC_INIT_VLC_STATIC(10, 462);
931 ff_dsputil_init(&ac->dsp, avctx);
932 ff_fmt_convert_init(&ac->fmt_conv, avctx);
933 avpriv_float_dsp_init(&ac->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
935 ac->random_state = 0x1f2e3d4c;
939 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
940 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
941 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
944 ff_mdct_init(&ac->mdct, 11, 1, 1.0 / (32768.0 * 1024.0));
945 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0 / (32768.0 * 128.0));
946 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0 * 32768.0);
947 // window initialization
948 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
949 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
950 ff_init_ff_sine_windows(10);
951 ff_init_ff_sine_windows( 7);
955 avcodec_get_frame_defaults(&ac->frame);
956 avctx->coded_frame = &ac->frame;
962 * Skip data_stream_element; reference: table 4.10.
964 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
966 int byte_align = get_bits1(gb);
967 int count = get_bits(gb, 8);
969 count += get_bits(gb, 8);
973 if (get_bits_left(gb) < 8 * count) {
974 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
977 skip_bits_long(gb, 8 * count);
981 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
986 ics->predictor_reset_group = get_bits(gb, 5);
987 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
988 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
992 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
993 ics->prediction_used[sfb] = get_bits1(gb);
999 * Decode Long Term Prediction data; reference: table 4.xx.
1001 static void decode_ltp(LongTermPrediction *ltp,
1002 GetBitContext *gb, uint8_t max_sfb)
1006 ltp->lag = get_bits(gb, 11);
1007 ltp->coef = ltp_coef[get_bits(gb, 3)];
1008 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1009 ltp->used[sfb] = get_bits1(gb);
1013 * Decode Individual Channel Stream info; reference: table 4.6.
1015 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
1018 if (get_bits1(gb)) {
1019 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1020 return AVERROR_INVALIDDATA;
1022 ics->window_sequence[1] = ics->window_sequence[0];
1023 ics->window_sequence[0] = get_bits(gb, 2);
1024 ics->use_kb_window[1] = ics->use_kb_window[0];
1025 ics->use_kb_window[0] = get_bits1(gb);
1026 ics->num_window_groups = 1;
1027 ics->group_len[0] = 1;
1028 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1030 ics->max_sfb = get_bits(gb, 4);
1031 for (i = 0; i < 7; i++) {
1032 if (get_bits1(gb)) {
1033 ics->group_len[ics->num_window_groups - 1]++;
1035 ics->num_window_groups++;
1036 ics->group_len[ics->num_window_groups - 1] = 1;
1039 ics->num_windows = 8;
1040 ics->swb_offset = ff_swb_offset_128[ac->oc[1].m4ac.sampling_index];
1041 ics->num_swb = ff_aac_num_swb_128[ac->oc[1].m4ac.sampling_index];
1042 ics->tns_max_bands = ff_tns_max_bands_128[ac->oc[1].m4ac.sampling_index];
1043 ics->predictor_present = 0;
1045 ics->max_sfb = get_bits(gb, 6);
1046 ics->num_windows = 1;
1047 ics->swb_offset = ff_swb_offset_1024[ac->oc[1].m4ac.sampling_index];
1048 ics->num_swb = ff_aac_num_swb_1024[ac->oc[1].m4ac.sampling_index];
1049 ics->tns_max_bands = ff_tns_max_bands_1024[ac->oc[1].m4ac.sampling_index];
1050 ics->predictor_present = get_bits1(gb);
1051 ics->predictor_reset_group = 0;
1052 if (ics->predictor_present) {
1053 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1054 if (decode_prediction(ac, ics, gb)) {
1057 } else if (ac->oc[1].m4ac.object_type == AOT_AAC_LC) {
1058 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
1061 if ((ics->ltp.present = get_bits(gb, 1)))
1062 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1067 if (ics->max_sfb > ics->num_swb) {
1068 av_log(ac->avctx, AV_LOG_ERROR,
1069 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
1070 ics->max_sfb, ics->num_swb);
1077 return AVERROR_INVALIDDATA;
1081 * Decode band types (section_data payload); reference: table 4.46.
1083 * @param band_type array of the used band type
1084 * @param band_type_run_end array of the last scalefactor band of a band type run
1086 * @return Returns error status. 0 - OK, !0 - error
1088 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1089 int band_type_run_end[120], GetBitContext *gb,
1090 IndividualChannelStream *ics)
1093 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1094 for (g = 0; g < ics->num_window_groups; g++) {
1096 while (k < ics->max_sfb) {
1097 uint8_t sect_end = k;
1099 int sect_band_type = get_bits(gb, 4);
1100 if (sect_band_type == 12) {
1101 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1105 sect_len_incr = get_bits(gb, bits);
1106 sect_end += sect_len_incr;
1107 if (get_bits_left(gb) < 0) {
1108 av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1111 if (sect_end > ics->max_sfb) {
1112 av_log(ac->avctx, AV_LOG_ERROR,
1113 "Number of bands (%d) exceeds limit (%d).\n",
1114 sect_end, ics->max_sfb);
1117 } while (sect_len_incr == (1 << bits) - 1);
1118 for (; k < sect_end; k++) {
1119 band_type [idx] = sect_band_type;
1120 band_type_run_end[idx++] = sect_end;
1128 * Decode scalefactors; reference: table 4.47.
