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);
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]++;
455 // Try to sniff a reasonable channel order, otherwise output the
456 // channels in the order the PCE declared them.
457 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
458 layout = sniff_channel_order(layout_map, tags);
459 for (i = 0; i < tags; i++) {
460 int type = layout_map[i][0];
461 int id = layout_map[i][1];
462 int iid = id_map[type][id];
463 int position = layout_map[i][2];
464 // Allocate or free elements depending on if they are in the
465 // current program configuration.
466 ret = che_configure(ac, position, type, iid, &channels);
469 ac->tag_che_map[type][id] = ac->che[type][iid];
471 if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
472 if (layout == AV_CH_FRONT_CENTER) {
473 layout = AV_CH_FRONT_LEFT|AV_CH_FRONT_RIGHT;
479 if (layout) avctx->channel_layout = layout;
480 ac->oc[1].channel_layout = layout;
481 avctx->channels = ac->oc[1].channels = channels;
482 ac->oc[1].status = oc_type;
485 if ((ret = frame_configure_elements(ac->avctx)) < 0)
492 static void flush(AVCodecContext *avctx)
494 AACContext *ac= avctx->priv_data;
497 for (type = 3; type >= 0; type--) {
498 for (i = 0; i < MAX_ELEM_ID; i++) {
499 ChannelElement *che = ac->che[type][i];
501 for (j = 0; j <= 1; j++) {
502 memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
510 * Set up channel positions based on a default channel configuration
511 * as specified in table 1.17.
513 * @return Returns error status. 0 - OK, !0 - error
515 static int set_default_channel_config(AVCodecContext *avctx,
516 uint8_t (*layout_map)[3],
520 if (channel_config < 1 || (channel_config > 7 && channel_config < 11) ||
521 channel_config > 12) {
522 av_log(avctx, AV_LOG_ERROR,
523 "invalid default channel configuration (%d)\n",
525 return AVERROR_INVALIDDATA;
527 *tags = tags_per_config[channel_config];
528 memcpy(layout_map, aac_channel_layout_map[channel_config - 1],
529 *tags * sizeof(*layout_map));
532 * AAC specification has 7.1(wide) as a default layout for 8-channel streams.
533 * However, at least Nero AAC encoder encodes 7.1 streams using the default
534 * channel config 7, mapping the side channels of the original audio stream
535 * to the second AAC_CHANNEL_FRONT pair in the AAC stream. Similarly, e.g. FAAD
536 * decodes the second AAC_CHANNEL_FRONT pair as side channels, therefore decoding
537 * the incorrect streams as if they were correct (and as the encoder intended).
539 * As actual intended 7.1(wide) streams are very rare, default to assuming a
540 * 7.1 layout was intended.
542 if (channel_config == 7 && avctx->strict_std_compliance < FF_COMPLIANCE_STRICT) {
543 av_log(avctx, AV_LOG_INFO, "Assuming an incorrectly encoded 7.1 channel layout"
544 " instead of a spec-compliant 7.1(wide) layout, use -strict %d to decode"
545 " according to the specification instead.\n", FF_COMPLIANCE_STRICT);
546 layout_map[2][2] = AAC_CHANNEL_SIDE;
552 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
554 /* For PCE based channel configurations map the channels solely based
556 if (!ac->oc[1].m4ac.chan_config) {
557 return ac->tag_che_map[type][elem_id];
559 // Allow single CPE stereo files to be signalled with mono configuration.
560 if (!ac->tags_mapped && type == TYPE_CPE &&
561 ac->oc[1].m4ac.chan_config == 1) {
562 uint8_t layout_map[MAX_ELEM_ID*4][3];
564 push_output_configuration(ac);
566 av_log(ac->avctx, AV_LOG_DEBUG, "mono with CPE\n");
568 if (set_default_channel_config(ac->avctx, layout_map,
569 &layout_map_tags, 2) < 0)
571 if (output_configure(ac, layout_map, layout_map_tags,
572 OC_TRIAL_FRAME, 1) < 0)
575 ac->oc[1].m4ac.chan_config = 2;
576 ac->oc[1].m4ac.ps = 0;
579 if (!ac->tags_mapped && type == TYPE_SCE &&
580 ac->oc[1].m4ac.chan_config == 2) {
581 uint8_t layout_map[MAX_ELEM_ID * 4][3];
583 push_output_configuration(ac);
585 av_log(ac->avctx, AV_LOG_DEBUG, "stereo with SCE\n");
587 if (set_default_channel_config(ac->avctx, layout_map,
588 &layout_map_tags, 1) < 0)
590 if (output_configure(ac, layout_map, layout_map_tags,
591 OC_TRIAL_FRAME, 1) < 0)
594 ac->oc[1].m4ac.chan_config = 1;
595 if (ac->oc[1].m4ac.sbr)
596 ac->oc[1].m4ac.ps = -1;
598 /* For indexed channel configurations map the channels solely based
600 switch (ac->oc[1].m4ac.chan_config) {
603 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
605 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
608 if (ac->tags_mapped == 2 &&
609 ac->oc[1].m4ac.chan_config == 11 &&
612 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
615 /* Some streams incorrectly code 5.1 audio as
616 * SCE[0] CPE[0] CPE[1] SCE[1]
618 * SCE[0] CPE[0] CPE[1] LFE[0].
619 * If we seem to have encountered such a stream, transfer
620 * the LFE[0] element to the SCE[1]'s mapping */
621 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
622 if (!ac->warned_remapping_once && (type != TYPE_LFE || elem_id != 0)) {
623 av_log(ac->avctx, AV_LOG_WARNING,
624 "This stream seems to incorrectly report its last channel as %s[%d], mapping to LFE[0]\n",
625 type == TYPE_SCE ? "SCE" : "LFE", elem_id);
626 ac->warned_remapping_once++;
629 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
632 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
634 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
637 /* Some streams incorrectly code 4.0 audio as
638 * SCE[0] CPE[0] LFE[0]
640 * SCE[0] CPE[0] SCE[1].
641 * If we seem to have encountered such a stream, transfer
642 * the SCE[1] element to the LFE[0]'s mapping */
643 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
644 if (!ac->warned_remapping_once && (type != TYPE_SCE || elem_id != 1)) {
645 av_log(ac->avctx, AV_LOG_WARNING,
646 "This stream seems to incorrectly report its last channel as %s[%d], mapping to SCE[1]\n",
647 type == TYPE_SCE ? "SCE" : "LFE", elem_id);
648 ac->warned_remapping_once++;
651 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_SCE][1];
653 if (ac->tags_mapped == 2 &&
654 ac->oc[1].m4ac.chan_config == 4 &&
657 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
661 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) &&
664 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
665 } else if (ac->oc[1].m4ac.chan_config == 2) {
669 if (!ac->tags_mapped && type == TYPE_SCE) {
671 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
679 * Decode an array of 4 bit element IDs, optionally interleaved with a
680 * stereo/mono switching bit.
682 * @param type speaker type/position for these channels
684 static void decode_channel_map(uint8_t layout_map[][3],
685 enum ChannelPosition type,
686 GetBitContext *gb, int n)
689 enum RawDataBlockType syn_ele;
691 case AAC_CHANNEL_FRONT:
692 case AAC_CHANNEL_BACK:
693 case AAC_CHANNEL_SIDE:
694 syn_ele = get_bits1(gb);
700 case AAC_CHANNEL_LFE:
704 // AAC_CHANNEL_OFF has no channel map
707 layout_map[0][0] = syn_ele;
708 layout_map[0][1] = get_bits(gb, 4);
709 layout_map[0][2] = type;
715 * Decode program configuration element; reference: table 4.2.
717 * @return Returns error status. 0 - OK, !0 - error
719 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
720 uint8_t (*layout_map)[3],
723 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc;
728 skip_bits(gb, 2); // object_type
730 sampling_index = get_bits(gb, 4);
731 if (m4ac->sampling_index != sampling_index)
732 av_log(avctx, AV_LOG_WARNING,
733 "Sample rate index in program config element does not "
734 "match the sample rate index configured by the container.\n");
736 num_front = get_bits(gb, 4);
737 num_side = get_bits(gb, 4);
738 num_back = get_bits(gb, 4);
739 num_lfe = get_bits(gb, 2);
740 num_assoc_data = get_bits(gb, 3);
741 num_cc = get_bits(gb, 4);
744 skip_bits(gb, 4); // mono_mixdown_tag
746 skip_bits(gb, 4); // stereo_mixdown_tag
749 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
751 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
752 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
755 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
757 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
759 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
761 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
764 skip_bits_long(gb, 4 * num_assoc_data);
766 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
771 /* comment field, first byte is length */
772 comment_len = get_bits(gb, 8) * 8;
773 if (get_bits_left(gb) < comment_len) {
774 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
775 return AVERROR_INVALIDDATA;
777 skip_bits_long(gb, comment_len);
782 * Decode GA "General Audio" specific configuration; reference: table 4.1.
