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,
779 av_dlog(avctx, "audio specific config size %d\n", bit_size >> 3);
780 for (i = 0; i < bit_size >> 3; i++)
781 av_dlog(avctx, "%02x ", data[i]);
782 av_dlog(avctx, "\n");
784 if ((ret = init_get_bits(&gb, data, bit_size)) < 0)
787 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size, sync_extension)) < 0)
789 if (m4ac->sampling_index > 12) {
790 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
794 skip_bits_long(&gb, i);
796 switch (m4ac->object_type) {
800 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
804 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
805 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
809 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
810 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
811 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
813 return get_bits_count(&gb);
817 * linear congruential pseudorandom number generator
819 * @param previous_val pointer to the current state of the generator
821 * @return Returns a 32-bit pseudorandom integer
823 static av_always_inline int lcg_random(unsigned previous_val)
825 union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
829 static av_always_inline void reset_predict_state(PredictorState *ps)
839 static void reset_all_predictors(PredictorState *ps)
842 for (i = 0; i < MAX_PREDICTORS; i++)
843 reset_predict_state(&ps[i]);
846 static int sample_rate_idx (int rate)
848 if (92017 <= rate) return 0;
849 else if (75132 <= rate) return 1;
850 else if (55426 <= rate) return 2;
851 else if (46009 <= rate) return 3;
852 else if (37566 <= rate) return 4;
853 else if (27713 <= rate) return 5;
854 else if (23004 <= rate) return 6;
855 else if (18783 <= rate) return 7;
856 else if (13856 <= rate) return 8;
857 else if (11502 <= rate) return 9;
858 else if (9391 <= rate) return 10;
862 static void reset_predictor_group(PredictorState *ps, int group_num)
865 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
866 reset_predict_state(&ps[i]);
869 #define AAC_INIT_VLC_STATIC(num, size) \
870 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
871 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
872 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
875 static av_cold int aac_decode_init(AVCodecContext *avctx)
877 AACContext *ac = avctx->priv_data;
880 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
882 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
884 if (avctx->extradata_size > 0) {
885 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
887 avctx->extradata_size*8, 1) < 0)
891 uint8_t layout_map[MAX_ELEM_ID*4][3];
894 sr = sample_rate_idx(avctx->sample_rate);
895 ac->oc[1].m4ac.sampling_index = sr;
896 ac->oc[1].m4ac.channels = avctx->channels;
897 ac->oc[1].m4ac.sbr = -1;
898 ac->oc[1].m4ac.ps = -1;
900 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
901 if (ff_mpeg4audio_channels[i] == avctx->channels)
903 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
906 ac->oc[1].m4ac.chan_config = i;
908 if (ac->oc[1].m4ac.chan_config) {
909 int ret = set_default_channel_config(avctx, layout_map,
910 &layout_map_tags, ac->oc[1].m4ac.chan_config);
912 output_configure(ac, layout_map, layout_map_tags,
914 else if (avctx->err_recognition & AV_EF_EXPLODE)
915 return AVERROR_INVALIDDATA;
919 if (avctx->channels > MAX_CHANNELS) {
920 av_log(avctx, AV_LOG_ERROR, "Too many channels\n");
921 return AVERROR_INVALIDDATA;
924 AAC_INIT_VLC_STATIC( 0, 304);
925 AAC_INIT_VLC_STATIC( 1, 270);
926 AAC_INIT_VLC_STATIC( 2, 550);
927 AAC_INIT_VLC_STATIC( 3, 300);
928 AAC_INIT_VLC_STATIC( 4, 328);
929 AAC_INIT_VLC_STATIC( 5, 294);
930 AAC_INIT_VLC_STATIC( 6, 306);
931 AAC_INIT_VLC_STATIC( 7, 268);
932 AAC_INIT_VLC_STATIC( 8, 510);
933 AAC_INIT_VLC_STATIC( 9, 366);
934 AAC_INIT_VLC_STATIC(10, 462);
938 ff_fmt_convert_init(&ac->fmt_conv, avctx);
939 avpriv_float_dsp_init(&ac->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
941 ac->random_state = 0x1f2e3d4c;
945 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
946 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
947 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
950 ff_mdct_init(&ac->mdct, 11, 1, 1.0 / (32768.0 * 1024.0));
951 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0 / (32768.0 * 128.0));
952 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0 * 32768.0);
953 // window initialization
954 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
955 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
956 ff_init_ff_sine_windows(10);
957 ff_init_ff_sine_windows( 7);
961 avcodec_get_frame_defaults(&ac->frame);
962 avctx->coded_frame = &ac->frame;
968 * Skip data_stream_element; reference: table 4.10.
970 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
972 int byte_align = get_bits1(gb);
973 int count = get_bits(gb, 8);
975 count += get_bits(gb, 8);
979 if (get_bits_left(gb) < 8 * count) {
980 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
983 skip_bits_long(gb, 8 * count);
987 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
992 ics->predictor_reset_group = get_bits(gb, 5);
993 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
994 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
998 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
999 ics->prediction_used[sfb] = get_bits1(gb);
1005 * Decode Long Term Prediction data; reference: table 4.xx.
1007 static void decode_ltp(LongTermPrediction *ltp,
1008 GetBitContext *gb, uint8_t max_sfb)
1012 ltp->lag = get_bits(gb, 11);
1013 ltp->coef = ltp_coef[get_bits(gb, 3)];
1014 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1015 ltp->used[sfb] = get_bits1(gb);
1019 * Decode Individual Channel Stream info; reference: table 4.6.
