2 * Copyright (C) 2016 foo86
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 #define UNCHECKED_BITSTREAM_READER 1
22 #define BITSTREAM_READER_LE
24 #include "libavutil/channel_layout.h"
29 #include "dca_syncwords.h"
30 #include "bytestream.h"
35 LBR_HEADER_SYNC_ONLY = 1,
36 LBR_HEADER_DECODER_INIT = 2
40 LBR_FLAG_24_BIT = 0x01,
41 LBR_FLAG_LFE_PRESENT = 0x02,
42 LBR_FLAG_BAND_LIMIT_2_3 = 0x04,
43 LBR_FLAG_BAND_LIMIT_1_2 = 0x08,
44 LBR_FLAG_BAND_LIMIT_1_3 = 0x0c,
45 LBR_FLAG_BAND_LIMIT_1_4 = 0x10,
46 LBR_FLAG_BAND_LIMIT_1_8 = 0x18,
47 LBR_FLAG_BAND_LIMIT_NONE = 0x14,
48 LBR_FLAG_BAND_LIMIT_MASK = 0x1c,
49 LBR_FLAG_DMIX_STEREO = 0x20,
50 LBR_FLAG_DMIX_MULTI_CH = 0x40
54 LBR_CHUNK_NULL = 0x00,
56 LBR_CHUNK_FRAME = 0x04,
57 LBR_CHUNK_FRAME_NO_CSUM = 0x06,
60 LBR_CHUNK_RESERVED_1 = 0x0c,
61 LBR_CHUNK_RESERVED_2 = 0x0d,
63 LBR_CHUNK_TONAL = 0x10,
64 LBR_CHUNK_TONAL_GRP_1 = 0x11,
65 LBR_CHUNK_TONAL_GRP_2 = 0x12,
66 LBR_CHUNK_TONAL_GRP_3 = 0x13,
67 LBR_CHUNK_TONAL_GRP_4 = 0x14,
68 LBR_CHUNK_TONAL_GRP_5 = 0x15,
69 LBR_CHUNK_TONAL_SCF = 0x16,
70 LBR_CHUNK_TONAL_SCF_GRP_1 = 0x17,
71 LBR_CHUNK_TONAL_SCF_GRP_2 = 0x18,
72 LBR_CHUNK_TONAL_SCF_GRP_3 = 0x19,
73 LBR_CHUNK_TONAL_SCF_GRP_4 = 0x1a,
74 LBR_CHUNK_TONAL_SCF_GRP_5 = 0x1b,
75 LBR_CHUNK_RES_GRID_LR = 0x30,
76 LBR_CHUNK_RES_GRID_LR_LAST = 0x3f,
77 LBR_CHUNK_RES_GRID_HR = 0x40,
78 LBR_CHUNK_RES_GRID_HR_LAST = 0x4f,
79 LBR_CHUNK_RES_TS_1 = 0x50,
80 LBR_CHUNK_RES_TS_1_LAST = 0x5f,
81 LBR_CHUNK_RES_TS_2 = 0x60,
82 LBR_CHUNK_RES_TS_2_LAST = 0x6f,
83 LBR_CHUNK_EXTENSION = 0x7f
86 typedef struct LBRChunk {
91 static const int8_t channel_reorder_nolfe[7][5] = {
92 { 0, -1, -1, -1, -1 }, // C
93 { 0, 1, -1, -1, -1 }, // LR
94 { 0, 1, 2, -1, -1 }, // LR C
95 { 0, 1, -1, -1, -1 }, // LsRs
96 { 1, 2, 0, -1, -1 }, // LsRs C
97 { 0, 1, 2, 3, -1 }, // LR LsRs
98 { 0, 1, 3, 4, 2 }, // LR LsRs C
101 static const int8_t channel_reorder_lfe[7][5] = {
102 { 0, -1, -1, -1, -1 }, // C
103 { 0, 1, -1, -1, -1 }, // LR
104 { 0, 1, 2, -1, -1 }, // LR C
105 { 1, 2, -1, -1, -1 }, // LsRs
106 { 2, 3, 0, -1, -1 }, // LsRs C
107 { 0, 1, 3, 4, -1 }, // LR LsRs
108 { 0, 1, 4, 5, 2 }, // LR LsRs C
111 static const uint8_t lfe_index[7] = {
115 static const uint8_t channel_counts[7] = {
119 static const uint16_t channel_layouts[7] = {
122 AV_CH_LAYOUT_SURROUND,
123 AV_CH_SIDE_LEFT | AV_CH_SIDE_RIGHT,
124 AV_CH_FRONT_CENTER | AV_CH_SIDE_LEFT | AV_CH_SIDE_RIGHT,
129 static float cos_tab[256];
130 static float lpc_tab[16];
132 static av_cold void init_tables(void)
134 static int initialized;
140 for (i = 0; i < 256; i++)
141 cos_tab[i] = cos(M_PI * i / 128);
143 for (i = 0; i < 16; i++)
144 lpc_tab[i] = sin((i - 8) * (M_PI / ((i < 8) ? 17 : 15)));
149 static int parse_lfe_24(DCALbrDecoder *s)
151 int step_max = FF_ARRAY_ELEMS(ff_dca_lfe_step_size_24) - 1;
152 int i, ps, si, code, step_i;
153 float step, value, delta;
155 ps = get_bits(&s->gb, 24);
158 value = (((ps & 0x7fffff) ^ -si) + si) * (1.0f / 0x7fffff);
160 step_i = get_bits(&s->gb, 8);
161 if (step_i > step_max) {
162 av_log(s->avctx, AV_LOG_ERROR, "Invalid LFE step size index\n");
166 step = ff_dca_lfe_step_size_24[step_i];
168 for (i = 0; i < 64; i++) {
169 code = get_bits(&s->gb, 6);
171 delta = step * 0.03125f;
175 delta += step * 0.5f;
177 delta += step * 0.25f;
179 delta += step * 0.125f;
181 delta += step * 0.0625f;
193 step_i += ff_dca_lfe_delta_index_24[code & 31];
194 step_i = av_clip(step_i, 0, step_max);
196 step = ff_dca_lfe_step_size_24[step_i];
197 s->lfe_data[i] = value * s->lfe_scale;
203 static int parse_lfe_16(DCALbrDecoder *s)
205 int step_max = FF_ARRAY_ELEMS(ff_dca_lfe_step_size_16) - 1;
206 int i, ps, si, code, step_i;
207 float step, value, delta;
209 ps = get_bits(&s->gb, 16);
212 value = (((ps & 0x7fff) ^ -si) + si) * (1.0f / 0x7fff);
214 step_i = get_bits(&s->gb, 8);
215 if (step_i > step_max) {
216 av_log(s->avctx, AV_LOG_ERROR, "Invalid LFE step size index\n");
220 step = ff_dca_lfe_step_size_16[step_i];
222 for (i = 0; i < 64; i++) {
223 code = get_bits(&s->gb, 4);
225 delta = step * 0.125f;
229 delta += step * 0.5f;
231 delta += step * 0.25f;
243 step_i += ff_dca_lfe_delta_index_16[code & 7];
244 step_i = av_clip(step_i, 0, step_max);
246 step = ff_dca_lfe_step_size_16[step_i];
247 s->lfe_data[i] = value * s->lfe_scale;
253 static int parse_lfe_chunk(DCALbrDecoder *s, LBRChunk *chunk)
255 if (!(s->flags & LBR_FLAG_LFE_PRESENT))
261 if (init_get_bits8(&s->gb, chunk->data, chunk->len) < 0)
264 // Determine bit depth from chunk size
265 if (chunk->len >= 52)
266 return parse_lfe_24(s);
267 if (chunk->len >= 35)
268 return parse_lfe_16(s);
270 av_log(s->avctx, AV_LOG_ERROR, "LFE chunk too short\n");
274 static inline int parse_vlc(GetBitContext *s, VLC *vlc, int max_depth)
276 int v = get_vlc2(s, vlc->table, vlc->bits, max_depth);
280 return get_bits(s, get_bits(s, 3) + 1);
283 static int parse_tonal(DCALbrDecoder *s, int group)
285 unsigned int amp[DCA_LBR_CHANNELS_TOTAL];
286 unsigned int phs[DCA_LBR_CHANNELS_TOTAL];
287 unsigned int diff, main_amp, shift;
288 int sf, sf_idx, ch, main_ch, freq;
289 int ch_nbits = av_ceil_log2(s->nchannels_total);
291 // Parse subframes for this group
292 for (sf = 0; sf < 1 << group; sf += diff ? 8 : 1) {