784 * @param ac pointer to AACContext, may be null
785 * @param avctx pointer to AVCCodecContext, used for logging
787 * @return Returns error status. 0 - OK, !0 - error
789 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
791 MPEG4AudioConfig *m4ac,
794 int extension_flag, ret, ep_config, res_flags;
795 uint8_t layout_map[MAX_ELEM_ID*4][3];
798 if (get_bits1(gb)) { // frameLengthFlag
799 avpriv_request_sample(avctx, "960/120 MDCT window");
800 return AVERROR_PATCHWELCOME;
802 m4ac->frame_length_short = 0;
804 if (get_bits1(gb)) // dependsOnCoreCoder
805 skip_bits(gb, 14); // coreCoderDelay
806 extension_flag = get_bits1(gb);
808 if (m4ac->object_type == AOT_AAC_SCALABLE ||
809 m4ac->object_type == AOT_ER_AAC_SCALABLE)
810 skip_bits(gb, 3); // layerNr
812 if (channel_config == 0) {
813 skip_bits(gb, 4); // element_instance_tag
814 tags = decode_pce(avctx, m4ac, layout_map, gb);
818 if ((ret = set_default_channel_config(avctx, layout_map,
819 &tags, channel_config)))
823 if (count_channels(layout_map, tags) > 1) {
825 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
828 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
831 if (extension_flag) {
832 switch (m4ac->object_type) {
834 skip_bits(gb, 5); // numOfSubFrame
835 skip_bits(gb, 11); // layer_length
839 case AOT_ER_AAC_SCALABLE:
841 res_flags = get_bits(gb, 3);
843 avpriv_report_missing_feature(avctx,
844 "AAC data resilience (flags %x)",
846 return AVERROR_PATCHWELCOME;
850 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
852 switch (m4ac->object_type) {
855 case AOT_ER_AAC_SCALABLE:
857 ep_config = get_bits(gb, 2);
859 avpriv_report_missing_feature(avctx,
860 "epConfig %d", ep_config);
861 return AVERROR_PATCHWELCOME;
867 static int decode_eld_specific_config(AACContext *ac, AVCodecContext *avctx,
869 MPEG4AudioConfig *m4ac,
872 int ret, ep_config, res_flags;
873 uint8_t layout_map[MAX_ELEM_ID*4][3];
875 const int ELDEXT_TERM = 0;
880 if (get_bits1(gb)) { // frameLengthFlag
881 avpriv_request_sample(avctx, "960/120 MDCT window");
882 return AVERROR_PATCHWELCOME;
885 m4ac->frame_length_short = get_bits1(gb);
887 res_flags = get_bits(gb, 3);
889 avpriv_report_missing_feature(avctx,
890 "AAC data resilience (flags %x)",
892 return AVERROR_PATCHWELCOME;
895 if (get_bits1(gb)) { // ldSbrPresentFlag
896 avpriv_report_missing_feature(avctx,
898 return AVERROR_PATCHWELCOME;
901 while (get_bits(gb, 4) != ELDEXT_TERM) {
902 int len = get_bits(gb, 4);
904 len += get_bits(gb, 8);
906 len += get_bits(gb, 16);
907 if (get_bits_left(gb) < len * 8 + 4) {
908 av_log(avctx, AV_LOG_ERROR, overread_err);
909 return AVERROR_INVALIDDATA;
911 skip_bits_long(gb, 8 * len);
914 if ((ret = set_default_channel_config(avctx, layout_map,
915 &tags, channel_config)))
918 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
921 ep_config = get_bits(gb, 2);
923 avpriv_report_missing_feature(avctx,
924 "epConfig %d", ep_config);
925 return AVERROR_PATCHWELCOME;
931 * Decode audio specific configuration; reference: table 1.13.
933 * @param ac pointer to AACContext, may be null
934 * @param avctx pointer to AVCCodecContext, used for logging
935 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
936 * @param data pointer to buffer holding an audio specific config
937 * @param bit_size size of audio specific config or data in bits
938 * @param sync_extension look for an appended sync extension
940 * @return Returns error status or number of consumed bits. <0 - error
942 static int decode_audio_specific_config(AACContext *ac,
943 AVCodecContext *avctx,
944 MPEG4AudioConfig *m4ac,
945 const uint8_t *data, int64_t bit_size,
951 if (bit_size < 0 || bit_size > INT_MAX) {
952 av_log(avctx, AV_LOG_ERROR, "Audio specific config size is invalid\n");
953 return AVERROR_INVALIDDATA;
956 ff_dlog(avctx, "audio specific config size %d\n", (int)bit_size >> 3);
957 for (i = 0; i < bit_size >> 3; i++)
958 ff_dlog(avctx, "%02x ", data[i]);
959 ff_dlog(avctx, "\n");
961 if ((ret = init_get_bits(&gb, data, bit_size)) < 0)
964 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size,
965 sync_extension)) < 0)
966 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,
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);
1016 * linear congruential pseudorandom number generator
1018 * @param previous_val pointer to the current state of the generator
1020 * @return Returns a 32-bit pseudorandom integer
1022 static av_always_inline int lcg_random(unsigned previous_val)
1024 union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
1028 static void reset_all_predictors(PredictorState *ps)
1031 for (i = 0; i < MAX_PREDICTORS; i++)
1032 reset_predict_state(&ps[i]);
1035 static int sample_rate_idx (int rate)
1037 if (92017 <= rate) return 0;
1038 else if (75132 <= rate) return 1;
1039 else if (55426 <= rate) return 2;
1040 else if (46009 <= rate) return 3;
1041 else if (37566 <= rate) return 4;
1042 else if (27713 <= rate) return 5;
1043 else if (23004 <= rate) return 6;
1044 else if (18783 <= rate) return 7;
1045 else if (13856 <= rate) return 8;
1046 else if (11502 <= rate) return 9;
1047 else if (9391 <= rate) return 10;
1051 static void reset_predictor_group(PredictorState *ps, int group_num)
1054 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
1055 reset_predict_state(&ps[i]);
1058 #define AAC_INIT_VLC_STATIC(num, size) \
1059 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
1060 ff_aac_spectral_bits[num], sizeof(ff_aac_spectral_bits[num][0]), \
1061 sizeof(ff_aac_spectral_bits[num][0]), \
1062 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), \
1063 sizeof(ff_aac_spectral_codes[num][0]), \
1066 static void aacdec_init(AACContext *ac);
1068 static av_cold void aac_static_table_init(void)
1070 AAC_INIT_VLC_STATIC( 0, 304);
1071 AAC_INIT_VLC_STATIC( 1, 270);
1072 AAC_INIT_VLC_STATIC( 2, 550);
1073 AAC_INIT_VLC_STATIC( 3, 300);
1074 AAC_INIT_VLC_STATIC( 4, 328);
1075 AAC_INIT_VLC_STATIC( 5, 294);
1076 AAC_INIT_VLC_STATIC( 6, 306);
1077 AAC_INIT_VLC_STATIC( 7, 268);
1078 AAC_INIT_VLC_STATIC( 8, 510);
1079 AAC_INIT_VLC_STATIC( 9, 366);
1080 AAC_INIT_VLC_STATIC(10, 462);
1082 AAC_RENAME(ff_aac_sbr_init)();
1086 INIT_VLC_STATIC(&vlc_scalefactors, 7,
1087 FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
1088 ff_aac_scalefactor_bits,
1089 sizeof(ff_aac_scalefactor_bits[0]),
1090 sizeof(ff_aac_scalefactor_bits[0]),
1091 ff_aac_scalefactor_code,
1092 sizeof(ff_aac_scalefactor_code[0]),
1093 sizeof(ff_aac_scalefactor_code[0]),
1096 // window initialization
1097 AAC_RENAME(ff_kbd_window_init)(AAC_RENAME(ff_aac_kbd_long_1024), 4.0, 1024);
1098 AAC_RENAME(ff_kbd_window_init)(AAC_RENAME(ff_aac_kbd_short_128), 6.0, 128);
1099 AAC_RENAME(ff_init_ff_sine_windows)(10);
1100 AAC_RENAME(ff_init_ff_sine_windows)( 9);
1101 AAC_RENAME(ff_init_ff_sine_windows)( 7);
1103 AAC_RENAME(cbrt_tableinit)();
1106 static AVOnce aac_table_init = AV_ONCE_INIT;
1108 static av_cold int aac_decode_init(AVCodecContext *avctx)
1110 AACContext *ac = avctx->priv_data;
1113 ret = ff_thread_once(&aac_table_init, &aac_static_table_init);
1115 return AVERROR_UNKNOWN;
1118 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
1122 avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
1124 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1125 #endif /* USE_FIXED */
1127 if (avctx->extradata_size > 0) {
1128 if ((ret = decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
1130 avctx->extradata_size * 8LL,
1135 uint8_t layout_map[MAX_ELEM_ID*4][3];
1136 int layout_map_tags;
1138 sr = sample_rate_idx(avctx->sample_rate);
1139 ac->oc[1].m4ac.sampling_index = sr;
1140 ac->oc[1].m4ac.channels = avctx->channels;
1141 ac->oc[1].m4ac.sbr = -1;
1142 ac->oc[1].m4ac.ps = -1;
1144 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
1145 if (ff_mpeg4audio_channels[i] == avctx->channels)
1147 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
1150 ac->oc[1].m4ac.chan_config = i;
1152 if (ac->oc[1].m4ac.chan_config) {
1153 int ret = set_default_channel_config(avctx, layout_map,
1154 &layout_map_tags, ac->oc[1].m4ac.chan_config);
1156 output_configure(ac, layout_map, layout_map_tags,
1158 else if (avctx->err_recognition & AV_EF_EXPLODE)
1159 return AVERROR_INVALIDDATA;
1163 if (avctx->channels > MAX_CHANNELS) {
1164 av_log(avctx, AV_LOG_ERROR, "Too many channels\n");
1165 return AVERROR_INVALIDDATA;
1169 ac->fdsp = avpriv_alloc_fixed_dsp(avctx->flags & AV_CODEC_FLAG_BITEXACT);
1171 ac->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
1172 #endif /* USE_FIXED */
1174 return AVERROR(ENOMEM);
1177 ac->random_state = 0x1f2e3d4c;
1179 AAC_RENAME_32(ff_mdct_init)(&ac->mdct, 11, 1, 1.0 / RANGE15(1024.0));
1180 AAC_RENAME_32(ff_mdct_init)(&ac->mdct_ld, 10, 1, 1.0 / RANGE15(512.0));
1181 AAC_RENAME_32(ff_mdct_init)(&ac->mdct_small, 8, 1, 1.0 / RANGE15(128.0));
1182 AAC_RENAME_32(ff_mdct_init)(&ac->mdct_ltp, 11, 0, RANGE15(-2.0));
1184 ret = ff_imdct15_init(&ac->mdct480, 5);
1193 * Skip data_stream_element; reference: table 4.10.