1130 * @param global_gain first scalefactor value as scalefactors are differentially coded
1131 * @param band_type array of the used band type
1132 * @param band_type_run_end array of the last scalefactor band of a band type run
1133 * @param sf array of scalefactors or intensity stereo positions
1135 * @return Returns error status. 0 - OK, !0 - error
1137 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1138 unsigned int global_gain,
1139 IndividualChannelStream *ics,
1140 enum BandType band_type[120],
1141 int band_type_run_end[120])
1144 int offset[3] = { global_gain, global_gain - 90, 0 };
1147 for (g = 0; g < ics->num_window_groups; g++) {
1148 for (i = 0; i < ics->max_sfb;) {
1149 int run_end = band_type_run_end[idx];
1150 if (band_type[idx] == ZERO_BT) {
1151 for (; i < run_end; i++, idx++)
1153 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
1154 for (; i < run_end; i++, idx++) {
1155 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1156 clipped_offset = av_clip(offset[2], -155, 100);
1157 if (offset[2] != clipped_offset) {
1158 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
1159 "position clipped (%d -> %d).\nIf you heard an "
1160 "audible artifact, there may be a bug in the "
1161 "decoder. ", offset[2], clipped_offset);
1163 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1165 } else if (band_type[idx] == NOISE_BT) {
1166 for (; i < run_end; i++, idx++) {
1167 if (noise_flag-- > 0)
1168 offset[1] += get_bits(gb, 9) - 256;
1170 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1171 clipped_offset = av_clip(offset[1], -100, 155);
1172 if (offset[1] != clipped_offset) {
1173 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
1174 "(%d -> %d).\nIf you heard an audible "
1175 "artifact, there may be a bug in the decoder. ",
1176 offset[1], clipped_offset);
1178 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1181 for (; i < run_end; i++, idx++) {
1182 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1183 if (offset[0] > 255U) {
1184 av_log(ac->avctx, AV_LOG_ERROR,
1185 "Scalefactor (%d) out of range.\n", offset[0]);
1188 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1197 * Decode pulse data; reference: table 4.7.
1199 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1200 const uint16_t *swb_offset, int num_swb)
1203 pulse->num_pulse = get_bits(gb, 2) + 1;
1204 pulse_swb = get_bits(gb, 6);
1205 if (pulse_swb >= num_swb)
1207 pulse->pos[0] = swb_offset[pulse_swb];
1208 pulse->pos[0] += get_bits(gb, 5);
1209 if (pulse->pos[0] > 1023)
1211 pulse->amp[0] = get_bits(gb, 4);
1212 for (i = 1; i < pulse->num_pulse; i++) {
1213 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1214 if (pulse->pos[i] > 1023)
1216 pulse->amp[i] = get_bits(gb, 4);
1222 * Decode Temporal Noise Shaping data; reference: table 4.48.
1224 * @return Returns error status. 0 - OK, !0 - error
1226 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1227 GetBitContext *gb, const IndividualChannelStream *ics)
1229 int w, filt, i, coef_len, coef_res, coef_compress;
1230 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1231 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1232 for (w = 0; w < ics->num_windows; w++) {
1233 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1234 coef_res = get_bits1(gb);
1236 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1238 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1240 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1241 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
1242 tns->order[w][filt], tns_max_order);
1243 tns->order[w][filt] = 0;
1246 if (tns->order[w][filt]) {
1247 tns->direction[w][filt] = get_bits1(gb);
1248 coef_compress = get_bits1(gb);
1249 coef_len = coef_res + 3 - coef_compress;
1250 tmp2_idx = 2 * coef_compress + coef_res;
1252 for (i = 0; i < tns->order[w][filt]; i++)
1253 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1262 * Decode Mid/Side data; reference: table 4.54.
1264 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1265 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1266 * [3] reserved for scalable AAC
1268 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1272 if (ms_present == 1) {
1273 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
1274 cpe->ms_mask[idx] = get_bits1(gb);
1275 } else if (ms_present == 2) {
1276 memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask[0]) * cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb);
1281 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1285 *dst++ = v[idx & 15] * s;
1286 *dst++ = v[idx>>4 & 15] * s;
1292 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1296 *dst++ = v[idx & 3] * s;
1297 *dst++ = v[idx>>2 & 3] * s;
1298 *dst++ = v[idx>>4 & 3] * s;
1299 *dst++ = v[idx>>6 & 3] * s;
1305 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1306 unsigned sign, const float *scale)
1308 union av_intfloat32 s0, s1;
1310 s0.f = s1.f = *scale;
1311 s0.i ^= sign >> 1 << 31;
1314 *dst++ = v[idx & 15] * s0.f;
1315 *dst++ = v[idx>>4 & 15] * s1.f;
1322 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1323 unsigned sign, const float *scale)
1325 unsigned nz = idx >> 12;
1326 union av_intfloat32 s = { .f = *scale };
1327 union av_intfloat32 t;
1329 t.i = s.i ^ (sign & 1U<<31);
1330 *dst++ = v[idx & 3] * t.f;
1332 sign <<= nz & 1; nz >>= 1;
1333 t.i = s.i ^ (sign & 1U<<31);
1334 *dst++ = v[idx>>2 & 3] * t.f;
1336 sign <<= nz & 1; nz >>= 1;
1337 t.i = s.i ^ (sign & 1U<<31);
1338 *dst++ = v[idx>>4 & 3] * t.f;
1341 t.i = s.i ^ (sign & 1U<<31);
1342 *dst++ = v[idx>>6 & 3] * t.f;
1349 * Decode spectral data; reference: table 4.50.
1350 * Dequantize and scale spectral data; reference: 4.6.3.3.