293 sf_idx = ((s->framenum << group) + sf) & 31;
294 s->tonal_bounds[group][sf_idx][0] = s->ntones;
296 // Parse tones for this subframe
297 for (freq = 1;; freq++) {
298 if (get_bits_left(&s->gb) < 1) {
299 av_log(s->avctx, AV_LOG_ERROR, "Tonal group chunk too short\n");
303 diff = parse_vlc(&s->gb, &ff_dca_vlc_tnl_grp[group], 2);
304 if (diff >= FF_ARRAY_ELEMS(ff_dca_fst_amp)) {
305 av_log(s->avctx, AV_LOG_ERROR, "Invalid tonal frequency diff\n");
309 diff = get_bitsz(&s->gb, diff >> 2) + ff_dca_fst_amp[diff];
311 break; // End of subframe
314 if (freq >> (5 - group) > s->nsubbands * 4 - 5) {
315 av_log(s->avctx, AV_LOG_ERROR, "Invalid spectral line offset\n");
320 main_ch = get_bitsz(&s->gb, ch_nbits);
321 main_amp = parse_vlc(&s->gb, &ff_dca_vlc_tnl_scf, 2)
322 + s->tonal_scf[ff_dca_freq_to_sb[freq >> (7 - group)]]
323 + s->limited_range - 2;
324 amp[main_ch] = main_amp < AMP_MAX ? main_amp : 0;
325 phs[main_ch] = get_bits(&s->gb, 3);
327 // Secondary channels
328 for (ch = 0; ch < s->nchannels_total; ch++) {
331 if (get_bits1(&s->gb)) {
332 amp[ch] = amp[main_ch] - parse_vlc(&s->gb, &ff_dca_vlc_damp, 1);
333 phs[ch] = phs[main_ch] - parse_vlc(&s->gb, &ff_dca_vlc_dph, 1);
342 DCALbrTone *t = &s->tones[s->ntones];
343 s->ntones = (s->ntones + 1) & (DCA_LBR_TONES - 1);
345 t->x_freq = freq >> (5 - group);
346 t->f_delt = (freq & ((1 << (5 - group)) - 1)) << group;
347 t->ph_rot = 256 - (t->x_freq & 1) * 128 - t->f_delt * 4;
349 shift = ff_dca_ph0_shift[(t->x_freq & 3) * 2 + (freq & 1)]
350 - ((t->ph_rot << (5 - group)) - t->ph_rot);
352 for (ch = 0; ch < s->nchannels; ch++) {
353 t->amp[ch] = amp[ch] < AMP_MAX ? amp[ch] : 0;
354 t->phs[ch] = 128 - phs[ch] * 32 + shift;
359 s->tonal_bounds[group][sf_idx][1] = s->ntones;
365 static int parse_tonal_chunk(DCALbrDecoder *s, LBRChunk *chunk)
372 if (init_get_bits8(&s->gb, chunk->data, chunk->len) < 0)
376 if (chunk->id == LBR_CHUNK_SCF || chunk->id == LBR_CHUNK_TONAL_SCF) {
377 if (get_bits_left(&s->gb) < 36) {
378 av_log(s->avctx, AV_LOG_ERROR, "Tonal scale factor chunk too short\n");
381 for (sb = 0; sb < 6; sb++)
382 s->tonal_scf[sb] = get_bits(&s->gb, 6);
386 if (chunk->id == LBR_CHUNK_TONAL || chunk->id == LBR_CHUNK_TONAL_SCF)
387 for (group = 0; group < 5; group++)
388 if (parse_tonal(s, group) < 0)
394 static int parse_tonal_group(DCALbrDecoder *s, LBRChunk *chunk)
399 if (init_get_bits8(&s->gb, chunk->data, chunk->len) < 0)
402 return parse_tonal(s, chunk->id);
406 * Check point to ensure that enough bits are left. Aborts decoding
407 * by skipping to the end of chunk otherwise.
409 static int ensure_bits(GetBitContext *s, int n)
411 int left = get_bits_left(s);
415 skip_bits_long(s, left);
421 static int parse_scale_factors(DCALbrDecoder *s, uint8_t *scf)
423 int i, sf, prev, next, dist;
425 // Truncated scale factors remain zero
426 if (ensure_bits(&s->gb, 20))
429 // Initial scale factor
430 prev = parse_vlc(&s->gb, &ff_dca_vlc_fst_rsd_amp, 2);
432 for (sf = 0; sf < 7; sf += dist) {
433 scf[sf] = prev; // Store previous value
435 if (ensure_bits(&s->gb, 20))
438 // Interpolation distance
439 dist = parse_vlc(&s->gb, &ff_dca_vlc_rsd_apprx, 1) + 1;
441 av_log(s->avctx, AV_LOG_ERROR, "Invalid scale factor distance\n");
445 if (ensure_bits(&s->gb, 20))
448 // Final interpolation point
449 next = parse_vlc(&s->gb, &ff_dca_vlc_rsd_amp, 2);
452 next = prev + ((next + 1) >> 1);
454 next = prev - ( next >> 1);
460 scf[sf + 1] = prev + ((next - prev) >> 1);
462 scf[sf + 1] = prev - ((prev - next) >> 1);
467 scf[sf + 1] = prev + ( (next - prev) >> 2);
468 scf[sf + 2] = prev + ( (next - prev) >> 1);
469 scf[sf + 3] = prev + (((next - prev) * 3) >> 2);
471 scf[sf + 1] = prev - ( (prev - next) >> 2);
472 scf[sf + 2] = prev - ( (prev - next) >> 1);
473 scf[sf + 3] = prev - (((prev - next) * 3) >> 2);
478 for (i = 1; i < dist; i++)
479 scf[sf + i] = prev + (next - prev) * i / dist;
486 scf[sf] = next; // Store final value
491 static int parse_st_code(GetBitContext *s, int min_v)
493 unsigned int v = parse_vlc(s, &ff_dca_vlc_st_grid, 2) + min_v;
500 if (v >= FF_ARRAY_ELEMS(ff_dca_st_coeff))
505 static int parse_grid_1_chunk(DCALbrDecoder *s, LBRChunk *chunk, int ch1, int ch2)
507 int ch, sb, sf, nsubbands;
512 if (init_get_bits8(&s->gb, chunk->data, chunk->len) < 0)
516 nsubbands = ff_dca_scf_to_grid_1[s->nsubbands - 1] + 1;
517 for (sb = 2; sb < nsubbands; sb++) {
518 if (parse_scale_factors(s, s->grid_1_scf[ch1][sb]) < 0)
520 if (ch1 != ch2 && ff_dca_grid_1_to_scf[sb] < s->min_mono_subband
521 && parse_scale_factors(s, s->grid_1_scf[ch2][sb]) < 0)
525 if (get_bits_left(&s->gb) < 1)
526 return 0; // Should not happen, but a sample exists that proves otherwise
528 // Average values for third grid
529 for (sb = 0; sb < s->nsubbands - 4; sb++) {
530 s->grid_3_avg[ch1][sb] = parse_vlc(&s->gb, &ff_dca_vlc_avg_g3, 2) - 16;
532 if (sb + 4 < s->min_mono_subband)
533 s->grid_3_avg[ch2][sb] = parse_vlc(&s->gb, &ff_dca_vlc_avg_g3, 2) - 16;
535 s->grid_3_avg[ch2][sb] = s->grid_3_avg[ch1][sb];
539 if (get_bits_left(&s->gb) < 0) {
540 av_log(s->avctx, AV_LOG_ERROR, "First grid chunk too short\n");
544 // Stereo image for partial mono mode
548 if (ensure_bits(&s->gb, 8))
