3 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
5 * Copyright (c) 2008-2013 Alex Converse <alex.converse@gmail.com>
8 * Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
9 * Copyright (c) 2010 Janne Grunau <janne-libav@jannau.net>
11 * AAC decoder fixed-point implementation
13 * MIPS Technologies, Inc., California.
15 * This file is part of FFmpeg.
17 * FFmpeg is free software; you can redistribute it and/or
18 * modify it under the terms of the GNU Lesser General Public
19 * License as published by the Free Software Foundation; either
20 * version 2.1 of the License, or (at your option) any later version.
22 * FFmpeg is distributed in the hope that it will be useful,
23 * but WITHOUT ANY WARRANTY; without even the implied warranty of
24 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
25 * Lesser General Public License for more details.
27 * You should have received a copy of the GNU Lesser General Public
28 * License along with FFmpeg; if not, write to the Free Software
29 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
35 * @author Oded Shimon ( ods15 ods15 dyndns org )
36 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
38 * AAC decoder fixed-point implementation
39 * @author Stanislav Ocovaj ( stanislav.ocovaj imgtec com )
40 * @author Nedeljko Babic ( nedeljko.babic imgtec com )
47 * N (code in SoC repo) gain control
49 * Y window shapes - standard
50 * N window shapes - Low Delay
51 * Y filterbank - standard
52 * N (code in SoC repo) filterbank - Scalable Sample Rate
53 * Y Temporal Noise Shaping
54 * Y Long Term Prediction
57 * Y frequency domain prediction
58 * Y Perceptual Noise Substitution
60 * N Scalable Inverse AAC Quantization
61 * N Frequency Selective Switch
63 * Y quantization & coding - AAC
64 * N quantization & coding - TwinVQ
65 * N quantization & coding - BSAC
66 * N AAC Error Resilience tools
67 * N Error Resilience payload syntax
68 * N Error Protection tool
70 * N Silence Compression
73 * N Structured Audio tools
74 * N Structured Audio Sample Bank Format
76 * N Harmonic and Individual Lines plus Noise
77 * N Text-To-Speech Interface
78 * Y Spectral Band Replication
79 * Y (not in this code) Layer-1
80 * Y (not in this code) Layer-2
81 * Y (not in this code) Layer-3
82 * N SinuSoidal Coding (Transient, Sinusoid, Noise)
84 * N Direct Stream Transfer
85 * Y (not in fixed point code) Enhanced AAC Low Delay (ER AAC ELD)
87 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
88 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
92 #include "libavutil/thread.h"
94 static VLC vlc_scalefactors;
95 static VLC vlc_spectral[11];
97 static int output_configure(AACContext *ac,
98 uint8_t layout_map[MAX_ELEM_ID*4][3], int tags,
99 enum OCStatus oc_type, int get_new_frame);
101 #define overread_err "Input buffer exhausted before END element found\n"
103 static int count_channels(uint8_t (*layout)[3], int tags)
106 for (i = 0; i < tags; i++) {
107 int syn_ele = layout[i][0];
108 int pos = layout[i][2];
109 sum += (1 + (syn_ele == TYPE_CPE)) *
110 (pos != AAC_CHANNEL_OFF && pos != AAC_CHANNEL_CC);
116 * Check for the channel element in the current channel position configuration.
117 * If it exists, make sure the appropriate element is allocated and map the
118 * channel order to match the internal FFmpeg channel layout.
120 * @param che_pos current channel position configuration
121 * @param type channel element type
122 * @param id channel element id
123 * @param channels count of the number of channels in the configuration
125 * @return Returns error status. 0 - OK, !0 - error
127 static av_cold int che_configure(AACContext *ac,
128 enum ChannelPosition che_pos,
129 int type, int id, int *channels)
131 if (*channels >= MAX_CHANNELS)
132 return AVERROR_INVALIDDATA;
134 if (!ac->che[type][id]) {
135 if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
136 return AVERROR(ENOMEM);
137 AAC_RENAME(ff_aac_sbr_ctx_init)(ac, &ac->che[type][id]->sbr, type);
139 if (type != TYPE_CCE) {
140 if (*channels >= MAX_CHANNELS - (type == TYPE_CPE || (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1))) {
141 av_log(ac->avctx, AV_LOG_ERROR, "Too many channels\n");
142 return AVERROR_INVALIDDATA;
144 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[0];
145 if (type == TYPE_CPE ||
146 (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1)) {
147 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[1];
151 if (ac->che[type][id])
152 AAC_RENAME(ff_aac_sbr_ctx_close)(&ac->che[type][id]->sbr);
153 av_freep(&ac->che[type][id]);
158 static int frame_configure_elements(AVCodecContext *avctx)
160 AACContext *ac = avctx->priv_data;
161 int type, id, ch, ret;
163 /* set channel pointers to internal buffers by default */
164 for (type = 0; type < 4; type++) {
165 for (id = 0; id < MAX_ELEM_ID; id++) {
166 ChannelElement *che = ac->che[type][id];
168 che->ch[0].ret = che->ch[0].ret_buf;
169 che->ch[1].ret = che->ch[1].ret_buf;
174 /* get output buffer */
175 av_frame_unref(ac->frame);
176 if (!avctx->channels)
179 ac->frame->nb_samples = 2048;
180 if ((ret = ff_get_buffer(avctx, ac->frame, 0)) < 0)
183 /* map output channel pointers to AVFrame data */
184 for (ch = 0; ch < avctx->channels; ch++) {
185 if (ac->output_element[ch])
186 ac->output_element[ch]->ret = (INTFLOAT *)ac->frame->extended_data[ch];
192 struct elem_to_channel {
193 uint64_t av_position;
196 uint8_t aac_position;
199 static int assign_pair(struct elem_to_channel e2c_vec[MAX_ELEM_ID],
200 uint8_t (*layout_map)[3], int offset, uint64_t left,
201 uint64_t right, int pos)
203 if (layout_map[offset][0] == TYPE_CPE) {
204 e2c_vec[offset] = (struct elem_to_channel) {
205 .av_position = left | right,
207 .elem_id = layout_map[offset][1],
212 e2c_vec[offset] = (struct elem_to_channel) {
215 .elem_id = layout_map[offset][1],
218 e2c_vec[offset + 1] = (struct elem_to_channel) {
219 .av_position = right,
221 .elem_id = layout_map[offset + 1][1],
228 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos,
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)
284 if (num_side_channels == 0 && num_back_channels >= 4) {
285 num_side_channels = 2;
286 num_back_channels -= 2;
290 if (num_front_channels & 1) {
291 e2c_vec[i] = (struct elem_to_channel) {
292 .av_position = AV_CH_FRONT_CENTER,
294 .elem_id = layout_map[i][1],
295 .aac_position = AAC_CHANNEL_FRONT
298 num_front_channels--;
300 if (num_front_channels >= 4) {
301 i += assign_pair(e2c_vec, layout_map, i,
302 AV_CH_FRONT_LEFT_OF_CENTER,
303 AV_CH_FRONT_RIGHT_OF_CENTER,
305 num_front_channels -= 2;
307 if (num_front_channels >= 2) {
308 i += assign_pair(e2c_vec, layout_map, i,
312 num_front_channels -= 2;
314 while (num_front_channels >= 2) {
315 i += assign_pair(e2c_vec, layout_map, i,
319 num_front_channels -= 2;
322 if (num_side_channels >= 2) {
323 i += assign_pair(e2c_vec, layout_map, i,
327 num_side_channels -= 2;
329 while (num_side_channels >= 2) {
330 i += assign_pair(e2c_vec, layout_map, i,
334 num_side_channels -= 2;
337 while (num_back_channels >= 4) {
338 i += assign_pair(e2c_vec, layout_map, i,
342 num_back_channels -= 2;
344 if (num_back_channels >= 2) {
345 i += assign_pair(e2c_vec, layout_map, i,
349 num_back_channels -= 2;
351 if (num_back_channels) {
352 e2c_vec[i] = (struct elem_to_channel) {
353 .av_position = AV_CH_BACK_CENTER,
355 .elem_id = layout_map[i][1],
356 .aac_position = AAC_CHANNEL_BACK
362 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
363 e2c_vec[i] = (struct elem_to_channel) {
364 .av_position = AV_CH_LOW_FREQUENCY,
366 .elem_id = layout_map[i][1],
367 .aac_position = AAC_CHANNEL_LFE
371 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
372 e2c_vec[i] = (struct elem_to_channel) {
373 .av_position = UINT64_MAX,
375 .elem_id = layout_map[i][1],
376 .aac_position = AAC_CHANNEL_LFE
381 // Must choose a stable sort
382 total_non_cc_elements = n = i;
385 for (i = 1; i < n; i++)
386 if (e2c_vec[i - 1].av_position > e2c_vec[i].av_position) {
387 FFSWAP(struct elem_to_channel, e2c_vec[i - 1], e2c_vec[i]);
394 for (i = 0; i < total_non_cc_elements; i++) {
395 layout_map[i][0] = e2c_vec[i].syn_ele;
396 layout_map[i][1] = e2c_vec[i].elem_id;
397 layout_map[i][2] = e2c_vec[i].aac_position;
398 if (e2c_vec[i].av_position != UINT64_MAX) {
399 layout |= e2c_vec[i].av_position;
407 * Save current output configuration if and only if it has been locked.
409 static void push_output_configuration(AACContext *ac) {
410 if (ac->oc[1].status == OC_LOCKED || ac->oc[0].status == OC_NONE) {
411 ac->oc[0] = ac->oc[1];
413 ac->oc[1].status = OC_NONE;
417 * Restore the previous output configuration if and only if the current
418 * configuration is unlocked.
420 static void pop_output_configuration(AACContext *ac) {
421 if (ac->oc[1].status != OC_LOCKED && ac->oc[0].status != OC_NONE) {
422 ac->oc[1] = ac->oc[0];
423 ac->avctx->channels = ac->oc[1].channels;
424 ac->avctx->channel_layout = ac->oc[1].channel_layout;
425 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
426 ac->oc[1].status, 0);
431 * Configure output channel order based on the current program
432 * configuration element.
434 * @return Returns error status. 0 - OK, !0 - error
436 static int output_configure(AACContext *ac,
437 uint8_t layout_map[MAX_ELEM_ID * 4][3], int tags,
438 enum OCStatus oc_type, int get_new_frame)
440 AVCodecContext *avctx = ac->avctx;
441 int i, channels = 0, ret;
443 uint8_t id_map[TYPE_END][MAX_ELEM_ID] = {{ 0 }};
444 uint8_t type_counts[TYPE_END] = { 0 };
446 if (ac->oc[1].layout_map != layout_map) {
447 memcpy(ac->oc[1].layout_map, layout_map, tags * sizeof(layout_map[0]));
448 ac->oc[1].layout_map_tags = tags;
450 for (i = 0; i < tags; i++) {
451 int type = layout_map[i][0];
452 int id = layout_map[i][1];
453 id_map[type][id] = type_counts[type]++;
454 if (id_map[type][id] >= MAX_ELEM_ID) {
455 avpriv_request_sample(ac->avctx, "Too large remapped id");
456 return AVERROR_PATCHWELCOME;
459 // Try to sniff a reasonable channel order, otherwise output the
460 // channels in the order the PCE declared them.