1021 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
1024 if (get_bits1(gb)) {
1025 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1026 return AVERROR_INVALIDDATA;
1028 ics->window_sequence[1] = ics->window_sequence[0];
1029 ics->window_sequence[0] = get_bits(gb, 2);
1030 ics->use_kb_window[1] = ics->use_kb_window[0];
1031 ics->use_kb_window[0] = get_bits1(gb);
1032 ics->num_window_groups = 1;
1033 ics->group_len[0] = 1;
1034 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1036 ics->max_sfb = get_bits(gb, 4);
1037 for (i = 0; i < 7; i++) {
1038 if (get_bits1(gb)) {
1039 ics->group_len[ics->num_window_groups - 1]++;
1041 ics->num_window_groups++;
1042 ics->group_len[ics->num_window_groups - 1] = 1;
1045 ics->num_windows = 8;
1046 ics->swb_offset = ff_swb_offset_128[ac->oc[1].m4ac.sampling_index];
1047 ics->num_swb = ff_aac_num_swb_128[ac->oc[1].m4ac.sampling_index];
1048 ics->tns_max_bands = ff_tns_max_bands_128[ac->oc[1].m4ac.sampling_index];
1049 ics->predictor_present = 0;
1051 ics->max_sfb = get_bits(gb, 6);
1052 ics->num_windows = 1;
1053 ics->swb_offset = ff_swb_offset_1024[ac->oc[1].m4ac.sampling_index];
1054 ics->num_swb = ff_aac_num_swb_1024[ac->oc[1].m4ac.sampling_index];
1055 ics->tns_max_bands = ff_tns_max_bands_1024[ac->oc[1].m4ac.sampling_index];
1056 ics->predictor_present = get_bits1(gb);
1057 ics->predictor_reset_group = 0;
1058 if (ics->predictor_present) {
1059 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1060 if (decode_prediction(ac, ics, gb)) {
1063 } else if (ac->oc[1].m4ac.object_type == AOT_AAC_LC) {
1064 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
1067 if ((ics->ltp.present = get_bits(gb, 1)))
1068 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1073 if (ics->max_sfb > ics->num_swb) {
1074 av_log(ac->avctx, AV_LOG_ERROR,
1075 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
1076 ics->max_sfb, ics->num_swb);
1083 return AVERROR_INVALIDDATA;
1087 * Decode band types (section_data payload); reference: table 4.46.
1089 * @param band_type array of the used band type
1090 * @param band_type_run_end array of the last scalefactor band of a band type run
1092 * @return Returns error status. 0 - OK, !0 - error
1094 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1095 int band_type_run_end[120], GetBitContext *gb,
1096 IndividualChannelStream *ics)
1099 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1100 for (g = 0; g < ics->num_window_groups; g++) {
1102 while (k < ics->max_sfb) {
1103 uint8_t sect_end = k;
1105 int sect_band_type = get_bits(gb, 4);
1106 if (sect_band_type == 12) {
1107 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1111 sect_len_incr = get_bits(gb, bits);
1112 sect_end += sect_len_incr;
1113 if (get_bits_left(gb) < 0) {
1114 av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1117 if (sect_end > ics->max_sfb) {
1118 av_log(ac->avctx, AV_LOG_ERROR,
1119 "Number of bands (%d) exceeds limit (%d).\n",
1120 sect_end, ics->max_sfb);
1123 } while (sect_len_incr == (1 << bits) - 1);
1124 for (; k < sect_end; k++) {
1125 band_type [idx] = sect_band_type;
1126 band_type_run_end[idx++] = sect_end;
1134 * Decode scalefactors; reference: table 4.47.
1136 * @param global_gain first scalefactor value as scalefactors are differentially coded
1137 * @param band_type array of the used band type
1138 * @param band_type_run_end array of the last scalefactor band of a band type run
1139 * @param sf array of scalefactors or intensity stereo positions
1141 * @return Returns error status. 0 - OK, !0 - error
1143 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1144 unsigned int global_gain,
1145 IndividualChannelStream *ics,
1146 enum BandType band_type[120],
1147 int band_type_run_end[120])
1150 int offset[3] = { global_gain, global_gain - 90, 0 };
1153 for (g = 0; g < ics->num_window_groups; g++) {
1154 for (i = 0; i < ics->max_sfb;) {
1155 int run_end = band_type_run_end[idx];
1156 if (band_type[idx] == ZERO_BT) {
1157 for (; i < run_end; i++, idx++)
1159 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
1160 for (; i < run_end; i++, idx++) {
1161 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1162 clipped_offset = av_clip(offset[2], -155, 100);
1163 if (offset[2] != clipped_offset) {
1164 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
1165 "position clipped (%d -> %d).\nIf you heard an "
1166 "audible artifact, there may be a bug in the "
1167 "decoder. ", offset[2], clipped_offset);
1169 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1171 } else if (band_type[idx] == NOISE_BT) {
1172 for (; i < run_end; i++, idx++) {
1173 if (noise_flag-- > 0)
1174 offset[1] += get_bits(gb, 9) - 256;
1176 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1177 clipped_offset = av_clip(offset[1], -100, 155);
1178 if (offset[1] != clipped_offset) {
1179 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
1180 "(%d -> %d).\nIf you heard an audible "
1181 "artifact, there may be a bug in the decoder. ",
1182 offset[1], clipped_offset);
1184 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1187 for (; i < run_end; i++, idx++) {
1188 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1189 if (offset[0] > 255U) {
1190 av_log(ac->avctx, AV_LOG_ERROR,
1191 "Scalefactor (%d) out of range.\n", offset[0]);
1194 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1203 * Decode pulse data; reference: table 4.7.
1205 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1206 const uint16_t *swb_offset, int num_swb)
1209 pulse->num_pulse = get_bits(gb, 2) + 1;
1210 pulse_swb = get_bits(gb, 6);
1211 if (pulse_swb >= num_swb)
1213 pulse->pos[0] = swb_offset[pulse_swb];
1214 pulse->pos[0] += get_bits(gb, 5);
1215 if (pulse->pos[0] > 1023)
1217 pulse->amp[0] = get_bits(gb, 4);
1218 for (i = 1; i < pulse->num_pulse; i++) {
1219 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1220 if (pulse->pos[i] > 1023)
1222 pulse->amp[i] = get_bits(gb, 4);
1228 * Decode Temporal Noise Shaping data; reference: table 4.48.
1230 * @return Returns error status. 0 - OK, !0 - error
1232 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1233 GetBitContext *gb, const IndividualChannelStream *ics)
1235 int w, filt, i, coef_len, coef_res, coef_compress;
1236 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1237 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1238 for (w = 0; w < ics->num_windows; w++) {
1239 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1240 coef_res = get_bits1(gb);
1242 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1244 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1246 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1247 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
1248 tns->order[w][filt], tns_max_order);
1249 tns->order[w][filt] = 0;
1252 if (tns->order[w][filt]) {
1253 tns->direction[w][filt] = get_bits1(gb);
1254 coef_compress = get_bits1(gb);
1255 coef_len = coef_res + 3 - coef_compress;
1256 tmp2_idx = 2 * coef_compress + coef_res;
1258 for (i = 0; i < tns->order[w][filt]; i++)
1259 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1268 * Decode Mid/Side data; reference: table 4.54.