551 min_v[0] = get_bits(&s->gb, 4);
552 min_v[1] = get_bits(&s->gb, 4);
554 nsubbands = (s->nsubbands - s->min_mono_subband + 3) / 4;
555 for (sb = 0; sb < nsubbands; sb++)
556 for (ch = ch1; ch <= ch2; ch++)
557 for (sf = 1; sf <= 4; sf++)
558 s->part_stereo[ch][sb][sf] = parse_st_code(&s->gb, min_v[ch - ch1]);
560 if (get_bits_left(&s->gb) >= 0)
561 s->part_stereo_pres |= 1 << ch1;
564 // Low resolution spatial information is not decoded
569 static int parse_grid_1_sec_ch(DCALbrDecoder *s, int ch2)
574 nsubbands = ff_dca_scf_to_grid_1[s->nsubbands - 1] + 1;
575 for (sb = 2; sb < nsubbands; sb++) {
576 if (ff_dca_grid_1_to_scf[sb] >= s->min_mono_subband
577 && parse_scale_factors(s, s->grid_1_scf[ch2][sb]) < 0)
581 // Average values for third grid
582 for (sb = 0; sb < s->nsubbands - 4; sb++) {
583 if (sb + 4 >= s->min_mono_subband) {
584 if (ensure_bits(&s->gb, 20))
586 s->grid_3_avg[ch2][sb] = parse_vlc(&s->gb, &ff_dca_vlc_avg_g3, 2) - 16;
593 static void parse_grid_3(DCALbrDecoder *s, int ch1, int ch2, int sb, int flag)
597 for (ch = ch1; ch <= ch2; ch++) {
598 if ((ch != ch1 && sb + 4 >= s->min_mono_subband) != flag)
601 if (s->grid_3_pres[ch] & (1U << sb))
602 continue; // Already parsed
604 for (i = 0; i < 8; i++) {
605 if (ensure_bits(&s->gb, 20))
607 s->grid_3_scf[ch][sb][i] = parse_vlc(&s->gb, &ff_dca_vlc_grid_3, 2) - 16;
610 // Flag scale factors for this subband parsed
611 s->grid_3_pres[ch] |= 1U << sb;
615 static float lbr_rand(DCALbrDecoder *s, int sb)
617 s->lbr_rand = 1103515245U * s->lbr_rand + 12345U;
618 return s->lbr_rand * s->sb_scf[sb];
622 * Parse time samples for one subband, filling truncated samples with randomness
624 static void parse_ch(DCALbrDecoder *s, int ch, int sb, int quant_level, int flag)
626 float *samples = s->time_samples[ch][sb];
627 int i, j, code, nblocks, coding_method;
629 if (ensure_bits(&s->gb, 20))
630 return; // Too few bits left
632 coding_method = get_bits1(&s->gb);
634 switch (quant_level) {
636 nblocks = FFMIN(get_bits_left(&s->gb) / 8, DCA_LBR_TIME_SAMPLES / 8);
637 for (i = 0; i < nblocks; i++, samples += 8) {
638 code = get_bits(&s->gb, 8);
639 for (j = 0; j < 8; j++)
640 samples[j] = ff_dca_rsd_level_2a[(code >> j) & 1];
647 for (i = 0; i < DCA_LBR_TIME_SAMPLES && get_bits_left(&s->gb) >= 2; i++) {
648 if (get_bits1(&s->gb))
649 samples[i] = ff_dca_rsd_level_2b[get_bits1(&s->gb)];
654 nblocks = FFMIN(get_bits_left(&s->gb) / 8, (DCA_LBR_TIME_SAMPLES + 4) / 5);
655 for (i = 0; i < nblocks; i++, samples += 5) {
656 code = ff_dca_rsd_pack_5_in_8[get_bits(&s->gb, 8)];
657 for (j = 0; j < 5; j++)
658 samples[j] = ff_dca_rsd_level_3[(code >> j * 2) & 3];
665 nblocks = FFMIN(get_bits_left(&s->gb) / 7, (DCA_LBR_TIME_SAMPLES + 2) / 3);
666 for (i = 0; i < nblocks; i++, samples += 3) {
667 code = get_bits(&s->gb, 7);
668 for (j = 0; j < 3; j++)
669 samples[j] = ff_dca_rsd_level_5[ff_dca_rsd_pack_3_in_7[code][j]];
675 for (i = 0; i < DCA_LBR_TIME_SAMPLES && get_bits_left(&s->gb) >= 6; i++)
676 samples[i] = ff_dca_rsd_level_8[get_vlc2(&s->gb, ff_dca_vlc_rsd.table, 6, 1)];
680 nblocks = FFMIN(get_bits_left(&s->gb) / 4, DCA_LBR_TIME_SAMPLES);
681 for (i = 0; i < nblocks; i++)
682 samples[i] = ff_dca_rsd_level_16[get_bits(&s->gb, 4)];
689 if (flag && get_bits_left(&s->gb) < 20)
690 return; // Skip incomplete mono subband
692 for (; i < DCA_LBR_TIME_SAMPLES; i++)
693 s->time_samples[ch][sb][i] = lbr_rand(s, sb);
695 s->ch_pres[ch] |= 1U << sb;
698 static int parse_ts(DCALbrDecoder *s, int ch1, int ch2,
699 int start_sb, int end_sb, int flag)
701 int sb, sb_g3, sb_reorder, quant_level;
703 for (sb = start_sb; sb < end_sb; sb++) {
704 // Subband number before reordering
707 } else if (flag && sb < s->max_mono_subband) {
708 sb_reorder = s->sb_indices[sb];
710 if (ensure_bits(&s->gb, 28))
712 sb_reorder = get_bits(&s->gb, s->limited_range + 3);
715 s->sb_indices[sb] = sb_reorder;
717 if (sb_reorder >= s->nsubbands)
720 // Third grid scale factors
722 for (sb_g3 = 0; sb_g3 < s->g3_avg_only_start_sb - 4; sb_g3++)
723 parse_grid_3(s, ch1, ch2, sb_g3, flag);
724 } else if (sb < 12 && sb_reorder >= 4) {
725 parse_grid_3(s, ch1, ch2, sb_reorder - 4, flag);
728 // Secondary channel flags
730 if (ensure_bits(&s->gb, 20))
732 if (!flag || sb_reorder >= s->max_mono_subband)
733 s->sec_ch_sbms[ch1 / 2][sb_reorder] = get_bits(&s->gb, 8);
734 if (flag && sb_reorder >= s->min_mono_subband)
735 s->sec_ch_lrms[ch1 / 2][sb_reorder] = get_bits(&s->gb, 8);
738 quant_level = s->quant_levels[ch1 / 2][sb];
742 // Time samples for one or both channels
743 if (sb < s->max_mono_subband && sb_reorder >= s->min_mono_subband) {
745 parse_ch(s, ch1, sb_reorder, quant_level, 0);
747 parse_ch(s, ch2, sb_reorder, quant_level, 1);
749 parse_ch(s, ch1, sb_reorder, quant_level, 0);
751 parse_ch(s, ch2, sb_reorder, quant_level, 0);
759 * Convert from reflection coefficients to direct form coefficients
761 static void convert_lpc(float *coeff, const int *codes)
765 for (i = 0; i < 8; i++) {
766 float rc = lpc_tab[codes[i]];
767 for (j = 0; j < (i + 1) / 2; j++) {
768 float tmp1 = coeff[ j ];
769 float tmp2 = coeff[i - j - 1];
770 coeff[ j ] = tmp1 + rc * tmp2;
771 coeff[i - j - 1] = tmp2 + rc * tmp1;