461 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
462 layout = sniff_channel_order(layout_map, tags);
463 for (i = 0; i < tags; i++) {
464 int type = layout_map[i][0];
465 int id = layout_map[i][1];
466 int iid = id_map[type][id];
467 int position = layout_map[i][2];
468 // Allocate or free elements depending on if they are in the
469 // current program configuration.
470 ret = che_configure(ac, position, type, iid, &channels);
473 ac->tag_che_map[type][id] = ac->che[type][iid];
475 if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
476 if (layout == AV_CH_FRONT_CENTER) {
477 layout = AV_CH_FRONT_LEFT|AV_CH_FRONT_RIGHT;
483 if (layout) avctx->channel_layout = layout;
484 ac->oc[1].channel_layout = layout;
485 avctx->channels = ac->oc[1].channels = channels;
486 ac->oc[1].status = oc_type;
489 if ((ret = frame_configure_elements(ac->avctx)) < 0)
496 static void flush(AVCodecContext *avctx)
498 AACContext *ac= avctx->priv_data;
501 for (type = 3; type >= 0; type--) {
502 for (i = 0; i < MAX_ELEM_ID; i++) {
503 ChannelElement *che = ac->che[type][i];
505 for (j = 0; j <= 1; j++) {
506 memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
514 * Set up channel positions based on a default channel configuration
515 * as specified in table 1.17.
517 * @return Returns error status. 0 - OK, !0 - error
519 static int set_default_channel_config(AVCodecContext *avctx,
520 uint8_t (*layout_map)[3],
524 if (channel_config < 1 || (channel_config > 7 && channel_config < 11) ||
525 channel_config > 12) {
526 av_log(avctx, AV_LOG_ERROR,
527 "invalid default channel configuration (%d)\n",
529 return AVERROR_INVALIDDATA;
531 *tags = tags_per_config[channel_config];
532 memcpy(layout_map, aac_channel_layout_map[channel_config - 1],
533 *tags * sizeof(*layout_map));
536 * AAC specification has 7.1(wide) as a default layout for 8-channel streams.
537 * However, at least Nero AAC encoder encodes 7.1 streams using the default
538 * channel config 7, mapping the side channels of the original audio stream
539 * to the second AAC_CHANNEL_FRONT pair in the AAC stream. Similarly, e.g. FAAD
540 * decodes the second AAC_CHANNEL_FRONT pair as side channels, therefore decoding
541 * the incorrect streams as if they were correct (and as the encoder intended).
543 * As actual intended 7.1(wide) streams are very rare, default to assuming a
544 * 7.1 layout was intended.
546 if (channel_config == 7 && avctx->strict_std_compliance < FF_COMPLIANCE_STRICT) {
547 av_log(avctx, AV_LOG_INFO, "Assuming an incorrectly encoded 7.1 channel layout"
548 " instead of a spec-compliant 7.1(wide) layout, use -strict %d to decode"
549 " according to the specification instead.\n", FF_COMPLIANCE_STRICT);
550 layout_map[2][2] = AAC_CHANNEL_SIDE;
556 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
558 /* For PCE based channel configurations map the channels solely based
560 if (!ac->oc[1].m4ac.chan_config) {
561 return ac->tag_che_map[type][elem_id];
563 // Allow single CPE stereo files to be signalled with mono configuration.
564 if (!ac->tags_mapped && type == TYPE_CPE &&
565 ac->oc[1].m4ac.chan_config == 1) {
566 uint8_t layout_map[MAX_ELEM_ID*4][3];
568 push_output_configuration(ac);
570 av_log(ac->avctx, AV_LOG_DEBUG, "mono with CPE\n");
572 if (set_default_channel_config(ac->avctx, layout_map,
573 &layout_map_tags, 2) < 0)
575 if (output_configure(ac, layout_map, layout_map_tags,
576 OC_TRIAL_FRAME, 1) < 0)
579 ac->oc[1].m4ac.chan_config = 2;
580 ac->oc[1].m4ac.ps = 0;
583 if (!ac->tags_mapped && type == TYPE_SCE &&
584 ac->oc[1].m4ac.chan_config == 2) {
585 uint8_t layout_map[MAX_ELEM_ID * 4][3];
587 push_output_configuration(ac);
589 av_log(ac->avctx, AV_LOG_DEBUG, "stereo with SCE\n");
591 if (set_default_channel_config(ac->avctx, layout_map,
592 &layout_map_tags, 1) < 0)
594 if (output_configure(ac, layout_map, layout_map_tags,
595 OC_TRIAL_FRAME, 1) < 0)
598 ac->oc[1].m4ac.chan_config = 1;
599 if (ac->oc[1].m4ac.sbr)
600 ac->oc[1].m4ac.ps = -1;
602 /* For indexed channel configurations map the channels solely based
604 switch (ac->oc[1].m4ac.chan_config) {
607 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
609 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
612 if (ac->tags_mapped == 2 &&
613 ac->oc[1].m4ac.chan_config == 11 &&
616 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
619 /* Some streams incorrectly code 5.1 audio as
620 * SCE[0] CPE[0] CPE[1] SCE[1]
622 * SCE[0] CPE[0] CPE[1] LFE[0].
623 * If we seem to have encountered such a stream, transfer
624 * the LFE[0] element to the SCE[1]'s mapping */
625 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
626 if (!ac->warned_remapping_once && (type != TYPE_LFE || elem_id != 0)) {
627 av_log(ac->avctx, AV_LOG_WARNING,
628 "This stream seems to incorrectly report its last channel as %s[%d], mapping to LFE[0]\n",
629 type == TYPE_SCE ? "SCE" : "LFE", elem_id);
630 ac->warned_remapping_once++;
633 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
636 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
638 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
641 /* Some streams incorrectly code 4.0 audio as
642 * SCE[0] CPE[0] LFE[0]
644 * SCE[0] CPE[0] SCE[1].
645 * If we seem to have encountered such a stream, transfer
646 * the SCE[1] element to the LFE[0]'s mapping */
647 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
648 if (!ac->warned_remapping_once && (type != TYPE_SCE || elem_id != 1)) {
649 av_log(ac->avctx, AV_LOG_WARNING,
650 "This stream seems to incorrectly report its last channel as %s[%d], mapping to SCE[1]\n",
651 type == TYPE_SCE ? "SCE" : "LFE", elem_id);
652 ac->warned_remapping_once++;
655 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_SCE][1];
657 if (ac->tags_mapped == 2 &&
658 ac->oc[1].m4ac.chan_config == 4 &&
661 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
665 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) &&
668 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
669 } else if (ac->oc[1].m4ac.chan_config == 2) {
673 if (!ac->tags_mapped && type == TYPE_SCE) {
675 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
683 * Decode an array of 4 bit element IDs, optionally interleaved with a
684 * stereo/mono switching bit.
686 * @param type speaker type/position for these channels
688 static void decode_channel_map(uint8_t layout_map[][3],
689 enum ChannelPosition type,
690 GetBitContext *gb, int n)
693 enum RawDataBlockType syn_ele;
695 case AAC_CHANNEL_FRONT:
696 case AAC_CHANNEL_BACK:
697 case AAC_CHANNEL_SIDE:
698 syn_ele = get_bits1(gb);
704 case AAC_CHANNEL_LFE:
708 // AAC_CHANNEL_OFF has no channel map
711 layout_map[0][0] = syn_ele;
712 layout_map[0][1] = get_bits(gb, 4);
713 layout_map[0][2] = type;
718 static inline void relative_align_get_bits(GetBitContext *gb,
719 int reference_position) {
720 int n = (reference_position - get_bits_count(gb) & 7);
726 * Decode program configuration element; reference: table 4.2.
728 * @return Returns error status. 0 - OK, !0 - error
730 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
731 uint8_t (*layout_map)[3],
732 GetBitContext *gb, int byte_align_ref)
734 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc;
739 skip_bits(gb, 2); // object_type
741 sampling_index = get_bits(gb, 4);
742 if (m4ac->sampling_index != sampling_index)
743 av_log(avctx, AV_LOG_WARNING,
744 "Sample rate index in program config element does not "
745 "match the sample rate index configured by the container.\n");
747 num_front = get_bits(gb, 4);
748 num_side = get_bits(gb, 4);
749 num_back = get_bits(gb, 4);
750 num_lfe = get_bits(gb, 2);
751 num_assoc_data = get_bits(gb, 3);
752 num_cc = get_bits(gb, 4);
755 skip_bits(gb, 4); // mono_mixdown_tag
757 skip_bits(gb, 4); // stereo_mixdown_tag
760 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
762 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
763 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
766 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
768 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
770 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
772 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
775 skip_bits_long(gb, 4 * num_assoc_data);
777 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
780 relative_align_get_bits(gb, byte_align_ref);
782 /* comment field, first byte is length */
783 comment_len = get_bits(gb, 8) * 8;
784 if (get_bits_left(gb) < comment_len) {
785 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
786 return AVERROR_INVALIDDATA;
788 skip_bits_long(gb, comment_len);
793 * Decode GA "General Audio" specific configuration; reference: table 4.1.
795 * @param ac pointer to AACContext, may be null
796 * @param avctx pointer to AVCCodecContext, used for logging
798 * @return Returns error status. 0 - OK, !0 - error
800 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
802 int get_bit_alignment,
803 MPEG4AudioConfig *m4ac,
806 int extension_flag, ret, ep_config, res_flags;
807 uint8_t layout_map[MAX_ELEM_ID*4][3];
810 if (get_bits1(gb)) { // frameLengthFlag
811 avpriv_request_sample(avctx, "960/120 MDCT window");
812 return AVERROR_PATCHWELCOME;
814 m4ac->frame_length_short = 0;
816 if (get_bits1(gb)) // dependsOnCoreCoder
817 skip_bits(gb, 14); // coreCoderDelay
818 extension_flag = get_bits1(gb);
820 if (m4ac->object_type == AOT_AAC_SCALABLE ||
821 m4ac->object_type == AOT_ER_AAC_SCALABLE)
822 skip_bits(gb, 3); // layerNr
824 if (channel_config == 0) {
825 skip_bits(gb, 4); // element_instance_tag
826 tags = decode_pce(avctx, m4ac, layout_map, gb, get_bit_alignment);
830 if ((ret = set_default_channel_config(avctx, layout_map,
831 &tags, channel_config)))
835 if (count_channels(layout_map, tags) > 1) {
837 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
840 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
843 if (extension_flag) {
844 switch (m4ac->object_type) {
846 skip_bits(gb, 5); // numOfSubFrame
847 skip_bits(gb, 11); // layer_length
851 case AOT_ER_AAC_SCALABLE:
853 res_flags = get_bits(gb, 3);
855 avpriv_report_missing_feature(avctx,
856 "AAC data resilience (flags %x)",
858 return AVERROR_PATCHWELCOME;
862 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
864 switch (m4ac->object_type) {
867 case AOT_ER_AAC_SCALABLE:
869 ep_config = get_bits(gb, 2);
871 avpriv_report_missing_feature(avctx,
872 "epConfig %d", ep_config);
873 return AVERROR_PATCHWELCOME;
879 static int decode_eld_specific_config(AACContext *ac, AVCodecContext *avctx,
881 MPEG4AudioConfig *m4ac,
884 int ret, ep_config, res_flags;
885 uint8_t layout_map[MAX_ELEM_ID*4][3];
887 const int ELDEXT_TERM = 0;
892 if (get_bits1(gb)) { // frameLengthFlag
893 avpriv_request_sample(avctx, "960/120 MDCT window");
894 return AVERROR_PATCHWELCOME;
897 m4ac->frame_length_short = get_bits1(gb);
899 res_flags = get_bits(gb, 3);
901 avpriv_report_missing_feature(avctx,
902 "AAC data resilience (flags %x)",
904 return AVERROR_PATCHWELCOME;
907 if (get_bits1(gb)) { // ldSbrPresentFlag
908 avpriv_report_missing_feature(avctx,
910 return AVERROR_PATCHWELCOME;
913 while (get_bits(gb, 4) != ELDEXT_TERM) {
914 int len = get_bits(gb, 4);
916 len += get_bits(gb, 8);
918 len += get_bits(gb, 16);
919 if (get_bits_left(gb) < len * 8 + 4) {
920 av_log(avctx, AV_LOG_ERROR, overread_err);
921 return AVERROR_INVALIDDATA;
923 skip_bits_long(gb, 8 * len);
926 if ((ret = set_default_channel_config(avctx, layout_map,
927 &tags, channel_config)))
930 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
933 ep_config = get_bits(gb, 2);
935 avpriv_report_missing_feature(avctx,
936 "epConfig %d", ep_config);
937 return AVERROR_PATCHWELCOME;
943 * Decode audio specific configuration; reference: table 1.13.
