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 Libav.
12 * Libav 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 * Libav 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 Libav; 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"
38 #include "aac_ac3_parser.h"
39 #include "ac3_parser.h"
41 #include "ac3dec_data.h"
45 * table for ungrouping 3 values in 7 bits.
46 * used for exponents and bap=2 mantissas
48 static uint8_t ungroup_3_in_7_bits_tab[128][3];
50 /** tables for ungrouping mantissas */
51 static int b1_mantissas[32][3];
52 static int b2_mantissas[128][3];
53 static int b3_mantissas[8];
54 static int b4_mantissas[128][2];
55 static int b5_mantissas[16];
58 * Quantization table: levels for symmetric. bits for asymmetric.
59 * reference: Table 7.18 Mapping of bap to Quantizer
61 static const uint8_t quantization_tab[16] = {
63 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
66 /** dynamic range table. converts codes to scale factors. */
67 static float dynamic_range_tab[256];
69 /** Adjustments in dB gain */
70 static const float gain_levels[9] = {
74 LEVEL_MINUS_1POINT5DB,
76 LEVEL_MINUS_4POINT5DB,
82 /** Adjustments in dB gain (LFE, +10 to -21 dB) */
83 static const float gain_levels_lfe[32] = {
84 3.162275, 2.818382, 2.511886, 2.238719, 1.995261, 1.778278, 1.584893,
85 1.412536, 1.258924, 1.122018, 1.000000, 0.891251, 0.794328, 0.707946,
86 0.630957, 0.562341, 0.501187, 0.446683, 0.398107, 0.354813, 0.316227,
87 0.281838, 0.251188, 0.223872, 0.199526, 0.177828, 0.158489, 0.141253,
88 0.125892, 0.112201, 0.100000, 0.089125
92 * Table for default stereo downmixing coefficients
93 * reference: Section 7.8.2 Downmixing Into Two Channels
95 static const uint8_t ac3_default_coeffs[8][5][2] = {
96 { { 2, 7 }, { 7, 2 }, },
98 { { 2, 7 }, { 7, 2 }, },
99 { { 2, 7 }, { 5, 5 }, { 7, 2 }, },
100 { { 2, 7 }, { 7, 2 }, { 6, 6 }, },
101 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 8, 8 }, },
102 { { 2, 7 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
103 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
107 * Symmetrical Dequantization
108 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
109 * Tables 7.19 to 7.23
112 symmetric_dequant(int code, int levels)
114 return ((code - (levels >> 1)) << 24) / levels;
118 * Initialize tables at runtime.
120 static av_cold void ac3_tables_init(void)
124 /* generate table for ungrouping 3 values in 7 bits
125 reference: Section 7.1.3 Exponent Decoding */
126 for (i = 0; i < 128; i++) {
127 ungroup_3_in_7_bits_tab[i][0] = i / 25;
128 ungroup_3_in_7_bits_tab[i][1] = (i % 25) / 5;
129 ungroup_3_in_7_bits_tab[i][2] = (i % 25) % 5;
132 /* generate grouped mantissa tables
133 reference: Section 7.3.5 Ungrouping of Mantissas */
134 for (i = 0; i < 32; i++) {
135 /* bap=1 mantissas */
136 b1_mantissas[i][0] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][0], 3);
137 b1_mantissas[i][1] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][1], 3);
138 b1_mantissas[i][2] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][2], 3);
140 for (i = 0; i < 128; i++) {
141 /* bap=2 mantissas */
142 b2_mantissas[i][0] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][0], 5);
143 b2_mantissas[i][1] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][1], 5);