1195 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
1197 int byte_align = get_bits1(gb);
1198 int count = get_bits(gb, 8);
1200 count += get_bits(gb, 8);
1204 if (get_bits_left(gb) < 8 * count) {
1205 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
1206 return AVERROR_INVALIDDATA;
1208 skip_bits_long(gb, 8 * count);
1212 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
1216 if (get_bits1(gb)) {
1217 ics->predictor_reset_group = get_bits(gb, 5);
1218 if (ics->predictor_reset_group == 0 ||
1219 ics->predictor_reset_group > 30) {
1220 av_log(ac->avctx, AV_LOG_ERROR,
1221 "Invalid Predictor Reset Group.\n");
1222 return AVERROR_INVALIDDATA;
1225 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
1226 ics->prediction_used[sfb] = get_bits1(gb);
1232 * Decode Long Term Prediction data; reference: table 4.xx.
1234 static void decode_ltp(LongTermPrediction *ltp,
1235 GetBitContext *gb, uint8_t max_sfb)
1239 ltp->lag = get_bits(gb, 11);
1240 ltp->coef = ltp_coef[get_bits(gb, 3)];
1241 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1242 ltp->used[sfb] = get_bits1(gb);
1246 * Decode Individual Channel Stream info; reference: table 4.6.
1248 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
1251 const MPEG4AudioConfig *const m4ac = &ac->oc[1].m4ac;
1252 const int aot = m4ac->object_type;
1253 const int sampling_index = m4ac->sampling_index;
1254 if (aot != AOT_ER_AAC_ELD) {
1255 if (get_bits1(gb)) {
1256 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1257 if (ac->avctx->err_recognition & AV_EF_BITSTREAM)
1258 return AVERROR_INVALIDDATA;
1260 ics->window_sequence[1] = ics->window_sequence[0];
1261 ics->window_sequence[0] = get_bits(gb, 2);
1262 if (aot == AOT_ER_AAC_LD &&
1263 ics->window_sequence[0] != ONLY_LONG_SEQUENCE) {
1264 av_log(ac->avctx, AV_LOG_ERROR,
1265 "AAC LD is only defined for ONLY_LONG_SEQUENCE but "
1266 "window sequence %d found.\n", ics->window_sequence[0]);
1267 ics->window_sequence[0] = ONLY_LONG_SEQUENCE;
1268 return AVERROR_INVALIDDATA;
1270 ics->use_kb_window[1] = ics->use_kb_window[0];
1271 ics->use_kb_window[0] = get_bits1(gb);
1273 ics->num_window_groups = 1;
1274 ics->group_len[0] = 1;
1275 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1277 ics->max_sfb = get_bits(gb, 4);
1278 for (i = 0; i < 7; i++) {
1279 if (get_bits1(gb)) {
1280 ics->group_len[ics->num_window_groups - 1]++;
1282 ics->num_window_groups++;
1283 ics->group_len[ics->num_window_groups - 1] = 1;
1286 ics->num_windows = 8;
1287 ics->swb_offset = ff_swb_offset_128[sampling_index];
1288 ics->num_swb = ff_aac_num_swb_128[sampling_index];
1289 ics->tns_max_bands = ff_tns_max_bands_128[sampling_index];
1290 ics->predictor_present = 0;
1292 ics->max_sfb = get_bits(gb, 6);
1293 ics->num_windows = 1;
1294 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD) {
1295 if (m4ac->frame_length_short) {
1296 ics->swb_offset = ff_swb_offset_480[sampling_index];
1297 ics->num_swb = ff_aac_num_swb_480[sampling_index];
1298 ics->tns_max_bands = ff_tns_max_bands_480[sampling_index];
1300 ics->swb_offset = ff_swb_offset_512[sampling_index];
1301 ics->num_swb = ff_aac_num_swb_512[sampling_index];
1302 ics->tns_max_bands = ff_tns_max_bands_512[sampling_index];
1304 if (!ics->num_swb || !ics->swb_offset)
1307 ics->swb_offset = ff_swb_offset_1024[sampling_index];
1308 ics->num_swb = ff_aac_num_swb_1024[sampling_index];
1309 ics->tns_max_bands = ff_tns_max_bands_1024[sampling_index];
1311 if (aot != AOT_ER_AAC_ELD) {
1312 ics->predictor_present = get_bits1(gb);
1313 ics->predictor_reset_group = 0;
1315 if (ics->predictor_present) {
1316 if (aot == AOT_AAC_MAIN) {
1317 if (decode_prediction(ac, ics, gb)) {
1320 } else if (aot == AOT_AAC_LC ||
1321 aot == AOT_ER_AAC_LC) {
1322 av_log(ac->avctx, AV_LOG_ERROR,
1323 "Prediction is not allowed in AAC-LC.\n");
1326 if (aot == AOT_ER_AAC_LD) {
1327 av_log(ac->avctx, AV_LOG_ERROR,
1328 "LTP in ER AAC LD not yet implemented.\n");
1329 return AVERROR_PATCHWELCOME;
1331 if ((ics->ltp.present = get_bits(gb, 1)))
1332 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1337 if (ics->max_sfb > ics->num_swb) {
1338 av_log(ac->avctx, AV_LOG_ERROR,
1339 "Number of scalefactor bands in group (%d) "
1340 "exceeds limit (%d).\n",
1341 ics->max_sfb, ics->num_swb);
1348 return AVERROR_INVALIDDATA;
1352 * Decode band types (section_data payload); reference: table 4.46.
1354 * @param band_type array of the used band type
1355 * @param band_type_run_end array of the last scalefactor band of a band type run
1357 * @return Returns error status. 0 - OK, !0 - error
1359 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1360 int band_type_run_end[120], GetBitContext *gb,
1361 IndividualChannelStream *ics)
1364 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1365 for (g = 0; g < ics->num_window_groups; g++) {
1367 while (k < ics->max_sfb) {
1368 uint8_t sect_end = k;
1370 int sect_band_type = get_bits(gb, 4);
1371 if (sect_band_type == 12) {
1372 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1373 return AVERROR_INVALIDDATA;
1376 sect_len_incr = get_bits(gb, bits);
1377 sect_end += sect_len_incr;
1378 if (get_bits_left(gb) < 0) {
1379 av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1380 return AVERROR_INVALIDDATA;
1382 if (sect_end > ics->max_sfb) {
1383 av_log(ac->avctx, AV_LOG_ERROR,
1384 "Number of bands (%d) exceeds limit (%d).\n",
1385 sect_end, ics->max_sfb);
1386 return AVERROR_INVALIDDATA;
1388 } while (sect_len_incr == (1 << bits) - 1);
1389 for (; k < sect_end; k++) {
1390 band_type [idx] = sect_band_type;
1391 band_type_run_end[idx++] = sect_end;
1399 * Decode scalefactors; reference: table 4.47.
1401 * @param global_gain first scalefactor value as scalefactors are differentially coded
1402 * @param band_type array of the used band type
1403 * @param band_type_run_end array of the last scalefactor band of a band type run
1404 * @param sf array of scalefactors or intensity stereo positions
1406 * @return Returns error status. 0 - OK, !0 - error
1408 static int decode_scalefactors(AACContext *ac, INTFLOAT sf[120], GetBitContext *gb,
1409 unsigned int global_gain,
1410 IndividualChannelStream *ics,
1411 enum BandType band_type[120],
1412 int band_type_run_end[120])
1415 int offset[3] = { global_gain, global_gain - NOISE_OFFSET, 0 };
1418 for (g = 0; g < ics->num_window_groups; g++) {
1419 for (i = 0; i < ics->max_sfb;) {
1420 int run_end = band_type_run_end[idx];
1421 if (band_type[idx] == ZERO_BT) {
1422 for (; i < run_end; i++, idx++)
1424 } else if ((band_type[idx] == INTENSITY_BT) ||
1425 (band_type[idx] == INTENSITY_BT2)) {
1426 for (; i < run_end; i++, idx++) {
1427 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - SCALE_DIFF_ZERO;
1428 clipped_offset = av_clip(offset[2], -155, 100);
1429 if (offset[2] != clipped_offset) {
1430 avpriv_request_sample(ac->avctx,
1431 "If you heard an audible artifact, there may be a bug in the decoder. "
1432 "Clipped intensity stereo position (%d -> %d)",
1433 offset[2], clipped_offset);
1436 sf[idx] = 100 - clipped_offset;
1438 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1439 #endif /* USE_FIXED */
1441 } else if (band_type[idx] == NOISE_BT) {
1442 for (; i < run_end; i++, idx++) {
1443 if (noise_flag-- > 0)
1444 offset[1] += get_bits(gb, NOISE_PRE_BITS) - NOISE_PRE;
1446 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - SCALE_DIFF_ZERO;
1447 clipped_offset = av_clip(offset[1], -100, 155);
1448 if (offset[1] != clipped_offset) {
1449 avpriv_request_sample(ac->avctx,
1450 "If you heard an audible artifact, there may be a bug in the decoder. "
1451 "Clipped noise gain (%d -> %d)",
1452 offset[1], clipped_offset);
1455 sf[idx] = -(100 + clipped_offset);
1457 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1458 #endif /* USE_FIXED */
1461 for (; i < run_end; i++, idx++) {
1462 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - SCALE_DIFF_ZERO;
1463 if (offset[0] > 255U) {
1464 av_log(ac->avctx, AV_LOG_ERROR,
1465 "Scalefactor (%d) out of range.\n", offset[0]);
1466 return AVERROR_INVALIDDATA;
1469 sf[idx] = -offset[0];
1471 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1472 #endif /* USE_FIXED */
1481 * Decode pulse data; reference: table 4.7.