945 * @param ac pointer to AACContext, may be null
946 * @param avctx pointer to AVCCodecContext, used for logging
947 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
948 * @param gb buffer holding an audio specific config
949 * @param get_bit_alignment relative alignment for byte align operations
950 * @param sync_extension look for an appended sync extension
952 * @return Returns error status or number of consumed bits. <0 - error
954 static int decode_audio_specific_config_gb(AACContext *ac,
955 AVCodecContext *avctx,
956 MPEG4AudioConfig *m4ac,
958 int get_bit_alignment,
962 GetBitContext gbc = *gb;
964 if ((i = ff_mpeg4audio_get_config_gb(m4ac, &gbc, sync_extension)) < 0)
965 return AVERROR_INVALIDDATA;
967 if (m4ac->sampling_index > 12) {
968 av_log(avctx, AV_LOG_ERROR,
969 "invalid sampling rate index %d\n",
970 m4ac->sampling_index);
971 return AVERROR_INVALIDDATA;
973 if (m4ac->object_type == AOT_ER_AAC_LD &&
974 (m4ac->sampling_index < 3 || m4ac->sampling_index > 7)) {
975 av_log(avctx, AV_LOG_ERROR,
976 "invalid low delay sampling rate index %d\n",
977 m4ac->sampling_index);
978 return AVERROR_INVALIDDATA;
981 skip_bits_long(gb, i);
983 switch (m4ac->object_type) {
989 if ((ret = decode_ga_specific_config(ac, avctx, gb, get_bit_alignment,
990 m4ac, m4ac->chan_config)) < 0)
994 if ((ret = decode_eld_specific_config(ac, avctx, gb,
995 m4ac, m4ac->chan_config)) < 0)
999 avpriv_report_missing_feature(avctx,
1000 "Audio object type %s%d",
1001 m4ac->sbr == 1 ? "SBR+" : "",
1003 return AVERROR(ENOSYS);
1007 "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
1008 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
1009 m4ac->sample_rate, m4ac->sbr,
1012 return get_bits_count(gb);
1015 static int decode_audio_specific_config(AACContext *ac,
1016 AVCodecContext *avctx,
1017 MPEG4AudioConfig *m4ac,
1018 const uint8_t *data, int64_t bit_size,
1024 if (bit_size < 0 || bit_size > INT_MAX) {
1025 av_log(avctx, AV_LOG_ERROR, "Audio specific config size is invalid\n");
1026 return AVERROR_INVALIDDATA;
1029 ff_dlog(avctx, "audio specific config size %d\n", (int)bit_size >> 3);
1030 for (i = 0; i < bit_size >> 3; i++)
1031 ff_dlog(avctx, "%02x ", data[i]);
1032 ff_dlog(avctx, "\n");
1034 if ((ret = init_get_bits(&gb, data, bit_size)) < 0)
1037 return decode_audio_specific_config_gb(ac, avctx, m4ac, &gb, 0,
1042 * linear congruential pseudorandom number generator
1044 * @param previous_val pointer to the current state of the generator
1046 * @return Returns a 32-bit pseudorandom integer
1048 static av_always_inline int lcg_random(unsigned previous_val)
1050 union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
1054 static void reset_all_predictors(PredictorState *ps)
1057 for (i = 0; i < MAX_PREDICTORS; i++)
1058 reset_predict_state(&ps[i]);
1061 static int sample_rate_idx (int rate)
1063 if (92017 <= rate) return 0;
1064 else if (75132 <= rate) return 1;
1065 else if (55426 <= rate) return 2;
1066 else if (46009 <= rate) return 3;
1067 else if (37566 <= rate) return 4;
1068 else if (27713 <= rate) return 5;
1069 else if (23004 <= rate) return 6;
1070 else if (18783 <= rate) return 7;
1071 else if (13856 <= rate) return 8;
1072 else if (11502 <= rate) return 9;
1073 else if (9391 <= rate) return 10;
1077 static void reset_predictor_group(PredictorState *ps, int group_num)
1080 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
1081 reset_predict_state(&ps[i]);
1084 #define AAC_INIT_VLC_STATIC(num, size) \
1085 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
1086 ff_aac_spectral_bits[num], sizeof(ff_aac_spectral_bits[num][0]), \
1087 sizeof(ff_aac_spectral_bits[num][0]), \
1088 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), \
1089 sizeof(ff_aac_spectral_codes[num][0]), \
1092 static void aacdec_init(AACContext *ac);
1094 static av_cold void aac_static_table_init(void)
1096 AAC_INIT_VLC_STATIC( 0, 304);
1097 AAC_INIT_VLC_STATIC( 1, 270);
1098 AAC_INIT_VLC_STATIC( 2, 550);
1099 AAC_INIT_VLC_STATIC( 3, 300);
1100 AAC_INIT_VLC_STATIC( 4, 328);
1101 AAC_INIT_VLC_STATIC( 5, 294);
1102 AAC_INIT_VLC_STATIC( 6, 306);
1103 AAC_INIT_VLC_STATIC( 7, 268);
1104 AAC_INIT_VLC_STATIC( 8, 510);
1105 AAC_INIT_VLC_STATIC( 9, 366);
1106 AAC_INIT_VLC_STATIC(10, 462);
1108 AAC_RENAME(ff_aac_sbr_init)();
1112 INIT_VLC_STATIC(&vlc_scalefactors, 7,
1113 FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
1114 ff_aac_scalefactor_bits,
1115 sizeof(ff_aac_scalefactor_bits[0]),
1116 sizeof(ff_aac_scalefactor_bits[0]),
1117 ff_aac_scalefactor_code,
1118 sizeof(ff_aac_scalefactor_code[0]),
1119 sizeof(ff_aac_scalefactor_code[0]),
1122 // window initialization
1123 AAC_RENAME(ff_kbd_window_init)(AAC_RENAME(ff_aac_kbd_long_1024), 4.0, 1024);
1124 AAC_RENAME(ff_kbd_window_init)(AAC_RENAME(ff_aac_kbd_short_128), 6.0, 128);
1125 AAC_RENAME(ff_init_ff_sine_windows)(10);
1126 AAC_RENAME(ff_init_ff_sine_windows)( 9);
1127 AAC_RENAME(ff_init_ff_sine_windows)( 7);
1129 AAC_RENAME(ff_cbrt_tableinit)();
1132 static AVOnce aac_table_init = AV_ONCE_INIT;
1134 static av_cold int aac_decode_init(AVCodecContext *avctx)
1136 AACContext *ac = avctx->priv_data;
1139 ret = ff_thread_once(&aac_table_init, &aac_static_table_init);
1141 return AVERROR_UNKNOWN;
1144 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
1148 avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
1150 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1151 #endif /* USE_FIXED */
1153 if (avctx->extradata_size > 0) {
1154 if ((ret = decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
1156 avctx->extradata_size * 8LL,
1161 uint8_t layout_map[MAX_ELEM_ID*4][3];
1162 int layout_map_tags;
1164 sr = sample_rate_idx(avctx->sample_rate);
1165 ac->oc[1].m4ac.sampling_index = sr;
1166 ac->oc[1].m4ac.channels = avctx->channels;
1167 ac->oc[1].m4ac.sbr = -1;
1168 ac->oc[1].m4ac.ps = -1;
1170 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
1171 if (ff_mpeg4audio_channels[i] == avctx->channels)
1173 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
1176 ac->oc[1].m4ac.chan_config = i;
1178 if (ac->oc[1].m4ac.chan_config) {
1179 int ret = set_default_channel_config(avctx, layout_map,
1180 &layout_map_tags, ac->oc[1].m4ac.chan_config);
1182 output_configure(ac, layout_map, layout_map_tags,
1184 else if (avctx->err_recognition & AV_EF_EXPLODE)
1185 return AVERROR_INVALIDDATA;
1189 if (avctx->channels > MAX_CHANNELS) {
1190 av_log(avctx, AV_LOG_ERROR, "Too many channels\n");
1191 return AVERROR_INVALIDDATA;
1195 ac->fdsp = avpriv_alloc_fixed_dsp(avctx->flags & AV_CODEC_FLAG_BITEXACT);
1197 ac->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
1198 #endif /* USE_FIXED */
1200 return AVERROR(ENOMEM);
1203 ac->random_state = 0x1f2e3d4c;
1205 AAC_RENAME_32(ff_mdct_init)(&ac->mdct, 11, 1, 1.0 / RANGE15(1024.0));
1206 AAC_RENAME_32(ff_mdct_init)(&ac->mdct_ld, 10, 1, 1.0 / RANGE15(512.0));
1207 AAC_RENAME_32(ff_mdct_init)(&ac->mdct_small, 8, 1, 1.0 / RANGE15(128.0));
1208 AAC_RENAME_32(ff_mdct_init)(&ac->mdct_ltp, 11, 0, RANGE15(-2.0));
1210 ret = ff_mdct15_init(&ac->mdct480, 1, 5, -1.0f);
1219 * Skip data_stream_element; reference: table 4.10.
1221 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
1223 int byte_align = get_bits1(gb);
1224 int count = get_bits(gb, 8);
1226 count += get_bits(gb, 8);
1230 if (get_bits_left(gb) < 8 * count) {
1231 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
1232 return AVERROR_INVALIDDATA;
1234 skip_bits_long(gb, 8 * count);
1238 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
1242 if (get_bits1(gb)) {
1243 ics->predictor_reset_group = get_bits(gb, 5);
1244 if (ics->predictor_reset_group == 0 ||
1245 ics->predictor_reset_group > 30) {
1246 av_log(ac->avctx, AV_LOG_ERROR,
1247 "Invalid Predictor Reset Group.\n");
1248 return AVERROR_INVALIDDATA;
1251 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
1252 ics->prediction_used[sfb] = get_bits1(gb);
1258 * Decode Long Term Prediction data; reference: table 4.xx.
1260 static void decode_ltp(LongTermPrediction *ltp,
1261 GetBitContext *gb, uint8_t max_sfb)
1265 ltp->lag = get_bits(gb, 11);
1266 ltp->coef = ltp_coef[get_bits(gb, 3)];
1267 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1268 ltp->used[sfb] = get_bits1(gb);
1272 * Decode Individual Channel Stream info; reference: table 4.6.