1270 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1271 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1272 * [3] reserved for scalable AAC
1274 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1278 if (ms_present == 1) {
1279 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
1280 cpe->ms_mask[idx] = get_bits1(gb);
1281 } else if (ms_present == 2) {
1282 memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask[0]) * cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb);
1287 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1291 *dst++ = v[idx & 15] * s;
1292 *dst++ = v[idx>>4 & 15] * s;
1298 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1302 *dst++ = v[idx & 3] * s;
1303 *dst++ = v[idx>>2 & 3] * s;
1304 *dst++ = v[idx>>4 & 3] * s;
1305 *dst++ = v[idx>>6 & 3] * s;
1311 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1312 unsigned sign, const float *scale)
1314 union av_intfloat32 s0, s1;
1316 s0.f = s1.f = *scale;
1317 s0.i ^= sign >> 1 << 31;
1320 *dst++ = v[idx & 15] * s0.f;
1321 *dst++ = v[idx>>4 & 15] * s1.f;
1328 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1329 unsigned sign, const float *scale)
1331 unsigned nz = idx >> 12;
1332 union av_intfloat32 s = { .f = *scale };
1333 union av_intfloat32 t;
1335 t.i = s.i ^ (sign & 1U<<31);
1336 *dst++ = v[idx & 3] * t.f;
1338 sign <<= nz & 1; nz >>= 1;
1339 t.i = s.i ^ (sign & 1U<<31);
1340 *dst++ = v[idx>>2 & 3] * t.f;
1342 sign <<= nz & 1; nz >>= 1;
1343 t.i = s.i ^ (sign & 1U<<31);
1344 *dst++ = v[idx>>4 & 3] * t.f;
1347 t.i = s.i ^ (sign & 1U<<31);
1348 *dst++ = v[idx>>6 & 3] * t.f;
1355 * Decode spectral data; reference: table 4.50.
1356 * Dequantize and scale spectral data; reference: 4.6.3.3.
1358 * @param coef array of dequantized, scaled spectral data
1359 * @param sf array of scalefactors or intensity stereo positions
1360 * @param pulse_present set if pulses are present
1361 * @param pulse pointer to pulse data struct
1362 * @param band_type array of the used band type
1364 * @return Returns error status. 0 - OK, !0 - error
1366 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1367 GetBitContext *gb, const float sf[120],
1368 int pulse_present, const Pulse *pulse,
1369 const IndividualChannelStream *ics,
1370 enum BandType band_type[120])
1372 int i, k, g, idx = 0;
1373 const int c = 1024 / ics->num_windows;
1374 const uint16_t *offsets = ics->swb_offset;
1375 float *coef_base = coef;
1377 for (g = 0; g < ics->num_windows; g++)
1378 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1380 for (g = 0; g < ics->num_window_groups; g++) {
1381 unsigned g_len = ics->group_len[g];
1383 for (i = 0; i < ics->max_sfb; i++, idx++) {
1384 const unsigned cbt_m1 = band_type[idx] - 1;
1385 float *cfo = coef + offsets[i];
1386 int off_len = offsets[i + 1] - offsets[i];
1389 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1390 for (group = 0; group < g_len; group++, cfo+=128) {
1391 memset(cfo, 0, off_len * sizeof(float));
1393 } else if (cbt_m1 == NOISE_BT - 1) {
1394 for (group = 0; group < g_len; group++, cfo+=128) {
1398 for (k = 0; k < off_len; k++) {
1399 ac->random_state = lcg_random(ac->random_state);
1400 cfo[k] = ac->random_state;
1403 band_energy = ac->fdsp.scalarproduct_float(cfo, cfo, off_len);
1404 scale = sf[idx] / sqrtf(band_energy);
1405 ac->fdsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1408 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1409 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1410 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1411 OPEN_READER(re, gb);
1413 switch (cbt_m1 >> 1) {
1415 for (group = 0; group < g_len; group++, cfo+=128) {
1423 UPDATE_CACHE(re, gb);
1424 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1425 cb_idx = cb_vector_idx[code];
1426 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1432 for (group = 0; group < g_len; group++, cfo+=128) {
1442 UPDATE_CACHE(re, gb);
1443 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1444 cb_idx = cb_vector_idx[code];
1445 nnz = cb_idx >> 8 & 15;
1446 bits = nnz ? GET_CACHE(re, gb) : 0;
1447 LAST_SKIP_BITS(re, gb, nnz);
1448 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1454 for (group = 0; group < g_len; group++, cfo+=128) {
1462 UPDATE_CACHE(re, gb);
1463 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1464 cb_idx = cb_vector_idx[code];
1465 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1472 for (group = 0; group < g_len; group++, cfo+=128) {
1482 UPDATE_CACHE(re, gb);
1483 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1484 cb_idx = cb_vector_idx[code];
1485 nnz = cb_idx >> 8 & 15;
1486 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1487 LAST_SKIP_BITS(re, gb, nnz);
1488 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1494 for (group = 0; group < g_len; group++, cfo+=128) {
1496 uint32_t *icf = (uint32_t *) cf;
1506 UPDATE_CACHE(re, gb);
1507 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1515 cb_idx = cb_vector_idx[code];
1518 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1519 LAST_SKIP_BITS(re, gb, nnz);
1521 for (j = 0; j < 2; j++) {
1525 /* The total length of escape_sequence must be < 22 bits according
1526 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1527 UPDATE_CACHE(re, gb);
1528 b = GET_CACHE(re, gb);
1529 b = 31 - av_log2(~b);
1532 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1536 SKIP_BITS(re, gb, b + 1);
1538 n = (1 << b) + SHOW_UBITS(re, gb, b);
1539 LAST_SKIP_BITS(re, gb, b);
1540 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1543 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1544 *icf++ = (bits & 1U<<31) | v;
1551 ac->fdsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1555 CLOSE_READER(re, gb);
1561 if (pulse_present) {
1563 for (i = 0; i < pulse->num_pulse; i++) {
1564 float co = coef_base[ pulse->pos[i] ];
1565 while (offsets[idx + 1] <= pulse->pos[i])
1567 if (band_type[idx] != NOISE_BT && sf[idx]) {
1568 float ico = -pulse->amp[i];
1571 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1573 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1580 static av_always_inline float flt16_round(float pf)
1582 union av_intfloat32 tmp;
1584 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1588 static av_always_inline float flt16_even(float pf)
1590 union av_intfloat32 tmp;
1592 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1596 static av_always_inline float flt16_trunc(float pf)
1598 union av_intfloat32 pun;
1600 pun.i &= 0xFFFF0000U;
1604 static av_always_inline void predict(PredictorState *ps, float *coef,
1607 const float a = 0.953125; // 61.0 / 64
1608 const float alpha = 0.90625; // 29.0 / 32
1612 float r0 = ps->r0, r1 = ps->r1;
1613 float cor0 = ps->cor0, cor1 = ps->cor1;
1614 float var0 = ps->var0, var1 = ps->var1;
1616 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1617 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1619 pv = flt16_round(k1 * r0 + k2 * r1);
1626 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1627 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1628 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1629 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1631 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1632 ps->r0 = flt16_trunc(a * e0);
1636 * Apply AAC-Main style frequency domain prediction.