1352 * @param coef array of dequantized, scaled spectral data
1353 * @param sf array of scalefactors or intensity stereo positions
1354 * @param pulse_present set if pulses are present
1355 * @param pulse pointer to pulse data struct
1356 * @param band_type array of the used band type
1358 * @return Returns error status. 0 - OK, !0 - error
1360 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1361 GetBitContext *gb, const float sf[120],
1362 int pulse_present, const Pulse *pulse,
1363 const IndividualChannelStream *ics,
1364 enum BandType band_type[120])
1366 int i, k, g, idx = 0;
1367 const int c = 1024 / ics->num_windows;
1368 const uint16_t *offsets = ics->swb_offset;
1369 float *coef_base = coef;
1371 for (g = 0; g < ics->num_windows; g++)
1372 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1374 for (g = 0; g < ics->num_window_groups; g++) {
1375 unsigned g_len = ics->group_len[g];
1377 for (i = 0; i < ics->max_sfb; i++, idx++) {
1378 const unsigned cbt_m1 = band_type[idx] - 1;
1379 float *cfo = coef + offsets[i];
1380 int off_len = offsets[i + 1] - offsets[i];
1383 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1384 for (group = 0; group < g_len; group++, cfo+=128) {
1385 memset(cfo, 0, off_len * sizeof(float));
1387 } else if (cbt_m1 == NOISE_BT - 1) {
1388 for (group = 0; group < g_len; group++, cfo+=128) {
1392 for (k = 0; k < off_len; k++) {
1393 ac->random_state = lcg_random(ac->random_state);
1394 cfo[k] = ac->random_state;
1397 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1398 scale = sf[idx] / sqrtf(band_energy);
1399 ac->fdsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1402 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1403 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1404 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1405 OPEN_READER(re, gb);
1407 switch (cbt_m1 >> 1) {
1409 for (group = 0; group < g_len; group++, cfo+=128) {
1417 UPDATE_CACHE(re, gb);
1418 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1419 cb_idx = cb_vector_idx[code];
1420 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1426 for (group = 0; group < g_len; group++, cfo+=128) {
1436 UPDATE_CACHE(re, gb);
1437 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1438 cb_idx = cb_vector_idx[code];
1439 nnz = cb_idx >> 8 & 15;
1440 bits = nnz ? GET_CACHE(re, gb) : 0;
1441 LAST_SKIP_BITS(re, gb, nnz);
1442 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1448 for (group = 0; group < g_len; group++, cfo+=128) {
1456 UPDATE_CACHE(re, gb);
1457 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1458 cb_idx = cb_vector_idx[code];
1459 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1466 for (group = 0; group < g_len; group++, cfo+=128) {
1476 UPDATE_CACHE(re, gb);
1477 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1478 cb_idx = cb_vector_idx[code];
1479 nnz = cb_idx >> 8 & 15;
1480 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1481 LAST_SKIP_BITS(re, gb, nnz);
1482 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1488 for (group = 0; group < g_len; group++, cfo+=128) {
1490 uint32_t *icf = (uint32_t *) cf;
1500 UPDATE_CACHE(re, gb);
1501 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1509 cb_idx = cb_vector_idx[code];
1512 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1513 LAST_SKIP_BITS(re, gb, nnz);
1515 for (j = 0; j < 2; j++) {
1519 /* The total length of escape_sequence must be < 22 bits according
1520 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1521 UPDATE_CACHE(re, gb);
1522 b = GET_CACHE(re, gb);
1523 b = 31 - av_log2(~b);
1526 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1530 SKIP_BITS(re, gb, b + 1);
1532 n = (1 << b) + SHOW_UBITS(re, gb, b);
1533 LAST_SKIP_BITS(re, gb, b);
1534 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1537 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1538 *icf++ = (bits & 1U<<31) | v;
1545 ac->fdsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1549 CLOSE_READER(re, gb);
1555 if (pulse_present) {
1557 for (i = 0; i < pulse->num_pulse; i++) {
1558 float co = coef_base[ pulse->pos[i] ];
1559 while (offsets[idx + 1] <= pulse->pos[i])
1561 if (band_type[idx] != NOISE_BT && sf[idx]) {
1562 float ico = -pulse->amp[i];
1565 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1567 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1574 static av_always_inline float flt16_round(float pf)
1576 union av_intfloat32 tmp;
1578 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1582 static av_always_inline float flt16_even(float pf)
1584 union av_intfloat32 tmp;
1586 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1590 static av_always_inline float flt16_trunc(float pf)
1592 union av_intfloat32 pun;
1594 pun.i &= 0xFFFF0000U;
1598 static av_always_inline void predict(PredictorState *ps, float *coef,
1601 const float a = 0.953125; // 61.0 / 64
1602 const float alpha = 0.90625; // 29.0 / 32
1606 float r0 = ps->r0, r1 = ps->r1;
1607 float cor0 = ps->cor0, cor1 = ps->cor1;
1608 float var0 = ps->var0, var1 = ps->var1;
1610 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1611 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1613 pv = flt16_round(k1 * r0 + k2 * r1);
1620 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1621 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1622 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1623 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1625 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1626 ps->r0 = flt16_trunc(a * e0);
1630 * Apply AAC-Main style frequency domain prediction.
1632 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1636 if (!sce->ics.predictor_initialized) {
1637 reset_all_predictors(sce->predictor_state);
1638 sce->ics.predictor_initialized = 1;
1641 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1642 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]; sfb++) {
1643 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1644 predict(&sce->predictor_state[k], &sce->coeffs[k],
1645 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1648 if (sce->ics.predictor_reset_group)
1649 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1651 reset_all_predictors(sce->predictor_state);
1655 * Decode an individual_channel_stream payload; reference: table 4.44.
1657 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1658 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1660 * @return Returns error status. 0 - OK, !0 - error
1662 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1663 GetBitContext *gb, int common_window, int scale_flag)
1666 TemporalNoiseShaping *tns = &sce->tns;
1667 IndividualChannelStream *ics = &sce->ics;
1668 float *out = sce->coeffs;
1669 int global_gain, pulse_present = 0;
1671 /* This assignment is to silence a GCC warning about the variable being used
1672 * uninitialized when in fact it always is.
