3 * This code was developed as part of Google Summer of Code 2006.
4 * E-AC-3 support was added as part of Google Summer of Code 2007.
6 * Copyright (c) 2006 Kartikey Mahendra BHATT (bhattkm at gmail dot com)
7 * Copyright (c) 2007-2008 Bartlomiej Wolowiec <bartek.wolowiec@gmail.com>
8 * Copyright (c) 2007 Justin Ruggles <justin.ruggles@gmail.com>
10 * This file is part of FFmpeg.
12 * FFmpeg is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public
14 * License as published by the Free Software Foundation; either
15 * version 2.1 of the License, or (at your option) any later version.
17 * FFmpeg is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with FFmpeg; if not, write to the Free Software
24 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
32 #include "libavutil/channel_layout.h"
33 #include "libavutil/crc.h"
34 #include "libavutil/downmix_info.h"
35 #include "libavutil/opt.h"
37 #include "aac_ac3_parser.h"
38 #include "ac3_parser.h"
40 #include "ac3dec_data.h"
44 * table for ungrouping 3 values in 7 bits.
45 * used for exponents and bap=2 mantissas
47 static uint8_t ungroup_3_in_7_bits_tab[128][3];
49 /** tables for ungrouping mantissas */
50 static int b1_mantissas[32][3];
51 static int b2_mantissas[128][3];
52 static int b3_mantissas[8];
53 static int b4_mantissas[128][2];
54 static int b5_mantissas[16];
57 * Quantization table: levels for symmetric. bits for asymmetric.
58 * reference: Table 7.18 Mapping of bap to Quantizer
60 static const uint8_t quantization_tab[16] = {
62 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
65 /** dynamic range table. converts codes to scale factors. */
66 static float dynamic_range_tab[256];
68 /** Adjustments in dB gain */
69 static const float gain_levels[9] = {
73 LEVEL_MINUS_1POINT5DB,
75 LEVEL_MINUS_4POINT5DB,
81 /** Adjustments in dB gain (LFE, +10 to -21 dB) */
82 static const float gain_levels_lfe[32] = {
83 3.162275, 2.818382, 2.511886, 2.238719, 1.995261, 1.778278, 1.584893,
84 1.412536, 1.258924, 1.122018, 1.000000, 0.891251, 0.794328, 0.707946,
85 0.630957, 0.562341, 0.501187, 0.446683, 0.398107, 0.354813, 0.316227,
86 0.281838, 0.251188, 0.223872, 0.199526, 0.177828, 0.158489, 0.141253,
87 0.125892, 0.112201, 0.100000, 0.089125
91 * Table for default stereo downmixing coefficients
92 * reference: Section 7.8.2 Downmixing Into Two Channels
94 static const uint8_t ac3_default_coeffs[8][5][2] = {
95 { { 2, 7 }, { 7, 2 }, },
97 { { 2, 7 }, { 7, 2 }, },
98 { { 2, 7 }, { 5, 5 }, { 7, 2 }, },
99 { { 2, 7 }, { 7, 2 }, { 6, 6 }, },
100 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 8, 8 }, },
101 { { 2, 7 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
102 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
106 * Symmetrical Dequantization
107 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
108 * Tables 7.19 to 7.23
111 symmetric_dequant(int code, int levels)
113 return ((code - (levels >> 1)) << 24) / levels;
117 * Initialize tables at runtime.
119 static av_cold void ac3_tables_init(void)
123 /* generate table for ungrouping 3 values in 7 bits
124 reference: Section 7.1.3 Exponent Decoding */
125 for (i = 0; i < 128; i++) {
126 ungroup_3_in_7_bits_tab[i][0] = i / 25;
127 ungroup_3_in_7_bits_tab[i][1] = (i % 25) / 5;
128 ungroup_3_in_7_bits_tab[i][2] = (i % 25) % 5;
131 /* generate grouped mantissa tables
132 reference: Section 7.3.5 Ungrouping of Mantissas */
133 for (i = 0; i < 32; i++) {
134 /* bap=1 mantissas */
135 b1_mantissas[i][0] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][0], 3);
136 b1_mantissas[i][1] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][1], 3);
137 b1_mantissas[i][2] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][2], 3);
139 for (i = 0; i < 128; i++) {
140 /* bap=2 mantissas */
141 b2_mantissas[i][0] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][0], 5);
142 b2_mantissas[i][1] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][1], 5);
143 b2_mantissas[i][2] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][2], 5);
145 /* bap=4 mantissas */
146 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
147 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
149 /* generate ungrouped mantissa tables
150 reference: Tables 7.21 and 7.23 */
151 for (i = 0; i < 7; i++) {
152 /* bap=3 mantissas */
153 b3_mantissas[i] = symmetric_dequant(i, 7);
155 for (i = 0; i < 15; i++) {
156 /* bap=5 mantissas */
157 b5_mantissas[i] = symmetric_dequant(i, 15);
160 /* generate dynamic range table
161 reference: Section 7.7.1 Dynamic Range Control */
162 for (i = 0; i < 256; i++) {
163 int v = (i >> 5) - ((i >> 7) << 3) - 5;
164 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
169 * AVCodec initialization
171 static av_cold int ac3_decode_init(AVCodecContext *avctx)
173 AC3DecodeContext *s = avctx->priv_data;
178 ff_ac3_common_init();
180 ff_mdct_init(&s->imdct_256, 8, 1, 1.0);
181 ff_mdct_init(&s->imdct_512, 9, 1, 1.0);
182 ff_kbd_window_init(s->window, 5.0, 256);
183 ff_dsputil_init(&s->dsp, avctx);
184 avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
185 ff_ac3dsp_init(&s->ac3dsp, avctx->flags & CODEC_FLAG_BITEXACT);
186 ff_fmt_convert_init(&s->fmt_conv, avctx);
187 av_lfg_init(&s->dith_state, 0);
189 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
191 /* allow downmixing to stereo or mono */
192 #if FF_API_REQUEST_CHANNELS
193 FF_DISABLE_DEPRECATION_WARNINGS
194 if (avctx->request_channels == 1)
195 avctx->request_channel_layout = AV_CH_LAYOUT_MONO;
196 else if (avctx->request_channels == 2)
197 avctx->request_channel_layout = AV_CH_LAYOUT_STEREO;
198 FF_ENABLE_DEPRECATION_WARNINGS
200 if (avctx->channels > 1 &&
201 avctx->request_channel_layout == AV_CH_LAYOUT_MONO)
203 else if (avctx->channels > 2 &&
204 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO)
208 for (i = 0; i < AC3_MAX_CHANNELS; i++) {
209 s->xcfptr[i] = s->transform_coeffs[i];
210 s->dlyptr[i] = s->delay[i];
217 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
218 * GetBitContext within AC3DecodeContext must point to
219 * the start of the synchronized AC-3 bitstream.