1483 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1484 const uint16_t *swb_offset, int num_swb)
1487 pulse->num_pulse = get_bits(gb, 2) + 1;
1488 pulse_swb = get_bits(gb, 6);
1489 if (pulse_swb >= num_swb)
1491 pulse->pos[0] = swb_offset[pulse_swb];
1492 pulse->pos[0] += get_bits(gb, 5);
1493 if (pulse->pos[0] >= swb_offset[num_swb])
1495 pulse->amp[0] = get_bits(gb, 4);
1496 for (i = 1; i < pulse->num_pulse; i++) {
1497 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1498 if (pulse->pos[i] >= swb_offset[num_swb])
1500 pulse->amp[i] = get_bits(gb, 4);
1506 * Decode Temporal Noise Shaping data; reference: table 4.48.
1508 * @return Returns error status. 0 - OK, !0 - error
1510 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1511 GetBitContext *gb, const IndividualChannelStream *ics)
1513 int w, filt, i, coef_len, coef_res, coef_compress;
1514 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1515 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1516 for (w = 0; w < ics->num_windows; w++) {
1517 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1518 coef_res = get_bits1(gb);
1520 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1522 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1524 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1525 av_log(ac->avctx, AV_LOG_ERROR,
1526 "TNS filter order %d is greater than maximum %d.\n",
1527 tns->order[w][filt], tns_max_order);
1528 tns->order[w][filt] = 0;
1529 return AVERROR_INVALIDDATA;
1531 if (tns->order[w][filt]) {
1532 tns->direction[w][filt] = get_bits1(gb);
1533 coef_compress = get_bits1(gb);
1534 coef_len = coef_res + 3 - coef_compress;
1535 tmp2_idx = 2 * coef_compress + coef_res;
1537 for (i = 0; i < tns->order[w][filt]; i++)
1538 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1547 * Decode Mid/Side data; reference: table 4.54.
1549 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1550 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1551 * [3] reserved for scalable AAC
1553 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1557 int max_idx = cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb;
1558 if (ms_present == 1) {
1559 for (idx = 0; idx < max_idx; idx++)
1560 cpe->ms_mask[idx] = get_bits1(gb);
1561 } else if (ms_present == 2) {
1562 memset(cpe->ms_mask, 1, max_idx * sizeof(cpe->ms_mask[0]));
1567 * Decode spectral data; reference: table 4.50.
1568 * Dequantize and scale spectral data; reference: 4.6.3.3.
1570 * @param coef array of dequantized, scaled spectral data
1571 * @param sf array of scalefactors or intensity stereo positions
1572 * @param pulse_present set if pulses are present
1573 * @param pulse pointer to pulse data struct
1574 * @param band_type array of the used band type
1576 * @return Returns error status. 0 - OK, !0 - error
1578 static int decode_spectrum_and_dequant(AACContext *ac, INTFLOAT coef[1024],
1579 GetBitContext *gb, const INTFLOAT sf[120],
1580 int pulse_present, const Pulse *pulse,
1581 const IndividualChannelStream *ics,
1582 enum BandType band_type[120])
1584 int i, k, g, idx = 0;
1585 const int c = 1024 / ics->num_windows;
1586 const uint16_t *offsets = ics->swb_offset;
1587 INTFLOAT *coef_base = coef;
1589 for (g = 0; g < ics->num_windows; g++)
1590 memset(coef + g * 128 + offsets[ics->max_sfb], 0,
1591 sizeof(INTFLOAT) * (c - offsets[ics->max_sfb]));
1593 for (g = 0; g < ics->num_window_groups; g++) {
1594 unsigned g_len = ics->group_len[g];
1596 for (i = 0; i < ics->max_sfb; i++, idx++) {
1597 const unsigned cbt_m1 = band_type[idx] - 1;
1598 INTFLOAT *cfo = coef + offsets[i];
1599 int off_len = offsets[i + 1] - offsets[i];
1602 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1603 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1604 memset(cfo, 0, off_len * sizeof(*cfo));
1606 } else if (cbt_m1 == NOISE_BT - 1) {
1607 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1610 #endif /* !USE_FIXED */
1611 INTFLOAT band_energy;
1613 for (k = 0; k < off_len; k++) {
1614 ac->random_state = lcg_random(ac->random_state);
1616 cfo[k] = ac->random_state >> 3;
1618 cfo[k] = ac->random_state;
1619 #endif /* USE_FIXED */
1623 band_energy = ac->fdsp->scalarproduct_fixed(cfo, cfo, off_len);
1624 band_energy = fixed_sqrt(band_energy, 31);
1625 noise_scale(cfo, sf[idx], band_energy, off_len);
1627 band_energy = ac->fdsp->scalarproduct_float(cfo, cfo, off_len);
1628 scale = sf[idx] / sqrtf(band_energy);
1629 ac->fdsp->vector_fmul_scalar(cfo, cfo, scale, off_len);
1630 #endif /* USE_FIXED */
1634 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1635 #endif /* !USE_FIXED */
1636 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1637 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1638 OPEN_READER(re, gb);
1640 switch (cbt_m1 >> 1) {
1642 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1650 UPDATE_CACHE(re, gb);
1651 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1652 cb_idx = cb_vector_idx[code];
1654 cf = DEC_SQUAD(cf, cb_idx);
1656 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1657 #endif /* USE_FIXED */
1663 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1673 UPDATE_CACHE(re, gb);
1674 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1675 cb_idx = cb_vector_idx[code];
1676 nnz = cb_idx >> 8 & 15;
1677 bits = nnz ? GET_CACHE(re, gb) : 0;
1678 LAST_SKIP_BITS(re, gb, nnz);
1680 cf = DEC_UQUAD(cf, cb_idx, bits);
1682 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1683 #endif /* USE_FIXED */
1689 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1697 UPDATE_CACHE(re, gb);
1698 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1699 cb_idx = cb_vector_idx[code];
1701 cf = DEC_SPAIR(cf, cb_idx);
1703 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1704 #endif /* USE_FIXED */
1711 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1721 UPDATE_CACHE(re, gb);
1722 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1723 cb_idx = cb_vector_idx[code];
1724 nnz = cb_idx >> 8 & 15;
1725 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1726 LAST_SKIP_BITS(re, gb, nnz);
1728 cf = DEC_UPAIR(cf, cb_idx, sign);
1730 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1731 #endif /* USE_FIXED */
1737 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1743 uint32_t *icf = (uint32_t *) cf;
1744 #endif /* USE_FIXED */
1754 UPDATE_CACHE(re, gb);
1755 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1763 cb_idx = cb_vector_idx[code];
1766 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1767 LAST_SKIP_BITS(re, gb, nnz);
1769 for (j = 0; j < 2; j++) {
1773 /* The total length of escape_sequence must be < 22 bits according
1774 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1775 UPDATE_CACHE(re, gb);
1776 b = GET_CACHE(re, gb);
1777 b = 31 - av_log2(~b);
1780 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1781 return AVERROR_INVALIDDATA;
1784 SKIP_BITS(re, gb, b + 1);
1786 n = (1 << b) + SHOW_UBITS(re, gb, b);
1787 LAST_SKIP_BITS(re, gb, b);
1794 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1795 #endif /* USE_FIXED */
1804 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1805 *icf++ = (bits & 1U<<31) | v;
1806 #endif /* USE_FIXED */
1813 ac->fdsp->vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1814 #endif /* !USE_FIXED */
1818 CLOSE_READER(re, gb);
1824 if (pulse_present) {
1826 for (i = 0; i < pulse->num_pulse; i++) {
1827 INTFLOAT co = coef_base[ pulse->pos[i] ];
1828 while (offsets[idx + 1] <= pulse->pos[i])
1830 if (band_type[idx] != NOISE_BT && sf[idx]) {
1831 INTFLOAT ico = -pulse->amp[i];
1834 ico = co + (co > 0 ? -ico : ico);
1836 coef_base[ pulse->pos[i] ] = ico;
1840 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1842 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1843 #endif /* USE_FIXED */
1850 for (g = 0; g < ics->num_window_groups; g++) {
1851 unsigned g_len = ics->group_len[g];
1853 for (i = 0; i < ics->max_sfb; i++, idx++) {
1854 const unsigned cbt_m1 = band_type[idx] - 1;
1855 int *cfo = coef + offsets[i];
1856 int off_len = offsets[i + 1] - offsets[i];
1859 if (cbt_m1 < NOISE_BT - 1) {
1860 for (group = 0; group < (int)g_len; group++, cfo+=128) {
1861 ac->vector_pow43(cfo, off_len);
1862 ac->subband_scale(cfo, cfo, sf[idx], 34, off_len);
1868 #endif /* USE_FIXED */
1873 * Apply AAC-Main style frequency domain prediction.
1875 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1879 if (!sce->ics.predictor_initialized) {
1880 reset_all_predictors(sce->predictor_state);
1881 sce->ics.predictor_initialized = 1;
1884 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1886 sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index];
1888 for (k = sce->ics.swb_offset[sfb];
1889 k < sce->ics.swb_offset[sfb + 1];
1891 predict(&sce->predictor_state[k], &sce->coeffs[k],
1892 sce->ics.predictor_present &&
1893 sce->ics.prediction_used[sfb]);
1896 if (sce->ics.predictor_reset_group)
1897 reset_predictor_group(sce->predictor_state,
1898 sce->ics.predictor_reset_group);
1900 reset_all_predictors(sce->predictor_state);
1904 * Decode an individual_channel_stream payload; reference: table 4.44.
1906 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1907 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1909 * @return Returns error status. 0 - OK, !0 - error
1911 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1912 GetBitContext *gb, int common_window, int scale_flag)
1915 TemporalNoiseShaping *tns = &sce->tns;
1916 IndividualChannelStream *ics = &sce->ics;
1917 INTFLOAT *out = sce->coeffs;
1918 int global_gain, eld_syntax, er_syntax, pulse_present = 0;
1921 eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1922 er_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_LC ||
1923 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LTP ||
1924 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LD ||
1925 ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1927 /* This assignment is to silence a GCC warning about the variable being used
1928 * uninitialized when in fact it always is.