1638 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1642 if (!sce->ics.predictor_initialized) {
1643 reset_all_predictors(sce->predictor_state);
1644 sce->ics.predictor_initialized = 1;
1647 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1648 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]; sfb++) {
1649 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1650 predict(&sce->predictor_state[k], &sce->coeffs[k],
1651 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1654 if (sce->ics.predictor_reset_group)
1655 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1657 reset_all_predictors(sce->predictor_state);
1661 * Decode an individual_channel_stream payload; reference: table 4.44.
1663 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1664 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1666 * @return Returns error status. 0 - OK, !0 - error
1668 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1669 GetBitContext *gb, int common_window, int scale_flag)
1672 TemporalNoiseShaping *tns = &sce->tns;
1673 IndividualChannelStream *ics = &sce->ics;
1674 float *out = sce->coeffs;
1675 int global_gain, pulse_present = 0;
1677 /* This assignment is to silence a GCC warning about the variable being used
1678 * uninitialized when in fact it always is.
1680 pulse.num_pulse = 0;
1682 global_gain = get_bits(gb, 8);
1684 if (!common_window && !scale_flag) {
1685 if (decode_ics_info(ac, ics, gb) < 0)
1686 return AVERROR_INVALIDDATA;
1689 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1691 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1696 if ((pulse_present = get_bits1(gb))) {
1697 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1698 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1701 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1702 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1706 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1708 if (get_bits1(gb)) {
1709 av_log_missing_feature(ac->avctx, "SSR", 1);
1710 return AVERROR_PATCHWELCOME;
1714 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1717 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1718 apply_prediction(ac, sce);
1724 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1726 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1728 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1729 float *ch0 = cpe->ch[0].coeffs;
1730 float *ch1 = cpe->ch[1].coeffs;
1731 int g, i, group, idx = 0;
1732 const uint16_t *offsets = ics->swb_offset;
1733 for (g = 0; g < ics->num_window_groups; g++) {
1734 for (i = 0; i < ics->max_sfb; i++, idx++) {
1735 if (cpe->ms_mask[idx] &&
1736 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1737 for (group = 0; group < ics->group_len[g]; group++) {
1738 ac->fdsp.butterflies_float(ch0 + group * 128 + offsets[i],
1739 ch1 + group * 128 + offsets[i],
1740 offsets[i+1] - offsets[i]);
1744 ch0 += ics->group_len[g] * 128;
1745 ch1 += ics->group_len[g] * 128;
1750 * intensity stereo decoding; reference: 4.6.8.2.3
1752 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1753 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1754 * [3] reserved for scalable AAC
1756 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1758 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1759 SingleChannelElement *sce1 = &cpe->ch[1];
1760 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1761 const uint16_t *offsets = ics->swb_offset;
1762 int g, group, i, idx = 0;
1765 for (g = 0; g < ics->num_window_groups; g++) {
1766 for (i = 0; i < ics->max_sfb;) {
1767 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1768 const int bt_run_end = sce1->band_type_run_end[idx];
1769 for (; i < bt_run_end; i++, idx++) {
1770 c = -1 + 2 * (sce1->band_type[idx] - 14);
1772 c *= 1 - 2 * cpe->ms_mask[idx];
1773 scale = c * sce1->sf[idx];
1774 for (group = 0; group < ics->group_len[g]; group++)
1775 ac->fdsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1776 coef0 + group * 128 + offsets[i],
1778 offsets[i + 1] - offsets[i]);
1781 int bt_run_end = sce1->band_type_run_end[idx];
1782 idx += bt_run_end - i;
1786 coef0 += ics->group_len[g] * 128;
1787 coef1 += ics->group_len[g] * 128;
1792 * Decode a channel_pair_element; reference: table 4.4.
1794 * @return Returns error status. 0 - OK, !0 - error
1796 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1798 int i, ret, common_window, ms_present = 0;
1800 common_window = get_bits1(gb);
1801 if (common_window) {
1802 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1803 return AVERROR_INVALIDDATA;
1804 i = cpe->ch[1].ics.use_kb_window[0];
1805 cpe->ch[1].ics = cpe->ch[0].ics;
1806 cpe->ch[1].ics.use_kb_window[1] = i;
1807 if (cpe->ch[1].ics.predictor_present && (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
1808 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1809 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1810 ms_present = get_bits(gb, 2);
1811 if (ms_present == 3) {
1812 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1814 } else if (ms_present)
1815 decode_mid_side_stereo(cpe, gb, ms_present);
1817 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1819 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1822 if (common_window) {
1824 apply_mid_side_stereo(ac, cpe);
1825 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1826 apply_prediction(ac, &cpe->ch[0]);
1827 apply_prediction(ac, &cpe->ch[1]);
1831 apply_intensity_stereo(ac, cpe, ms_present);
1835 static const float cce_scale[] = {
1836 1.09050773266525765921, //2^(1/8)
1837 1.18920711500272106672, //2^(1/4)
1843 * Decode coupling_channel_element; reference: table 4.8.
1845 * @return Returns error status. 0 - OK, !0 - error
1847 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1853 SingleChannelElement *sce = &che->ch[0];
1854 ChannelCoupling *coup = &che->coup;
1856 coup->coupling_point = 2 * get_bits1(gb);
1857 coup->num_coupled = get_bits(gb, 3);
1858 for (c = 0; c <= coup->num_coupled; c++) {
1860 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1861 coup->id_select[c] = get_bits(gb, 4);
1862 if (coup->type[c] == TYPE_CPE) {
1863 coup->ch_select[c] = get_bits(gb, 2);
1864 if (coup->ch_select[c] == 3)
1867 coup->ch_select[c] = 2;
1869 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1871 sign = get_bits(gb, 1);
1872 scale = cce_scale[get_bits(gb, 2)];
1874 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1877 for (c = 0; c < num_gain; c++) {
1881 float gain_cache = 1.;
1883 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1884 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1885 gain_cache = powf(scale, -gain);
1887 if (coup->coupling_point == AFTER_IMDCT) {
1888 coup->gain[c][0] = gain_cache;
1890 for (g = 0; g < sce->ics.num_window_groups; g++) {
1891 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1892 if (sce->band_type[idx] != ZERO_BT) {
1894 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1902 gain_cache = powf(scale, -t) * s;
1905 coup->gain[c][idx] = gain_cache;
1915 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1917 * @return Returns number of bytes consumed.
1919 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1923 int num_excl_chan = 0;
1926 for (i = 0; i < 7; i++)
1927 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1928 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1930 return num_excl_chan / 7;
1934 * Decode dynamic range information; reference: table 4.52.