221 static int ac3_parse_header(AC3DecodeContext *s)
223 GetBitContext *gbc = &s->gbc;
226 /* read the rest of the bsi. read twice for dual mono mode. */
227 i = !s->channel_mode;
229 skip_bits(gbc, 5); // skip dialog normalization
231 skip_bits(gbc, 8); //skip compression
233 skip_bits(gbc, 8); //skip language code
235 skip_bits(gbc, 7); //skip audio production information
238 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
240 /* skip the timecodes or parse the Alternate Bit Stream Syntax */
241 if (s->bitstream_id != 6) {
243 skip_bits(gbc, 14); //skip timecode1
245 skip_bits(gbc, 14); //skip timecode2
247 if (get_bits1(gbc)) {
248 s->preferred_downmix = get_bits(gbc, 2);
249 s->center_mix_level_ltrt = get_bits(gbc, 3);
250 s->surround_mix_level_ltrt = av_clip(get_bits(gbc, 3), 3, 7);
251 s->center_mix_level = get_bits(gbc, 3);
252 s->surround_mix_level = av_clip(get_bits(gbc, 3), 3, 7);
254 if (get_bits1(gbc)) {
255 s->dolby_surround_ex_mode = get_bits(gbc, 2);
256 s->dolby_headphone_mode = get_bits(gbc, 2);
257 skip_bits(gbc, 10); // skip adconvtyp (1), xbsi2 (8), encinfo (1)
261 /* skip additional bitstream info */
262 if (get_bits1(gbc)) {
263 i = get_bits(gbc, 6);
273 * Common function to parse AC-3 or E-AC-3 frame header
275 static int parse_frame_header(AC3DecodeContext *s)
277 AC3HeaderInfo hdr, *phdr=&hdr;
280 err = avpriv_ac3_parse_header2(&s->gbc, &phdr);
284 /* get decoding parameters from header info */
285 s->bit_alloc_params.sr_code = hdr.sr_code;
286 s->bitstream_id = hdr.bitstream_id;
287 s->bitstream_mode = hdr.bitstream_mode;
288 s->channel_mode = hdr.channel_mode;
289 s->lfe_on = hdr.lfe_on;
290 s->bit_alloc_params.sr_shift = hdr.sr_shift;
291 s->sample_rate = hdr.sample_rate;
292 s->bit_rate = hdr.bit_rate;
293 s->channels = hdr.channels;
294 s->fbw_channels = s->channels - s->lfe_on;
295 s->lfe_ch = s->fbw_channels + 1;
296 s->frame_size = hdr.frame_size;
297 s->preferred_downmix = AC3_DMIXMOD_NOTINDICATED;
298 s->center_mix_level = hdr.center_mix_level;
299 s->center_mix_level_ltrt = 4; // -3.0dB
300 s->surround_mix_level = hdr.surround_mix_level;
301 s->surround_mix_level_ltrt = 4; // -3.0dB
302 s->lfe_mix_level_exists = 0;
303 s->num_blocks = hdr.num_blocks;
304 s->frame_type = hdr.frame_type;
305 s->substreamid = hdr.substreamid;
306 s->dolby_surround_mode = hdr.dolby_surround_mode;
307 s->dolby_surround_ex_mode = AC3_DSUREXMOD_NOTINDICATED;
308 s->dolby_headphone_mode = AC3_DHEADPHONMOD_NOTINDICATED;
311 s->start_freq[s->lfe_ch] = 0;
312 s->end_freq[s->lfe_ch] = 7;
313 s->num_exp_groups[s->lfe_ch] = 2;
314 s->channel_in_cpl[s->lfe_ch] = 0;
317 if (s->bitstream_id <= 10) {
319 s->snr_offset_strategy = 2;
320 s->block_switch_syntax = 1;
321 s->dither_flag_syntax = 1;
322 s->bit_allocation_syntax = 1;
323 s->fast_gain_syntax = 0;
324 s->first_cpl_leak = 0;
327 memset(s->channel_uses_aht, 0, sizeof(s->channel_uses_aht));
328 return ac3_parse_header(s);
329 } else if (CONFIG_EAC3_DECODER) {
331 return ff_eac3_parse_header(s);
333 av_log(s->avctx, AV_LOG_ERROR, "E-AC-3 support not compiled in\n");
334 return AVERROR(ENOSYS);
339 * Set stereo downmixing coefficients based on frame header info.
340 * reference: Section 7.8.2 Downmixing Into Two Channels
342 static void set_downmix_coeffs(AC3DecodeContext *s)
345 float cmix = gain_levels[s-> center_mix_level];
346 float smix = gain_levels[s->surround_mix_level];
349 for (i = 0; i < s->fbw_channels; i++) {
350 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
351 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
353 if (s->channel_mode > 1 && s->channel_mode & 1) {
354 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
356 if (s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
357 int nf = s->channel_mode - 2;
358 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
360 if (s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
361 int nf = s->channel_mode - 4;
362 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
367 for (i = 0; i < s->fbw_channels; i++) {
368 norm0 += s->downmix_coeffs[i][0];
369 norm1 += s->downmix_coeffs[i][1];
371 norm0 = 1.0f / norm0;
372 norm1 = 1.0f / norm1;
373 for (i = 0; i < s->fbw_channels; i++) {
374 s->downmix_coeffs[i][0] *= norm0;
375 s->downmix_coeffs[i][1] *= norm1;
378 if (s->output_mode == AC3_CHMODE_MONO) {
379 for (i = 0; i < s->fbw_channels; i++)
380 s->downmix_coeffs[i][0] = (s->downmix_coeffs[i][0] +
381 s->downmix_coeffs[i][1]) * LEVEL_MINUS_3DB;
386 * Decode the grouped exponents according to exponent strategy.
387 * reference: Section 7.1.3 Exponent Decoding
389 static int decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
390 uint8_t absexp, int8_t *dexps)
392 int i, j, grp, group_size;
397 group_size = exp_strategy + (exp_strategy == EXP_D45);
398 for (grp = 0, i = 0; grp < ngrps; grp++) {
399 expacc = get_bits(gbc, 7);
400 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][0];
401 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][1];
402 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][2];
405 /* convert to absolute exps and expand groups */
407 for (i = 0, j = 0; i < ngrps * 3; i++) {
408 prevexp += dexp[i] - 2;
411 switch (group_size) {
412 case 4: dexps[j++] = prevexp;
413 dexps[j++] = prevexp;
414 case 2: dexps[j++] = prevexp;
415 case 1: dexps[j++] = prevexp;
422 * Generate transform coefficients for each coupled channel in the coupling
423 * range using the coupling coefficients and coupling coordinates.