1930 pulse.num_pulse = 0;
1932 global_gain = get_bits(gb, 8);
1934 if (!common_window && !scale_flag) {
1935 if (decode_ics_info(ac, ics, gb) < 0)
1936 return AVERROR_INVALIDDATA;
1939 if ((ret = decode_band_types(ac, sce->band_type,
1940 sce->band_type_run_end, gb, ics)) < 0)
1942 if ((ret = decode_scalefactors(ac, sce->sf, gb, global_gain, ics,
1943 sce->band_type, sce->band_type_run_end)) < 0)
1948 if (!eld_syntax && (pulse_present = get_bits1(gb))) {
1949 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1950 av_log(ac->avctx, AV_LOG_ERROR,
1951 "Pulse tool not allowed in eight short sequence.\n");
1952 return AVERROR_INVALIDDATA;
1954 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1955 av_log(ac->avctx, AV_LOG_ERROR,
1956 "Pulse data corrupt or invalid.\n");
1957 return AVERROR_INVALIDDATA;
1960 tns->present = get_bits1(gb);
1961 if (tns->present && !er_syntax)
1962 if (decode_tns(ac, tns, gb, ics) < 0)
1963 return AVERROR_INVALIDDATA;
1964 if (!eld_syntax && get_bits1(gb)) {
1965 avpriv_request_sample(ac->avctx, "SSR");
1966 return AVERROR_PATCHWELCOME;
1968 // I see no textual basis in the spec for this occurring after SSR gain
1969 // control, but this is what both reference and real implmentations do
1970 if (tns->present && er_syntax)
1971 if (decode_tns(ac, tns, gb, ics) < 0)
1972 return AVERROR_INVALIDDATA;
1975 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present,
1976 &pulse, ics, sce->band_type) < 0)
1977 return AVERROR_INVALIDDATA;
1979 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1980 apply_prediction(ac, sce);
1986 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1988 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1990 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1991 INTFLOAT *ch0 = cpe->ch[0].coeffs;
1992 INTFLOAT *ch1 = cpe->ch[1].coeffs;
1993 int g, i, group, idx = 0;
1994 const uint16_t *offsets = ics->swb_offset;
1995 for (g = 0; g < ics->num_window_groups; g++) {
1996 for (i = 0; i < ics->max_sfb; i++, idx++) {
1997 if (cpe->ms_mask[idx] &&
1998 cpe->ch[0].band_type[idx] < NOISE_BT &&
1999 cpe->ch[1].band_type[idx] < NOISE_BT) {
2001 for (group = 0; group < ics->group_len[g]; group++) {
2002 ac->fdsp->butterflies_fixed(ch0 + group * 128 + offsets[i],
2003 ch1 + group * 128 + offsets[i],
2004 offsets[i+1] - offsets[i]);
2006 for (group = 0; group < ics->group_len[g]; group++) {
2007 ac->fdsp->butterflies_float(ch0 + group * 128 + offsets[i],
2008 ch1 + group * 128 + offsets[i],
2009 offsets[i+1] - offsets[i]);
2010 #endif /* USE_FIXED */
2014 ch0 += ics->group_len[g] * 128;
2015 ch1 += ics->group_len[g] * 128;
2020 * intensity stereo decoding; reference: 4.6.8.2.3
2022 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
2023 * [1] mask is decoded from bitstream; [2] mask is all 1s;
2024 * [3] reserved for scalable AAC
2026 static void apply_intensity_stereo(AACContext *ac,
2027 ChannelElement *cpe, int ms_present)
2029 const IndividualChannelStream *ics = &cpe->ch[1].ics;
2030 SingleChannelElement *sce1 = &cpe->ch[1];
2031 INTFLOAT *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
2032 const uint16_t *offsets = ics->swb_offset;
2033 int g, group, i, idx = 0;
2036 for (g = 0; g < ics->num_window_groups; g++) {
2037 for (i = 0; i < ics->max_sfb;) {
2038 if (sce1->band_type[idx] == INTENSITY_BT ||
2039 sce1->band_type[idx] == INTENSITY_BT2) {
2040 const int bt_run_end = sce1->band_type_run_end[idx];
2041 for (; i < bt_run_end; i++, idx++) {
2042 c = -1 + 2 * (sce1->band_type[idx] - 14);
2044 c *= 1 - 2 * cpe->ms_mask[idx];
2045 scale = c * sce1->sf[idx];
2046 for (group = 0; group < ics->group_len[g]; group++)
2048 ac->subband_scale(coef1 + group * 128 + offsets[i],
2049 coef0 + group * 128 + offsets[i],
2052 offsets[i + 1] - offsets[i]);
2054 ac->fdsp->vector_fmul_scalar(coef1 + group * 128 + offsets[i],
2055 coef0 + group * 128 + offsets[i],
2057 offsets[i + 1] - offsets[i]);
2058 #endif /* USE_FIXED */
2061 int bt_run_end = sce1->band_type_run_end[idx];
2062 idx += bt_run_end - i;
2066 coef0 += ics->group_len[g] * 128;
2067 coef1 += ics->group_len[g] * 128;
2072 * Decode a channel_pair_element; reference: table 4.4.
2074 * @return Returns error status. 0 - OK, !0 - error
2076 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
2078 int i, ret, common_window, ms_present = 0;
2079 int eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
2081 common_window = eld_syntax || get_bits1(gb);
2082 if (common_window) {
2083 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
2084 return AVERROR_INVALIDDATA;
2085 i = cpe->ch[1].ics.use_kb_window[0];
2086 cpe->ch[1].ics = cpe->ch[0].ics;
2087 cpe->ch[1].ics.use_kb_window[1] = i;
2088 if (cpe->ch[1].ics.predictor_present &&
2089 (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
2090 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
2091 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
2092 ms_present = get_bits(gb, 2);
2093 if (ms_present == 3) {
2094 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
2095 return AVERROR_INVALIDDATA;
2096 } else if (ms_present)
2097 decode_mid_side_stereo(cpe, gb, ms_present);
2099 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
2101 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
2104 if (common_window) {
2106 apply_mid_side_stereo(ac, cpe);
2107 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
2108 apply_prediction(ac, &cpe->ch[0]);
2109 apply_prediction(ac, &cpe->ch[1]);
2113 apply_intensity_stereo(ac, cpe, ms_present);
2117 static const float cce_scale[] = {
2118 1.09050773266525765921, //2^(1/8)
2119 1.18920711500272106672, //2^(1/4)
2125 * Decode coupling_channel_element; reference: table 4.8.
2127 * @return Returns error status. 0 - OK, !0 - error
2129 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
2135 SingleChannelElement *sce = &che->ch[0];
2136 ChannelCoupling *coup = &che->coup;
2138 coup->coupling_point = 2 * get_bits1(gb);
2139 coup->num_coupled = get_bits(gb, 3);
2140 for (c = 0; c <= coup->num_coupled; c++) {
2142 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
2143 coup->id_select[c] = get_bits(gb, 4);
2144 if (coup->type[c] == TYPE_CPE) {
2145 coup->ch_select[c] = get_bits(gb, 2);
2146 if (coup->ch_select[c] == 3)
2149 coup->ch_select[c] = 2;
2151 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
2153 sign = get_bits(gb, 1);
2154 scale = AAC_RENAME(cce_scale)[get_bits(gb, 2)];
2156 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
2159 for (c = 0; c < num_gain; c++) {
2163 INTFLOAT gain_cache = FIXR10(1.);
2165 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
2166 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
2167 gain_cache = GET_GAIN(scale, gain);
2169 if (coup->coupling_point == AFTER_IMDCT) {
2170 coup->gain[c][0] = gain_cache;
2172 for (g = 0; g < sce->ics.num_window_groups; g++) {
2173 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
2174 if (sce->band_type[idx] != ZERO_BT) {
2176 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
2184 gain_cache = GET_GAIN(scale, t) * s;
2187 coup->gain[c][idx] = gain_cache;
2197 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
2199 * @return Returns number of bytes consumed.
2201 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
2205 int num_excl_chan = 0;
2208 for (i = 0; i < 7; i++)
2209 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
2210 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
2212 return num_excl_chan / 7;
2216 * Decode dynamic range information; reference: table 4.52.
2218 * @return Returns number of bytes consumed.
2220 static int decode_dynamic_range(DynamicRangeControl *che_drc,
2224 int drc_num_bands = 1;
2227 /* pce_tag_present? */
2228 if (get_bits1(gb)) {
2229 che_drc->pce_instance_tag = get_bits(gb, 4);
2230 skip_bits(gb, 4); // tag_reserved_bits
2234 /* excluded_chns_present? */
2235 if (get_bits1(gb)) {
2236 n += decode_drc_channel_exclusions(che_drc, gb);
2239 /* drc_bands_present? */
2240 if (get_bits1(gb)) {
2241 che_drc->band_incr = get_bits(gb, 4);
2242 che_drc->interpolation_scheme = get_bits(gb, 4);
2244 drc_num_bands += che_drc->band_incr;
2245 for (i = 0; i < drc_num_bands; i++) {
2246 che_drc->band_top[i] = get_bits(gb, 8);
2251 /* prog_ref_level_present? */
2252 if (get_bits1(gb)) {
2253 che_drc->prog_ref_level = get_bits(gb, 7);
2254 skip_bits1(gb); // prog_ref_level_reserved_bits
2258 for (i = 0; i < drc_num_bands; i++) {
2259 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
2260 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
2267 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
2269 int i, major, minor;
2274 get_bits(gb, 13); len -= 13;
2276 for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
2277 buf[i] = get_bits(gb, 8);
2280 if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
2281 av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
2283 if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
2284 ac->avctx->internal->skip_samples = 1024;
2288 skip_bits_long(gb, len);
2294 * Decode extension data (incomplete); reference: table 4.51.