777 static int parse_lpc(DCALbrDecoder *s, int ch1, int ch2, int start_sb, int end_sb)
779 int f = s->framenum & 1;
780 int i, sb, ch, codes[16];
782 // First two subbands have two sets of coefficients, third subband has one
783 for (sb = start_sb; sb < end_sb; sb++) {
784 int ncodes = 8 * (1 + (sb < 2));
785 for (ch = ch1; ch <= ch2; ch++) {
786 if (ensure_bits(&s->gb, 4 * ncodes))
788 for (i = 0; i < ncodes; i++)
789 codes[i] = get_bits(&s->gb, 4);
790 for (i = 0; i < ncodes / 8; i++)
791 convert_lpc(s->lpc_coeff[f][ch][sb][i], &codes[i * 8]);
798 static int parse_high_res_grid(DCALbrDecoder *s, LBRChunk *chunk, int ch1, int ch2)
800 int quant_levels[DCA_LBR_SUBBANDS];
801 int sb, ch, ol, st, max_sb, profile;
806 if (init_get_bits8(&s->gb, chunk->data, chunk->len) < 0)
810 profile = get_bits(&s->gb, 8);
812 ol = (profile >> 3) & 7;
815 // Max energy subband
816 max_sb = profile & 7;
818 // Calculate quantization levels
819 for (sb = 0; sb < s->nsubbands; sb++) {
820 int f = sb * s->limited_rate / s->nsubbands;
821 int a = 18000 / (12 * f / 1000 + 100 + 40 * st) + 20 * ol;
823 quant_levels[sb] = 1;
825 quant_levels[sb] = 2;
827 quant_levels[sb] = 3;
829 quant_levels[sb] = 4;
831 quant_levels[sb] = 5;
834 // Reorder quantization levels for lower subbands
835 for (sb = 0; sb < 8; sb++)
836 s->quant_levels[ch1 / 2][sb] = quant_levels[ff_dca_sb_reorder[max_sb][sb]];
837 for (; sb < s->nsubbands; sb++)
838 s->quant_levels[ch1 / 2][sb] = quant_levels[sb];
840 // LPC for the first two subbands
841 if (parse_lpc(s, ch1, ch2, 0, 2) < 0)
844 // Time-samples for the first two subbands of main channel
845 if (parse_ts(s, ch1, ch2, 0, 2, 0) < 0)
848 // First two bands of the first grid
849 for (sb = 0; sb < 2; sb++)
850 for (ch = ch1; ch <= ch2; ch++)
851 if (parse_scale_factors(s, s->grid_1_scf[ch][sb]) < 0)
857 static int parse_grid_2(DCALbrDecoder *s, int ch1, int ch2,
858 int start_sb, int end_sb, int flag)
860 int i, j, sb, ch, nsubbands;
862 nsubbands = ff_dca_scf_to_grid_2[s->nsubbands - 1] + 1;
863 if (end_sb > nsubbands)
866 for (sb = start_sb; sb < end_sb; sb++) {
867 for (ch = ch1; ch <= ch2; ch++) {
868 uint8_t *g2_scf = s->grid_2_scf[ch][sb];
870 if ((ch != ch1 && ff_dca_grid_2_to_scf[sb] >= s->min_mono_subband) != flag) {
872 memcpy(g2_scf, s->grid_2_scf[ch1][sb], 64);
876 // Scale factors in groups of 8
877 for (i = 0; i < 8; i++, g2_scf += 8) {
878 if (get_bits_left(&s->gb) < 1) {
879 memset(g2_scf, 0, 64 - i * 8);
882 // Bit indicating if whole group has zero values
883 if (get_bits1(&s->gb)) {
884 for (j = 0; j < 8; j++) {
885 if (ensure_bits(&s->gb, 20))
887 g2_scf[j] = parse_vlc(&s->gb, &ff_dca_vlc_grid_2, 2);
890 memset(g2_scf, 0, 8);
899 static int parse_ts1_chunk(DCALbrDecoder *s, LBRChunk *chunk, int ch1, int ch2)
903 if (init_get_bits8(&s->gb, chunk->data, chunk->len) < 0)
905 if (parse_lpc(s, ch1, ch2, 2, 3) < 0)
907 if (parse_ts(s, ch1, ch2, 2, 4, 0) < 0)
909 if (parse_grid_2(s, ch1, ch2, 0, 1, 0) < 0)
911 if (parse_ts(s, ch1, ch2, 4, 6, 0) < 0)
916 static int parse_ts2_chunk(DCALbrDecoder *s, LBRChunk *chunk, int ch1, int ch2)
920 if (init_get_bits8(&s->gb, chunk->data, chunk->len) < 0)
922 if (parse_grid_2(s, ch1, ch2, 1, 3, 0) < 0)
924 if (parse_ts(s, ch1, ch2, 6, s->max_mono_subband, 0) < 0)
927 if (parse_grid_1_sec_ch(s, ch2) < 0)
929 if (parse_grid_2(s, ch1, ch2, 0, 3, 1) < 0)
932 if (parse_ts(s, ch1, ch2, s->min_mono_subband, s->nsubbands, 1) < 0)
937 static int init_sample_rate(DCALbrDecoder *s)
939 double scale = (-1.0 / (1 << 17)) * sqrt(1 << (2 - s->limited_range));
940 int i, br_per_ch = s->bit_rate_scaled / s->nchannels_total;
942 ff_mdct_end(&s->imdct);
944 if (ff_mdct_init(&s->imdct, s->freq_range + 6, 1, scale) < 0)
947 for (i = 0; i < 32 << s->freq_range; i++)
948 s->window[i] = ff_dca_long_window[i << (2 - s->freq_range)];
950 if (br_per_ch < 14000)
952 else if (br_per_ch < 32000)
953 scale = (br_per_ch - 14000) * (1.0 / 120000) + 0.85;
957 scale *= 1.0 / INT_MAX;
959 for (i = 0; i < s->nsubbands; i++) {
961 s->sb_scf[i] = 0; // The first two subbands are always zero
963 s->sb_scf[i] = (i - 1) * 0.25 * 0.785 * scale;
965 s->sb_scf[i] = 0.785 * scale;
968 s->lfe_scale = (16 << s->freq_range) * 0.0000078265894;
973 static int alloc_sample_buffer(DCALbrDecoder *s)
975 // Reserve space for history and padding
976 int nchsamples = DCA_LBR_TIME_SAMPLES + DCA_LBR_TIME_HISTORY * 2;
977 int nsamples = nchsamples * s->nchannels * s->nsubbands;
981 // Reallocate time sample buffer
982 av_fast_mallocz(&s->ts_buffer, &s->ts_size, nsamples * sizeof(float));
986 ptr = s->ts_buffer + DCA_LBR_TIME_HISTORY;
987 for (ch = 0; ch < s->nchannels; ch++) {
988 for (sb = 0; sb < s->nsubbands; sb++) {
989 s->time_samples[ch][sb] = ptr;
997 static int parse_decoder_init(DCALbrDecoder *s, GetByteContext *gb)
999 int old_rate = s->sample_rate;
1000 int old_band_limit = s->band_limit;
1001 int old_nchannels = s->nchannels;
1002 int version, bit_rate_hi;
1005 // Sample rate of LBR audio
1006 code = bytestream2_get_byte(gb);
1007 if (code >= FF_ARRAY_ELEMS(ff_dca_sampling_freqs)) {
1008 av_log(s->avctx, AV_LOG_ERROR, "Invalid LBR sample rate\n");
1009 return AVERROR_INVALIDDATA;
1011 s->sample_rate = ff_dca_sampling_freqs[code];