1274 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
1277 const MPEG4AudioConfig *const m4ac = &ac->oc[1].m4ac;
1278 const int aot = m4ac->object_type;
1279 const int sampling_index = m4ac->sampling_index;
1280 if (aot != AOT_ER_AAC_ELD) {
1281 if (get_bits1(gb)) {
1282 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1283 if (ac->avctx->err_recognition & AV_EF_BITSTREAM)
1284 return AVERROR_INVALIDDATA;
1286 ics->window_sequence[1] = ics->window_sequence[0];
1287 ics->window_sequence[0] = get_bits(gb, 2);
1288 if (aot == AOT_ER_AAC_LD &&
1289 ics->window_sequence[0] != ONLY_LONG_SEQUENCE) {
1290 av_log(ac->avctx, AV_LOG_ERROR,
1291 "AAC LD is only defined for ONLY_LONG_SEQUENCE but "
1292 "window sequence %d found.\n", ics->window_sequence[0]);
1293 ics->window_sequence[0] = ONLY_LONG_SEQUENCE;
1294 return AVERROR_INVALIDDATA;
1296 ics->use_kb_window[1] = ics->use_kb_window[0];
1297 ics->use_kb_window[0] = get_bits1(gb);
1299 ics->num_window_groups = 1;
1300 ics->group_len[0] = 1;
1301 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1303 ics->max_sfb = get_bits(gb, 4);
1304 for (i = 0; i < 7; i++) {
1305 if (get_bits1(gb)) {
1306 ics->group_len[ics->num_window_groups - 1]++;
1308 ics->num_window_groups++;
1309 ics->group_len[ics->num_window_groups - 1] = 1;
1312 ics->num_windows = 8;
1313 ics->swb_offset = ff_swb_offset_128[sampling_index];
1314 ics->num_swb = ff_aac_num_swb_128[sampling_index];
1315 ics->tns_max_bands = ff_tns_max_bands_128[sampling_index];
1316 ics->predictor_present = 0;
1318 ics->max_sfb = get_bits(gb, 6);
1319 ics->num_windows = 1;
1320 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD) {
1321 if (m4ac->frame_length_short) {
1322 ics->swb_offset = ff_swb_offset_480[sampling_index];
1323 ics->num_swb = ff_aac_num_swb_480[sampling_index];
1324 ics->tns_max_bands = ff_tns_max_bands_480[sampling_index];
1326 ics->swb_offset = ff_swb_offset_512[sampling_index];
1327 ics->num_swb = ff_aac_num_swb_512[sampling_index];
1328 ics->tns_max_bands = ff_tns_max_bands_512[sampling_index];
1330 if (!ics->num_swb || !ics->swb_offset)
1333 ics->swb_offset = ff_swb_offset_1024[sampling_index];
1334 ics->num_swb = ff_aac_num_swb_1024[sampling_index];
1335 ics->tns_max_bands = ff_tns_max_bands_1024[sampling_index];
1337 if (aot != AOT_ER_AAC_ELD) {
1338 ics->predictor_present = get_bits1(gb);
1339 ics->predictor_reset_group = 0;
1341 if (ics->predictor_present) {
1342 if (aot == AOT_AAC_MAIN) {
1343 if (decode_prediction(ac, ics, gb)) {
1346 } else if (aot == AOT_AAC_LC ||
1347 aot == AOT_ER_AAC_LC) {
1348 av_log(ac->avctx, AV_LOG_ERROR,
1349 "Prediction is not allowed in AAC-LC.\n");
1352 if (aot == AOT_ER_AAC_LD) {
1353 av_log(ac->avctx, AV_LOG_ERROR,
1354 "LTP in ER AAC LD not yet implemented.\n");
1355 return AVERROR_PATCHWELCOME;
1357 if ((ics->ltp.present = get_bits(gb, 1)))
1358 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1363 if (ics->max_sfb > ics->num_swb) {
1364 av_log(ac->avctx, AV_LOG_ERROR,
1365 "Number of scalefactor bands in group (%d) "
1366 "exceeds limit (%d).\n",
1367 ics->max_sfb, ics->num_swb);
1374 return AVERROR_INVALIDDATA;
1378 * Decode band types (section_data payload); reference: table 4.46.
1380 * @param band_type array of the used band type
1381 * @param band_type_run_end array of the last scalefactor band of a band type run
1383 * @return Returns error status. 0 - OK, !0 - error
1385 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1386 int band_type_run_end[120], GetBitContext *gb,
1387 IndividualChannelStream *ics)
1390 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1391 for (g = 0; g < ics->num_window_groups; g++) {
1393 while (k < ics->max_sfb) {
1394 uint8_t sect_end = k;
1396 int sect_band_type = get_bits(gb, 4);
1397 if (sect_band_type == 12) {
1398 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1399 return AVERROR_INVALIDDATA;
1402 sect_len_incr = get_bits(gb, bits);
1403 sect_end += sect_len_incr;
1404 if (get_bits_left(gb) < 0) {
1405 av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1406 return AVERROR_INVALIDDATA;
1408 if (sect_end > ics->max_sfb) {
1409 av_log(ac->avctx, AV_LOG_ERROR,
1410 "Number of bands (%d) exceeds limit (%d).\n",
1411 sect_end, ics->max_sfb);
1412 return AVERROR_INVALIDDATA;
1414 } while (sect_len_incr == (1 << bits) - 1);
1415 for (; k < sect_end; k++) {
1416 band_type [idx] = sect_band_type;
1417 band_type_run_end[idx++] = sect_end;
1425 * Decode scalefactors; reference: table 4.47.
1427 * @param global_gain first scalefactor value as scalefactors are differentially coded
1428 * @param band_type array of the used band type
1429 * @param band_type_run_end array of the last scalefactor band of a band type run
1430 * @param sf array of scalefactors or intensity stereo positions
1432 * @return Returns error status. 0 - OK, !0 - error
1434 static int decode_scalefactors(AACContext *ac, INTFLOAT sf[120], GetBitContext *gb,
1435 unsigned int global_gain,
1436 IndividualChannelStream *ics,
1437 enum BandType band_type[120],
1438 int band_type_run_end[120])
1441 int offset[3] = { global_gain, global_gain - NOISE_OFFSET, 0 };
1444 for (g = 0; g < ics->num_window_groups; g++) {
1445 for (i = 0; i < ics->max_sfb;) {
1446 int run_end = band_type_run_end[idx];
1447 if (band_type[idx] == ZERO_BT) {
1448 for (; i < run_end; i++, idx++)
1450 } else if ((band_type[idx] == INTENSITY_BT) ||
1451 (band_type[idx] == INTENSITY_BT2)) {
1452 for (; i < run_end; i++, idx++) {
1453 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - SCALE_DIFF_ZERO;
1454 clipped_offset = av_clip(offset[2], -155, 100);
1455 if (offset[2] != clipped_offset) {
1456 avpriv_request_sample(ac->avctx,
1457 "If you heard an audible artifact, there may be a bug in the decoder. "
1458 "Clipped intensity stereo position (%d -> %d)",
1459 offset[2], clipped_offset);
1462 sf[idx] = 100 - clipped_offset;
1464 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1465 #endif /* USE_FIXED */
1467 } else if (band_type[idx] == NOISE_BT) {
1468 for (; i < run_end; i++, idx++) {
1469 if (noise_flag-- > 0)
1470 offset[1] += get_bits(gb, NOISE_PRE_BITS) - NOISE_PRE;
1472 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - SCALE_DIFF_ZERO;
1473 clipped_offset = av_clip(offset[1], -100, 155);
1474 if (offset[1] != clipped_offset) {
1475 avpriv_request_sample(ac->avctx,
1476 "If you heard an audible artifact, there may be a bug in the decoder. "
1477 "Clipped noise gain (%d -> %d)",
1478 offset[1], clipped_offset);
1481 sf[idx] = -(100 + clipped_offset);
1483 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1484 #endif /* USE_FIXED */
1487 for (; i < run_end; i++, idx++) {
1488 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - SCALE_DIFF_ZERO;
1489 if (offset[0] > 255U) {
1490 av_log(ac->avctx, AV_LOG_ERROR,
1491 "Scalefactor (%d) out of range.\n", offset[0]);
1492 return AVERROR_INVALIDDATA;
1495 sf[idx] = -offset[0];
1497 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1498 #endif /* USE_FIXED */
1507 * Decode pulse data; reference: table 4.7.
1509 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1510 const uint16_t *swb_offset, int num_swb)
1513 pulse->num_pulse = get_bits(gb, 2) + 1;
1514 pulse_swb = get_bits(gb, 6);
1515 if (pulse_swb >= num_swb)
1517 pulse->pos[0] = swb_offset[pulse_swb];
1518 pulse->pos[0] += get_bits(gb, 5);
1519 if (pulse->pos[0] >= swb_offset[num_swb])
1521 pulse->amp[0] = get_bits(gb, 4);
1522 for (i = 1; i < pulse->num_pulse; i++) {
1523 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1524 if (pulse->pos[i] >= swb_offset[num_swb])
1526 pulse->amp[i] = get_bits(gb, 4);
1532 * Decode Temporal Noise Shaping data; reference: table 4.48.
1534 * @return Returns error status. 0 - OK, !0 - error
1536 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1537 GetBitContext *gb, const IndividualChannelStream *ics)
1539 int w, filt, i, coef_len, coef_res, coef_compress;
1540 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1541 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1542 for (w = 0; w < ics->num_windows; w++) {
1543 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1544 coef_res = get_bits1(gb);
1546 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1548 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1550 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1551 av_log(ac->avctx, AV_LOG_ERROR,
1552 "TNS filter order %d is greater than maximum %d.\n",
1553 tns->order[w][filt], tns_max_order);
1554 tns->order[w][filt] = 0;
1555 return AVERROR_INVALIDDATA;
1557 if (tns->order[w][filt]) {
1558 tns->direction[w][filt] = get_bits1(gb);
1559 coef_compress = get_bits1(gb);
1560 coef_len = coef_res + 3 - coef_compress;
1561 tmp2_idx = 2 * coef_compress + coef_res;
1563 for (i = 0; i < tns->order[w][filt]; i++)
1564 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1573 * Decode Mid/Side data; reference: table 4.54.
1575 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1576 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1577 * [3] reserved for scalable AAC
1579 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1583 int max_idx = cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb;
1584 if (ms_present == 1) {
1585 for (idx = 0; idx < max_idx; idx++)
1586 cpe->ms_mask[idx] = get_bits1(gb);
1587 } else if (ms_present == 2) {
1588 memset(cpe->ms_mask, 1, max_idx * sizeof(cpe->ms_mask[0]));
1593 * Decode spectral data; reference: table 4.50.
1594 * Dequantize and scale spectral data; reference: 4.6.3.3.