424 * reference: Section 7.4.3 Coupling Coordinate Format
426 static void calc_transform_coeffs_cpl(AC3DecodeContext *s)
430 bin = s->start_freq[CPL_CH];
431 for (band = 0; band < s->num_cpl_bands; band++) {
432 int band_start = bin;
433 int band_end = bin + s->cpl_band_sizes[band];
434 for (ch = 1; ch <= s->fbw_channels; ch++) {
435 if (s->channel_in_cpl[ch]) {
436 int cpl_coord = s->cpl_coords[ch][band] << 5;
437 for (bin = band_start; bin < band_end; bin++) {
438 s->fixed_coeffs[ch][bin] =
439 MULH(s->fixed_coeffs[CPL_CH][bin] << 4, cpl_coord);
441 if (ch == 2 && s->phase_flags[band]) {
442 for (bin = band_start; bin < band_end; bin++)
443 s->fixed_coeffs[2][bin] = -s->fixed_coeffs[2][bin];
452 * Grouped mantissas for 3-level 5-level and 11-level quantization
464 * Decode the transform coefficients for a particular channel
465 * reference: Section 7.3 Quantization and Decoding of Mantissas
467 static void ac3_decode_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
469 int start_freq = s->start_freq[ch_index];
470 int end_freq = s->end_freq[ch_index];
471 uint8_t *baps = s->bap[ch_index];
472 int8_t *exps = s->dexps[ch_index];
473 int32_t *coeffs = s->fixed_coeffs[ch_index];
474 int dither = (ch_index == CPL_CH) || s->dither_flag[ch_index];
475 GetBitContext *gbc = &s->gbc;
478 for (freq = start_freq; freq < end_freq; freq++) {
479 int bap = baps[freq];
483 /* random noise with approximate range of -0.707 to 0.707 */
485 mantissa = (((av_lfg_get(&s->dith_state)>>8)*181)>>8) - 5931008;
492 mantissa = m->b1_mant[m->b1];
494 int bits = get_bits(gbc, 5);
495 mantissa = b1_mantissas[bits][0];
496 m->b1_mant[1] = b1_mantissas[bits][1];
497 m->b1_mant[0] = b1_mantissas[bits][2];
504 mantissa = m->b2_mant[m->b2];
506 int bits = get_bits(gbc, 7);
507 mantissa = b2_mantissas[bits][0];
508 m->b2_mant[1] = b2_mantissas[bits][1];
509 m->b2_mant[0] = b2_mantissas[bits][2];
514 mantissa = b3_mantissas[get_bits(gbc, 3)];
519 mantissa = m->b4_mant;
521 int bits = get_bits(gbc, 7);
522 mantissa = b4_mantissas[bits][0];
523 m->b4_mant = b4_mantissas[bits][1];
528 mantissa = b5_mantissas[get_bits(gbc, 4)];
530 default: /* 6 to 15 */
531 /* Shift mantissa and sign-extend it. */
533 av_log(s->avctx, AV_LOG_ERROR, "bap %d is invalid in plain AC-3\n", bap);
536 mantissa = get_sbits(gbc, quantization_tab[bap]);
537 mantissa <<= 24 - quantization_tab[bap];
540 coeffs[freq] = mantissa >> exps[freq];
545 * Remove random dithering from coupling range coefficients with zero-bit
546 * mantissas for coupled channels which do not use dithering.
547 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
549 static void remove_dithering(AC3DecodeContext *s) {
552 for (ch = 1; ch <= s->fbw_channels; ch++) {
553 if (!s->dither_flag[ch] && s->channel_in_cpl[ch]) {
554 for (i = s->start_freq[CPL_CH]; i < s->end_freq[CPL_CH]; i++) {
555 if (!s->bap[CPL_CH][i])
556 s->fixed_coeffs[ch][i] = 0;
562 static void decode_transform_coeffs_ch(AC3DecodeContext *s, int blk, int ch,
565 if (!s->channel_uses_aht[ch]) {
566 ac3_decode_transform_coeffs_ch(s, ch, m);
568 /* if AHT is used, mantissas for all blocks are encoded in the first
569 block of the frame. */
571 if (!blk && CONFIG_EAC3_DECODER)
572 ff_eac3_decode_transform_coeffs_aht_ch(s, ch);
573 for (bin = s->start_freq[ch]; bin < s->end_freq[ch]; bin++) {
574 s->fixed_coeffs[ch][bin] = s->pre_mantissa[ch][bin][blk] >> s->dexps[ch][bin];
580 * Decode the transform coefficients.
582 static void decode_transform_coeffs(AC3DecodeContext *s, int blk)
588 m.b1 = m.b2 = m.b4 = 0;
590 for (ch = 1; ch <= s->channels; ch++) {
591 /* transform coefficients for full-bandwidth channel */
592 decode_transform_coeffs_ch(s, blk, ch, &m);
593 /* transform coefficients for coupling channel come right after the
594 coefficients for the first coupled channel*/
595 if (s->channel_in_cpl[ch]) {
597 decode_transform_coeffs_ch(s, blk, CPL_CH, &m);
598 calc_transform_coeffs_cpl(s);
601 end = s->end_freq[CPL_CH];
603 end = s->end_freq[ch];
606 s->fixed_coeffs[ch][end] = 0;
610 /* zero the dithered coefficients for appropriate channels */
615 * Stereo rematrixing.
616 * reference: Section 7.5.4 Rematrixing : Decoding Technique
618 static void do_rematrixing(AC3DecodeContext *s)
623 end = FFMIN(s->end_freq[1], s->end_freq[2]);
625 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++) {
626 if (s->rematrixing_flags[bnd]) {
627 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd + 1]);
628 for (i = ff_ac3_rematrix_band_tab[bnd]; i < bndend; i++) {
629 int tmp0 = s->fixed_coeffs[1][i];
630 s->fixed_coeffs[1][i] += s->fixed_coeffs[2][i];
631 s->fixed_coeffs[2][i] = tmp0 - s->fixed_coeffs[2][i];
638 * Inverse MDCT Transform.
639 * Convert frequency domain coefficients to time-domain audio samples.
640 * reference: Section 7.9.4 Transformation Equations
642 static inline void do_imdct(AC3DecodeContext *s, int channels)
646 for (ch = 1; ch <= channels; ch++) {
647 if (s->block_switch[ch]) {
649 float *x = s->tmp_output + 128;
650 for (i = 0; i < 128; i++)
651 x[i] = s->transform_coeffs[ch][2 * i];
652 s->imdct_256.imdct_half(&s->imdct_256, s->tmp_output, x);
653 s->fdsp.vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1],
654 s->tmp_output, s->window, 128);
655 for (i = 0; i < 128; i++)
656 x[i] = s->transform_coeffs[ch][2 * i + 1];
657 s->imdct_256.imdct_half(&s->imdct_256, s->delay[ch - 1], x);
659 s->imdct_512.imdct_half(&s->imdct_512, s->tmp_output, s->transform_coeffs[ch]);
660 s->fdsp.vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1],
661 s->tmp_output, s->window, 128);
662 memcpy(s->delay[ch - 1], s->tmp_output + 128, 128 * sizeof(float));
668 * Upmix delay samples from stereo to original channel layout.