1012 if (s->sample_rate > 48000) {
1013 avpriv_report_missing_feature(s->avctx, "%d Hz LBR sample rate", s->sample_rate);
1014 return AVERROR_PATCHWELCOME;
1018 s->ch_mask = bytestream2_get_le16(gb);
1019 if (!(s->ch_mask & 0x7)) {
1020 avpriv_report_missing_feature(s->avctx, "LBR channel mask %#x", s->ch_mask);
1021 return AVERROR_PATCHWELCOME;
1023 if ((s->ch_mask & 0xfff0) && !(s->warned & 1)) {
1024 avpriv_report_missing_feature(s->avctx, "LBR channel mask %#x", s->ch_mask);
1028 // LBR bitstream version
1029 version = bytestream2_get_le16(gb);
1030 if ((version & 0xff00) != 0x0800) {
1031 avpriv_report_missing_feature(s->avctx, "LBR stream version %#x", version);
1032 return AVERROR_PATCHWELCOME;
1035 // Flags for LBR decoder initialization
1036 s->flags = bytestream2_get_byte(gb);
1037 if (s->flags & LBR_FLAG_DMIX_MULTI_CH) {
1038 avpriv_report_missing_feature(s->avctx, "LBR multi-channel downmix");
1039 return AVERROR_PATCHWELCOME;
1041 if ((s->flags & LBR_FLAG_LFE_PRESENT) && s->sample_rate != 48000) {
1042 if (!(s->warned & 2)) {
1043 avpriv_report_missing_feature(s->avctx, "%d Hz LFE interpolation", s->sample_rate);
1046 s->flags &= ~LBR_FLAG_LFE_PRESENT;
1049 // Most significant bit rate nibbles
1050 bit_rate_hi = bytestream2_get_byte(gb);
1052 // Least significant original bit rate word
1053 s->bit_rate_orig = bytestream2_get_le16(gb) | ((bit_rate_hi & 0x0F) << 16);
1055 // Least significant scaled bit rate word
1056 s->bit_rate_scaled = bytestream2_get_le16(gb) | ((bit_rate_hi & 0xF0) << 12);
1058 // Setup number of fullband channels
1059 s->nchannels_total = ff_dca_count_chs_for_mask(s->ch_mask & ~DCA_SPEAKER_PAIR_LFE1);
1060 s->nchannels = FFMIN(s->nchannels_total, DCA_LBR_CHANNELS);
1063 switch (s->flags & LBR_FLAG_BAND_LIMIT_MASK) {
1064 case LBR_FLAG_BAND_LIMIT_NONE:
1067 case LBR_FLAG_BAND_LIMIT_1_2:
1070 case LBR_FLAG_BAND_LIMIT_1_4:
1074 avpriv_report_missing_feature(s->avctx, "LBR band limit %#x", s->flags & LBR_FLAG_BAND_LIMIT_MASK);
1075 return AVERROR_PATCHWELCOME;
1078 // Setup frequency range
1079 if (s->sample_rate < 14000)
1081 else if (s->sample_rate < 28000)
1086 // Setup resolution profile
1087 if (s->bit_rate_orig >= 44000 * (s->nchannels_total + 2))
1089 else if (s->bit_rate_orig >= 25000 * (s->nchannels_total + 2))
1094 // Setup limited sample rate, number of subbands, etc
1095 s->limited_rate = s->sample_rate >> s->band_limit;
1096 s->limited_range = s->freq_range - s->band_limit;
1097 if (s->limited_range < 0) {
1098 av_log(s->avctx, AV_LOG_ERROR, "Invalid LBR band limit for frequency range\n");
1099 return AVERROR_INVALIDDATA;
1102 s->nsubbands = 8 << s->limited_range;
1104 s->g3_avg_only_start_sb = s->nsubbands * ff_dca_avg_g3_freqs[s->res_profile] / (s->limited_rate / 2);
1105 if (s->g3_avg_only_start_sb > s->nsubbands)
1106 s->g3_avg_only_start_sb = s->nsubbands;
1108 s->min_mono_subband = s->nsubbands * 2000 / (s->limited_rate / 2);
1109 if (s->min_mono_subband > s->nsubbands)
1110 s->min_mono_subband = s->nsubbands;
1112 s->max_mono_subband = s->nsubbands * 14000 / (s->limited_rate / 2);
1113 if (s->max_mono_subband > s->nsubbands)
1114 s->max_mono_subband = s->nsubbands;
1116 // Handle change of sample rate
1117 if ((old_rate != s->sample_rate || old_band_limit != s->band_limit) && init_sample_rate(s) < 0)
1118 return AVERROR(ENOMEM);
1120 // Setup stereo downmix
1121 if (s->flags & LBR_FLAG_DMIX_STEREO) {
1122 DCAContext *dca = s->avctx->priv_data;
1124 if (s->nchannels_total < 3 || s->nchannels_total > DCA_LBR_CHANNELS_TOTAL - 2) {
1125 av_log(s->avctx, AV_LOG_ERROR, "Invalid number of channels for LBR stereo downmix\n");
1126 return AVERROR_INVALIDDATA;
1129 // This decoder doesn't support ECS chunk
1130 if (dca->request_channel_layout != DCA_SPEAKER_LAYOUT_STEREO && !(s->warned & 4)) {
1131 avpriv_report_missing_feature(s->avctx, "Embedded LBR stereo downmix");
1135 // Account for extra downmixed channel pair
1136 s->nchannels_total += 2;
1138 s->ch_mask = DCA_SPEAKER_PAIR_LR;
1139 s->flags &= ~LBR_FLAG_LFE_PRESENT;
1142 // Handle change of sample rate or number of channels
1143 if (old_rate != s->sample_rate
1144 || old_band_limit != s->band_limit
1145 || old_nchannels != s->nchannels) {
1146 if (alloc_sample_buffer(s) < 0)
1147 return AVERROR(ENOMEM);
1148 ff_dca_lbr_flush(s);
1154 int ff_dca_lbr_parse(DCALbrDecoder *s, uint8_t *data, DCAExssAsset *asset)
1159 LBRChunk tonal_grp[5];
1160 LBRChunk grid1[DCA_LBR_CHANNELS / 2];
1161 LBRChunk hr_grid[DCA_LBR_CHANNELS / 2];
1162 LBRChunk ts1[DCA_LBR_CHANNELS / 2];
1163 LBRChunk ts2[DCA_LBR_CHANNELS / 2];
1168 int i, ch, sb, sf, ret, group, chunk_id, chunk_len;
1170 bytestream2_init(&gb, data + asset->lbr_offset, asset->lbr_size);
1173 if (bytestream2_get_be32(&gb) != DCA_SYNCWORD_LBR) {
1174 av_log(s->avctx, AV_LOG_ERROR, "Invalid LBR sync word\n");
1175 return AVERROR_INVALIDDATA;
1179 switch (bytestream2_get_byte(&gb)) {
1180 case LBR_HEADER_SYNC_ONLY:
1181 if (!s->sample_rate) {
1182 av_log(s->avctx, AV_LOG_ERROR, "LBR decoder not initialized\n");
1183 return AVERROR_INVALIDDATA;
1186 case LBR_HEADER_DECODER_INIT:
1187 if ((ret = parse_decoder_init(s, &gb)) < 0) {
1193 av_log(s->avctx, AV_LOG_ERROR, "Invalid LBR header type\n");