1674 pulse.num_pulse = 0;
1676 global_gain = get_bits(gb, 8);
1678 if (!common_window && !scale_flag) {
1679 if (decode_ics_info(ac, ics, gb) < 0)
1680 return AVERROR_INVALIDDATA;
1683 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1685 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1690 if ((pulse_present = get_bits1(gb))) {
1691 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1692 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1695 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1696 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1700 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1702 if (get_bits1(gb)) {
1703 av_log_missing_feature(ac->avctx, "SSR", 1);
1704 return AVERROR_PATCHWELCOME;
1708 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1711 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1712 apply_prediction(ac, sce);
1718 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1720 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1722 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1723 float *ch0 = cpe->ch[0].coeffs;
1724 float *ch1 = cpe->ch[1].coeffs;
1725 int g, i, group, idx = 0;
1726 const uint16_t *offsets = ics->swb_offset;
1727 for (g = 0; g < ics->num_window_groups; g++) {
1728 for (i = 0; i < ics->max_sfb; i++, idx++) {
1729 if (cpe->ms_mask[idx] &&
1730 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1731 for (group = 0; group < ics->group_len[g]; group++) {
1732 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1733 ch1 + group * 128 + offsets[i],
1734 offsets[i+1] - offsets[i]);
1738 ch0 += ics->group_len[g] * 128;
1739 ch1 += ics->group_len[g] * 128;
1744 * intensity stereo decoding; reference: 4.6.8.2.3
1746 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1747 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1748 * [3] reserved for scalable AAC
1750 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1752 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1753 SingleChannelElement *sce1 = &cpe->ch[1];
1754 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1755 const uint16_t *offsets = ics->swb_offset;
1756 int g, group, i, idx = 0;
1759 for (g = 0; g < ics->num_window_groups; g++) {
1760 for (i = 0; i < ics->max_sfb;) {
1761 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1762 const int bt_run_end = sce1->band_type_run_end[idx];
1763 for (; i < bt_run_end; i++, idx++) {
1764 c = -1 + 2 * (sce1->band_type[idx] - 14);
1766 c *= 1 - 2 * cpe->ms_mask[idx];
1767 scale = c * sce1->sf[idx];
1768 for (group = 0; group < ics->group_len[g]; group++)
1769 ac->fdsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1770 coef0 + group * 128 + offsets[i],
1772 offsets[i + 1] - offsets[i]);
1775 int bt_run_end = sce1->band_type_run_end[idx];
1776 idx += bt_run_end - i;
1780 coef0 += ics->group_len[g] * 128;
1781 coef1 += ics->group_len[g] * 128;
1786 * Decode a channel_pair_element; reference: table 4.4.
1788 * @return Returns error status. 0 - OK, !0 - error
1790 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1792 int i, ret, common_window, ms_present = 0;
1794 common_window = get_bits1(gb);
1795 if (common_window) {
1796 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1797 return AVERROR_INVALIDDATA;
1798 i = cpe->ch[1].ics.use_kb_window[0];
1799 cpe->ch[1].ics = cpe->ch[0].ics;
1800 cpe->ch[1].ics.use_kb_window[1] = i;
1801 if (cpe->ch[1].ics.predictor_present && (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
1802 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1803 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1804 ms_present = get_bits(gb, 2);
1805 if (ms_present == 3) {
1806 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1808 } else if (ms_present)
1809 decode_mid_side_stereo(cpe, gb, ms_present);
1811 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1813 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1816 if (common_window) {
1818 apply_mid_side_stereo(ac, cpe);
1819 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1820 apply_prediction(ac, &cpe->ch[0]);
1821 apply_prediction(ac, &cpe->ch[1]);
1825 apply_intensity_stereo(ac, cpe, ms_present);
1829 static const float cce_scale[] = {
1830 1.09050773266525765921, //2^(1/8)
1831 1.18920711500272106672, //2^(1/4)
1837 * Decode coupling_channel_element; reference: table 4.8.
1839 * @return Returns error status. 0 - OK, !0 - error
1841 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1847 SingleChannelElement *sce = &che->ch[0];
1848 ChannelCoupling *coup = &che->coup;
1850 coup->coupling_point = 2 * get_bits1(gb);
1851 coup->num_coupled = get_bits(gb, 3);
1852 for (c = 0; c <= coup->num_coupled; c++) {
1854 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1855 coup->id_select[c] = get_bits(gb, 4);
1856 if (coup->type[c] == TYPE_CPE) {
1857 coup->ch_select[c] = get_bits(gb, 2);
1858 if (coup->ch_select[c] == 3)
1861 coup->ch_select[c] = 2;
1863 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1865 sign = get_bits(gb, 1);
1866 scale = cce_scale[get_bits(gb, 2)];
1868 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1871 for (c = 0; c < num_gain; c++) {
1875 float gain_cache = 1.;
1877 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1878 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1879 gain_cache = powf(scale, -gain);
1881 if (coup->coupling_point == AFTER_IMDCT) {
1882 coup->gain[c][0] = gain_cache;
1884 for (g = 0; g < sce->ics.num_window_groups; g++) {
1885 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1886 if (sce->band_type[idx] != ZERO_BT) {
1888 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1896 gain_cache = powf(scale, -t) * s;
1899 coup->gain[c][idx] = gain_cache;
1909 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1911 * @return Returns number of bytes consumed.
1913 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1917 int num_excl_chan = 0;
1920 for (i = 0; i < 7; i++)
1921 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1922 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1924 return num_excl_chan / 7;
1928 * Decode dynamic range information; reference: table 4.52.
1930 * @return Returns number of bytes consumed.
1932 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1936 int drc_num_bands = 1;
1939 /* pce_tag_present? */
1940 if (get_bits1(gb)) {
1941 che_drc->pce_instance_tag = get_bits(gb, 4);
1942 skip_bits(gb, 4); // tag_reserved_bits
1946 /* excluded_chns_present? */
1947 if (get_bits1(gb)) {
1948 n += decode_drc_channel_exclusions(che_drc, gb);
1951 /* drc_bands_present? */
1952 if (get_bits1(gb)) {
1953 che_drc->band_incr = get_bits(gb, 4);
1954 che_drc->interpolation_scheme = get_bits(gb, 4);
1956 drc_num_bands += che_drc->band_incr;
1957 for (i = 0; i < drc_num_bands; i++) {
1958 che_drc->band_top[i] = get_bits(gb, 8);
1963 /* prog_ref_level_present? */
1964 if (get_bits1(gb)) {
1965 che_drc->prog_ref_level = get_bits(gb, 7);
1966 skip_bits1(gb); // prog_ref_level_reserved_bits
1970 for (i = 0; i < drc_num_bands; i++) {
1971 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1972 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1979 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
1981 int i, major, minor;
1986 get_bits(gb, 13); len -= 13;
1988 for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
1989 buf[i] = get_bits(gb, 8);
1992 if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
1993 av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
1995 if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
1996 ac->avctx->internal->skip_samples = 1024;
2000 skip_bits_long(gb, len);
2006 * Decode extension data (incomplete); reference: table 4.51.