144 b2_mantissas[i][2] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][2], 5);
146 /* bap=4 mantissas */
147 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
148 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
150 /* generate ungrouped mantissa tables
151 reference: Tables 7.21 and 7.23 */
152 for (i = 0; i < 7; i++) {
153 /* bap=3 mantissas */
154 b3_mantissas[i] = symmetric_dequant(i, 7);
156 for (i = 0; i < 15; i++) {
157 /* bap=5 mantissas */
158 b5_mantissas[i] = symmetric_dequant(i, 15);
161 /* generate dynamic range table
162 reference: Section 7.7.1 Dynamic Range Control */
163 for (i = 0; i < 256; i++) {
164 int v = (i >> 5) - ((i >> 7) << 3) - 5;
165 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
170 * AVCodec initialization
172 static av_cold int ac3_decode_init(AVCodecContext *avctx)
174 AC3DecodeContext *s = avctx->priv_data;
179 ff_ac3_common_init();
181 ff_mdct_init(&s->imdct_256, 8, 1, 1.0);
182 ff_mdct_init(&s->imdct_512, 9, 1, 1.0);
183 ff_kbd_window_init(s->window, 5.0, 256);
184 ff_bswapdsp_init(&s->bdsp);
185 avpriv_float_dsp_init(&s->fdsp, avctx->flags & AV_CODEC_FLAG_BITEXACT);
186 ff_ac3dsp_init(&s->ac3dsp, avctx->flags & AV_CODEC_FLAG_BITEXACT);
187 ff_fmt_convert_init(&s->fmt_conv, avctx);
188 av_lfg_init(&s->dith_state, 0);
190 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
192 /* allow downmixing to stereo or mono */
193 if (avctx->channels > 1 &&
194 avctx->request_channel_layout == AV_CH_LAYOUT_MONO)
196 else if (avctx->channels > 2 &&
197 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO)
201 for (i = 0; i < AC3_MAX_CHANNELS; i++) {
202 s->xcfptr[i] = s->transform_coeffs[i];
203 s->dlyptr[i] = s->delay[i];
210 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
211 * GetBitContext within AC3DecodeContext must point to
212 * the start of the synchronized AC-3 bitstream.
214 static int ac3_parse_header(AC3DecodeContext *s)
216 GetBitContext *gbc = &s->gbc;
219 /* read the rest of the bsi. read twice for dual mono mode. */
220 i = !s->channel_mode;
222 skip_bits(gbc, 5); // skip dialog normalization
224 skip_bits(gbc, 8); //skip compression
226 skip_bits(gbc, 8); //skip language code
228 skip_bits(gbc, 7); //skip audio production information
231 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
233 /* skip the timecodes or parse the Alternate Bit Stream Syntax */
234 if (s->bitstream_id != 6) {
236 skip_bits(gbc, 14); //skip timecode1
238 skip_bits(gbc, 14); //skip timecode2
240 if (get_bits1(gbc)) {
241 s->preferred_downmix = get_bits(gbc, 2);
242 s->center_mix_level_ltrt = get_bits(gbc, 3);
243 s->surround_mix_level_ltrt = av_clip(get_bits(gbc, 3), 3, 7);
244 s->center_mix_level = get_bits(gbc, 3);
245 s->surround_mix_level = av_clip(get_bits(gbc, 3), 3, 7);
247 if (get_bits1(gbc)) {
248 s->dolby_surround_ex_mode = get_bits(gbc, 2);
249 s->dolby_headphone_mode = get_bits(gbc, 2);
250 skip_bits(gbc, 10); // skip adconvtyp (1), xbsi2 (8), encinfo (1)
254 /* skip additional bitstream info */
255 if (get_bits1(gbc)) {
256 i = get_bits(gbc, 6);
266 * Common function to parse AC-3 or E-AC-3 frame header
268 static int parse_frame_header(AC3DecodeContext *s)
273 err = avpriv_ac3_parse_header(&s->gbc, &hdr);
277 /* get decoding parameters from header info */
278 s->bit_alloc_params.sr_code = hdr.sr_code;
279 s->bitstream_id = hdr.bitstream_id;