670 static void ac3_upmix_delay(AC3DecodeContext *s)
672 int channel_data_size = sizeof(s->delay[0]);
673 switch (s->channel_mode) {
674 case AC3_CHMODE_DUALMONO:
675 case AC3_CHMODE_STEREO:
676 /* upmix mono to stereo */
677 memcpy(s->delay[1], s->delay[0], channel_data_size);
679 case AC3_CHMODE_2F2R:
680 memset(s->delay[3], 0, channel_data_size);
681 case AC3_CHMODE_2F1R:
682 memset(s->delay[2], 0, channel_data_size);
684 case AC3_CHMODE_3F2R:
685 memset(s->delay[4], 0, channel_data_size);
686 case AC3_CHMODE_3F1R:
687 memset(s->delay[3], 0, channel_data_size);
689 memcpy(s->delay[2], s->delay[1], channel_data_size);
690 memset(s->delay[1], 0, channel_data_size);
696 * Decode band structure for coupling, spectral extension, or enhanced coupling.
697 * The band structure defines how many subbands are in each band. For each
698 * subband in the range, 1 means it is combined with the previous band, and 0
699 * means that it starts a new band.
701 * @param[in] gbc bit reader context
702 * @param[in] blk block number
703 * @param[in] eac3 flag to indicate E-AC-3
704 * @param[in] ecpl flag to indicate enhanced coupling
705 * @param[in] start_subband subband number for start of range
706 * @param[in] end_subband subband number for end of range
707 * @param[in] default_band_struct default band structure table
708 * @param[out] num_bands number of bands (optionally NULL)
709 * @param[out] band_sizes array containing the number of bins in each band (optionally NULL)
711 static void decode_band_structure(GetBitContext *gbc, int blk, int eac3,
712 int ecpl, int start_subband, int end_subband,
713 const uint8_t *default_band_struct,
714 int *num_bands, uint8_t *band_sizes)
716 int subbnd, bnd, n_subbands, n_bands=0;
718 uint8_t coded_band_struct[22];
719 const uint8_t *band_struct;
721 n_subbands = end_subband - start_subband;
723 /* decode band structure from bitstream or use default */
724 if (!eac3 || get_bits1(gbc)) {
725 for (subbnd = 0; subbnd < n_subbands - 1; subbnd++) {
726 coded_band_struct[subbnd] = get_bits1(gbc);
728 band_struct = coded_band_struct;
730 band_struct = &default_band_struct[start_subband+1];
732 /* no change in band structure */
736 /* calculate number of bands and band sizes based on band structure.
737 note that the first 4 subbands in enhanced coupling span only 6 bins
739 if (num_bands || band_sizes ) {
740 n_bands = n_subbands;
741 bnd_sz[0] = ecpl ? 6 : 12;
742 for (bnd = 0, subbnd = 1; subbnd < n_subbands; subbnd++) {
743 int subbnd_size = (ecpl && subbnd < 4) ? 6 : 12;
744 if (band_struct[subbnd - 1]) {
746 bnd_sz[bnd] += subbnd_size;
748 bnd_sz[++bnd] = subbnd_size;
753 /* set optional output params */
755 *num_bands = n_bands;
757 memcpy(band_sizes, bnd_sz, n_bands);
761 * Decode a single audio block from the AC-3 bitstream.
763 static int decode_audio_block(AC3DecodeContext *s, int blk)
765 int fbw_channels = s->fbw_channels;
766 int channel_mode = s->channel_mode;
768 int different_transforms;
771 GetBitContext *gbc = &s->gbc;
772 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS] = { 0 };
774 /* block switch flags */
775 different_transforms = 0;
776 if (s->block_switch_syntax) {
777 for (ch = 1; ch <= fbw_channels; ch++) {
778 s->block_switch[ch] = get_bits1(gbc);
779 if (ch > 1 && s->block_switch[ch] != s->block_switch[1])
780 different_transforms = 1;
784 /* dithering flags */
785 if (s->dither_flag_syntax) {
786 for (ch = 1; ch <= fbw_channels; ch++) {
787 s->dither_flag[ch] = get_bits1(gbc);
792 i = !s->channel_mode;
794 if (get_bits1(gbc)) {
795 /* Allow asymmetric application of DRC when drc_scale > 1.
796 Amplification of quiet sounds is enhanced */
797 float range = dynamic_range_tab[get_bits(gbc, 8)];
798 if (range > 1.0 || s->drc_scale <= 1.0)
799 s->dynamic_range[i] = powf(range, s->drc_scale);
801 s->dynamic_range[i] = range;
802 } else if (blk == 0) {
803 s->dynamic_range[i] = 1.0f;
807 /* spectral extension strategy */
808 if (s->eac3 && (!blk || get_bits1(gbc))) {
809 s->spx_in_use = get_bits1(gbc);
811 int dst_start_freq, dst_end_freq, src_start_freq,
812 start_subband, end_subband;
814 /* determine which channels use spx */
815 if (s->channel_mode == AC3_CHMODE_MONO) {
816 s->channel_uses_spx[1] = 1;
818 for (ch = 1; ch <= fbw_channels; ch++)
819 s->channel_uses_spx[ch] = get_bits1(gbc);
822 /* get the frequency bins of the spx copy region and the spx start
824 dst_start_freq = get_bits(gbc, 2);
825 start_subband = get_bits(gbc, 3) + 2;
826 if (start_subband > 7)
827 start_subband += start_subband - 7;
828 end_subband = get_bits(gbc, 3) + 5;
830 end_subband += end_subband - 7;
831 dst_start_freq = dst_start_freq * 12 + 25;
832 src_start_freq = start_subband * 12 + 25;
833 dst_end_freq = end_subband * 12 + 25;
835 /* check validity of spx ranges */
836 if (start_subband >= end_subband) {
837 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
838 "range (%d >= %d)\n", start_subband, end_subband);
839 return AVERROR_INVALIDDATA;
841 if (dst_start_freq >= src_start_freq) {
842 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
843 "copy start bin (%d >= %d)\n", dst_start_freq, src_start_freq);
844 return AVERROR_INVALIDDATA;
847 s->spx_dst_start_freq = dst_start_freq;
848 s->spx_src_start_freq = src_start_freq;
849 s->spx_dst_end_freq = dst_end_freq;
851 decode_band_structure(gbc, blk, s->eac3, 0,
852 start_subband, end_subband,
853 ff_eac3_default_spx_band_struct,
857 for (ch = 1; ch <= fbw_channels; ch++) {
858 s->channel_uses_spx[ch] = 0;
859 s->first_spx_coords[ch] = 1;
864 /* spectral extension coordinates */
866 for (ch = 1; ch <= fbw_channels; ch++) {
867 if (s->channel_uses_spx[ch]) {
868 if (s->first_spx_coords[ch] || get_bits1(gbc)) {
870 int bin, master_spx_coord;
872 s->first_spx_coords[ch] = 0;
873 spx_blend = get_bits(gbc, 5) * (1.0f/32);
874 master_spx_coord = get_bits(gbc, 2) * 3;