2008 * @param cnt length of TYPE_FIL syntactic element in bytes
2010 * @return Returns number of bytes consumed
2012 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2013 ChannelElement *che, enum RawDataBlockType elem_type)
2017 switch (get_bits(gb, 4)) { // extension type
2018 case EXT_SBR_DATA_CRC:
2022 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2024 } else if (!ac->oc[1].m4ac.sbr) {
2025 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2026 skip_bits_long(gb, 8 * cnt - 4);
2028 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2029 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2030 skip_bits_long(gb, 8 * cnt - 4);
2032 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2033 ac->oc[1].m4ac.sbr = 1;
2034 ac->oc[1].m4ac.ps = 1;
2035 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2036 ac->oc[1].status, 1);
2038 ac->oc[1].m4ac.sbr = 1;
2040 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2042 case EXT_DYNAMIC_RANGE:
2043 res = decode_dynamic_range(&ac->che_drc, gb);
2046 decode_fill(ac, gb, 8 * cnt - 4);
2049 case EXT_DATA_ELEMENT:
2051 skip_bits_long(gb, 8 * cnt - 4);
2058 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2060 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2061 * @param coef spectral coefficients
2063 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2064 IndividualChannelStream *ics, int decode)
2066 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2068 int bottom, top, order, start, end, size, inc;
2069 float lpc[TNS_MAX_ORDER];
2070 float tmp[TNS_MAX_ORDER+1];
2072 for (w = 0; w < ics->num_windows; w++) {
2073 bottom = ics->num_swb;
2074 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2076 bottom = FFMAX(0, top - tns->length[w][filt]);
2077 order = tns->order[w][filt];
2082 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2084 start = ics->swb_offset[FFMIN(bottom, mmm)];
2085 end = ics->swb_offset[FFMIN( top, mmm)];
2086 if ((size = end - start) <= 0)
2088 if (tns->direction[w][filt]) {
2098 for (m = 0; m < size; m++, start += inc)
2099 for (i = 1; i <= FFMIN(m, order); i++)
2100 coef[start] -= coef[start - i * inc] * lpc[i - 1];
2103 for (m = 0; m < size; m++, start += inc) {
2104 tmp[0] = coef[start];
2105 for (i = 1; i <= FFMIN(m, order); i++)
2106 coef[start] += tmp[i] * lpc[i - 1];
2107 for (i = order; i > 0; i--)
2108 tmp[i] = tmp[i - 1];
2116 * Apply windowing and MDCT to obtain the spectral
2117 * coefficient from the predicted sample by LTP.
2119 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2120 float *in, IndividualChannelStream *ics)
2122 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2123 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2124 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2125 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2127 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2128 ac->fdsp.vector_fmul(in, in, lwindow_prev, 1024);
2130 memset(in, 0, 448 * sizeof(float));
2131 ac->fdsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2133 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2134 ac->fdsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2136 ac->fdsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2137 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2139 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2143 * Apply the long term prediction
2145 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2147 const LongTermPrediction *ltp = &sce->ics.ltp;
2148 const uint16_t *offsets = sce->ics.swb_offset;
2151 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2152 float *predTime = sce->ret;
2153 float *predFreq = ac->buf_mdct;
2154 int16_t num_samples = 2048;
2156 if (ltp->lag < 1024)
2157 num_samples = ltp->lag + 1024;
2158 for (i = 0; i < num_samples; i++)
2159 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2160 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2162 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2164 if (sce->tns.present)
2165 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2167 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2169 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2170 sce->coeffs[i] += predFreq[i];
2175 * Update the LTP buffer for next frame
2177 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2179 IndividualChannelStream *ics = &sce->ics;
2180 float *saved = sce->saved;
2181 float *saved_ltp = sce->coeffs;
2182 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2183 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2186 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2187 memcpy(saved_ltp, saved, 512 * sizeof(float));
2188 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2189 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2190 for (i = 0; i < 64; i++)
2191 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2192 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2193 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2194 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2195 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2196 for (i = 0; i < 64; i++)
2197 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2198 } else { // LONG_STOP or ONLY_LONG
2199 ac->fdsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2200 for (i = 0; i < 512; i++)
2201 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2204 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2205 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2206 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2210 * Conduct IMDCT and windowing.
2212 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2214 IndividualChannelStream *ics = &sce->ics;
2215 float *in = sce->coeffs;
2216 float *out = sce->ret;
2217 float *saved = sce->saved;
2218 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2219 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2220 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2221 float *buf = ac->buf_mdct;
2222 float *temp = ac->temp;
2226 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2227 for (i = 0; i < 1024; i += 128)
2228 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2230 ac->mdct.imdct_half(&ac->mdct, buf, in);
2232 /* window overlapping
2233 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2234 * and long to short transitions are considered to be short to short
2235 * transitions. This leaves just two cases (long to long and short to short)
2236 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2238 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2239 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2240 ac->fdsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2242 memcpy( out, saved, 448 * sizeof(float));
2244 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2245 ac->fdsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2246 ac->fdsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2247 ac->fdsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2248 ac->fdsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2249 ac->fdsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2250 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2252 ac->fdsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2253 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2258 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2259 memcpy( saved, temp + 64, 64 * sizeof(float));
2260 ac->fdsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2261 ac->fdsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2262 ac->fdsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2263 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2264 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2265 memcpy( saved, buf + 512, 448 * sizeof(float));
2266 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2267 } else { // LONG_STOP or ONLY_LONG
2268 memcpy( saved, buf + 512, 512 * sizeof(float));
2273 * Apply dependent channel coupling (applied before IMDCT).