1596 * @param coef array of dequantized, scaled spectral data
1597 * @param sf array of scalefactors or intensity stereo positions
1598 * @param pulse_present set if pulses are present
1599 * @param pulse pointer to pulse data struct
1600 * @param band_type array of the used band type
1602 * @return Returns error status. 0 - OK, !0 - error
1604 static int decode_spectrum_and_dequant(AACContext *ac, INTFLOAT coef[1024],
1605 GetBitContext *gb, const INTFLOAT sf[120],
1606 int pulse_present, const Pulse *pulse,
1607 const IndividualChannelStream *ics,
1608 enum BandType band_type[120])
1610 int i, k, g, idx = 0;
1611 const int c = 1024 / ics->num_windows;
1612 const uint16_t *offsets = ics->swb_offset;
1613 INTFLOAT *coef_base = coef;
1615 for (g = 0; g < ics->num_windows; g++)
1616 memset(coef + g * 128 + offsets[ics->max_sfb], 0,
1617 sizeof(INTFLOAT) * (c - offsets[ics->max_sfb]));
1619 for (g = 0; g < ics->num_window_groups; g++) {
1620 unsigned g_len = ics->group_len[g];
1622 for (i = 0; i < ics->max_sfb; i++, idx++) {
1623 const unsigned cbt_m1 = band_type[idx] - 1;
1624 INTFLOAT *cfo = coef + offsets[i];
1625 int off_len = offsets[i + 1] - offsets[i];
1628 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1629 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1630 memset(cfo, 0, off_len * sizeof(*cfo));
1632 } else if (cbt_m1 == NOISE_BT - 1) {
1633 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1636 #endif /* !USE_FIXED */
1637 INTFLOAT band_energy;
1639 for (k = 0; k < off_len; k++) {
1640 ac->random_state = lcg_random(ac->random_state);
1642 cfo[k] = ac->random_state >> 3;
1644 cfo[k] = ac->random_state;
1645 #endif /* USE_FIXED */
1649 band_energy = ac->fdsp->scalarproduct_fixed(cfo, cfo, off_len);
1650 band_energy = fixed_sqrt(band_energy, 31);
1651 noise_scale(cfo, sf[idx], band_energy, off_len);
1653 band_energy = ac->fdsp->scalarproduct_float(cfo, cfo, off_len);
1654 scale = sf[idx] / sqrtf(band_energy);
1655 ac->fdsp->vector_fmul_scalar(cfo, cfo, scale, off_len);
1656 #endif /* USE_FIXED */
1660 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1661 #endif /* !USE_FIXED */
1662 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1663 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1664 OPEN_READER(re, gb);
1666 switch (cbt_m1 >> 1) {
1668 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1676 UPDATE_CACHE(re, gb);
1677 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1678 cb_idx = cb_vector_idx[code];
1680 cf = DEC_SQUAD(cf, cb_idx);
1682 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1683 #endif /* USE_FIXED */
1689 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1699 UPDATE_CACHE(re, gb);
1700 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1701 cb_idx = cb_vector_idx[code];
1702 nnz = cb_idx >> 8 & 15;
1703 bits = nnz ? GET_CACHE(re, gb) : 0;
1704 LAST_SKIP_BITS(re, gb, nnz);
1706 cf = DEC_UQUAD(cf, cb_idx, bits);
1708 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1709 #endif /* USE_FIXED */
1715 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1723 UPDATE_CACHE(re, gb);
1724 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1725 cb_idx = cb_vector_idx[code];
1727 cf = DEC_SPAIR(cf, cb_idx);
1729 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1730 #endif /* USE_FIXED */
1737 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1747 UPDATE_CACHE(re, gb);
1748 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1749 cb_idx = cb_vector_idx[code];
1750 nnz = cb_idx >> 8 & 15;
1751 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1752 LAST_SKIP_BITS(re, gb, nnz);
1754 cf = DEC_UPAIR(cf, cb_idx, sign);
1756 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1757 #endif /* USE_FIXED */
1763 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1769 uint32_t *icf = (uint32_t *) cf;
1770 #endif /* USE_FIXED */
1780 UPDATE_CACHE(re, gb);
1781 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1789 cb_idx = cb_vector_idx[code];
1792 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1793 LAST_SKIP_BITS(re, gb, nnz);
1795 for (j = 0; j < 2; j++) {
1799 /* The total length of escape_sequence must be < 22 bits according
1800 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1801 UPDATE_CACHE(re, gb);
1802 b = GET_CACHE(re, gb);
1803 b = 31 - av_log2(~b);
1806 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1807 return AVERROR_INVALIDDATA;
1810 SKIP_BITS(re, gb, b + 1);
1812 n = (1 << b) + SHOW_UBITS(re, gb, b);
1813 LAST_SKIP_BITS(re, gb, b);
1820 *icf++ = ff_cbrt_tab[n] | (bits & 1U<<31);
1821 #endif /* USE_FIXED */
1830 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1831 *icf++ = (bits & 1U<<31) | v;
1832 #endif /* USE_FIXED */
1839 ac->fdsp->vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1840 #endif /* !USE_FIXED */
1844 CLOSE_READER(re, gb);
1850 if (pulse_present) {
1852 for (i = 0; i < pulse->num_pulse; i++) {
1853 INTFLOAT co = coef_base[ pulse->pos[i] ];
1854 while (offsets[idx + 1] <= pulse->pos[i])
1856 if (band_type[idx] != NOISE_BT && sf[idx]) {
1857 INTFLOAT ico = -pulse->amp[i];
1860 ico = co + (co > 0 ? -ico : ico);
1862 coef_base[ pulse->pos[i] ] = ico;
1866 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1868 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1869 #endif /* USE_FIXED */
1876 for (g = 0; g < ics->num_window_groups; g++) {
1877 unsigned g_len = ics->group_len[g];
1879 for (i = 0; i < ics->max_sfb; i++, idx++) {
1880 const unsigned cbt_m1 = band_type[idx] - 1;
1881 int *cfo = coef + offsets[i];
1882 int off_len = offsets[i + 1] - offsets[i];
1885 if (cbt_m1 < NOISE_BT - 1) {
1886 for (group = 0; group < (int)g_len; group++, cfo+=128) {
1887 ac->vector_pow43(cfo, off_len);
1888 ac->subband_scale(cfo, cfo, sf[idx], 34, off_len);
1894 #endif /* USE_FIXED */
1899 * Apply AAC-Main style frequency domain prediction.
1901 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1905 if (!sce->ics.predictor_initialized) {
1906 reset_all_predictors(sce->predictor_state);
1907 sce->ics.predictor_initialized = 1;
1910 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1912 sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index];
1914 for (k = sce->ics.swb_offset[sfb];
1915 k < sce->ics.swb_offset[sfb + 1];
1917 predict(&sce->predictor_state[k], &sce->coeffs[k],
1918 sce->ics.predictor_present &&
1919 sce->ics.prediction_used[sfb]);
1922 if (sce->ics.predictor_reset_group)
1923 reset_predictor_group(sce->predictor_state,
1924 sce->ics.predictor_reset_group);
1926 reset_all_predictors(sce->predictor_state);
1930 * Decode an individual_channel_stream payload; reference: table 4.44.
1932 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1933 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1935 * @return Returns error status. 0 - OK, !0 - error
1937 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1938 GetBitContext *gb, int common_window, int scale_flag)
1941 TemporalNoiseShaping *tns = &sce->tns;
1942 IndividualChannelStream *ics = &sce->ics;
1943 INTFLOAT *out = sce->coeffs;
1944 int global_gain, eld_syntax, er_syntax, pulse_present = 0;
1947 eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1948 er_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_LC ||
1949 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LTP ||
1950 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LD ||
1951 ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1953 /* This assignment is to silence a GCC warning about the variable being used
1954 * uninitialized when in fact it always is.
1956 pulse.num_pulse = 0;
1958 global_gain = get_bits(gb, 8);
1960 if (!common_window && !scale_flag) {
1961 if (decode_ics_info(ac, ics, gb) < 0)
1962 return AVERROR_INVALIDDATA;
1965 if ((ret = decode_band_types(ac, sce->band_type,
1966 sce->band_type_run_end, gb, ics)) < 0)
1968 if ((ret = decode_scalefactors(ac, sce->sf, gb, global_gain, ics,
1969 sce->band_type, sce->band_type_run_end)) < 0)
1974 if (!eld_syntax && (pulse_present = get_bits1(gb))) {
1975 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1976 av_log(ac->avctx, AV_LOG_ERROR,
1977 "Pulse tool not allowed in eight short sequence.\n");
1978 return AVERROR_INVALIDDATA;
1980 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1981 av_log(ac->avctx, AV_LOG_ERROR,
1982 "Pulse data corrupt or invalid.\n");
1983 return AVERROR_INVALIDDATA;
1986 tns->present = get_bits1(gb);
1987 if (tns->present && !er_syntax)
1988 if (decode_tns(ac, tns, gb, ics) < 0)
1989 return AVERROR_INVALIDDATA;
1990 if (!eld_syntax && get_bits1(gb)) {
1991 avpriv_request_sample(ac->avctx, "SSR");
1992 return AVERROR_PATCHWELCOME;
1994 // I see no textual basis in the spec for this occurring after SSR gain
1995 // control, but this is what both reference and real implmentations do
1996 if (tns->present && er_syntax)
1997 if (decode_tns(ac, tns, gb, ics) < 0)
1998 return AVERROR_INVALIDDATA;
2001 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present,
2002 &pulse, ics, sce->band_type) < 0)
2003 return AVERROR_INVALIDDATA;
2005 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
2006 apply_prediction(ac, sce);
2012 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
2014 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
2016 const IndividualChannelStream *ics = &cpe->ch[0].ics;
2017 INTFLOAT *ch0 = cpe->ch[0].coeffs;
2018 INTFLOAT *ch1 = cpe->ch[1].coeffs;
2019 int g, i, group, idx = 0;
2020 const uint16_t *offsets = ics->swb_offset;
2021 for (g = 0; g < ics->num_window_groups; g++) {
2022 for (i = 0; i < ics->max_sfb; i++, idx++) {
2023 if (cpe->ms_mask[idx] &&
2024 cpe->ch[0].band_type[idx] < NOISE_BT &&
2025 cpe->ch[1].band_type[idx] < NOISE_BT) {
2027 for (group = 0; group < ics->group_len[g]; group++) {
2028 ac->fdsp->butterflies_fixed(ch0 + group * 128 + offsets[i],
2029 ch1 + group * 128 + offsets[i],
2030 offsets[i+1] - offsets[i]);
2032 for (group = 0; group < ics->group_len[g]; group++) {
2033 ac->fdsp->butterflies_float(ch0 + group * 128 + offsets[i],
2034 ch1 + group * 128 + offsets[i],
2035 offsets[i+1] - offsets[i]);
2036 #endif /* USE_FIXED */
2040 ch0 += ics->group_len[g] * 128;
2041 ch1 += ics->group_len[g] * 128;
2046 * intensity stereo decoding; reference: 4.6.8.2.3
2048 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
2049 * [1] mask is decoded from bitstream; [2] mask is all 1s;
2050 * [3] reserved for scalable AAC
2052 static void apply_intensity_stereo(AACContext *ac,
2053 ChannelElement *cpe, int ms_present)
2055 const IndividualChannelStream *ics = &cpe->ch[1].ics;
2056 SingleChannelElement *sce1 = &cpe->ch[1];
2057 INTFLOAT *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
2058 const uint16_t *offsets = ics->swb_offset;
2059 int g, group, i, idx = 0;
2062 for (g = 0; g < ics->num_window_groups; g++) {
2063 for (i = 0; i < ics->max_sfb;) {
2064 if (sce1->band_type[idx] == INTENSITY_BT ||
2065 sce1->band_type[idx] == INTENSITY_BT2) {
2066 const int bt_run_end = sce1->band_type_run_end[idx];
2067 for (; i < bt_run_end; i++, idx++) {
2068 c = -1 + 2 * (sce1->band_type[idx] - 14);
2070 c *= 1 - 2 * cpe->ms_mask[idx];
2071 scale = c * sce1->sf[idx];
2072 for (group = 0; group < ics->group_len[g]; group++)
2074 ac->subband_scale(coef1 + group * 128 + offsets[i],
2075 coef0 + group * 128 + offsets[i],
2078 offsets[i + 1] - offsets[i]);
2080 ac->fdsp->vector_fmul_scalar(coef1 + group * 128 + offsets[i],
2081 coef0 + group * 128 + offsets[i],
2083 offsets[i + 1] - offsets[i]);
2084 #endif /* USE_FIXED */
2087 int bt_run_end = sce1->band_type_run_end[idx];
2088 idx += bt_run_end - i;
2092 coef0 += ics->group_len[g] * 128;
2093 coef1 += ics->group_len[g] * 128;
2098 * Decode a channel_pair_element; reference: table 4.4.