876 bin = s->spx_src_start_freq;
877 for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
879 int spx_coord_exp, spx_coord_mant;
880 float nratio, sblend, nblend, spx_coord;
882 /* calculate blending factors */
883 bandsize = s->spx_band_sizes[bnd];
884 nratio = ((float)((bin + (bandsize >> 1))) / s->spx_dst_end_freq) - spx_blend;
885 nratio = av_clipf(nratio, 0.0f, 1.0f);
886 nblend = sqrtf(3.0f * nratio); // noise is scaled by sqrt(3)
887 // to give unity variance
888 sblend = sqrtf(1.0f - nratio);
891 /* decode spx coordinates */
892 spx_coord_exp = get_bits(gbc, 4);
893 spx_coord_mant = get_bits(gbc, 2);
894 if (spx_coord_exp == 15) spx_coord_mant <<= 1;
895 else spx_coord_mant += 4;
896 spx_coord_mant <<= (25 - spx_coord_exp - master_spx_coord);
897 spx_coord = spx_coord_mant * (1.0f / (1 << 23));
899 /* multiply noise and signal blending factors by spx coordinate */
900 s->spx_noise_blend [ch][bnd] = nblend * spx_coord;
901 s->spx_signal_blend[ch][bnd] = sblend * spx_coord;
905 s->first_spx_coords[ch] = 1;
910 /* coupling strategy */
911 if (s->eac3 ? s->cpl_strategy_exists[blk] : get_bits1(gbc)) {
912 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
914 s->cpl_in_use[blk] = get_bits1(gbc);
915 if (s->cpl_in_use[blk]) {
916 /* coupling in use */
917 int cpl_start_subband, cpl_end_subband;
919 if (channel_mode < AC3_CHMODE_STEREO) {
920 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
921 return AVERROR_INVALIDDATA;
924 /* check for enhanced coupling */
925 if (s->eac3 && get_bits1(gbc)) {
926 /* TODO: parse enhanced coupling strategy info */
927 avpriv_request_sample(s->avctx, "Enhanced coupling");
928 return AVERROR_PATCHWELCOME;
931 /* determine which channels are coupled */
932 if (s->eac3 && s->channel_mode == AC3_CHMODE_STEREO) {
933 s->channel_in_cpl[1] = 1;
934 s->channel_in_cpl[2] = 1;
936 for (ch = 1; ch <= fbw_channels; ch++)
937 s->channel_in_cpl[ch] = get_bits1(gbc);
940 /* phase flags in use */
941 if (channel_mode == AC3_CHMODE_STEREO)
942 s->phase_flags_in_use = get_bits1(gbc);
944 /* coupling frequency range */
945 cpl_start_subband = get_bits(gbc, 4);
946 cpl_end_subband = s->spx_in_use ? (s->spx_src_start_freq - 37) / 12 :
947 get_bits(gbc, 4) + 3;
948 if (cpl_start_subband >= cpl_end_subband) {
949 av_log(s->avctx, AV_LOG_ERROR, "invalid coupling range (%d >= %d)\n",
950 cpl_start_subband, cpl_end_subband);
951 return AVERROR_INVALIDDATA;
953 s->start_freq[CPL_CH] = cpl_start_subband * 12 + 37;
954 s->end_freq[CPL_CH] = cpl_end_subband * 12 + 37;
956 decode_band_structure(gbc, blk, s->eac3, 0, cpl_start_subband,
958 ff_eac3_default_cpl_band_struct,
959 &s->num_cpl_bands, s->cpl_band_sizes);
961 /* coupling not in use */
962 for (ch = 1; ch <= fbw_channels; ch++) {
963 s->channel_in_cpl[ch] = 0;
964 s->first_cpl_coords[ch] = 1;
966 s->first_cpl_leak = s->eac3;
967 s->phase_flags_in_use = 0;
969 } else if (!s->eac3) {
971 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must "
972 "be present in block 0\n");
973 return AVERROR_INVALIDDATA;
975 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
978 cpl_in_use = s->cpl_in_use[blk];
980 /* coupling coordinates */
982 int cpl_coords_exist = 0;
984 for (ch = 1; ch <= fbw_channels; ch++) {
985 if (s->channel_in_cpl[ch]) {
986 if ((s->eac3 && s->first_cpl_coords[ch]) || get_bits1(gbc)) {
987 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
988 s->first_cpl_coords[ch] = 0;
989 cpl_coords_exist = 1;
990 master_cpl_coord = 3 * get_bits(gbc, 2);
991 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
992 cpl_coord_exp = get_bits(gbc, 4);
993 cpl_coord_mant = get_bits(gbc, 4);
994 if (cpl_coord_exp == 15)
995 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
997 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
998 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
1001 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must "
1002 "be present in block 0\n");
1003 return AVERROR_INVALIDDATA;
1006 /* channel not in coupling */
1007 s->first_cpl_coords[ch] = 1;
1011 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
1012 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
1013 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
1018 /* stereo rematrixing strategy and band structure */
1019 if (channel_mode == AC3_CHMODE_STEREO) {
1020 if ((s->eac3 && !blk) || get_bits1(gbc)) {
1021 s->num_rematrixing_bands = 4;
1022 if (cpl_in_use && s->start_freq[CPL_CH] <= 61) {
1023 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
1024 } else if (s->spx_in_use && s->spx_src_start_freq <= 61) {
1025 s->num_rematrixing_bands--;
1027 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++)
1028 s->rematrixing_flags[bnd] = get_bits1(gbc);
1030 av_log(s->avctx, AV_LOG_WARNING, "Warning: "
1031 "new rematrixing strategy not present in block 0\n");
1032 s->num_rematrixing_bands = 0;
1036 /* exponent strategies for each channel */
1037 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1039 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
1040 if (s->exp_strategy[blk][ch] != EXP_REUSE)
1041 bit_alloc_stages[ch] = 3;
1044 /* channel bandwidth */
1045 for (ch = 1; ch <= fbw_channels; ch++) {
1046 s->start_freq[ch] = 0;
1047 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1049 int prev = s->end_freq[ch];
1050 if (s->channel_in_cpl[ch])
1051 s->end_freq[ch] = s->start_freq[CPL_CH];
1052 else if (s->channel_uses_spx[ch])
1053 s->end_freq[ch] = s->spx_src_start_freq;
1055 int bandwidth_code = get_bits(gbc, 6);
1056 if (bandwidth_code > 60) {
1057 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60\n", bandwidth_code);
1058 return AVERROR_INVALIDDATA;
1060 s->end_freq[ch] = bandwidth_code * 3 + 73;
1062 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
1063 s->num_exp_groups[ch] = (s->end_freq[ch] + group_size-4) / group_size;