1936 * @return Returns number of bytes consumed.
1938 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1942 int drc_num_bands = 1;
1945 /* pce_tag_present? */
1946 if (get_bits1(gb)) {
1947 che_drc->pce_instance_tag = get_bits(gb, 4);
1948 skip_bits(gb, 4); // tag_reserved_bits
1952 /* excluded_chns_present? */
1953 if (get_bits1(gb)) {
1954 n += decode_drc_channel_exclusions(che_drc, gb);
1957 /* drc_bands_present? */
1958 if (get_bits1(gb)) {
1959 che_drc->band_incr = get_bits(gb, 4);
1960 che_drc->interpolation_scheme = get_bits(gb, 4);
1962 drc_num_bands += che_drc->band_incr;
1963 for (i = 0; i < drc_num_bands; i++) {
1964 che_drc->band_top[i] = get_bits(gb, 8);
1969 /* prog_ref_level_present? */
1970 if (get_bits1(gb)) {
1971 che_drc->prog_ref_level = get_bits(gb, 7);
1972 skip_bits1(gb); // prog_ref_level_reserved_bits
1976 for (i = 0; i < drc_num_bands; i++) {
1977 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1978 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1985 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
1987 int i, major, minor;
1992 get_bits(gb, 13); len -= 13;
1994 for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
1995 buf[i] = get_bits(gb, 8);
1998 if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
1999 av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
2001 if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
2002 ac->avctx->internal->skip_samples = 1024;
2006 skip_bits_long(gb, len);
2012 * Decode extension data (incomplete); reference: table 4.51.
2014 * @param cnt length of TYPE_FIL syntactic element in bytes
2016 * @return Returns number of bytes consumed
2018 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2019 ChannelElement *che, enum RawDataBlockType elem_type)
2023 switch (get_bits(gb, 4)) { // extension type
2024 case EXT_SBR_DATA_CRC:
2028 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2030 } else if (!ac->oc[1].m4ac.sbr) {
2031 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2032 skip_bits_long(gb, 8 * cnt - 4);
2034 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2035 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2036 skip_bits_long(gb, 8 * cnt - 4);
2038 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2039 ac->oc[1].m4ac.sbr = 1;
2040 ac->oc[1].m4ac.ps = 1;
2041 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2042 ac->oc[1].status, 1);
2044 ac->oc[1].m4ac.sbr = 1;
2046 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2048 case EXT_DYNAMIC_RANGE:
2049 res = decode_dynamic_range(&ac->che_drc, gb);
2052 decode_fill(ac, gb, 8 * cnt - 4);
2055 case EXT_DATA_ELEMENT:
2057 skip_bits_long(gb, 8 * cnt - 4);
2064 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2066 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2067 * @param coef spectral coefficients
2069 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2070 IndividualChannelStream *ics, int decode)
2072 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2074 int bottom, top, order, start, end, size, inc;
2075 float lpc[TNS_MAX_ORDER];
2076 float tmp[TNS_MAX_ORDER+1];
2078 for (w = 0; w < ics->num_windows; w++) {
2079 bottom = ics->num_swb;
2080 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2082 bottom = FFMAX(0, top - tns->length[w][filt]);
2083 order = tns->order[w][filt];
2088 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2090 start = ics->swb_offset[FFMIN(bottom, mmm)];
2091 end = ics->swb_offset[FFMIN( top, mmm)];
2092 if ((size = end - start) <= 0)
2094 if (tns->direction[w][filt]) {
2104 for (m = 0; m < size; m++, start += inc)
2105 for (i = 1; i <= FFMIN(m, order); i++)
2106 coef[start] -= coef[start - i * inc] * lpc[i - 1];
2109 for (m = 0; m < size; m++, start += inc) {
2110 tmp[0] = coef[start];
2111 for (i = 1; i <= FFMIN(m, order); i++)
2112 coef[start] += tmp[i] * lpc[i - 1];
2113 for (i = order; i > 0; i--)
2114 tmp[i] = tmp[i - 1];
2122 * Apply windowing and MDCT to obtain the spectral
2123 * coefficient from the predicted sample by LTP.
2125 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2126 float *in, IndividualChannelStream *ics)
2128 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2129 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2130 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2131 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2133 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2134 ac->fdsp.vector_fmul(in, in, lwindow_prev, 1024);
2136 memset(in, 0, 448 * sizeof(float));
2137 ac->fdsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2139 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2140 ac->fdsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2142 ac->fdsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2143 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2145 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2149 * Apply the long term prediction
2151 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2153 const LongTermPrediction *ltp = &sce->ics.ltp;
2154 const uint16_t *offsets = sce->ics.swb_offset;
2157 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2158 float *predTime = sce->ret;
2159 float *predFreq = ac->buf_mdct;
2160 int16_t num_samples = 2048;
2162 if (ltp->lag < 1024)
2163 num_samples = ltp->lag + 1024;
2164 for (i = 0; i < num_samples; i++)
2165 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2166 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2168 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2170 if (sce->tns.present)
2171 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2173 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2175 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2176 sce->coeffs[i] += predFreq[i];
2181 * Update the LTP buffer for next frame
2183 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2185 IndividualChannelStream *ics = &sce->ics;
2186 float *saved = sce->saved;
2187 float *saved_ltp = sce->coeffs;
2188 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2189 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2192 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2193 memcpy(saved_ltp, saved, 512 * 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 if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2199 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2200 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2201 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2202 for (i = 0; i < 64; i++)
2203 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2204 } else { // LONG_STOP or ONLY_LONG
2205 ac->fdsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2206 for (i = 0; i < 512; i++)
2207 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2210 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2211 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2212 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2216 * Conduct IMDCT and windowing.
2218 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2220 IndividualChannelStream *ics = &sce->ics;
2221 float *in = sce->coeffs;
2222 float *out = sce->ret;
2223 float *saved = sce->saved;
2224 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2225 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2226 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2227 float *buf = ac->buf_mdct;
2228 float *temp = ac->temp;
2232 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2233 for (i = 0; i < 1024; i += 128)
2234 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2236 ac->mdct.imdct_half(&ac->mdct, buf, in);
2238 /* window overlapping
2239 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2240 * and long to short transitions are considered to be short to short
2241 * transitions. This leaves just two cases (long to long and short to short)
2242 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2244 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2245 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2246 ac->fdsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2248 memcpy( out, saved, 448 * sizeof(float));
2250 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2251 ac->fdsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2252 ac->fdsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2253 ac->fdsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2254 ac->fdsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2255 ac->fdsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2256 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2258 ac->fdsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2259 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2264 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2265 memcpy( saved, temp + 64, 64 * sizeof(float));
2266 ac->fdsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2267 ac->fdsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2268 ac->fdsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2269 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2270 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2271 memcpy( saved, buf + 512, 448 * sizeof(float));
2272 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2273 } else { // LONG_STOP or ONLY_LONG
2274 memcpy( saved, buf + 512, 512 * sizeof(float));
2279 * Apply dependent channel coupling (applied before IMDCT).