1194 return AVERROR_INVALIDDATA;
1197 // LBR frame chunk header
1198 chunk_id = bytestream2_get_byte(&gb);
1199 chunk_len = (chunk_id & 0x80) ? bytestream2_get_be16(&gb) : bytestream2_get_byte(&gb);
1201 if (chunk_len > bytestream2_get_bytes_left(&gb)) {
1202 chunk_len = bytestream2_get_bytes_left(&gb);
1203 av_log(s->avctx, AV_LOG_WARNING, "LBR frame chunk was truncated\n");
1204 if (s->avctx->err_recognition & AV_EF_EXPLODE)
1205 return AVERROR_INVALIDDATA;
1208 bytestream2_init(&gb, gb.buffer, chunk_len);
1210 switch (chunk_id & 0x7f) {
1211 case LBR_CHUNK_FRAME:
1212 if (s->avctx->err_recognition & (AV_EF_CRCCHECK | AV_EF_CAREFUL)) {
1213 int checksum = bytestream2_get_be16(&gb);
1214 uint16_t res = chunk_id;
1215 res += (chunk_len >> 8) & 0xff;
1216 res += chunk_len & 0xff;
1217 for (i = 0; i < chunk_len - 2; i++)
1218 res += gb.buffer[i];
1219 if (checksum != res) {
1220 av_log(s->avctx, AV_LOG_WARNING, "Invalid LBR checksum\n");
1221 if (s->avctx->err_recognition & AV_EF_EXPLODE)
1222 return AVERROR_INVALIDDATA;
1225 bytestream2_skip(&gb, 2);
1228 case LBR_CHUNK_FRAME_NO_CSUM:
1231 av_log(s->avctx, AV_LOG_ERROR, "Invalid LBR frame chunk ID\n");
1232 return AVERROR_INVALIDDATA;
1235 // Clear current frame
1236 memset(s->quant_levels, 0, sizeof(s->quant_levels));
1237 memset(s->sb_indices, 0xff, sizeof(s->sb_indices));
1238 memset(s->sec_ch_sbms, 0, sizeof(s->sec_ch_sbms));
1239 memset(s->sec_ch_lrms, 0, sizeof(s->sec_ch_lrms));
1240 memset(s->ch_pres, 0, sizeof(s->ch_pres));
1241 memset(s->grid_1_scf, 0, sizeof(s->grid_1_scf));
1242 memset(s->grid_2_scf, 0, sizeof(s->grid_2_scf));
1243 memset(s->grid_3_avg, 0, sizeof(s->grid_3_avg));
1244 memset(s->grid_3_scf, 0, sizeof(s->grid_3_scf));
1245 memset(s->grid_3_pres, 0, sizeof(s->grid_3_pres));
1246 memset(s->tonal_scf, 0, sizeof(s->tonal_scf));
1247 memset(s->lfe_data, 0, sizeof(s->lfe_data));
1248 s->part_stereo_pres = 0;
1249 s->framenum = (s->framenum + 1) & 31;
1251 for (ch = 0; ch < s->nchannels; ch++) {
1252 for (sb = 0; sb < s->nsubbands / 4; sb++) {
1253 s->part_stereo[ch][sb][0] = s->part_stereo[ch][sb][4];
1254 s->part_stereo[ch][sb][4] = 16;
1258 memset(s->lpc_coeff[s->framenum & 1], 0, sizeof(s->lpc_coeff[0]));
1260 for (group = 0; group < 5; group++) {
1261 for (sf = 0; sf < 1 << group; sf++) {
1262 int sf_idx = ((s->framenum << group) + sf) & 31;
1263 s->tonal_bounds[group][sf_idx][0] =
1264 s->tonal_bounds[group][sf_idx][1] = s->ntones;
1268 // Parse chunk headers
1269 while (bytestream2_get_bytes_left(&gb) > 0) {
1270 chunk_id = bytestream2_get_byte(&gb);
1271 chunk_len = (chunk_id & 0x80) ? bytestream2_get_be16(&gb) : bytestream2_get_byte(&gb);
1274 if (chunk_len > bytestream2_get_bytes_left(&gb)) {
1275 chunk_len = bytestream2_get_bytes_left(&gb);
1276 av_log(s->avctx, AV_LOG_WARNING, "LBR chunk %#x was truncated\n", chunk_id);
1277 if (s->avctx->err_recognition & AV_EF_EXPLODE)
1278 return AVERROR_INVALIDDATA;
1283 chunk.lfe.len = chunk_len;
1284 chunk.lfe.data = gb.buffer;
1288 case LBR_CHUNK_TONAL:
1289 case LBR_CHUNK_TONAL_SCF:
1290 chunk.tonal.id = chunk_id;
1291 chunk.tonal.len = chunk_len;
1292 chunk.tonal.data = gb.buffer;
1295 case LBR_CHUNK_TONAL_GRP_1:
1296 case LBR_CHUNK_TONAL_GRP_2:
1297 case LBR_CHUNK_TONAL_GRP_3:
1298 case LBR_CHUNK_TONAL_GRP_4:
1299 case LBR_CHUNK_TONAL_GRP_5:
1300 i = LBR_CHUNK_TONAL_GRP_5 - chunk_id;
1301 chunk.tonal_grp[i].id = i;
1302 chunk.tonal_grp[i].len = chunk_len;
1303 chunk.tonal_grp[i].data = gb.buffer;
1306 case LBR_CHUNK_TONAL_SCF_GRP_1:
1307 case LBR_CHUNK_TONAL_SCF_GRP_2:
1308 case LBR_CHUNK_TONAL_SCF_GRP_3:
1309 case LBR_CHUNK_TONAL_SCF_GRP_4:
1310 case LBR_CHUNK_TONAL_SCF_GRP_5:
1311 i = LBR_CHUNK_TONAL_SCF_GRP_5 - chunk_id;
1312 chunk.tonal_grp[i].id = i;
1313 chunk.tonal_grp[i].len = chunk_len;
1314 chunk.tonal_grp[i].data = gb.buffer;
1317 case LBR_CHUNK_RES_GRID_LR:
1318 case LBR_CHUNK_RES_GRID_LR + 1:
1319 case LBR_CHUNK_RES_GRID_LR + 2:
1320 i = chunk_id - LBR_CHUNK_RES_GRID_LR;
1321 chunk.grid1[i].len = chunk_len;
1322 chunk.grid1[i].data = gb.buffer;
1325 case LBR_CHUNK_RES_GRID_HR:
1326 case LBR_CHUNK_RES_GRID_HR + 1:
1327 case LBR_CHUNK_RES_GRID_HR + 2:
1328 i = chunk_id - LBR_CHUNK_RES_GRID_HR;
1329 chunk.hr_grid[i].len = chunk_len;
1330 chunk.hr_grid[i].data = gb.buffer;
1333 case LBR_CHUNK_RES_TS_1:
1334 case LBR_CHUNK_RES_TS_1 + 1:
1335 case LBR_CHUNK_RES_TS_1 + 2:
1336 i = chunk_id - LBR_CHUNK_RES_TS_1;
1337 chunk.ts1[i].len = chunk_len;
1338 chunk.ts1[i].data = gb.buffer;
1341 case LBR_CHUNK_RES_TS_2:
1342 case LBR_CHUNK_RES_TS_2 + 1:
1343 case LBR_CHUNK_RES_TS_2 + 2:
1344 i = chunk_id - LBR_CHUNK_RES_TS_2;
1345 chunk.ts2[i].len = chunk_len;
1346 chunk.ts2[i].data = gb.buffer;
1350 bytestream2_skip(&gb, chunk_len);
1354 ret = parse_lfe_chunk(s, &chunk.lfe);
1356 ret |= parse_tonal_chunk(s, &chunk.tonal);
1358 for (i = 0; i < 5; i++)
1359 ret |= parse_tonal_group(s, &chunk.tonal_grp[i]);
1361 for (i = 0; i < (s->nchannels + 1) / 2; i++) {
1363 int ch2 = FFMIN(ch1 + 1, s->nchannels - 1);
1365 if (parse_grid_1_chunk (s, &chunk.grid1 [i], ch1, ch2) < 0 ||
1366 parse_high_res_grid(s, &chunk.hr_grid[i], ch1, ch2) < 0) {
1371 // TS chunks depend on both grids. TS_2 depends on TS_1.