2100 * @return Returns error status. 0 - OK, !0 - error
2102 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
2104 int i, ret, common_window, ms_present = 0;
2105 int eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
2107 common_window = eld_syntax || get_bits1(gb);
2108 if (common_window) {
2109 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
2110 return AVERROR_INVALIDDATA;
2111 i = cpe->ch[1].ics.use_kb_window[0];
2112 cpe->ch[1].ics = cpe->ch[0].ics;
2113 cpe->ch[1].ics.use_kb_window[1] = i;
2114 if (cpe->ch[1].ics.predictor_present &&
2115 (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
2116 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
2117 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
2118 ms_present = get_bits(gb, 2);
2119 if (ms_present == 3) {
2120 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
2121 return AVERROR_INVALIDDATA;
2122 } else if (ms_present)
2123 decode_mid_side_stereo(cpe, gb, ms_present);
2125 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
2127 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
2130 if (common_window) {
2132 apply_mid_side_stereo(ac, cpe);
2133 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
2134 apply_prediction(ac, &cpe->ch[0]);
2135 apply_prediction(ac, &cpe->ch[1]);
2139 apply_intensity_stereo(ac, cpe, ms_present);
2143 static const float cce_scale[] = {
2144 1.09050773266525765921, //2^(1/8)
2145 1.18920711500272106672, //2^(1/4)
2151 * Decode coupling_channel_element; reference: table 4.8.
2153 * @return Returns error status. 0 - OK, !0 - error
2155 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
2161 SingleChannelElement *sce = &che->ch[0];
2162 ChannelCoupling *coup = &che->coup;
2164 coup->coupling_point = 2 * get_bits1(gb);
2165 coup->num_coupled = get_bits(gb, 3);
2166 for (c = 0; c <= coup->num_coupled; c++) {
2168 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
2169 coup->id_select[c] = get_bits(gb, 4);
2170 if (coup->type[c] == TYPE_CPE) {
2171 coup->ch_select[c] = get_bits(gb, 2);
2172 if (coup->ch_select[c] == 3)
2175 coup->ch_select[c] = 2;
2177 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
2179 sign = get_bits(gb, 1);
2180 scale = AAC_RENAME(cce_scale)[get_bits(gb, 2)];
2182 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
2185 for (c = 0; c < num_gain; c++) {
2189 INTFLOAT gain_cache = FIXR10(1.);
2191 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
2192 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
2193 gain_cache = GET_GAIN(scale, gain);
2195 if (coup->coupling_point == AFTER_IMDCT) {
2196 coup->gain[c][0] = gain_cache;
2198 for (g = 0; g < sce->ics.num_window_groups; g++) {
2199 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
2200 if (sce->band_type[idx] != ZERO_BT) {
2202 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
2210 gain_cache = GET_GAIN(scale, t) * s;
2213 coup->gain[c][idx] = gain_cache;
2223 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
2225 * @return Returns number of bytes consumed.
2227 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
2231 int num_excl_chan = 0;
2234 for (i = 0; i < 7; i++)
2235 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
2236 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
2238 return num_excl_chan / 7;
2242 * Decode dynamic range information; reference: table 4.52.
2244 * @return Returns number of bytes consumed.
2246 static int decode_dynamic_range(DynamicRangeControl *che_drc,
2250 int drc_num_bands = 1;
2253 /* pce_tag_present? */
2254 if (get_bits1(gb)) {
2255 che_drc->pce_instance_tag = get_bits(gb, 4);
2256 skip_bits(gb, 4); // tag_reserved_bits
2260 /* excluded_chns_present? */
2261 if (get_bits1(gb)) {
2262 n += decode_drc_channel_exclusions(che_drc, gb);
2265 /* drc_bands_present? */
2266 if (get_bits1(gb)) {
2267 che_drc->band_incr = get_bits(gb, 4);
2268 che_drc->interpolation_scheme = get_bits(gb, 4);
2270 drc_num_bands += che_drc->band_incr;
2271 for (i = 0; i < drc_num_bands; i++) {
2272 che_drc->band_top[i] = get_bits(gb, 8);
2277 /* prog_ref_level_present? */
2278 if (get_bits1(gb)) {
2279 che_drc->prog_ref_level = get_bits(gb, 7);
2280 skip_bits1(gb); // prog_ref_level_reserved_bits
2284 for (i = 0; i < drc_num_bands; i++) {
2285 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
2286 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
2293 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
2295 int i, major, minor;
2300 get_bits(gb, 13); len -= 13;
2302 for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
2303 buf[i] = get_bits(gb, 8);
2306 if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
2307 av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
2309 if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
2310 ac->avctx->internal->skip_samples = 1024;
2314 skip_bits_long(gb, len);
2320 * Decode extension data (incomplete); reference: table 4.51.
2322 * @param cnt length of TYPE_FIL syntactic element in bytes
2324 * @return Returns number of bytes consumed
2326 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2327 ChannelElement *che, enum RawDataBlockType elem_type)
2331 int type = get_bits(gb, 4);
2333 if (ac->avctx->debug & FF_DEBUG_STARTCODE)
2334 av_log(ac->avctx, AV_LOG_DEBUG, "extension type: %d len:%d\n", type, cnt);
2336 switch (type) { // extension type
2337 case EXT_SBR_DATA_CRC:
2341 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2343 } else if (!ac->oc[1].m4ac.sbr) {
2344 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2345 skip_bits_long(gb, 8 * cnt - 4);
2347 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2348 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2349 skip_bits_long(gb, 8 * cnt - 4);
2351 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2352 ac->oc[1].m4ac.sbr = 1;
2353 ac->oc[1].m4ac.ps = 1;
2354 ac->avctx->profile = FF_PROFILE_AAC_HE_V2;
2355 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2356 ac->oc[1].status, 1);
2358 ac->oc[1].m4ac.sbr = 1;
2359 ac->avctx->profile = FF_PROFILE_AAC_HE;
2361 res = AAC_RENAME(ff_decode_sbr_extension)(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2363 case EXT_DYNAMIC_RANGE:
2364 res = decode_dynamic_range(&ac->che_drc, gb);
2367 decode_fill(ac, gb, 8 * cnt - 4);
2370 case EXT_DATA_ELEMENT:
2372 skip_bits_long(gb, 8 * cnt - 4);
2379 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2381 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2382 * @param coef spectral coefficients
2384 static void apply_tns(INTFLOAT coef[1024], TemporalNoiseShaping *tns,
2385 IndividualChannelStream *ics, int decode)
2387 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2389 int bottom, top, order, start, end, size, inc;
2390 INTFLOAT lpc[TNS_MAX_ORDER];
2391 INTFLOAT tmp[TNS_MAX_ORDER+1];
2393 for (w = 0; w < ics->num_windows; w++) {
2394 bottom = ics->num_swb;
2395 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2397 bottom = FFMAX(0, top - tns->length[w][filt]);
2398 order = tns->order[w][filt];
2403 AAC_RENAME(compute_lpc_coefs)(tns->coef[w][filt], order, lpc, 0, 0, 0);
2405 start = ics->swb_offset[FFMIN(bottom, mmm)];
2406 end = ics->swb_offset[FFMIN( top, mmm)];
2407 if ((size = end - start) <= 0)
2409 if (tns->direction[w][filt]) {
2419 for (m = 0; m < size; m++, start += inc)
2420 for (i = 1; i <= FFMIN(m, order); i++)
2421 coef[start] -= AAC_MUL26(coef[start - i * inc], lpc[i - 1]);
2424 for (m = 0; m < size; m++, start += inc) {
2425 tmp[0] = coef[start];
2426 for (i = 1; i <= FFMIN(m, order); i++)
2427 coef[start] += AAC_MUL26(tmp[i], lpc[i - 1]);
2428 for (i = order; i > 0; i--)
2429 tmp[i] = tmp[i - 1];
2437 * Apply windowing and MDCT to obtain the spectral
2438 * coefficient from the predicted sample by LTP.
2440 static void windowing_and_mdct_ltp(AACContext *ac, INTFLOAT *out,
2441 INTFLOAT *in, IndividualChannelStream *ics)
2443 const INTFLOAT *lwindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_long_1024) : AAC_RENAME(ff_sine_1024);
2444 const INTFLOAT *swindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2445 const INTFLOAT *lwindow_prev = ics->use_kb_window[1] ? AAC_RENAME(ff_aac_kbd_long_1024) : AAC_RENAME(ff_sine_1024);
2446 const INTFLOAT *swindow_prev = ics->use_kb_window[1] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2448 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2449 ac->fdsp->vector_fmul(in, in, lwindow_prev, 1024);
2451 memset(in, 0, 448 * sizeof(*in));
2452 ac->fdsp->vector_fmul(in + 448, in + 448, swindow_prev, 128);
2454 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2455 ac->fdsp->vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2457 ac->fdsp->vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2458 memset(in + 1024 + 576, 0, 448 * sizeof(*in));
2460 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2464 * Apply the long term prediction
2466 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2468 const LongTermPrediction *ltp = &sce->ics.ltp;
2469 const uint16_t *offsets = sce->ics.swb_offset;
2472 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2473 INTFLOAT *predTime = sce->ret;
2474 INTFLOAT *predFreq = ac->buf_mdct;
2475 int16_t num_samples = 2048;
2477 if (ltp->lag < 1024)
2478 num_samples = ltp->lag + 1024;
2479 for (i = 0; i < num_samples; i++)
2480 predTime[i] = AAC_MUL30(sce->ltp_state[i + 2048 - ltp->lag], ltp->coef);
2481 memset(&predTime[i], 0, (2048 - i) * sizeof(*predTime));
2483 ac->windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2485 if (sce->tns.present)
2486 ac->apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2488 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2490 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2491 sce->coeffs[i] += predFreq[i];
2496 * Update the LTP buffer for next frame
2498 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2500 IndividualChannelStream *ics = &sce->ics;
2501 INTFLOAT *saved = sce->saved;
2502 INTFLOAT *saved_ltp = sce->coeffs;
2503 const INTFLOAT *lwindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_long_1024) : AAC_RENAME(ff_sine_1024);
2504 const INTFLOAT *swindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2507 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2508 memcpy(saved_ltp, saved, 512 * sizeof(*saved_ltp));
2509 memset(saved_ltp + 576, 0, 448 * sizeof(*saved_ltp));
2510 ac->fdsp->vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2512 for (i = 0; i < 64; i++)
2513 saved_ltp[i + 512] = AAC_MUL31(ac->buf_mdct[1023 - i], swindow[63 - i]);
2514 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2515 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(*saved_ltp));
2516 memset(saved_ltp + 576, 0, 448 * sizeof(*saved_ltp));
2517 ac->fdsp->vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2519 for (i = 0; i < 64; i++)
2520 saved_ltp[i + 512] = AAC_MUL31(ac->buf_mdct[1023 - i], swindow[63 - i]);
2521 } else { // LONG_STOP or ONLY_LONG
2522 ac->fdsp->vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2524 for (i = 0; i < 512; i++)
2525 saved_ltp[i + 512] = AAC_MUL31(ac->buf_mdct[1023 - i], lwindow[511 - i]);
2528 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2529 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2530 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2534 * Conduct IMDCT and windowing.