2296 * @param cnt length of TYPE_FIL syntactic element in bytes
2298 * @return Returns number of bytes consumed
2300 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2301 ChannelElement *che, enum RawDataBlockType elem_type)
2305 int type = get_bits(gb, 4);
2307 if (ac->avctx->debug & FF_DEBUG_STARTCODE)
2308 av_log(ac->avctx, AV_LOG_DEBUG, "extension type: %d len:%d\n", type, cnt);
2310 switch (type) { // extension type
2311 case EXT_SBR_DATA_CRC:
2315 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2317 } else if (!ac->oc[1].m4ac.sbr) {
2318 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2319 skip_bits_long(gb, 8 * cnt - 4);
2321 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2322 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2323 skip_bits_long(gb, 8 * cnt - 4);
2325 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2326 ac->oc[1].m4ac.sbr = 1;
2327 ac->oc[1].m4ac.ps = 1;
2328 ac->avctx->profile = FF_PROFILE_AAC_HE_V2;
2329 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2330 ac->oc[1].status, 1);
2332 ac->oc[1].m4ac.sbr = 1;
2333 ac->avctx->profile = FF_PROFILE_AAC_HE;
2335 res = AAC_RENAME(ff_decode_sbr_extension)(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2337 case EXT_DYNAMIC_RANGE:
2338 res = decode_dynamic_range(&ac->che_drc, gb);
2341 decode_fill(ac, gb, 8 * cnt - 4);
2344 case EXT_DATA_ELEMENT:
2346 skip_bits_long(gb, 8 * cnt - 4);
2353 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2355 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2356 * @param coef spectral coefficients
2358 static void apply_tns(INTFLOAT coef[1024], TemporalNoiseShaping *tns,
2359 IndividualChannelStream *ics, int decode)
2361 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2363 int bottom, top, order, start, end, size, inc;
2364 INTFLOAT lpc[TNS_MAX_ORDER];
2365 INTFLOAT tmp[TNS_MAX_ORDER+1];
2367 for (w = 0; w < ics->num_windows; w++) {
2368 bottom = ics->num_swb;
2369 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2371 bottom = FFMAX(0, top - tns->length[w][filt]);
2372 order = tns->order[w][filt];
2377 AAC_RENAME(compute_lpc_coefs)(tns->coef[w][filt], order, lpc, 0, 0, 0);
2379 start = ics->swb_offset[FFMIN(bottom, mmm)];
2380 end = ics->swb_offset[FFMIN( top, mmm)];
2381 if ((size = end - start) <= 0)
2383 if (tns->direction[w][filt]) {
2393 for (m = 0; m < size; m++, start += inc)
2394 for (i = 1; i <= FFMIN(m, order); i++)
2395 coef[start] -= AAC_MUL26(coef[start - i * inc], lpc[i - 1]);
2398 for (m = 0; m < size; m++, start += inc) {
2399 tmp[0] = coef[start];
2400 for (i = 1; i <= FFMIN(m, order); i++)
2401 coef[start] += AAC_MUL26(tmp[i], lpc[i - 1]);
2402 for (i = order; i > 0; i--)
2403 tmp[i] = tmp[i - 1];
2411 * Apply windowing and MDCT to obtain the spectral
2412 * coefficient from the predicted sample by LTP.
2414 static void windowing_and_mdct_ltp(AACContext *ac, INTFLOAT *out,
2415 INTFLOAT *in, IndividualChannelStream *ics)
2417 const INTFLOAT *lwindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_long_1024) : AAC_RENAME(ff_sine_1024);
2418 const INTFLOAT *swindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2419 const INTFLOAT *lwindow_prev = ics->use_kb_window[1] ? AAC_RENAME(ff_aac_kbd_long_1024) : AAC_RENAME(ff_sine_1024);
2420 const INTFLOAT *swindow_prev = ics->use_kb_window[1] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2422 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2423 ac->fdsp->vector_fmul(in, in, lwindow_prev, 1024);
2425 memset(in, 0, 448 * sizeof(*in));
2426 ac->fdsp->vector_fmul(in + 448, in + 448, swindow_prev, 128);
2428 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2429 ac->fdsp->vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2431 ac->fdsp->vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2432 memset(in + 1024 + 576, 0, 448 * sizeof(*in));
2434 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2438 * Apply the long term prediction
2440 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2442 const LongTermPrediction *ltp = &sce->ics.ltp;
2443 const uint16_t *offsets = sce->ics.swb_offset;
2446 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2447 INTFLOAT *predTime = sce->ret;
2448 INTFLOAT *predFreq = ac->buf_mdct;
2449 int16_t num_samples = 2048;
2451 if (ltp->lag < 1024)
2452 num_samples = ltp->lag + 1024;
2453 for (i = 0; i < num_samples; i++)
2454 predTime[i] = AAC_MUL30(sce->ltp_state[i + 2048 - ltp->lag], ltp->coef);
2455 memset(&predTime[i], 0, (2048 - i) * sizeof(*predTime));
2457 ac->windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2459 if (sce->tns.present)
2460 ac->apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2462 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2464 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2465 sce->coeffs[i] += predFreq[i];
2470 * Update the LTP buffer for next frame
2472 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2474 IndividualChannelStream *ics = &sce->ics;
2475 INTFLOAT *saved = sce->saved;
2476 INTFLOAT *saved_ltp = sce->coeffs;
2477 const INTFLOAT *lwindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_long_1024) : AAC_RENAME(ff_sine_1024);
2478 const INTFLOAT *swindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2481 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2482 memcpy(saved_ltp, saved, 512 * sizeof(*saved_ltp));
2483 memset(saved_ltp + 576, 0, 448 * sizeof(*saved_ltp));
2484 ac->fdsp->vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2486 for (i = 0; i < 64; i++)
2487 saved_ltp[i + 512] = AAC_MUL31(ac->buf_mdct[1023 - i], swindow[63 - i]);
2488 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2489 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(*saved_ltp));
2490 memset(saved_ltp + 576, 0, 448 * sizeof(*saved_ltp));
2491 ac->fdsp->vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2493 for (i = 0; i < 64; i++)
2494 saved_ltp[i + 512] = AAC_MUL31(ac->buf_mdct[1023 - i], swindow[63 - i]);
2495 } else { // LONG_STOP or ONLY_LONG
2496 ac->fdsp->vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2498 for (i = 0; i < 512; i++)
2499 saved_ltp[i + 512] = AAC_MUL31(ac->buf_mdct[1023 - i], lwindow[511 - i]);
2502 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2503 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2504 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2508 * Conduct IMDCT and windowing.
2510 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2512 IndividualChannelStream *ics = &sce->ics;
2513 INTFLOAT *in = sce->coeffs;
2514 INTFLOAT *out = sce->ret;
2515 INTFLOAT *saved = sce->saved;
2516 const INTFLOAT *swindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2517 const INTFLOAT *lwindow_prev = ics->use_kb_window[1] ? AAC_RENAME(ff_aac_kbd_long_1024) : AAC_RENAME(ff_sine_1024);
2518 const INTFLOAT *swindow_prev = ics->use_kb_window[1] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2519 INTFLOAT *buf = ac->buf_mdct;
2520 INTFLOAT *temp = ac->temp;
2524 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2525 for (i = 0; i < 1024; i += 128)
2526 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2528 ac->mdct.imdct_half(&ac->mdct, buf, in);
2530 for (i=0; i<1024; i++)
2531 buf[i] = (buf[i] + 4) >> 3;
2532 #endif /* USE_FIXED */
2535 /* window overlapping
2536 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2537 * and long to short transitions are considered to be short to short
2538 * transitions. This leaves just two cases (long to long and short to short)
2539 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2541 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2542 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2543 ac->fdsp->vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2545 memcpy( out, saved, 448 * sizeof(*out));
2547 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2548 ac->fdsp->vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2549 ac->fdsp->vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2550 ac->fdsp->vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2551 ac->fdsp->vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2552 ac->fdsp->vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2553 memcpy( out + 448 + 4*128, temp, 64 * sizeof(*out));
2555 ac->fdsp->vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2556 memcpy( out + 576, buf + 64, 448 * sizeof(*out));
2561 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2562 memcpy( saved, temp + 64, 64 * sizeof(*saved));
2563 ac->fdsp->vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2564 ac->fdsp->vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2565 ac->fdsp->vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2566 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(*saved));
2567 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2568 memcpy( saved, buf + 512, 448 * sizeof(*saved));
2569 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(*saved));
2570 } else { // LONG_STOP or ONLY_LONG
2571 memcpy( saved, buf + 512, 512 * sizeof(*saved));
2575 static void imdct_and_windowing_ld(AACContext *ac, SingleChannelElement *sce)
2577 IndividualChannelStream *ics = &sce->ics;
2578 INTFLOAT *in = sce->coeffs;
2579 INTFLOAT *out = sce->ret;
2580 INTFLOAT *saved = sce->saved;
2581 INTFLOAT *buf = ac->buf_mdct;
2584 #endif /* USE_FIXED */
2587 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2590 for (i = 0; i < 1024; i++)
2591 buf[i] = (buf[i] + 2) >> 2;
2592 #endif /* USE_FIXED */
2594 // window overlapping
2595 if (ics->use_kb_window[1]) {
2596 // AAC LD uses a low overlap sine window instead of a KBD window
2597 memcpy(out, saved, 192 * sizeof(*out));
2598 ac->fdsp->vector_fmul_window(out + 192, saved + 192, buf, AAC_RENAME(ff_sine_128), 64);
2599 memcpy( out + 320, buf + 64, 192 * sizeof(*out));
2601 ac->fdsp->vector_fmul_window(out, saved, buf, AAC_RENAME(ff_sine_512), 256);
2605 memcpy(saved, buf + 256, 256 * sizeof(*saved));
2608 static void imdct_and_windowing_eld(AACContext *ac, SingleChannelElement *sce)
2610 INTFLOAT *in = sce->coeffs;
2611 INTFLOAT *out = sce->ret;
2612 INTFLOAT *saved = sce->saved;
2613 INTFLOAT *buf = ac->buf_mdct;
2615 const int n = ac->oc[1].m4ac.frame_length_short ? 480 : 512;
2616 const int n2 = n >> 1;
2617 const int n4 = n >> 2;
2618 const INTFLOAT *const window = n == 480 ? AAC_RENAME(ff_aac_eld_window_480) :
2619 AAC_RENAME(ff_aac_eld_window_512);
2621 // Inverse transform, mapped to the conventional IMDCT by
2622 // Chivukula, R.K.; Reznik, Y.A.; Devarajan, V.,
2623 // "Efficient algorithms for MPEG-4 AAC-ELD, AAC-LD and AAC-LC filterbanks,"
2624 // International Conference on Audio, Language and Image Processing, ICALIP 2008.