1064 if (blk > 0 && s->end_freq[ch] != prev)
1065 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
1068 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
1069 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
1070 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
1073 /* decode exponents for each channel */
1074 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1075 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1076 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
1077 if (decode_exponents(gbc, s->exp_strategy[blk][ch],
1078 s->num_exp_groups[ch], s->dexps[ch][0],
1079 &s->dexps[ch][s->start_freq[ch]+!!ch])) {
1080 av_log(s->avctx, AV_LOG_ERROR, "exponent out-of-range\n");
1081 return AVERROR_INVALIDDATA;
1083 if (ch != CPL_CH && ch != s->lfe_ch)
1084 skip_bits(gbc, 2); /* skip gainrng */
1088 /* bit allocation information */
1089 if (s->bit_allocation_syntax) {
1090 if (get_bits1(gbc)) {
1091 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1092 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1093 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
1094 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
1095 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
1096 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1097 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1099 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must "
1100 "be present in block 0\n");
1101 return AVERROR_INVALIDDATA;
1105 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
1106 if (!s->eac3 || !blk) {
1107 if (s->snr_offset_strategy && get_bits1(gbc)) {
1110 csnr = (get_bits(gbc, 6) - 15) << 4;
1111 for (i = ch = !cpl_in_use; ch <= s->channels; ch++) {
1113 if (ch == i || s->snr_offset_strategy == 2)
1114 snr = (csnr + get_bits(gbc, 4)) << 2;
1115 /* run at least last bit allocation stage if snr offset changes */
1116 if (blk && s->snr_offset[ch] != snr) {
1117 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 1);
1119 s->snr_offset[ch] = snr;
1121 /* fast gain (normal AC-3 only) */
1123 int prev = s->fast_gain[ch];
1124 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1125 /* run last 2 bit allocation stages if fast gain changes */
1126 if (blk && prev != s->fast_gain[ch])
1127 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1130 } else if (!s->eac3 && !blk) {
1131 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
1132 return AVERROR_INVALIDDATA;
1136 /* fast gain (E-AC-3 only) */
1137 if (s->fast_gain_syntax && get_bits1(gbc)) {
1138 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1139 int prev = s->fast_gain[ch];
1140 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1141 /* run last 2 bit allocation stages if fast gain changes */
1142 if (blk && prev != s->fast_gain[ch])
1143 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1145 } else if (s->eac3 && !blk) {
1146 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1147 s->fast_gain[ch] = ff_ac3_fast_gain_tab[4];
1150 /* E-AC-3 to AC-3 converter SNR offset */
1151 if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT && get_bits1(gbc)) {
1152 skip_bits(gbc, 10); // skip converter snr offset
1155 /* coupling leak information */
1157 if (s->first_cpl_leak || get_bits1(gbc)) {
1158 int fl = get_bits(gbc, 3);
1159 int sl = get_bits(gbc, 3);
1160 /* run last 2 bit allocation stages for coupling channel if
1161 coupling leak changes */
1162 if (blk && (fl != s->bit_alloc_params.cpl_fast_leak ||
1163 sl != s->bit_alloc_params.cpl_slow_leak)) {
1164 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
1166 s->bit_alloc_params.cpl_fast_leak = fl;
1167 s->bit_alloc_params.cpl_slow_leak = sl;
1168 } else if (!s->eac3 && !blk) {
1169 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must "
1170 "be present in block 0\n");
1171 return AVERROR_INVALIDDATA;
1173 s->first_cpl_leak = 0;
1176 /* delta bit allocation information */
1177 if (s->dba_syntax && get_bits1(gbc)) {
1178 /* delta bit allocation exists (strategy) */
1179 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1180 s->dba_mode[ch] = get_bits(gbc, 2);
1181 if (s->dba_mode[ch] == DBA_RESERVED) {
1182 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1183 return AVERROR_INVALIDDATA;
1185 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1187 /* channel delta offset, len and bit allocation */
1188 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1189 if (s->dba_mode[ch] == DBA_NEW) {
1190 s->dba_nsegs[ch] = get_bits(gbc, 3) + 1;
1191 for (seg = 0; seg < s->dba_nsegs[ch]; seg++) {
1192 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1193 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1194 s->dba_values[ch][seg] = get_bits(gbc, 3);
1196 /* run last 2 bit allocation stages if new dba values */
1197 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1200 } else if (blk == 0) {
1201 for (ch = 0; ch <= s->channels; ch++) {
1202 s->dba_mode[ch] = DBA_NONE;
1206 /* Bit allocation */
1207 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1208 if (bit_alloc_stages[ch] > 2) {
1209 /* Exponent mapping into PSD and PSD integration */
1210 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1211 s->start_freq[ch], s->end_freq[ch],
1212 s->psd[ch], s->band_psd[ch]);
1214 if (bit_alloc_stages[ch] > 1) {
1215 /* Compute excitation function, Compute masking curve, and
1216 Apply delta bit allocation */
1217 if (ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1218 s->start_freq[ch], s->end_freq[ch],
1219 s->fast_gain[ch], (ch == s->lfe_ch),
1220 s->dba_mode[ch], s->dba_nsegs[ch],
1221 s->dba_offsets[ch], s->dba_lengths[ch],
1222 s->dba_values[ch], s->mask[ch])) {
1223 av_log(s->avctx, AV_LOG_ERROR, "error in bit allocation\n");
1224 return AVERROR_INVALIDDATA;
1227 if (bit_alloc_stages[ch] > 0) {
1228 /* Compute bit allocation */
1229 const uint8_t *bap_tab = s->channel_uses_aht[ch] ?