2536 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2538 IndividualChannelStream *ics = &sce->ics;
2539 INTFLOAT *in = sce->coeffs;
2540 INTFLOAT *out = sce->ret;
2541 INTFLOAT *saved = sce->saved;
2542 const INTFLOAT *swindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2543 const INTFLOAT *lwindow_prev = ics->use_kb_window[1] ? AAC_RENAME(ff_aac_kbd_long_1024) : AAC_RENAME(ff_sine_1024);
2544 const INTFLOAT *swindow_prev = ics->use_kb_window[1] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2545 INTFLOAT *buf = ac->buf_mdct;
2546 INTFLOAT *temp = ac->temp;
2550 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2551 for (i = 0; i < 1024; i += 128)
2552 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2554 ac->mdct.imdct_half(&ac->mdct, buf, in);
2556 for (i=0; i<1024; i++)
2557 buf[i] = (buf[i] + 4) >> 3;
2558 #endif /* USE_FIXED */
2561 /* window overlapping
2562 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2563 * and long to short transitions are considered to be short to short
2564 * transitions. This leaves just two cases (long to long and short to short)
2565 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2567 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2568 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2569 ac->fdsp->vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2571 memcpy( out, saved, 448 * sizeof(*out));
2573 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2574 ac->fdsp->vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2575 ac->fdsp->vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2576 ac->fdsp->vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2577 ac->fdsp->vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2578 ac->fdsp->vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2579 memcpy( out + 448 + 4*128, temp, 64 * sizeof(*out));
2581 ac->fdsp->vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2582 memcpy( out + 576, buf + 64, 448 * sizeof(*out));
2587 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2588 memcpy( saved, temp + 64, 64 * sizeof(*saved));
2589 ac->fdsp->vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2590 ac->fdsp->vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2591 ac->fdsp->vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2592 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(*saved));
2593 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2594 memcpy( saved, buf + 512, 448 * sizeof(*saved));
2595 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(*saved));
2596 } else { // LONG_STOP or ONLY_LONG
2597 memcpy( saved, buf + 512, 512 * sizeof(*saved));
2601 static void imdct_and_windowing_ld(AACContext *ac, SingleChannelElement *sce)
2603 IndividualChannelStream *ics = &sce->ics;
2604 INTFLOAT *in = sce->coeffs;
2605 INTFLOAT *out = sce->ret;
2606 INTFLOAT *saved = sce->saved;
2607 INTFLOAT *buf = ac->buf_mdct;
2610 #endif /* USE_FIXED */
2613 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2616 for (i = 0; i < 1024; i++)
2617 buf[i] = (buf[i] + 2) >> 2;
2618 #endif /* USE_FIXED */
2620 // window overlapping
2621 if (ics->use_kb_window[1]) {
2622 // AAC LD uses a low overlap sine window instead of a KBD window
2623 memcpy(out, saved, 192 * sizeof(*out));
2624 ac->fdsp->vector_fmul_window(out + 192, saved + 192, buf, AAC_RENAME(ff_sine_128), 64);
2625 memcpy( out + 320, buf + 64, 192 * sizeof(*out));
2627 ac->fdsp->vector_fmul_window(out, saved, buf, AAC_RENAME(ff_sine_512), 256);
2631 memcpy(saved, buf + 256, 256 * sizeof(*saved));
2634 static void imdct_and_windowing_eld(AACContext *ac, SingleChannelElement *sce)
2636 INTFLOAT *in = sce->coeffs;
2637 INTFLOAT *out = sce->ret;
2638 INTFLOAT *saved = sce->saved;
2639 INTFLOAT *buf = ac->buf_mdct;
2641 const int n = ac->oc[1].m4ac.frame_length_short ? 480 : 512;
2642 const int n2 = n >> 1;
2643 const int n4 = n >> 2;
2644 const INTFLOAT *const window = n == 480 ? AAC_RENAME(ff_aac_eld_window_480) :
2645 AAC_RENAME(ff_aac_eld_window_512);
2647 // Inverse transform, mapped to the conventional IMDCT by
2648 // Chivukula, R.K.; Reznik, Y.A.; Devarajan, V.,
2649 // "Efficient algorithms for MPEG-4 AAC-ELD, AAC-LD and AAC-LC filterbanks,"
2650 // International Conference on Audio, Language and Image Processing, ICALIP 2008.
2651 // URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4590245&isnumber=4589950
2652 for (i = 0; i < n2; i+=2) {
2654 temp = in[i ]; in[i ] = -in[n - 1 - i]; in[n - 1 - i] = temp;
2655 temp = -in[i + 1]; in[i + 1] = in[n - 2 - i]; in[n - 2 - i] = temp;
2659 ac->mdct480->imdct_half(ac->mdct480, buf, in, 1, -1.f/(16*1024*960));
2662 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2665 for (i = 0; i < 1024; i++)
2666 buf[i] = (buf[i] + 1) >> 1;
2667 #endif /* USE_FIXED */
2669 for (i = 0; i < n; i+=2) {
2672 // Like with the regular IMDCT at this point we still have the middle half
2673 // of a transform but with even symmetry on the left and odd symmetry on
2676 // window overlapping
2677 // The spec says to use samples [0..511] but the reference decoder uses
2678 // samples [128..639].
2679 for (i = n4; i < n2; i ++) {
2680 out[i - n4] = AAC_MUL31( buf[ n2 - 1 - i] , window[i - n4]) +
2681 AAC_MUL31( saved[ i + n2] , window[i + n - n4]) +
2682 AAC_MUL31(-saved[n + n2 - 1 - i] , window[i + 2*n - n4]) +
2683 AAC_MUL31(-saved[ 2*n + n2 + i] , window[i + 3*n - n4]);
2685 for (i = 0; i < n2; i ++) {
2686 out[n4 + i] = AAC_MUL31( buf[ i] , window[i + n2 - n4]) +
2687 AAC_MUL31(-saved[ n - 1 - i] , window[i + n2 + n - n4]) +
2688 AAC_MUL31(-saved[ n + i] , window[i + n2 + 2*n - n4]) +
2689 AAC_MUL31( saved[2*n + n - 1 - i] , window[i + n2 + 3*n - n4]);
2691 for (i = 0; i < n4; i ++) {
2692 out[n2 + n4 + i] = AAC_MUL31( buf[ i + n2] , window[i + n - n4]) +
2693 AAC_MUL31(-saved[n2 - 1 - i] , window[i + 2*n - n4]) +
2694 AAC_MUL31(-saved[n + n2 + i] , window[i + 3*n - n4]);
2698 memmove(saved + n, saved, 2 * n * sizeof(*saved));
2699 memcpy( saved, buf, n * sizeof(*saved));
2703 * channel coupling transformation interface
2705 * @param apply_coupling_method pointer to (in)dependent coupling function
2707 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2708 enum RawDataBlockType type, int elem_id,
2709 enum CouplingPoint coupling_point,
2710 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2714 for (i = 0; i < MAX_ELEM_ID; i++) {
2715 ChannelElement *cce = ac->che[TYPE_CCE][i];
2718 if (cce && cce->coup.coupling_point == coupling_point) {
2719 ChannelCoupling *coup = &cce->coup;
2721 for (c = 0; c <= coup->num_coupled; c++) {
2722 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2723 if (coup->ch_select[c] != 1) {
2724 apply_coupling_method(ac, &cc->ch[0], cce, index);
2725 if (coup->ch_select[c] != 0)
2728 if (coup->ch_select[c] != 2)
2729 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2731 index += 1 + (coup->ch_select[c] == 3);
2738 * Convert spectral data to samples, applying all supported tools as appropriate.
2740 static void spectral_to_sample(AACContext *ac, int samples)
2743 void (*imdct_and_window)(AACContext *ac, SingleChannelElement *sce);
2744 switch (ac->oc[1].m4ac.object_type) {
2746 imdct_and_window = imdct_and_windowing_ld;
2748 case AOT_ER_AAC_ELD:
2749 imdct_and_window = imdct_and_windowing_eld;
2752 imdct_and_window = ac->imdct_and_windowing;
2754 for (type = 3; type >= 0; type--) {
2755 for (i = 0; i < MAX_ELEM_ID; i++) {
2756 ChannelElement *che = ac->che[type][i];
2757 if (che && che->present) {
2758 if (type <= TYPE_CPE)
2759 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, AAC_RENAME(apply_dependent_coupling));
2760 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2761 if (che->ch[0].ics.predictor_present) {
2762 if (che->ch[0].ics.ltp.present)
2763 ac->apply_ltp(ac, &che->ch[0]);
2764 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2765 ac->apply_ltp(ac, &che->ch[1]);
2768 if (che->ch[0].tns.present)
2769 ac->apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2770 if (che->ch[1].tns.present)
2771 ac->apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2772 if (type <= TYPE_CPE)
2773 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, AAC_RENAME(apply_dependent_coupling));
2774 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2775 imdct_and_window(ac, &che->ch[0]);
2776 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2777 ac->update_ltp(ac, &che->ch[0]);
2778 if (type == TYPE_CPE) {
2779 imdct_and_window(ac, &che->ch[1]);
2780 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2781 ac->update_ltp(ac, &che->ch[1]);
2783 if (ac->oc[1].m4ac.sbr > 0) {
2784 AAC_RENAME(ff_sbr_apply)(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2787 if (type <= TYPE_CCE)
2788 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, AAC_RENAME(apply_independent_coupling));
2793 /* preparation for resampler */
2794 for(j = 0; j<samples; j++){
2795 che->ch[0].ret[j] = (int32_t)av_clipl_int32((int64_t)che->ch[0].ret[j]<<7)+0x8000;
2796 if(type == TYPE_CPE)
2797 che->ch[1].ret[j] = (int32_t)av_clipl_int32((int64_t)che->ch[1].ret[j]<<7)+0x8000;
2800 #endif /* USE_FIXED */
2803 av_log(ac->avctx, AV_LOG_VERBOSE, "ChannelElement %d.%d missing \n", type, i);
2809 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2812 AACADTSHeaderInfo hdr_info;
2813 uint8_t layout_map[MAX_ELEM_ID*4][3];
2814 int layout_map_tags, ret;
2816 size = avpriv_aac_parse_header(gb, &hdr_info);
2818 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2819 // This is 2 for "VLB " audio in NSV files.
2820 // See samples/nsv/vlb_audio.
2821 avpriv_report_missing_feature(ac->avctx,
2822 "More than one AAC RDB per ADTS frame");
2823 ac->warned_num_aac_frames = 1;
2825 push_output_configuration(ac);
2826 if (hdr_info.chan_config) {
2827 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2828 if ((ret = set_default_channel_config(ac->avctx,
2831 hdr_info.chan_config)) < 0)
2833 if ((ret = output_configure(ac, layout_map, layout_map_tags,
2834 FFMAX(ac->oc[1].status,
2835 OC_TRIAL_FRAME), 0)) < 0)
2838 ac->oc[1].m4ac.chan_config = 0;
2840 * dual mono frames in Japanese DTV can have chan_config 0
2841 * WITHOUT specifying PCE.