2625 // URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4590245&isnumber=4589950
2626 for (i = 0; i < n2; i+=2) {
2628 temp = in[i ]; in[i ] = -in[n - 1 - i]; in[n - 1 - i] = temp;
2629 temp = -in[i + 1]; in[i + 1] = in[n - 2 - i]; in[n - 2 - i] = temp;
2633 ac->mdct480->imdct_half(ac->mdct480, buf, in, 1, -1.f/(16*1024*960));
2636 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2639 for (i = 0; i < 1024; i++)
2640 buf[i] = (buf[i] + 1) >> 1;
2641 #endif /* USE_FIXED */
2643 for (i = 0; i < n; i+=2) {
2646 // Like with the regular IMDCT at this point we still have the middle half
2647 // of a transform but with even symmetry on the left and odd symmetry on
2650 // window overlapping
2651 // The spec says to use samples [0..511] but the reference decoder uses
2652 // samples [128..639].
2653 for (i = n4; i < n2; i ++) {
2654 out[i - n4] = AAC_MUL31( buf[ n2 - 1 - i] , window[i - n4]) +
2655 AAC_MUL31( saved[ i + n2] , window[i + n - n4]) +
2656 AAC_MUL31(-saved[n + n2 - 1 - i] , window[i + 2*n - n4]) +
2657 AAC_MUL31(-saved[ 2*n + n2 + i] , window[i + 3*n - n4]);
2659 for (i = 0; i < n2; i ++) {
2660 out[n4 + i] = AAC_MUL31( buf[ i] , window[i + n2 - n4]) +
2661 AAC_MUL31(-saved[ n - 1 - i] , window[i + n2 + n - n4]) +
2662 AAC_MUL31(-saved[ n + i] , window[i + n2 + 2*n - n4]) +
2663 AAC_MUL31( saved[2*n + n - 1 - i] , window[i + n2 + 3*n - n4]);
2665 for (i = 0; i < n4; i ++) {
2666 out[n2 + n4 + i] = AAC_MUL31( buf[ i + n2] , window[i + n - n4]) +
2667 AAC_MUL31(-saved[n2 - 1 - i] , window[i + 2*n - n4]) +
2668 AAC_MUL31(-saved[n + n2 + i] , window[i + 3*n - n4]);
2672 memmove(saved + n, saved, 2 * n * sizeof(*saved));
2673 memcpy( saved, buf, n * sizeof(*saved));
2677 * channel coupling transformation interface
2679 * @param apply_coupling_method pointer to (in)dependent coupling function
2681 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2682 enum RawDataBlockType type, int elem_id,
2683 enum CouplingPoint coupling_point,
2684 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2688 for (i = 0; i < MAX_ELEM_ID; i++) {
2689 ChannelElement *cce = ac->che[TYPE_CCE][i];
2692 if (cce && cce->coup.coupling_point == coupling_point) {
2693 ChannelCoupling *coup = &cce->coup;
2695 for (c = 0; c <= coup->num_coupled; c++) {
2696 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2697 if (coup->ch_select[c] != 1) {
2698 apply_coupling_method(ac, &cc->ch[0], cce, index);
2699 if (coup->ch_select[c] != 0)
2702 if (coup->ch_select[c] != 2)
2703 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2705 index += 1 + (coup->ch_select[c] == 3);
2712 * Convert spectral data to samples, applying all supported tools as appropriate.
2714 static void spectral_to_sample(AACContext *ac, int samples)
2717 void (*imdct_and_window)(AACContext *ac, SingleChannelElement *sce);
2718 switch (ac->oc[1].m4ac.object_type) {
2720 imdct_and_window = imdct_and_windowing_ld;
2722 case AOT_ER_AAC_ELD:
2723 imdct_and_window = imdct_and_windowing_eld;
2726 imdct_and_window = ac->imdct_and_windowing;
2728 for (type = 3; type >= 0; type--) {
2729 for (i = 0; i < MAX_ELEM_ID; i++) {
2730 ChannelElement *che = ac->che[type][i];
2731 if (che && che->present) {
2732 if (type <= TYPE_CPE)
2733 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, AAC_RENAME(apply_dependent_coupling));
2734 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2735 if (che->ch[0].ics.predictor_present) {
2736 if (che->ch[0].ics.ltp.present)
2737 ac->apply_ltp(ac, &che->ch[0]);
2738 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2739 ac->apply_ltp(ac, &che->ch[1]);
2742 if (che->ch[0].tns.present)
2743 ac->apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2744 if (che->ch[1].tns.present)
2745 ac->apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2746 if (type <= TYPE_CPE)
2747 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, AAC_RENAME(apply_dependent_coupling));
2748 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2749 imdct_and_window(ac, &che->ch[0]);
2750 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2751 ac->update_ltp(ac, &che->ch[0]);
2752 if (type == TYPE_CPE) {
2753 imdct_and_window(ac, &che->ch[1]);
2754 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2755 ac->update_ltp(ac, &che->ch[1]);
2757 if (ac->oc[1].m4ac.sbr > 0) {
2758 AAC_RENAME(ff_sbr_apply)(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2761 if (type <= TYPE_CCE)
2762 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, AAC_RENAME(apply_independent_coupling));
2767 /* preparation for resampler */
2768 for(j = 0; j<samples; j++){
2769 che->ch[0].ret[j] = (int32_t)av_clipl_int32((int64_t)che->ch[0].ret[j]<<7)+0x8000;
2770 if(type == TYPE_CPE)
2771 che->ch[1].ret[j] = (int32_t)av_clipl_int32((int64_t)che->ch[1].ret[j]<<7)+0x8000;
2774 #endif /* USE_FIXED */
2777 av_log(ac->avctx, AV_LOG_VERBOSE, "ChannelElement %d.%d missing \n", type, i);
2783 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2786 AACADTSHeaderInfo hdr_info;
2787 uint8_t layout_map[MAX_ELEM_ID*4][3];
2788 int layout_map_tags, ret;
2790 size = avpriv_aac_parse_header(gb, &hdr_info);
2792 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2793 // This is 2 for "VLB " audio in NSV files.
2794 // See samples/nsv/vlb_audio.
2795 avpriv_report_missing_feature(ac->avctx,
2796 "More than one AAC RDB per ADTS frame");
2797 ac->warned_num_aac_frames = 1;
2799 push_output_configuration(ac);
2800 if (hdr_info.chan_config) {
2801 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2802 if ((ret = set_default_channel_config(ac->avctx,
2805 hdr_info.chan_config)) < 0)
2807 if ((ret = output_configure(ac, layout_map, layout_map_tags,
2808 FFMAX(ac->oc[1].status,
2809 OC_TRIAL_FRAME), 0)) < 0)
2812 ac->oc[1].m4ac.chan_config = 0;
2814 * dual mono frames in Japanese DTV can have chan_config 0
2815 * WITHOUT specifying PCE.
2816 * thus, set dual mono as default.