2275 * @param index index into coupling gain array
2277 static void apply_dependent_coupling(AACContext *ac,
2278 SingleChannelElement *target,
2279 ChannelElement *cce, int index)
2281 IndividualChannelStream *ics = &cce->ch[0].ics;
2282 const uint16_t *offsets = ics->swb_offset;
2283 float *dest = target->coeffs;
2284 const float *src = cce->ch[0].coeffs;
2285 int g, i, group, k, idx = 0;
2286 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2287 av_log(ac->avctx, AV_LOG_ERROR,
2288 "Dependent coupling is not supported together with LTP\n");
2291 for (g = 0; g < ics->num_window_groups; g++) {
2292 for (i = 0; i < ics->max_sfb; i++, idx++) {
2293 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2294 const float gain = cce->coup.gain[index][idx];
2295 for (group = 0; group < ics->group_len[g]; group++) {
2296 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2298 dest[group * 128 + k] += gain * src[group * 128 + k];
2303 dest += ics->group_len[g] * 128;
2304 src += ics->group_len[g] * 128;
2309 * Apply independent channel coupling (applied after IMDCT).
2311 * @param index index into coupling gain array
2313 static void apply_independent_coupling(AACContext *ac,
2314 SingleChannelElement *target,
2315 ChannelElement *cce, int index)
2318 const float gain = cce->coup.gain[index][0];
2319 const float *src = cce->ch[0].ret;
2320 float *dest = target->ret;
2321 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2323 for (i = 0; i < len; i++)
2324 dest[i] += gain * src[i];
2328 * channel coupling transformation interface
2330 * @param apply_coupling_method pointer to (in)dependent coupling function
2332 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2333 enum RawDataBlockType type, int elem_id,
2334 enum CouplingPoint coupling_point,
2335 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2339 for (i = 0; i < MAX_ELEM_ID; i++) {
2340 ChannelElement *cce = ac->che[TYPE_CCE][i];
2343 if (cce && cce->coup.coupling_point == coupling_point) {
2344 ChannelCoupling *coup = &cce->coup;
2346 for (c = 0; c <= coup->num_coupled; c++) {
2347 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2348 if (coup->ch_select[c] != 1) {
2349 apply_coupling_method(ac, &cc->ch[0], cce, index);
2350 if (coup->ch_select[c] != 0)
2353 if (coup->ch_select[c] != 2)
2354 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2356 index += 1 + (coup->ch_select[c] == 3);
2363 * Convert spectral data to float samples, applying all supported tools as appropriate.
2365 static void spectral_to_sample(AACContext *ac)
2368 for (type = 3; type >= 0; type--) {
2369 for (i = 0; i < MAX_ELEM_ID; i++) {
2370 ChannelElement *che = ac->che[type][i];
2372 if (type <= TYPE_CPE)
2373 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2374 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2375 if (che->ch[0].ics.predictor_present) {
2376 if (che->ch[0].ics.ltp.present)
2377 apply_ltp(ac, &che->ch[0]);
2378 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2379 apply_ltp(ac, &che->ch[1]);
2382 if (che->ch[0].tns.present)
2383 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2384 if (che->ch[1].tns.present)
2385 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2386 if (type <= TYPE_CPE)
2387 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2388 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2389 imdct_and_windowing(ac, &che->ch[0]);
2390 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2391 update_ltp(ac, &che->ch[0]);
2392 if (type == TYPE_CPE) {
2393 imdct_and_windowing(ac, &che->ch[1]);
2394 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2395 update_ltp(ac, &che->ch[1]);
2397 if (ac->oc[1].m4ac.sbr > 0) {
2398 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2401 if (type <= TYPE_CCE)
2402 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2408 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2411 AACADTSHeaderInfo hdr_info;
2412 uint8_t layout_map[MAX_ELEM_ID*4][3];
2413 int layout_map_tags;
2415 size = avpriv_aac_parse_header(gb, &hdr_info);
2417 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2418 // This is 2 for "VLB " audio in NSV files.
2419 // See samples/nsv/vlb_audio.
2420 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame", 0);
2421 ac->warned_num_aac_frames = 1;
2423 push_output_configuration(ac);
2424 if (hdr_info.chan_config) {
2425 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2426 if (set_default_channel_config(ac->avctx, layout_map,
2427 &layout_map_tags, hdr_info.chan_config))
2429 if (output_configure(ac, layout_map, layout_map_tags,
2430 FFMAX(ac->oc[1].status, OC_TRIAL_FRAME), 0))
2433 ac->oc[1].m4ac.chan_config = 0;
2435 * dual mono frames in Japanese DTV can have chan_config 0
2436 * WITHOUT specifying PCE.
2437 * thus, set dual mono as default.