2842 * thus, set dual mono as default.
2844 if (ac->dmono_mode && ac->oc[0].status == OC_NONE) {
2845 layout_map_tags = 2;
2846 layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
2847 layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
2848 layout_map[0][1] = 0;
2849 layout_map[1][1] = 1;
2850 if (output_configure(ac, layout_map, layout_map_tags,
2855 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2856 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2857 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2858 ac->oc[1].m4ac.frame_length_short = 0;
2859 if (ac->oc[0].status != OC_LOCKED ||
2860 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2861 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2862 ac->oc[1].m4ac.sbr = -1;
2863 ac->oc[1].m4ac.ps = -1;
2865 if (!hdr_info.crc_absent)
2871 static int aac_decode_er_frame(AVCodecContext *avctx, void *data,
2872 int *got_frame_ptr, GetBitContext *gb)
2874 AACContext *ac = avctx->priv_data;
2875 const MPEG4AudioConfig *const m4ac = &ac->oc[1].m4ac;
2876 ChannelElement *che;
2878 int samples = m4ac->frame_length_short ? 960 : 1024;
2879 int chan_config = m4ac->chan_config;
2880 int aot = m4ac->object_type;
2882 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD)
2887 if ((err = frame_configure_elements(avctx)) < 0)
2890 // The FF_PROFILE_AAC_* defines are all object_type - 1
2891 // This may lead to an undefined profile being signaled
2892 ac->avctx->profile = aot - 1;
2894 ac->tags_mapped = 0;
2896 if (chan_config < 0 || (chan_config >= 8 && chan_config < 11) || chan_config >= 13) {
2897 avpriv_request_sample(avctx, "Unknown ER channel configuration %d",
2899 return AVERROR_INVALIDDATA;
2901 for (i = 0; i < tags_per_config[chan_config]; i++) {
2902 const int elem_type = aac_channel_layout_map[chan_config-1][i][0];
2903 const int elem_id = aac_channel_layout_map[chan_config-1][i][1];
2904 if (!(che=get_che(ac, elem_type, elem_id))) {
2905 av_log(ac->avctx, AV_LOG_ERROR,
2906 "channel element %d.%d is not allocated\n",
2907 elem_type, elem_id);
2908 return AVERROR_INVALIDDATA;
2911 if (aot != AOT_ER_AAC_ELD)
2913 switch (elem_type) {
2915 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2918 err = decode_cpe(ac, gb, che);
2921 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2928 spectral_to_sample(ac, samples);
2930 if (!ac->frame->data[0] && samples) {
2931 av_log(avctx, AV_LOG_ERROR, "no frame data found\n");
2932 return AVERROR_INVALIDDATA;
2935 ac->frame->nb_samples = samples;
2936 ac->frame->sample_rate = avctx->sample_rate;
2939 skip_bits_long(gb, get_bits_left(gb));
2943 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2944 int *got_frame_ptr, GetBitContext *gb, AVPacket *avpkt)
2946 AACContext *ac = avctx->priv_data;
2947 ChannelElement *che = NULL, *che_prev = NULL;
2948 enum RawDataBlockType elem_type, che_prev_type = TYPE_END;
2950 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2951 int is_dmono, sce_count = 0;
2952 int payload_alignment;
2956 if (show_bits(gb, 12) == 0xfff) {
2957 if ((err = parse_adts_frame_header(ac, gb)) < 0) {
2958 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2961 if (ac->oc[1].m4ac.sampling_index > 12) {
2962 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2963 err = AVERROR_INVALIDDATA;
2968 if ((err = frame_configure_elements(avctx)) < 0)
2971 // The FF_PROFILE_AAC_* defines are all object_type - 1
2972 // This may lead to an undefined profile being signaled
2973 ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2975 payload_alignment = get_bits_count(gb);
2976 ac->tags_mapped = 0;
2978 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2979 elem_id = get_bits(gb, 4);
2981 if (avctx->debug & FF_DEBUG_STARTCODE)
2982 av_log(avctx, AV_LOG_DEBUG, "Elem type:%x id:%x\n", elem_type, elem_id);
2984 if (!avctx->channels && elem_type != TYPE_PCE) {
2985 err = AVERROR_INVALIDDATA;
2989 if (elem_type < TYPE_DSE) {
2990 if (!(che=get_che(ac, elem_type, elem_id))) {
2991 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2992 elem_type, elem_id);
2993 err = AVERROR_INVALIDDATA;
3000 switch (elem_type) {
3003 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
3009 err = decode_cpe(ac, gb, che);
3014 err = decode_cce(ac, gb, che);
3018 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
3023 err = skip_data_stream_element(ac, gb);
3027 uint8_t layout_map[MAX_ELEM_ID*4][3];
3029 push_output_configuration(ac);
3030 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb,
3037 av_log(avctx, AV_LOG_ERROR,
3038 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
3039 pop_output_configuration(ac);
3041 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
3043 ac->oc[1].m4ac.chan_config = 0;
3051 elem_id += get_bits(gb, 8) - 1;
3052 if (get_bits_left(gb) < 8 * elem_id) {
3053 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
3054 err = AVERROR_INVALIDDATA;
3058 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, che_prev_type);
3059 err = 0; /* FIXME */
3063 err = AVERROR_BUG; /* should not happen, but keeps compiler happy */
3067 if (elem_type < TYPE_DSE) {
3069 che_prev_type = elem_type;
3075 if (get_bits_left(gb) < 3) {
3076 av_log(avctx, AV_LOG_ERROR, overread_err);
3077 err = AVERROR_INVALIDDATA;
3082 if (!avctx->channels) {
3087 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
3088 samples <<= multiplier;
3090 spectral_to_sample(ac, samples);
3092 if (ac->oc[1].status && audio_found) {
3093 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
3094 avctx->frame_size = samples;
3095 ac->oc[1].status = OC_LOCKED;
3099 avctx->internal->skip_samples_multiplier = 2;
3101 if (!ac->frame->data[0] && samples) {
3102 av_log(avctx, AV_LOG_ERROR, "no frame data found\n");
3103 err = AVERROR_INVALIDDATA;
3108 ac->frame->nb_samples = samples;
3109 ac->frame->sample_rate = avctx->sample_rate;
3111 av_frame_unref(ac->frame);
3112 *got_frame_ptr = !!samples;
3114 /* for dual-mono audio (SCE + SCE) */
3115 is_dmono = ac->dmono_mode && sce_count == 2 &&
3116 ac->oc[1].channel_layout == (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT);
3118 if (ac->dmono_mode == 1)
3119 ((AVFrame *)data)->data[1] =((AVFrame *)data)->data[0];
3120 else if (ac->dmono_mode == 2)
3121 ((AVFrame *)data)->data[0] =((AVFrame *)data)->data[1];
3126 pop_output_configuration(ac);
3130 static int aac_decode_frame(AVCodecContext *avctx, void *data,
3131 int *got_frame_ptr, AVPacket *avpkt)
3133 AACContext *ac = avctx->priv_data;
3134 const uint8_t *buf = avpkt->data;
3135 int buf_size = avpkt->size;
3140 int new_extradata_size;
3141 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
3142 AV_PKT_DATA_NEW_EXTRADATA,
3143 &new_extradata_size);
3144 int jp_dualmono_size;
3145 const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
3146 AV_PKT_DATA_JP_DUALMONO,
3149 if (new_extradata && 0) {
3150 av_free(avctx->extradata);
3151 avctx->extradata = av_mallocz(new_extradata_size +
3152 AV_INPUT_BUFFER_PADDING_SIZE);
3153 if (!avctx->extradata)
3154 return AVERROR(ENOMEM);
3155 avctx->extradata_size = new_extradata_size;
3156 memcpy(avctx->extradata, new_extradata, new_extradata_size);
3157 push_output_configuration(ac);
3158 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
3160 avctx->extradata_size*8LL, 1) < 0) {
3161 pop_output_configuration(ac);
3162 return AVERROR_INVALIDDATA;
3167 if (jp_dualmono && jp_dualmono_size > 0)
3168 ac->dmono_mode = 1 + *jp_dualmono;
3169 if (ac->force_dmono_mode >= 0)
3170 ac->dmono_mode = ac->force_dmono_mode;
3172 if (INT_MAX / 8 <= buf_size)
3173 return AVERROR_INVALIDDATA;
3175 if ((err = init_get_bits8(&gb, buf, buf_size)) < 0)
3178 switch (ac->oc[1].m4ac.object_type) {
3180 case AOT_ER_AAC_LTP:
3182 case AOT_ER_AAC_ELD:
3183 err = aac_decode_er_frame(avctx, data, got_frame_ptr, &gb);
3186 err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt);
3191 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
3192 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
3193 if (buf[buf_offset])
3196 return buf_size > buf_offset ? buf_consumed : buf_size;
3199 static av_cold int aac_decode_close(AVCodecContext *avctx)
3201 AACContext *ac = avctx->priv_data;
3204 for (i = 0; i < MAX_ELEM_ID; i++) {
3205 for (type = 0; type < 4; type++) {
3206 if (ac->che[type][i])
3207 AAC_RENAME(ff_aac_sbr_ctx_close)(&ac->che[type][i]->sbr);
3208 av_freep(&ac->che[type][i]);
3212 ff_mdct_end(&ac->mdct);
3213 ff_mdct_end(&ac->mdct_small);
3214 ff_mdct_end(&ac->mdct_ld);
3215 ff_mdct_end(&ac->mdct_ltp);
3217 ff_mdct15_uninit(&ac->mdct480);
3219 av_freep(&ac->fdsp);
3223 static void aacdec_init(AACContext *c)
3225 c->imdct_and_windowing = imdct_and_windowing;
3226 c->apply_ltp = apply_ltp;
3227 c->apply_tns = apply_tns;
3228 c->windowing_and_mdct_ltp = windowing_and_mdct_ltp;
3229 c->update_ltp = update_ltp;
3231 c->vector_pow43 = vector_pow43;
3232 c->subband_scale = subband_scale;
3237 ff_aacdec_init_mips(c);
3238 #endif /* !USE_FIXED */
3241 * AVOptions for Japanese DTV specific extensions (ADTS only)
3243 #define AACDEC_FLAGS AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
3244 static const AVOption options[] = {
3245 {"dual_mono_mode", "Select the channel to decode for dual mono",
3246 offsetof(AACContext, force_dmono_mode), AV_OPT_TYPE_INT, {.i64=-1}, -1, 2,
3247 AACDEC_FLAGS, "dual_mono_mode"},
3249 {"auto", "autoselection", 0, AV_OPT_TYPE_CONST, {.i64=-1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3250 {"main", "Select Main/Left channel", 0, AV_OPT_TYPE_CONST, {.i64= 1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3251 {"sub" , "Select Sub/Right channel", 0, AV_OPT_TYPE_CONST, {.i64= 2}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3252 {"both", "Select both channels", 0, AV_OPT_TYPE_CONST, {.i64= 0}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3257 static const AVClass aac_decoder_class = {
3258 .class_name = "AAC decoder",
3259 .item_name = av_default_item_name,
3261 .version = LIBAVUTIL_VERSION_INT,