2818 if (ac->dmono_mode && ac->oc[0].status == OC_NONE) {
2819 layout_map_tags = 2;
2820 layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
2821 layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
2822 layout_map[0][1] = 0;
2823 layout_map[1][1] = 1;
2824 if (output_configure(ac, layout_map, layout_map_tags,
2829 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2830 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2831 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2832 ac->oc[1].m4ac.frame_length_short = 0;
2833 if (ac->oc[0].status != OC_LOCKED ||
2834 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2835 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2836 ac->oc[1].m4ac.sbr = -1;
2837 ac->oc[1].m4ac.ps = -1;
2839 if (!hdr_info.crc_absent)
2845 static int aac_decode_er_frame(AVCodecContext *avctx, void *data,
2846 int *got_frame_ptr, GetBitContext *gb)
2848 AACContext *ac = avctx->priv_data;
2849 const MPEG4AudioConfig *const m4ac = &ac->oc[1].m4ac;
2850 ChannelElement *che;
2852 int samples = m4ac->frame_length_short ? 960 : 1024;
2853 int chan_config = m4ac->chan_config;
2854 int aot = m4ac->object_type;
2856 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD)
2861 if ((err = frame_configure_elements(avctx)) < 0)
2864 // The FF_PROFILE_AAC_* defines are all object_type - 1
2865 // This may lead to an undefined profile being signaled
2866 ac->avctx->profile = aot - 1;
2868 ac->tags_mapped = 0;
2870 if (chan_config < 0 || (chan_config >= 8 && chan_config < 11) || chan_config >= 13) {
2871 avpriv_request_sample(avctx, "Unknown ER channel configuration %d",
2873 return AVERROR_INVALIDDATA;
2875 for (i = 0; i < tags_per_config[chan_config]; i++) {
2876 const int elem_type = aac_channel_layout_map[chan_config-1][i][0];
2877 const int elem_id = aac_channel_layout_map[chan_config-1][i][1];
2878 if (!(che=get_che(ac, elem_type, elem_id))) {
2879 av_log(ac->avctx, AV_LOG_ERROR,
2880 "channel element %d.%d is not allocated\n",
2881 elem_type, elem_id);
2882 return AVERROR_INVALIDDATA;
2885 if (aot != AOT_ER_AAC_ELD)
2887 switch (elem_type) {
2889 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2892 err = decode_cpe(ac, gb, che);
2895 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2902 spectral_to_sample(ac, samples);
2904 if (!ac->frame->data[0] && samples) {
2905 av_log(avctx, AV_LOG_ERROR, "no frame data found\n");
2906 return AVERROR_INVALIDDATA;
2909 ac->frame->nb_samples = samples;
2910 ac->frame->sample_rate = avctx->sample_rate;
2913 skip_bits_long(gb, get_bits_left(gb));
2917 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2918 int *got_frame_ptr, GetBitContext *gb, AVPacket *avpkt)
2920 AACContext *ac = avctx->priv_data;
2921 ChannelElement *che = NULL, *che_prev = NULL;
2922 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2924 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2925 int is_dmono, sce_count = 0;
2929 if (show_bits(gb, 12) == 0xfff) {
2930 if ((err = parse_adts_frame_header(ac, gb)) < 0) {
2931 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2934 if (ac->oc[1].m4ac.sampling_index > 12) {
2935 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2936 err = AVERROR_INVALIDDATA;
2941 if ((err = frame_configure_elements(avctx)) < 0)
2944 // The FF_PROFILE_AAC_* defines are all object_type - 1
2945 // This may lead to an undefined profile being signaled
2946 ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2948 ac->tags_mapped = 0;
2950 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2951 elem_id = get_bits(gb, 4);
2953 if (avctx->debug & FF_DEBUG_STARTCODE)
2954 av_log(avctx, AV_LOG_DEBUG, "Elem type:%x id:%x\n", elem_type, elem_id);
2956 if (!avctx->channels && elem_type != TYPE_PCE) {
2957 err = AVERROR_INVALIDDATA;
2961 if (elem_type < TYPE_DSE) {
2962 if (!(che=get_che(ac, elem_type, elem_id))) {
2963 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2964 elem_type, elem_id);
2965 err = AVERROR_INVALIDDATA;
2972 switch (elem_type) {
2975 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2981 err = decode_cpe(ac, gb, che);
2986 err = decode_cce(ac, gb, che);
2990 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2995 err = skip_data_stream_element(ac, gb);
2999 uint8_t layout_map[MAX_ELEM_ID*4][3];
3001 push_output_configuration(ac);
3002 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
3008 av_log(avctx, AV_LOG_ERROR,
3009 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
3011 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
3013 ac->oc[1].m4ac.chan_config = 0;
3021 elem_id += get_bits(gb, 8) - 1;
3022 if (get_bits_left(gb) < 8 * elem_id) {
3023 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
3024 err = AVERROR_INVALIDDATA;
3028 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
3029 err = 0; /* FIXME */
3033 err = AVERROR_BUG; /* should not happen, but keeps compiler happy */
3038 elem_type_prev = elem_type;
3043 if (get_bits_left(gb) < 3) {
3044 av_log(avctx, AV_LOG_ERROR, overread_err);
3045 err = AVERROR_INVALIDDATA;
3050 if (!avctx->channels) {
3055 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
3056 samples <<= multiplier;
3058 spectral_to_sample(ac, samples);
3060 if (ac->oc[1].status && audio_found) {
3061 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
3062 avctx->frame_size = samples;
3063 ac->oc[1].status = OC_LOCKED;
3068 const uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, &side_size);
3069 if (side && side_size>=4)
3070 AV_WL32(side, 2*AV_RL32(side));
3073 if (!ac->frame->data[0] && samples) {
3074 av_log(avctx, AV_LOG_ERROR, "no frame data found\n");
3075 err = AVERROR_INVALIDDATA;
3080 ac->frame->nb_samples = samples;
3081 ac->frame->sample_rate = avctx->sample_rate;
3083 av_frame_unref(ac->frame);
3084 *got_frame_ptr = !!samples;
3086 /* for dual-mono audio (SCE + SCE) */
3087 is_dmono = ac->dmono_mode && sce_count == 2 &&
3088 ac->oc[1].channel_layout == (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT);
3090 if (ac->dmono_mode == 1)
3091 ((AVFrame *)data)->data[1] =((AVFrame *)data)->data[0];
3092 else if (ac->dmono_mode == 2)
3093 ((AVFrame *)data)->data[0] =((AVFrame *)data)->data[1];
3098 pop_output_configuration(ac);
3102 static int aac_decode_frame(AVCodecContext *avctx, void *data,
3103 int *got_frame_ptr, AVPacket *avpkt)
3105 AACContext *ac = avctx->priv_data;
3106 const uint8_t *buf = avpkt->data;
3107 int buf_size = avpkt->size;
3112 int new_extradata_size;
3113 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
3114 AV_PKT_DATA_NEW_EXTRADATA,
3115 &new_extradata_size);
3116 int jp_dualmono_size;
3117 const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
3118 AV_PKT_DATA_JP_DUALMONO,
3121 if (new_extradata && 0) {
3122 av_free(avctx->extradata);
3123 avctx->extradata = av_mallocz(new_extradata_size +
3124 AV_INPUT_BUFFER_PADDING_SIZE);
3125 if (!avctx->extradata)
3126 return AVERROR(ENOMEM);
3127 avctx->extradata_size = new_extradata_size;
3128 memcpy(avctx->extradata, new_extradata, new_extradata_size);
3129 push_output_configuration(ac);
3130 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
3132 avctx->extradata_size*8LL, 1) < 0) {
3133 pop_output_configuration(ac);
3134 return AVERROR_INVALIDDATA;
3139 if (jp_dualmono && jp_dualmono_size > 0)
3140 ac->dmono_mode = 1 + *jp_dualmono;
3141 if (ac->force_dmono_mode >= 0)
3142 ac->dmono_mode = ac->force_dmono_mode;
3144 if (INT_MAX / 8 <= buf_size)
3145 return AVERROR_INVALIDDATA;
3147 if ((err = init_get_bits8(&gb, buf, buf_size)) < 0)
3150 switch (ac->oc[1].m4ac.object_type) {
3152 case AOT_ER_AAC_LTP:
3154 case AOT_ER_AAC_ELD:
3155 err = aac_decode_er_frame(avctx, data, got_frame_ptr, &gb);
3158 err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt);
3163 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
3164 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
3165 if (buf[buf_offset])
3168 return buf_size > buf_offset ? buf_consumed : buf_size;
3171 static av_cold int aac_decode_close(AVCodecContext *avctx)
3173 AACContext *ac = avctx->priv_data;
3176 for (i = 0; i < MAX_ELEM_ID; i++) {
3177 for (type = 0; type < 4; type++) {
3178 if (ac->che[type][i])
3179 AAC_RENAME(ff_aac_sbr_ctx_close)(&ac->che[type][i]->sbr);
3180 av_freep(&ac->che[type][i]);
3184 ff_mdct_end(&ac->mdct);
3185 ff_mdct_end(&ac->mdct_small);
3186 ff_mdct_end(&ac->mdct_ld);
3187 ff_mdct_end(&ac->mdct_ltp);
3189 ff_imdct15_uninit(&ac->mdct480);
3191 av_freep(&ac->fdsp);
3195 static void aacdec_init(AACContext *c)
3197 c->imdct_and_windowing = imdct_and_windowing;
3198 c->apply_ltp = apply_ltp;
3199 c->apply_tns = apply_tns;
3200 c->windowing_and_mdct_ltp = windowing_and_mdct_ltp;
3201 c->update_ltp = update_ltp;
3203 c->vector_pow43 = vector_pow43;
3204 c->subband_scale = subband_scale;
3209 ff_aacdec_init_mips(c);
3210 #endif /* !USE_FIXED */
3213 * AVOptions for Japanese DTV specific extensions (ADTS only)
3215 #define AACDEC_FLAGS AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
3216 static const AVOption options[] = {
3217 {"dual_mono_mode", "Select the channel to decode for dual mono",
3218 offsetof(AACContext, force_dmono_mode), AV_OPT_TYPE_INT, {.i64=-1}, -1, 2,
3219 AACDEC_FLAGS, "dual_mono_mode"},
3221 {"auto", "autoselection", 0, AV_OPT_TYPE_CONST, {.i64=-1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3222 {"main", "Select Main/Left channel", 0, AV_OPT_TYPE_CONST, {.i64= 1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3223 {"sub" , "Select Sub/Right channel", 0, AV_OPT_TYPE_CONST, {.i64= 2}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3224 {"both", "Select both channels", 0, AV_OPT_TYPE_CONST, {.i64= 0}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3229 static const AVClass aac_decoder_class = {
3230 .class_name = "AAC decoder",
3231 .item_name = av_default_item_name,
3233 .version = LIBAVUTIL_VERSION_INT,
3236 static const AVProfile profiles[] = {
3237 { FF_PROFILE_AAC_MAIN, "Main" },
3238 { FF_PROFILE_AAC_LOW, "LC" },
3239 { FF_PROFILE_AAC_SSR, "SSR" },
3240 { FF_PROFILE_AAC_LTP, "LTP" },
3241 { FF_PROFILE_AAC_HE, "HE-AAC" },
3242 { FF_PROFILE_AAC_HE_V2, "HE-AACv2" },
3243 { FF_PROFILE_AAC_LD, "LD" },
3244 { FF_PROFILE_AAC_ELD, "ELD" },
3245 { FF_PROFILE_UNKNOWN },