2281 * @param index index into coupling gain array
2283 static void apply_dependent_coupling(AACContext *ac,
2284 SingleChannelElement *target,
2285 ChannelElement *cce, int index)
2287 IndividualChannelStream *ics = &cce->ch[0].ics;
2288 const uint16_t *offsets = ics->swb_offset;
2289 float *dest = target->coeffs;
2290 const float *src = cce->ch[0].coeffs;
2291 int g, i, group, k, idx = 0;
2292 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2293 av_log(ac->avctx, AV_LOG_ERROR,
2294 "Dependent coupling is not supported together with LTP\n");
2297 for (g = 0; g < ics->num_window_groups; g++) {
2298 for (i = 0; i < ics->max_sfb; i++, idx++) {
2299 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2300 const float gain = cce->coup.gain[index][idx];
2301 for (group = 0; group < ics->group_len[g]; group++) {
2302 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2304 dest[group * 128 + k] += gain * src[group * 128 + k];
2309 dest += ics->group_len[g] * 128;
2310 src += ics->group_len[g] * 128;
2315 * Apply independent channel coupling (applied after IMDCT).
2317 * @param index index into coupling gain array
2319 static void apply_independent_coupling(AACContext *ac,
2320 SingleChannelElement *target,
2321 ChannelElement *cce, int index)
2324 const float gain = cce->coup.gain[index][0];
2325 const float *src = cce->ch[0].ret;
2326 float *dest = target->ret;
2327 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2329 for (i = 0; i < len; i++)
2330 dest[i] += gain * src[i];
2334 * channel coupling transformation interface
2336 * @param apply_coupling_method pointer to (in)dependent coupling function
2338 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2339 enum RawDataBlockType type, int elem_id,
2340 enum CouplingPoint coupling_point,
2341 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2345 for (i = 0; i < MAX_ELEM_ID; i++) {
2346 ChannelElement *cce = ac->che[TYPE_CCE][i];
2349 if (cce && cce->coup.coupling_point == coupling_point) {
2350 ChannelCoupling *coup = &cce->coup;
2352 for (c = 0; c <= coup->num_coupled; c++) {
2353 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2354 if (coup->ch_select[c] != 1) {
2355 apply_coupling_method(ac, &cc->ch[0], cce, index);
2356 if (coup->ch_select[c] != 0)
2359 if (coup->ch_select[c] != 2)
2360 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2362 index += 1 + (coup->ch_select[c] == 3);
2369 * Convert spectral data to float samples, applying all supported tools as appropriate.
2371 static void spectral_to_sample(AACContext *ac)
2374 for (type = 3; type >= 0; type--) {
2375 for (i = 0; i < MAX_ELEM_ID; i++) {
2376 ChannelElement *che = ac->che[type][i];
2378 if (type <= TYPE_CPE)
2379 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2380 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2381 if (che->ch[0].ics.predictor_present) {
2382 if (che->ch[0].ics.ltp.present)
2383 apply_ltp(ac, &che->ch[0]);
2384 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2385 apply_ltp(ac, &che->ch[1]);
2388 if (che->ch[0].tns.present)
2389 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2390 if (che->ch[1].tns.present)
2391 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2392 if (type <= TYPE_CPE)
2393 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2394 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2395 imdct_and_windowing(ac, &che->ch[0]);
2396 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2397 update_ltp(ac, &che->ch[0]);
2398 if (type == TYPE_CPE) {
2399 imdct_and_windowing(ac, &che->ch[1]);
2400 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2401 update_ltp(ac, &che->ch[1]);
2403 if (ac->oc[1].m4ac.sbr > 0) {
2404 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2407 if (type <= TYPE_CCE)
2408 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2414 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2417 AACADTSHeaderInfo hdr_info;
2418 uint8_t layout_map[MAX_ELEM_ID*4][3];
2419 int layout_map_tags;
2421 size = avpriv_aac_parse_header(gb, &hdr_info);
2423 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2424 // This is 2 for "VLB " audio in NSV files.
2425 // See samples/nsv/vlb_audio.
2426 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame", 0);
2427 ac->warned_num_aac_frames = 1;
2429 push_output_configuration(ac);
2430 if (hdr_info.chan_config) {
2431 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2432 if (set_default_channel_config(ac->avctx, layout_map,
2433 &layout_map_tags, hdr_info.chan_config))
2435 if (output_configure(ac, layout_map, layout_map_tags,
2436 FFMAX(ac->oc[1].status, OC_TRIAL_FRAME), 0))
2439 ac->oc[1].m4ac.chan_config = 0;
2441 * dual mono frames in Japanese DTV can have chan_config 0
2442 * WITHOUT specifying PCE.
2443 * thus, set dual mono as default.