1372 if (!chunk.grid1[i].len || !chunk.hr_grid[i].len || !chunk.ts1[i].len)
1375 if (parse_ts1_chunk(s, &chunk.ts1[i], ch1, ch2) < 0 ||
1376 parse_ts2_chunk(s, &chunk.ts2[i], ch1, ch2) < 0) {
1382 if (ret < 0 && (s->avctx->err_recognition & AV_EF_EXPLODE))
1383 return AVERROR_INVALIDDATA;
1389 * Reconstruct high-frequency resolution grid from first and third grids
1391 static void decode_grid(DCALbrDecoder *s, int ch1, int ch2)
1395 for (ch = ch1; ch <= ch2; ch++) {
1396 for (sb = 0; sb < s->nsubbands; sb++) {
1397 int g1_sb = ff_dca_scf_to_grid_1[sb];
1399 uint8_t *g1_scf_a = s->grid_1_scf[ch][g1_sb ];
1400 uint8_t *g1_scf_b = s->grid_1_scf[ch][g1_sb + 1];
1402 int w1 = ff_dca_grid_1_weights[g1_sb ][sb];
1403 int w2 = ff_dca_grid_1_weights[g1_sb + 1][sb];
1405 uint8_t *hr_scf = s->high_res_scf[ch][sb];
1408 for (i = 0; i < 8; i++) {
1409 int scf = w1 * g1_scf_a[i] + w2 * g1_scf_b[i];
1410 hr_scf[i] = scf >> 7;
1413 int8_t *g3_scf = s->grid_3_scf[ch][sb - 4];
1414 int g3_avg = s->grid_3_avg[ch][sb - 4];
1416 for (i = 0; i < 8; i++) {
1417 int scf = w1 * g1_scf_a[i] + w2 * g1_scf_b[i];
1418 hr_scf[i] = (scf >> 7) - g3_avg - g3_scf[i];
1426 * Fill unallocated subbands with randomness
1428 static void random_ts(DCALbrDecoder *s, int ch1, int ch2)
1430 int i, j, k, ch, sb;
1432 for (ch = ch1; ch <= ch2; ch++) {
1433 for (sb = 0; sb < s->nsubbands; sb++) {
1434 float *samples = s->time_samples[ch][sb];
1436 if (s->ch_pres[ch] & (1U << sb))
1437 continue; // Skip allocated subband
1440 // The first two subbands are always zero
1441 memset(samples, 0, DCA_LBR_TIME_SAMPLES * sizeof(float));
1442 } else if (sb < 10) {
1443 for (i = 0; i < DCA_LBR_TIME_SAMPLES; i++)
1444 samples[i] = lbr_rand(s, sb);
1446 for (i = 0; i < DCA_LBR_TIME_SAMPLES / 8; i++, samples += 8) {
1447 float accum[8] = { 0 };
1449 // Modulate by subbands 2-5 in blocks of 8
1450 for (k = 2; k < 6; k++) {
1451 float *other = &s->time_samples[ch][k][i * 8];
1452 for (j = 0; j < 8; j++)
1453 accum[j] += fabs(other[j]);
1456 for (j = 0; j < 8; j++)
1457 samples[j] = (accum[j] * 0.25f + 0.5f) * lbr_rand(s, sb);
1464 static void predict(float *samples, const float *coeff, int nsamples)
1468 for (i = 0; i < nsamples; i++) {
1470 for (j = 0; j < 8; j++)
1471 res += coeff[j] * samples[i - j - 1];
1476 static void synth_lpc(DCALbrDecoder *s, int ch1, int ch2, int sb)
1478 int f = s->framenum & 1;
1481 for (ch = ch1; ch <= ch2; ch++) {
1482 float *samples = s->time_samples[ch][sb];
1484 if (!(s->ch_pres[ch] & (1U << sb)))
1488 predict(samples, s->lpc_coeff[f^1][ch][sb][1], 16);
1489 predict(samples + 16, s->lpc_coeff[f ][ch][sb][0], 64);
1490 predict(samples + 80, s->lpc_coeff[f ][ch][sb][1], 48);
1492 predict(samples, s->lpc_coeff[f^1][ch][sb][0], 16);
1493 predict(samples + 16, s->lpc_coeff[f ][ch][sb][0], 112);
1498 static void filter_ts(DCALbrDecoder *s, int ch1, int ch2)
1502 for (sb = 0; sb < s->nsubbands; sb++) {
1504 for (ch = ch1; ch <= ch2; ch++) {
1505 float *samples = s->time_samples[ch][sb];
1506 uint8_t *hr_scf = s->high_res_scf[ch][sb];
1508 for (i = 0; i < DCA_LBR_TIME_SAMPLES / 16; i++, samples += 16) {
1509 unsigned int scf = hr_scf[i];
1512 for (j = 0; j < 16; j++)
1513 samples[j] *= ff_dca_quant_amp[scf];
1516 uint8_t *g2_scf = s->grid_2_scf[ch][ff_dca_scf_to_grid_2[sb]];
1517 for (i = 0; i < DCA_LBR_TIME_SAMPLES / 2; i++, samples += 2) {
1518 unsigned int scf = hr_scf[i / 8] - g2_scf[i];
1521 samples[0] *= ff_dca_quant_amp[scf];
1522 samples[1] *= ff_dca_quant_amp[scf];
1529 float *samples_l = s->time_samples[ch1][sb];
1530 float *samples_r = s->time_samples[ch2][sb];
1531 int ch2_pres = s->ch_pres[ch2] & (1U << sb);
1533 for (i = 0; i < DCA_LBR_TIME_SAMPLES / 16; i++) {
1534 int sbms = (s->sec_ch_sbms[ch1 / 2][sb] >> i) & 1;
1535 int lrms = (s->sec_ch_lrms[ch1 / 2][sb] >> i) & 1;
1537 if (sb >= s->min_mono_subband) {
1538 if (lrms && ch2_pres) {
1540 for (j = 0; j < 16; j++) {
1541 float tmp = samples_l[j];
1542 samples_l[j] = samples_r[j];
1543 samples_r[j] = -tmp;
1546 for (j = 0; j < 16; j++) {
1547 float tmp = samples_l[j];
1548 samples_l[j] = samples_r[j];
1552 } else if (!ch2_pres) {
1553 if (sbms && (s->part_stereo_pres & (1 << ch1))) {
1554 for (j = 0; j < 16; j++)
1555 samples_r[j] = -samples_l[j];
1557 for (j = 0; j < 16; j++)
1558 samples_r[j] = samples_l[j];
1561 } else if (sbms && ch2_pres) {
1562 for (j = 0; j < 16; j++) {
1563 float tmp = samples_l[j];
1564 samples_l[j] = (tmp + samples_r[j]) * 0.5f;
1565 samples_r[j] = (tmp - samples_r[j]) * 0.5f;
1574 // Inverse prediction
1576 synth_lpc(s, ch1, ch2, sb);
1581 * Modulate by interpolated partial stereo coefficients
1583 static void decode_part_stereo(DCALbrDecoder *s, int ch1, int ch2)
1587 for (ch = ch1; ch <= ch2; ch++) {
1588 for (sb = s->min_mono_subband; sb < s->nsubbands; sb++) {
1589 uint8_t *pt_st = s->part_stereo[ch][(sb - s->min_mono_subband) / 4];
1590 float *samples = s->time_samples[ch][sb];
1592 if (s->ch_pres[ch2] & (1U << sb))
1595 for (sf = 1; sf <= 4; sf++, samples += 32) {
1596 float prev = ff_dca_st_coeff[pt_st[sf - 1]];
1597 float next = ff_dca_st_coeff[pt_st[sf ]];
1599 for (i = 0; i < 32; i++)
1600 samples[i] *= (32 - i) * prev + i * next;
1607 * Synthesise tones in the given group for the given tonal subframe
1609 static void synth_tones(DCALbrDecoder *s, int ch, float *values,
1610 int group, int group_sf, int synth_idx)
1612 int i, start, count;
1617 start = s->tonal_bounds[group][group_sf][0];