2439 if (ac->dmono_mode && ac->oc[0].status == OC_NONE) {
2440 layout_map_tags = 2;
2441 layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
2442 layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
2443 layout_map[0][1] = 0;
2444 layout_map[1][1] = 1;
2445 if (output_configure(ac, layout_map, layout_map_tags,
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, AVPacket *avpkt)
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;
2473 int is_dmono, sce_count = 0;
2475 if (show_bits(gb, 12) == 0xfff) {
2476 if (parse_adts_frame_header(ac, gb) < 0) {
2477 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2481 if (ac->oc[1].m4ac.sampling_index > 12) {
2482 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2488 if (frame_configure_elements(avctx) < 0) {
2493 ac->tags_mapped = 0;
2495 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2496 elem_id = get_bits(gb, 4);
2498 if (elem_type < TYPE_DSE) {
2499 if (!(che=get_che(ac, elem_type, elem_id))) {
2500 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2501 elem_type, elem_id);
2508 switch (elem_type) {
2511 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);
2550 ac->oc[1].m4ac.chan_config = 0;
2558 elem_id += get_bits(gb, 8) - 1;
2559 if (get_bits_left(gb) < 8 * elem_id) {
2560 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
2565 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2566 err = 0; /* FIXME */
2570 err = -1; /* should not happen, but keeps compiler happy */
2575 elem_type_prev = elem_type;
2580 if (get_bits_left(gb) < 3) {
2581 av_log(avctx, AV_LOG_ERROR, overread_err);
2587 spectral_to_sample(ac);
2589 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
2590 samples <<= multiplier;
2591 /* for dual-mono audio (SCE + SCE) */
2592 is_dmono = ac->dmono_mode && sce_count == 2 &&
2593 ac->oc[1].channel_layout == (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT);
2596 ac->frame.nb_samples = samples;
2597 *(AVFrame *)data = ac->frame;
2599 *got_frame_ptr = !!samples;
2602 if (ac->dmono_mode == 1)
2603 ((AVFrame *)data)->data[1] =((AVFrame *)data)->data[0];
2604 else if (ac->dmono_mode == 2)
2605 ((AVFrame *)data)->data[0] =((AVFrame *)data)->data[1];
2608 if (ac->oc[1].status && audio_found) {
2609 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
2610 avctx->frame_size = samples;
2611 ac->oc[1].status = OC_LOCKED;
2616 uint32_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, &side_size);
2617 if (side && side_size>=4)
2618 AV_WL32(side, 2*AV_RL32(side));
2622 pop_output_configuration(ac);
2626 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2627 int *got_frame_ptr, AVPacket *avpkt)
2629 AACContext *ac = avctx->priv_data;
2630 const uint8_t *buf = avpkt->data;
2631 int buf_size = avpkt->size;
2636 int new_extradata_size;
2637 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2638 AV_PKT_DATA_NEW_EXTRADATA,
2639 &new_extradata_size);
2640 int jp_dualmono_size;
2641 const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
2642 AV_PKT_DATA_JP_DUALMONO,
2645 if (new_extradata && 0) {
2646 av_free(avctx->extradata);
2647 avctx->extradata = av_mallocz(new_extradata_size +
2648 FF_INPUT_BUFFER_PADDING_SIZE);
2649 if (!avctx->extradata)
2650 return AVERROR(ENOMEM);
2651 avctx->extradata_size = new_extradata_size;
2652 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2653 push_output_configuration(ac);
2654 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
2656 avctx->extradata_size*8, 1) < 0) {
2657 pop_output_configuration(ac);
2658 return AVERROR_INVALIDDATA;
2663 if (jp_dualmono && jp_dualmono_size > 0)
2664 ac->dmono_mode = 1 + *jp_dualmono;
2665 if (ac->force_dmono_mode >= 0)
2666 ac->dmono_mode = ac->force_dmono_mode;
2668 if (INT_MAX / 8 <= buf_size)
2669 return AVERROR_INVALIDDATA;
2671 init_get_bits(&gb, buf, buf_size * 8);
2673 if ((err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt)) < 0)
2676 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2677 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2678 if (buf[buf_offset])
2681 return buf_size > buf_offset ? buf_consumed : buf_size;
2684 static av_cold int aac_decode_close(AVCodecContext *avctx)
2686 AACContext *ac = avctx->priv_data;
2689 for (i = 0; i < MAX_ELEM_ID; i++) {
2690 for (type = 0; type < 4; type++) {
2691 if (ac->che[type][i])
2692 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2693 av_freep(&ac->che[type][i]);
2697 ff_mdct_end(&ac->mdct);
2698 ff_mdct_end(&ac->mdct_small);
2699 ff_mdct_end(&ac->mdct_ltp);
2704 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2706 struct LATMContext {
2707 AACContext aac_ctx; ///< containing AACContext
2708 int initialized; ///< initialized after a valid extradata was seen
2711 int audio_mux_version_A; ///< LATM syntax version
2712 int frame_length_type; ///< 0/1 variable/fixed frame length
2713 int frame_length; ///< frame length for fixed frame length
2716 static inline uint32_t latm_get_value(GetBitContext *b)
2718 int length = get_bits(b, 2);
2720 return get_bits_long(b, (length+1)*8);
2723 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2724 GetBitContext *gb, int asclen)
2726 AACContext *ac = &latmctx->aac_ctx;
2727 AVCodecContext *avctx = ac->avctx;
2728 MPEG4AudioConfig m4ac = { 0 };
2729 int config_start_bit = get_bits_count(gb);
2730 int sync_extension = 0;
2731 int bits_consumed, esize;
2735 asclen = FFMIN(asclen, get_bits_left(gb));
2737 asclen = get_bits_left(gb);
2739 if (config_start_bit % 8) {
2740 av_log_missing_feature(latmctx->aac_ctx.avctx,
2741 "Non-byte-aligned audio-specific config", 1);
2742 return AVERROR_PATCHWELCOME;
2745 return AVERROR_INVALIDDATA;
2746 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
2747 gb->buffer + (config_start_bit / 8),
2748 asclen, sync_extension);
2750 if (bits_consumed < 0)
2751 return AVERROR_INVALIDDATA;
2753 if (!latmctx->initialized ||
2754 ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
2755 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
2757 if(latmctx->initialized) {
2758 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
2760 av_log(avctx, AV_LOG_DEBUG, "initializing latmctx\n");
2762 latmctx->initialized = 0;
2764 esize = (bits_consumed+7) / 8;
2766 if (avctx->extradata_size < esize) {
2767 av_free(avctx->extradata);
2768 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2769 if (!avctx->extradata)
2770 return AVERROR(ENOMEM);
2773 avctx->extradata_size = esize;
2774 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2775 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2777 skip_bits_long(gb, bits_consumed);
2779 return bits_consumed;
2782 static int read_stream_mux_config(struct LATMContext *latmctx,