2445 if (ac->dmono_mode && ac->oc[0].status == OC_NONE) {
2446 layout_map_tags = 2;
2447 layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
2448 layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
2449 layout_map[0][1] = 0;
2450 layout_map[1][1] = 1;
2451 if (output_configure(ac, layout_map, layout_map_tags,
2456 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2457 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2458 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2459 if (ac->oc[0].status != OC_LOCKED ||
2460 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2461 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2462 ac->oc[1].m4ac.sbr = -1;
2463 ac->oc[1].m4ac.ps = -1;
2465 if (!hdr_info.crc_absent)
2471 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2472 int *got_frame_ptr, GetBitContext *gb, AVPacket *avpkt)
2474 AACContext *ac = avctx->priv_data;
2475 ChannelElement *che = NULL, *che_prev = NULL;
2476 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2478 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2479 int is_dmono, sce_count = 0;
2481 if (show_bits(gb, 12) == 0xfff) {
2482 if (parse_adts_frame_header(ac, gb) < 0) {
2483 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2487 if (ac->oc[1].m4ac.sampling_index > 12) {
2488 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2494 if (frame_configure_elements(avctx) < 0) {
2499 ac->tags_mapped = 0;
2501 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2502 elem_id = get_bits(gb, 4);
2504 if (elem_type < TYPE_DSE) {
2505 if (!(che=get_che(ac, elem_type, elem_id))) {
2506 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2507 elem_type, elem_id);
2514 switch (elem_type) {
2517 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2523 err = decode_cpe(ac, gb, che);
2528 err = decode_cce(ac, gb, che);
2532 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2537 err = skip_data_stream_element(ac, gb);
2541 uint8_t layout_map[MAX_ELEM_ID*4][3];
2543 push_output_configuration(ac);
2544 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
2550 av_log(avctx, AV_LOG_ERROR,
2551 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2552 pop_output_configuration(ac);
2554 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
2556 ac->oc[1].m4ac.chan_config = 0;
2564 elem_id += get_bits(gb, 8) - 1;
2565 if (get_bits_left(gb) < 8 * elem_id) {
2566 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
2571 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2572 err = 0; /* FIXME */
2576 err = -1; /* should not happen, but keeps compiler happy */
2581 elem_type_prev = elem_type;
2586 if (get_bits_left(gb) < 3) {
2587 av_log(avctx, AV_LOG_ERROR, overread_err);
2593 spectral_to_sample(ac);
2595 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
2596 samples <<= multiplier;
2597 /* for dual-mono audio (SCE + SCE) */
2598 is_dmono = ac->dmono_mode && sce_count == 2 &&
2599 ac->oc[1].channel_layout == (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT);
2602 ac->frame.nb_samples = samples;
2603 *(AVFrame *)data = ac->frame;
2605 *got_frame_ptr = !!samples;
2608 if (ac->dmono_mode == 1)
2609 ((AVFrame *)data)->data[1] =((AVFrame *)data)->data[0];
2610 else if (ac->dmono_mode == 2)
2611 ((AVFrame *)data)->data[0] =((AVFrame *)data)->data[1];
2614 if (ac->oc[1].status && audio_found) {
2615 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
2616 avctx->frame_size = samples;
2617 ac->oc[1].status = OC_LOCKED;
2622 uint32_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, &side_size);
2623 if (side && side_size>=4)
2624 AV_WL32(side, 2*AV_RL32(side));
2628 pop_output_configuration(ac);
2632 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2633 int *got_frame_ptr, AVPacket *avpkt)
2635 AACContext *ac = avctx->priv_data;
2636 const uint8_t *buf = avpkt->data;
2637 int buf_size = avpkt->size;
2642 int new_extradata_size;
2643 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2644 AV_PKT_DATA_NEW_EXTRADATA,
2645 &new_extradata_size);
2646 int jp_dualmono_size;
2647 const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
2648 AV_PKT_DATA_JP_DUALMONO,
2651 if (new_extradata && 0) {
2652 av_free(avctx->extradata);
2653 avctx->extradata = av_mallocz(new_extradata_size +
2654 FF_INPUT_BUFFER_PADDING_SIZE);
2655 if (!avctx->extradata)
2656 return AVERROR(ENOMEM);
2657 avctx->extradata_size = new_extradata_size;
2658 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2659 push_output_configuration(ac);
2660 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
2662 avctx->extradata_size*8, 1) < 0) {
2663 pop_output_configuration(ac);
2664 return AVERROR_INVALIDDATA;
2669 if (jp_dualmono && jp_dualmono_size > 0)
2670 ac->dmono_mode = 1 + *jp_dualmono;
2671 if (ac->force_dmono_mode >= 0)
2672 ac->dmono_mode = ac->force_dmono_mode;
2674 if (INT_MAX / 8 <= buf_size)
2675 return AVERROR_INVALIDDATA;
2677 init_get_bits(&gb, buf, buf_size * 8);
2679 if ((err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt)) < 0)
2682 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2683 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2684 if (buf[buf_offset])
2687 return buf_size > buf_offset ? buf_consumed : buf_size;
2690 static av_cold int aac_decode_close(AVCodecContext *avctx)
2692 AACContext *ac = avctx->priv_data;
2695 for (i = 0; i < MAX_ELEM_ID; i++) {
2696 for (type = 0; type < 4; type++) {
2697 if (ac->che[type][i])
2698 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2699 av_freep(&ac->che[type][i]);
2703 ff_mdct_end(&ac->mdct);
2704 ff_mdct_end(&ac->mdct_small);
2705 ff_mdct_end(&ac->mdct_ltp);
2710 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2712 struct LATMContext {
2713 AACContext aac_ctx; ///< containing AACContext
2714 int initialized; ///< initialized after a valid extradata was seen
2717 int audio_mux_version_A; ///< LATM syntax version
2718 int frame_length_type; ///< 0/1 variable/fixed frame length
2719 int frame_length; ///< frame length for fixed frame length
2722 static inline uint32_t latm_get_value(GetBitContext *b)
2724 int length = get_bits(b, 2);
2726 return get_bits_long(b, (length+1)*8);
2729 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2730 GetBitContext *gb, int asclen)
2732 AACContext *ac = &latmctx->aac_ctx;
2733 AVCodecContext *avctx = ac->avctx;
2734 MPEG4AudioConfig m4ac = { 0 };
2735 int config_start_bit = get_bits_count(gb);
2736 int sync_extension = 0;
2737 int bits_consumed, esize;
2741 asclen = FFMIN(asclen, get_bits_left(gb));
2743 asclen = get_bits_left(gb);
2745 if (config_start_bit % 8) {
2746 av_log_missing_feature(latmctx->aac_ctx.avctx,
2747 "Non-byte-aligned audio-specific config", 1);
2748 return AVERROR_PATCHWELCOME;
2751 return AVERROR_INVALIDDATA;
2752 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
2753 gb->buffer + (config_start_bit / 8),
2754 asclen, sync_extension);
2756 if (bits_consumed < 0)
2757 return AVERROR_INVALIDDATA;
2759 if (!latmctx->initialized ||
2760 ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
2761 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
2763 if(latmctx->initialized) {
2764 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
2766 av_log(avctx, AV_LOG_DEBUG, "initializing latmctx\n");