1230 ff_eac3_hebap_tab : ff_ac3_bap_tab;
1231 s->ac3dsp.bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1232 s->start_freq[ch], s->end_freq[ch],
1234 s->bit_alloc_params.floor,
1235 bap_tab, s->bap[ch]);
1239 /* unused dummy data */
1240 if (s->skip_syntax && get_bits1(gbc)) {
1241 int skipl = get_bits(gbc, 9);
1246 /* unpack the transform coefficients
1247 this also uncouples channels if coupling is in use. */
1248 decode_transform_coeffs(s, blk);
1250 /* TODO: generate enhanced coupling coordinates and uncouple */
1252 /* recover coefficients if rematrixing is in use */
1253 if (s->channel_mode == AC3_CHMODE_STEREO)
1256 /* apply scaling to coefficients (headroom, dynrng) */
1257 for (ch = 1; ch <= s->channels; ch++) {
1258 float gain = 1.0 / 4194304.0f;
1259 if (s->channel_mode == AC3_CHMODE_DUALMONO) {
1260 gain *= s->dynamic_range[2 - ch];
1262 gain *= s->dynamic_range[0];
1264 s->fmt_conv.int32_to_float_fmul_scalar(s->transform_coeffs[ch],
1265 s->fixed_coeffs[ch], gain, 256);
1268 /* apply spectral extension to high frequency bins */
1269 if (s->spx_in_use && CONFIG_EAC3_DECODER) {
1270 ff_eac3_apply_spectral_extension(s);
1273 /* downmix and MDCT. order depends on whether block switching is used for
1274 any channel in this block. this is because coefficients for the long
1275 and short transforms cannot be mixed. */
1276 downmix_output = s->channels != s->out_channels &&
1277 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1278 s->fbw_channels == s->out_channels);
1279 if (different_transforms) {
1280 /* the delay samples have already been downmixed, so we upmix the delay
1281 samples in order to reconstruct all channels before downmixing. */
1287 do_imdct(s, s->channels);
1289 if (downmix_output) {
1290 s->ac3dsp.downmix(s->outptr, s->downmix_coeffs,
1291 s->out_channels, s->fbw_channels, 256);
1294 if (downmix_output) {
1295 s->ac3dsp.downmix(s->xcfptr + 1, s->downmix_coeffs,
1296 s->out_channels, s->fbw_channels, 256);
1299 if (downmix_output && !s->downmixed) {
1301 s->ac3dsp.downmix(s->dlyptr, s->downmix_coeffs, s->out_channels,
1302 s->fbw_channels, 128);
1305 do_imdct(s, s->out_channels);
1312 * Decode a single AC-3 frame.
1314 static int ac3_decode_frame(AVCodecContext * avctx, void *data,
1315 int *got_frame_ptr, AVPacket *avpkt)
1317 AVFrame *frame = data;
1318 const uint8_t *buf = avpkt->data;
1319 int buf_size = avpkt->size;
1320 AC3DecodeContext *s = avctx->priv_data;
1321 int blk, ch, err, ret;
1322 const uint8_t *channel_map;
1323 const float *output[AC3_MAX_CHANNELS];
1324 enum AVMatrixEncoding matrix_encoding;
1325 AVDownmixInfo *downmix_info;
1327 /* copy input buffer to decoder context to avoid reading past the end
1328 of the buffer, which can be caused by a damaged input stream. */
1329 if (buf_size >= 2 && AV_RB16(buf) == 0x770B) {
1330 // seems to be byte-swapped AC-3
1331 int cnt = FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE) >> 1;
1332 s->dsp.bswap16_buf((uint16_t *)s->input_buffer, (const uint16_t *)buf, cnt);
1334 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE));
1335 buf = s->input_buffer;
1336 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1337 init_get_bits(&s->gbc, buf, buf_size * 8);
1339 /* parse the syncinfo */
1340 err = parse_frame_header(s);
1344 case AAC_AC3_PARSE_ERROR_SYNC:
1345 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1346 return AVERROR_INVALIDDATA;
1347 case AAC_AC3_PARSE_ERROR_BSID:
1348 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1350 case AAC_AC3_PARSE_ERROR_SAMPLE_RATE:
1351 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1353 case AAC_AC3_PARSE_ERROR_FRAME_SIZE:
1354 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1356 case AAC_AC3_PARSE_ERROR_FRAME_TYPE:
1357 /* skip frame if CRC is ok. otherwise use error concealment. */
1358 /* TODO: add support for substreams and dependent frames */
1359 if (s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1360 av_log(avctx, AV_LOG_WARNING, "unsupported frame type : "
1361 "skipping frame\n");
1365 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1368 case AAC_AC3_PARSE_ERROR_CRC:
1369 case AAC_AC3_PARSE_ERROR_CHANNEL_CFG:
1371 default: // Normal AVERROR do not try to recover.
1376 /* check that reported frame size fits in input buffer */
1377 if (s->frame_size > buf_size) {
1378 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1379 err = AAC_AC3_PARSE_ERROR_FRAME_SIZE;
1380 } else if (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL)) {
1381 /* check for crc mismatch */
1382 if (av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2],
1383 s->frame_size - 2)) {
1384 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1385 if (avctx->err_recognition & AV_EF_EXPLODE)
1386 return AVERROR_INVALIDDATA;
1387 err = AAC_AC3_PARSE_ERROR_CRC;
1392 /* if frame is ok, set audio parameters */
1394 avctx->sample_rate = s->sample_rate;
1395 avctx->bit_rate = s->bit_rate;
1398 /* channel config */
1399 if (!err || (s->channels && s->out_channels != s->channels)) {
1400 s->out_channels = s->channels;
1401 s->output_mode = s->channel_mode;
1403 s->output_mode |= AC3_OUTPUT_LFEON;
1404 if (s->channels > 1 &&
1405 avctx->request_channel_layout == AV_CH_LAYOUT_MONO) {
1406 s->out_channels = 1;
1407 s->output_mode = AC3_CHMODE_MONO;
1408 } else if (s->channels > 2 &&
1409 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
1410 s->out_channels = 2;
1411 s->output_mode = AC3_CHMODE_STEREO;
1414 s->loro_center_mix_level = gain_levels[s-> center_mix_level];
1415 s->loro_surround_mix_level = gain_levels[s->surround_mix_level];
1416 s->ltrt_center_mix_level = LEVEL_MINUS_3DB;
1417 s->ltrt_surround_mix_level = LEVEL_MINUS_3DB;
1418 /* set downmixing coefficients if needed */
1419 if (s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1420 s->fbw_channels == s->out_channels)) {
1421 set_downmix_coeffs(s);
1423 } else if (!s->channels) {
1424 av_log(avctx, AV_LOG_ERROR, "unable to determine channel mode\n");
1425 return AVERROR_INVALIDDATA;
1427 avctx->channels = s->out_channels;
1428 avctx->channel_layout = avpriv_ac3_channel_layout_tab[s->output_mode & ~AC3_OUTPUT_LFEON];
1429 if (s->output_mode & AC3_OUTPUT_LFEON)
1430 avctx->channel_layout |= AV_CH_LOW_FREQUENCY;
1432 /* set audio service type based on bitstream mode for AC-3 */
1433 avctx->audio_service_type = s->bitstream_mode;
1434 if (s->bitstream_mode == 0x7 && s->channels > 1)
1435 avctx->audio_service_type = AV_AUDIO_SERVICE_TYPE_KARAOKE;
1437 /* get output buffer */
1438 frame->nb_samples = s->num_blocks * AC3_BLOCK_SIZE;
1439 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1442 /* decode the audio blocks */
1443 channel_map = ff_ac3_dec_channel_map[s->output_mode & ~AC3_OUTPUT_LFEON][s->lfe_on];
1444 for (ch = 0; ch < AC3_MAX_CHANNELS; ch++) {
1445 output[ch] = s->output[ch];
1446 s->outptr[ch] = s->output[ch];
1448 for (ch = 0; ch < s->channels; ch++) {
1449 if (ch < s->out_channels)
1450 s->outptr[channel_map[ch]] = (float *)frame->data[ch];
1452 for (blk = 0; blk < s->num_blocks; blk++) {
1453 if (!err && decode_audio_block(s, blk)) {
1454 av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n");
1458 for (ch = 0; ch < s->out_channels; ch++)
1459 memcpy(((float*)frame->data[ch]) + AC3_BLOCK_SIZE*blk, output[ch], sizeof(**output) * AC3_BLOCK_SIZE);
1460 for (ch = 0; ch < s->out_channels; ch++)
1461 output[ch] = s->outptr[channel_map[ch]];
1462 for (ch = 0; ch < s->out_channels; ch++) {
1463 if (!ch || channel_map[ch])
1464 s->outptr[channel_map[ch]] += AC3_BLOCK_SIZE;
1468 av_frame_set_decode_error_flags(frame, err ? FF_DECODE_ERROR_INVALID_BITSTREAM : 0);
1470 /* keep last block for error concealment in next frame */
1471 for (ch = 0; ch < s->out_channels; ch++)
1472 memcpy(s->output[ch], output[ch], sizeof(**output) * AC3_BLOCK_SIZE);
1477 * Check whether the input layout is compatible, and make sure we're not
1478 * downmixing (else the matrix encoding is no longer applicable).