2785 int ret, audio_mux_version = get_bits(gb, 1);
2787 latmctx->audio_mux_version_A = 0;
2788 if (audio_mux_version)
2789 latmctx->audio_mux_version_A = get_bits(gb, 1);
2791 if (!latmctx->audio_mux_version_A) {
2793 if (audio_mux_version)
2794 latm_get_value(gb); // taraFullness
2796 skip_bits(gb, 1); // allStreamSameTimeFraming
2797 skip_bits(gb, 6); // numSubFrames
2799 if (get_bits(gb, 4)) { // numPrograms
2800 av_log_missing_feature(latmctx->aac_ctx.avctx,
2801 "Multiple programs", 1);
2802 return AVERROR_PATCHWELCOME;
2805 // for each program (which there is only one in DVB)
2807 // for each layer (which there is only one in DVB)
2808 if (get_bits(gb, 3)) { // numLayer
2809 av_log_missing_feature(latmctx->aac_ctx.avctx,
2810 "Multiple layers", 1);
2811 return AVERROR_PATCHWELCOME;
2814 // for all but first stream: use_same_config = get_bits(gb, 1);
2815 if (!audio_mux_version) {
2816 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
2819 int ascLen = latm_get_value(gb);
2820 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
2823 skip_bits_long(gb, ascLen);
2826 latmctx->frame_length_type = get_bits(gb, 3);
2827 switch (latmctx->frame_length_type) {
2829 skip_bits(gb, 8); // latmBufferFullness
2832 latmctx->frame_length = get_bits(gb, 9);
2837 skip_bits(gb, 6); // CELP frame length table index
2841 skip_bits(gb, 1); // HVXC frame length table index
2845 if (get_bits(gb, 1)) { // other data
2846 if (audio_mux_version) {
2847 latm_get_value(gb); // other_data_bits
2851 esc = get_bits(gb, 1);
2857 if (get_bits(gb, 1)) // crc present
2858 skip_bits(gb, 8); // config_crc
2864 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2868 if (ctx->frame_length_type == 0) {
2869 int mux_slot_length = 0;
2871 tmp = get_bits(gb, 8);
2872 mux_slot_length += tmp;
2873 } while (tmp == 255);
2874 return mux_slot_length;
2875 } else if (ctx->frame_length_type == 1) {
2876 return ctx->frame_length;
2877 } else if (ctx->frame_length_type == 3 ||
2878 ctx->frame_length_type == 5 ||
2879 ctx->frame_length_type == 7) {
2880 skip_bits(gb, 2); // mux_slot_length_coded
2885 static int read_audio_mux_element(struct LATMContext *latmctx,
2889 uint8_t use_same_mux = get_bits(gb, 1);
2890 if (!use_same_mux) {
2891 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2893 } else if (!latmctx->aac_ctx.avctx->extradata) {
2894 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2895 "no decoder config found\n");
2896 return AVERROR(EAGAIN);
2898 if (latmctx->audio_mux_version_A == 0) {
2899 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2900 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2901 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2902 return AVERROR_INVALIDDATA;
2903 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2904 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2905 "frame length mismatch %d << %d\n",
2906 mux_slot_length_bytes * 8, get_bits_left(gb));
2907 return AVERROR_INVALIDDATA;
2914 static int latm_decode_frame(AVCodecContext *avctx, void *out,
2915 int *got_frame_ptr, AVPacket *avpkt)
2917 struct LATMContext *latmctx = avctx->priv_data;
2921 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2923 // check for LOAS sync word
2924 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2925 return AVERROR_INVALIDDATA;
2927 muxlength = get_bits(&gb, 13) + 3;
2928 // not enough data, the parser should have sorted this out
2929 if (muxlength > avpkt->size)
2930 return AVERROR_INVALIDDATA;
2932 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2935 if (!latmctx->initialized) {
2936 if (!avctx->extradata) {
2940 push_output_configuration(&latmctx->aac_ctx);
2941 if ((err = decode_audio_specific_config(
2942 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
2943 avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
2944 pop_output_configuration(&latmctx->aac_ctx);
2947 latmctx->initialized = 1;
2951 if (show_bits(&gb, 12) == 0xfff) {
2952 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2953 "ADTS header detected, probably as result of configuration "
2955 return AVERROR_INVALIDDATA;
2958 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb, avpkt)) < 0)
2964 static av_cold int latm_decode_init(AVCodecContext *avctx)
2966 struct LATMContext *latmctx = avctx->priv_data;
2967 int ret = aac_decode_init(avctx);
2969 if (avctx->extradata_size > 0)
2970 latmctx->initialized = !ret;
2976 * AVOptions for Japanese DTV specific extensions (ADTS only)
2978 #define AACDEC_FLAGS AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
2979 static const AVOption options[] = {
2980 {"dual_mono_mode", "Select the channel to decode for dual mono",
2981 offsetof(AACContext, force_dmono_mode), AV_OPT_TYPE_INT, {.i64=-1}, -1, 2,
2982 AACDEC_FLAGS, "dual_mono_mode"},
2984 {"auto", "autoselection", 0, AV_OPT_TYPE_CONST, {.i64=-1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
2985 {"main", "Select Main/Left channel", 0, AV_OPT_TYPE_CONST, {.i64= 1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
2986 {"sub" , "Select Sub/Right channel", 0, AV_OPT_TYPE_CONST, {.i64= 2}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
2987 {"both", "Select both channels", 0, AV_OPT_TYPE_CONST, {.i64= 0}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
2992 static const AVClass aac_decoder_class = {
2993 .class_name = "AAC decoder",
2994 .item_name = av_default_item_name,
2996 .version = LIBAVUTIL_VERSION_INT,
2999 AVCodec ff_aac_decoder = {
3001 .type = AVMEDIA_TYPE_AUDIO,
3002 .id = AV_CODEC_ID_AAC,
3003 .priv_data_size = sizeof(AACContext),
3004 .init = aac_decode_init,
3005 .close = aac_decode_close,
3006 .decode = aac_decode_frame,
3007 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
3008 .sample_fmts = (const enum AVSampleFormat[]) {
3009 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3011 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3012 .channel_layouts = aac_channel_layout,
3014 .priv_class = &aac_decoder_class,
3018 Note: This decoder filter is intended to decode LATM streams transferred
3019 in MPEG transport streams which only contain one program.
3020 To do a more complex LATM demuxing a separate LATM demuxer should be used.
3022 AVCodec ff_aac_latm_decoder = {
3024 .type = AVMEDIA_TYPE_AUDIO,
3025 .id = AV_CODEC_ID_AAC_LATM,
3026 .priv_data_size = sizeof(struct LATMContext),
3027 .init = latm_decode_init,
3028 .close = aac_decode_close,
3029 .decode = latm_decode_frame,
3030 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
3031 .sample_fmts = (const enum AVSampleFormat[]) {
3032 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3034 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3035 .channel_layouts = aac_channel_layout,