2768 latmctx->initialized = 0;
2770 esize = (bits_consumed+7) / 8;
2772 if (avctx->extradata_size < esize) {
2773 av_free(avctx->extradata);
2774 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2775 if (!avctx->extradata)
2776 return AVERROR(ENOMEM);
2779 avctx->extradata_size = esize;
2780 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2781 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2783 skip_bits_long(gb, bits_consumed);
2785 return bits_consumed;
2788 static int read_stream_mux_config(struct LATMContext *latmctx,
2791 int ret, audio_mux_version = get_bits(gb, 1);
2793 latmctx->audio_mux_version_A = 0;
2794 if (audio_mux_version)
2795 latmctx->audio_mux_version_A = get_bits(gb, 1);
2797 if (!latmctx->audio_mux_version_A) {
2799 if (audio_mux_version)
2800 latm_get_value(gb); // taraFullness
2802 skip_bits(gb, 1); // allStreamSameTimeFraming
2803 skip_bits(gb, 6); // numSubFrames
2805 if (get_bits(gb, 4)) { // numPrograms
2806 av_log_missing_feature(latmctx->aac_ctx.avctx,
2807 "Multiple programs", 1);
2808 return AVERROR_PATCHWELCOME;
2811 // for each program (which there is only one in DVB)
2813 // for each layer (which there is only one in DVB)
2814 if (get_bits(gb, 3)) { // numLayer
2815 av_log_missing_feature(latmctx->aac_ctx.avctx,
2816 "Multiple layers", 1);
2817 return AVERROR_PATCHWELCOME;
2820 // for all but first stream: use_same_config = get_bits(gb, 1);
2821 if (!audio_mux_version) {
2822 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
2825 int ascLen = latm_get_value(gb);
2826 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
2829 skip_bits_long(gb, ascLen);
2832 latmctx->frame_length_type = get_bits(gb, 3);
2833 switch (latmctx->frame_length_type) {
2835 skip_bits(gb, 8); // latmBufferFullness
2838 latmctx->frame_length = get_bits(gb, 9);
2843 skip_bits(gb, 6); // CELP frame length table index
2847 skip_bits(gb, 1); // HVXC frame length table index
2851 if (get_bits(gb, 1)) { // other data
2852 if (audio_mux_version) {
2853 latm_get_value(gb); // other_data_bits
2857 esc = get_bits(gb, 1);
2863 if (get_bits(gb, 1)) // crc present
2864 skip_bits(gb, 8); // config_crc
2870 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2874 if (ctx->frame_length_type == 0) {
2875 int mux_slot_length = 0;
2877 tmp = get_bits(gb, 8);
2878 mux_slot_length += tmp;
2879 } while (tmp == 255);
2880 return mux_slot_length;
2881 } else if (ctx->frame_length_type == 1) {
2882 return ctx->frame_length;
2883 } else if (ctx->frame_length_type == 3 ||
2884 ctx->frame_length_type == 5 ||
2885 ctx->frame_length_type == 7) {
2886 skip_bits(gb, 2); // mux_slot_length_coded
2891 static int read_audio_mux_element(struct LATMContext *latmctx,
2895 uint8_t use_same_mux = get_bits(gb, 1);
2896 if (!use_same_mux) {
2897 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2899 } else if (!latmctx->aac_ctx.avctx->extradata) {
2900 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2901 "no decoder config found\n");
2902 return AVERROR(EAGAIN);
2904 if (latmctx->audio_mux_version_A == 0) {
2905 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2906 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2907 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2908 return AVERROR_INVALIDDATA;
2909 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2910 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2911 "frame length mismatch %d << %d\n",
2912 mux_slot_length_bytes * 8, get_bits_left(gb));
2913 return AVERROR_INVALIDDATA;
2920 static int latm_decode_frame(AVCodecContext *avctx, void *out,
2921 int *got_frame_ptr, AVPacket *avpkt)
2923 struct LATMContext *latmctx = avctx->priv_data;
2927 if ((err = init_get_bits8(&gb, avpkt->data, avpkt->size)) < 0)
2930 // check for LOAS sync word
2931 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2932 return AVERROR_INVALIDDATA;
2934 muxlength = get_bits(&gb, 13) + 3;
2935 // not enough data, the parser should have sorted this out
2936 if (muxlength > avpkt->size)
2937 return AVERROR_INVALIDDATA;
2939 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2942 if (!latmctx->initialized) {
2943 if (!avctx->extradata) {
2947 push_output_configuration(&latmctx->aac_ctx);
2948 if ((err = decode_audio_specific_config(
2949 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
2950 avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
2951 pop_output_configuration(&latmctx->aac_ctx);
2954 latmctx->initialized = 1;
2958 if (show_bits(&gb, 12) == 0xfff) {
2959 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2960 "ADTS header detected, probably as result of configuration "
2962 return AVERROR_INVALIDDATA;
2965 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb, avpkt)) < 0)
2971 static av_cold int latm_decode_init(AVCodecContext *avctx)
2973 struct LATMContext *latmctx = avctx->priv_data;
2974 int ret = aac_decode_init(avctx);
2976 if (avctx->extradata_size > 0)
2977 latmctx->initialized = !ret;
2983 * AVOptions for Japanese DTV specific extensions (ADTS only)
2985 #define AACDEC_FLAGS AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
2986 static const AVOption options[] = {
2987 {"dual_mono_mode", "Select the channel to decode for dual mono",
2988 offsetof(AACContext, force_dmono_mode), AV_OPT_TYPE_INT, {.i64=-1}, -1, 2,
2989 AACDEC_FLAGS, "dual_mono_mode"},
2991 {"auto", "autoselection", 0, AV_OPT_TYPE_CONST, {.i64=-1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
2992 {"main", "Select Main/Left channel", 0, AV_OPT_TYPE_CONST, {.i64= 1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
2993 {"sub" , "Select Sub/Right channel", 0, AV_OPT_TYPE_CONST, {.i64= 2}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
2994 {"both", "Select both channels", 0, AV_OPT_TYPE_CONST, {.i64= 0}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
2999 static const AVClass aac_decoder_class = {
3000 .class_name = "AAC decoder",
3001 .item_name = av_default_item_name,
3003 .version = LIBAVUTIL_VERSION_INT,
3006 AVCodec ff_aac_decoder = {
3008 .type = AVMEDIA_TYPE_AUDIO,
3009 .id = AV_CODEC_ID_AAC,
3010 .priv_data_size = sizeof(AACContext),
3011 .init = aac_decode_init,
3012 .close = aac_decode_close,
3013 .decode = aac_decode_frame,
3014 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
3015 .sample_fmts = (const enum AVSampleFormat[]) {
3016 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3018 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3019 .channel_layouts = aac_channel_layout,
3021 .priv_class = &aac_decoder_class,
3025 Note: This decoder filter is intended to decode LATM streams transferred
3026 in MPEG transport streams which only contain one program.
3027 To do a more complex LATM demuxing a separate LATM demuxer should be used.
3029 AVCodec ff_aac_latm_decoder = {
3031 .type = AVMEDIA_TYPE_AUDIO,
3032 .id = AV_CODEC_ID_AAC_LATM,
3033 .priv_data_size = sizeof(struct LATMContext),
3034 .init = latm_decode_init,
3035 .close = aac_decode_close,
3036 .decode = latm_decode_frame,
3037 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
3038 .sample_fmts = (const enum AVSampleFormat[]) {
3039 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3041 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3042 .channel_layouts = aac_channel_layout,