1480 matrix_encoding = AV_MATRIX_ENCODING_NONE;
1481 if (s->channel_mode == AC3_CHMODE_STEREO &&
1482 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1483 if (s->dolby_surround_mode == AC3_DSURMOD_ON)
1484 matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
1485 else if (s->dolby_headphone_mode == AC3_DHEADPHONMOD_ON)
1486 matrix_encoding = AV_MATRIX_ENCODING_DOLBYHEADPHONE;
1487 } else if (s->channel_mode >= AC3_CHMODE_2F2R &&
1488 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1489 switch (s->dolby_surround_ex_mode) {
1490 case AC3_DSUREXMOD_ON: // EX or PLIIx
1491 matrix_encoding = AV_MATRIX_ENCODING_DOLBYEX;
1493 case AC3_DSUREXMOD_PLIIZ:
1494 matrix_encoding = AV_MATRIX_ENCODING_DPLIIZ;
1496 default: // not indicated or off
1500 if ((ret = ff_side_data_update_matrix_encoding(frame, matrix_encoding)) < 0)
1504 if ((downmix_info = av_downmix_info_update_side_data(frame))) {
1505 switch (s->preferred_downmix) {
1506 case AC3_DMIXMOD_LTRT:
1507 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LTRT;
1509 case AC3_DMIXMOD_LORO:
1510 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LORO;
1512 case AC3_DMIXMOD_DPLII:
1513 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_DPLII;
1516 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_UNKNOWN;
1519 downmix_info->center_mix_level = gain_levels[s-> center_mix_level];
1520 downmix_info->center_mix_level_ltrt = gain_levels[s-> center_mix_level_ltrt];
1521 downmix_info->surround_mix_level = gain_levels[s-> surround_mix_level];
1522 downmix_info->surround_mix_level_ltrt = gain_levels[s->surround_mix_level_ltrt];
1523 if (s->lfe_mix_level_exists)
1524 downmix_info->lfe_mix_level = gain_levels_lfe[s->lfe_mix_level];
1526 downmix_info->lfe_mix_level = 0.0; // -inf dB
1528 return AVERROR(ENOMEM);
1532 return FFMIN(buf_size, s->frame_size);
1536 * Uninitialize the AC-3 decoder.
1538 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1540 AC3DecodeContext *s = avctx->priv_data;
1541 ff_mdct_end(&s->imdct_512);
1542 ff_mdct_end(&s->imdct_256);
1547 #define OFFSET(x) offsetof(AC3DecodeContext, x)
1548 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)
1549 static const AVOption options[] = {
1550 { "drc_scale", "percentage of dynamic range compression to apply", OFFSET(drc_scale), AV_OPT_TYPE_FLOAT, {.dbl = 1.0}, 0.0, 6.0, PAR },
1552 {"dmix_mode", "Preferred Stereo Downmix Mode", OFFSET(preferred_stereo_downmix), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, 2, 0, "dmix_mode"},
1553 {"ltrt_cmixlev", "Lt/Rt Center Mix Level", OFFSET(ltrt_center_mix_level), AV_OPT_TYPE_FLOAT, {.dbl = -1.0 }, -1.0, 2.0, 0},
1554 {"ltrt_surmixlev", "Lt/Rt Surround Mix Level", OFFSET(ltrt_surround_mix_level), AV_OPT_TYPE_FLOAT, {.dbl = -1.0 }, -1.0, 2.0, 0},
1555 {"loro_cmixlev", "Lo/Ro Center Mix Level", OFFSET(loro_center_mix_level), AV_OPT_TYPE_FLOAT, {.dbl = -1.0 }, -1.0, 2.0, 0},
1556 {"loro_surmixlev", "Lo/Ro Surround Mix Level", OFFSET(loro_surround_mix_level), AV_OPT_TYPE_FLOAT, {.dbl = -1.0 }, -1.0, 2.0, 0},
1561 static const AVClass ac3_decoder_class = {
1562 .class_name = "AC3 decoder",
1563 .item_name = av_default_item_name,
1565 .version = LIBAVUTIL_VERSION_INT,
1568 AVCodec ff_ac3_decoder = {
1570 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52A (AC-3)"),
1571 .type = AVMEDIA_TYPE_AUDIO,
1572 .id = AV_CODEC_ID_AC3,
1573 .priv_data_size = sizeof (AC3DecodeContext),
1574 .init = ac3_decode_init,
1575 .close = ac3_decode_end,
1576 .decode = ac3_decode_frame,
1577 .capabilities = CODEC_CAP_DR1,
1578 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1579 AV_SAMPLE_FMT_NONE },
1580 .priv_class = &ac3_decoder_class,
1583 #if CONFIG_EAC3_DECODER
1584 static const AVClass eac3_decoder_class = {
1585 .class_name = "E-AC3 decoder",
1586 .item_name = av_default_item_name,
1588 .version = LIBAVUTIL_VERSION_INT,
1591 AVCodec ff_eac3_decoder = {
1593 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52B (AC-3, E-AC-3)"),
1594 .type = AVMEDIA_TYPE_AUDIO,
1595 .id = AV_CODEC_ID_EAC3,
1596 .priv_data_size = sizeof (AC3DecodeContext),
1597 .init = ac3_decode_init,
1598 .close = ac3_decode_end,
1599 .decode = ac3_decode_frame,
1600 .capabilities = CODEC_CAP_DR1,
1601 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1602 AV_SAMPLE_FMT_NONE },
1603 .priv_class = &eac3_decoder_class,