1618 count = (s->tonal_bounds[group][group_sf][1] - start) & (DCA_LBR_TONES - 1);
1620 for (i = 0; i < count; i++) {
1621 DCALbrTone *t = &s->tones[(start + i) & (DCA_LBR_TONES - 1)];
1624 float amp = ff_dca_synth_env[synth_idx] * ff_dca_quant_amp[t->amp[ch]];
1625 float c = amp * cos_tab[(t->phs[ch] ) & 255];
1626 float s = amp * cos_tab[(t->phs[ch] + 64) & 255];
1627 const float *cf = ff_dca_corr_cf[t->f_delt];
1628 int x_freq = t->x_freq;
1634 values[3] += cf[0] * -s;
1635 values[2] += cf[1] * c;
1636 values[1] += cf[2] * s;
1637 values[0] += cf[3] * -c;
1640 values[2] += cf[0] * -s;
1641 values[1] += cf[1] * c;
1642 values[0] += cf[2] * s;
1645 values[1] += cf[0] * -s;
1646 values[0] += cf[1] * c;
1649 values[0] += cf[0] * -s;
1653 values[x_freq - 5] += cf[ 0] * -s;
1654 p4: values[x_freq - 4] += cf[ 1] * c;
1655 p3: values[x_freq - 3] += cf[ 2] * s;
1656 p2: values[x_freq - 2] += cf[ 3] * -c;
1657 p1: values[x_freq - 1] += cf[ 4] * -s;
1658 p0: values[x_freq ] += cf[ 5] * c;
1659 values[x_freq + 1] += cf[ 6] * s;
1660 values[x_freq + 2] += cf[ 7] * -c;
1661 values[x_freq + 3] += cf[ 8] * -s;
1662 values[x_freq + 4] += cf[ 9] * c;
1663 values[x_freq + 5] += cf[10] * s;
1666 t->phs[ch] += t->ph_rot;
1671 * Synthesise all tones in all groups for the given residual subframe
1673 static void base_func_synth(DCALbrDecoder *s, int ch, float *values, int sf)
1677 // Tonal vs residual shift is 22 subframes
1678 for (group = 0; group < 5; group++) {
1679 int group_sf = (s->framenum << group) + ((sf - 22) >> (5 - group));
1680 int synth_idx = ((((sf - 22) & 31) << group) & 31) + (1 << group) - 1;
1682 synth_tones(s, ch, values, group, (group_sf - 1) & 31, 30 - synth_idx);
1683 synth_tones(s, ch, values, group, (group_sf ) & 31, synth_idx);
1687 static void transform_channel(DCALbrDecoder *s, int ch, float *output)
1689 LOCAL_ALIGNED_32(float, values, [DCA_LBR_SUBBANDS ], [4]);
1690 LOCAL_ALIGNED_32(float, result, [DCA_LBR_SUBBANDS * 2], [4]);
1691 int sf, sb, nsubbands = s->nsubbands, noutsubbands = 8 << s->freq_range;
1693 // Clear inactive subbands
1694 if (nsubbands < noutsubbands)
1695 memset(values[nsubbands], 0, (noutsubbands - nsubbands) * sizeof(values[0]));
1697 for (sf = 0; sf < DCA_LBR_TIME_SAMPLES / 4; sf++) {
1698 // Hybrid filterbank
1699 s->dcadsp->lbr_bank(values, s->time_samples[ch],
1700 ff_dca_bank_coeff, sf * 4, nsubbands);
1702 base_func_synth(s, ch, values[0], sf);
1704 s->imdct.imdct_calc(&s->imdct, result[0], values[0]);
1706 // Long window and overlap-add
1707 s->fdsp->vector_fmul_add(output, result[0], s->window,
1708 s->history[ch], noutsubbands * 4);
1709 s->fdsp->vector_fmul_reverse(s->history[ch], result[noutsubbands],
1710 s->window, noutsubbands * 4);
1711 output += noutsubbands * 4;
1714 // Update history for LPC and forward MDCT
1715 for (sb = 0; sb < nsubbands; sb++) {
1716 float *samples = s->time_samples[ch][sb] - DCA_LBR_TIME_HISTORY;
1717 memcpy(samples, samples + DCA_LBR_TIME_SAMPLES, DCA_LBR_TIME_HISTORY * sizeof(float));
1721 int ff_dca_lbr_filter_frame(DCALbrDecoder *s, AVFrame *frame)
1723 AVCodecContext *avctx = s->avctx;
1724 int i, ret, nchannels, ch_conf = (s->ch_mask & 0x7) - 1;
1725 const int8_t *reorder;
1727 avctx->channel_layout = channel_layouts[ch_conf];
1728 avctx->channels = nchannels = channel_counts[ch_conf];
1729 avctx->sample_rate = s->sample_rate;
1730 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1731 avctx->bits_per_raw_sample = 0;
1732 avctx->profile = FF_PROFILE_DTS_EXPRESS;
1733 avctx->bit_rate = s->bit_rate_scaled;
1735 if (s->flags & LBR_FLAG_LFE_PRESENT) {
1736 avctx->channel_layout |= AV_CH_LOW_FREQUENCY;
1738 reorder = channel_reorder_lfe[ch_conf];
1740 reorder = channel_reorder_nolfe[ch_conf];
1743 frame->nb_samples = 1024 << s->freq_range;
1744 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1747 // Filter fullband channels
1748 for (i = 0; i < (s->nchannels + 1) / 2; i++) {
1750 int ch2 = FFMIN(ch1 + 1, s->nchannels - 1);
1752 decode_grid(s, ch1, ch2);
1754 random_ts(s, ch1, ch2);
1756 filter_ts(s, ch1, ch2);
1758 if (ch1 != ch2 && (s->part_stereo_pres & (1 << ch1)))
1759 decode_part_stereo(s, ch1, ch2);
1761 if (ch1 < nchannels)
1762 transform_channel(s, ch1, (float *)frame->extended_data[reorder[ch1]]);
1764 if (ch1 != ch2 && ch2 < nchannels)
1765 transform_channel(s, ch2, (float *)frame->extended_data[reorder[ch2]]);
1768 // Interpolate LFE channel
1769 if (s->flags & LBR_FLAG_LFE_PRESENT) {
1770 s->dcadsp->lfe_iir((float *)frame->extended_data[lfe_index[ch_conf]],
1771 s->lfe_data, ff_dca_lfe_iir,
1772 s->lfe_history, 16 << s->freq_range);
1775 if ((ret = ff_side_data_update_matrix_encoding(frame, AV_MATRIX_ENCODING_NONE)) < 0)
1781 av_cold void ff_dca_lbr_flush(DCALbrDecoder *s)
1785 if (!s->sample_rate)
1789 memset(s->part_stereo, 16, sizeof(s->part_stereo));
1790 memset(s->lpc_coeff, 0, sizeof(s->lpc_coeff));
1791 memset(s->history, 0, sizeof(s->history));
1792 memset(s->tonal_bounds, 0, sizeof(s->tonal_bounds));
1793 memset(s->lfe_history, 0, sizeof(s->lfe_history));
1797 for (ch = 0; ch < s->nchannels; ch++) {
1798 for (sb = 0; sb < s->nsubbands; sb++) {
1799 float *samples = s->time_samples[ch][sb] - DCA_LBR_TIME_HISTORY;
1800 memset(samples, 0, DCA_LBR_TIME_HISTORY * sizeof(float));
1805 av_cold int ff_dca_lbr_init(DCALbrDecoder *s)
1809 if (!(s->fdsp = avpriv_float_dsp_alloc(0)))
1816 av_cold void ff_dca_lbr_close(DCALbrDecoder *s)
1820 av_freep(&s->ts_buffer);
1824 ff_mdct_end(&s->imdct);