280 s->bitstream_mode = hdr.bitstream_mode;
281 s->channel_mode = hdr.channel_mode;
282 s->lfe_on = hdr.lfe_on;
283 s->bit_alloc_params.sr_shift = hdr.sr_shift;
284 s->sample_rate = hdr.sample_rate;
285 s->bit_rate = hdr.bit_rate;
286 s->channels = hdr.channels;
287 s->fbw_channels = s->channels - s->lfe_on;
288 s->lfe_ch = s->fbw_channels + 1;
289 s->frame_size = hdr.frame_size;
290 s->preferred_downmix = AC3_DMIXMOD_NOTINDICATED;
291 s->center_mix_level = hdr.center_mix_level;
292 s->center_mix_level_ltrt = 4; // -3.0dB
293 s->surround_mix_level = hdr.surround_mix_level;
294 s->surround_mix_level_ltrt = 4; // -3.0dB
295 s->lfe_mix_level_exists = 0;
296 s->num_blocks = hdr.num_blocks;
297 s->frame_type = hdr.frame_type;
298 s->substreamid = hdr.substreamid;
299 s->dolby_surround_mode = hdr.dolby_surround_mode;
300 s->dolby_surround_ex_mode = AC3_DSUREXMOD_NOTINDICATED;
301 s->dolby_headphone_mode = AC3_DHEADPHONMOD_NOTINDICATED;
304 s->start_freq[s->lfe_ch] = 0;
305 s->end_freq[s->lfe_ch] = 7;
306 s->num_exp_groups[s->lfe_ch] = 2;
307 s->channel_in_cpl[s->lfe_ch] = 0;
310 if (s->bitstream_id <= 10) {
312 s->snr_offset_strategy = 2;
313 s->block_switch_syntax = 1;
314 s->dither_flag_syntax = 1;
315 s->bit_allocation_syntax = 1;
316 s->fast_gain_syntax = 0;
317 s->first_cpl_leak = 0;
320 memset(s->channel_uses_aht, 0, sizeof(s->channel_uses_aht));
321 return ac3_parse_header(s);
322 } else if (CONFIG_EAC3_DECODER) {
324 return ff_eac3_parse_header(s);
326 av_log(s->avctx, AV_LOG_ERROR, "E-AC-3 support not compiled in\n");
327 return AVERROR(ENOSYS);
332 * Set stereo downmixing coefficients based on frame header info.
333 * reference: Section 7.8.2 Downmixing Into Two Channels
335 static void set_downmix_coeffs(AC3DecodeContext *s)
338 float cmix = gain_levels[s-> center_mix_level];
339 float smix = gain_levels[s->surround_mix_level];
342 for (i = 0; i < s->fbw_channels; i++) {
343 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
344 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
346 if (s->channel_mode > 1 && s->channel_mode & 1) {
347 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
349 if (s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
350 int nf = s->channel_mode - 2;
351 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
353 if (s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
354 int nf = s->channel_mode - 4;
355 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
360 for (i = 0; i < s->fbw_channels; i++) {
361 norm0 += s->downmix_coeffs[i][0];
362 norm1 += s->downmix_coeffs[i][1];
364 norm0 = 1.0f / norm0;
365 norm1 = 1.0f / norm1;
366 for (i = 0; i < s->fbw_channels; i++) {
367 s->downmix_coeffs[i][0] *= norm0;
368 s->downmix_coeffs[i][1] *= norm1;
371 if (s->output_mode == AC3_CHMODE_MONO) {
372 for (i = 0; i < s->fbw_channels; i++)
373 s->downmix_coeffs[i][0] = (s->downmix_coeffs[i][0] +
374 s->downmix_coeffs[i][1]) * LEVEL_MINUS_3DB;
379 * Decode the grouped exponents according to exponent strategy.
380 * reference: Section 7.1.3 Exponent Decoding
382 static int decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
383 uint8_t absexp, int8_t *dexps)
385 int i, j, grp, group_size;
390 group_size = exp_strategy + (exp_strategy == EXP_D45);
391 for (grp = 0, i = 0; grp < ngrps; grp++) {
392 expacc = get_bits(gbc, 7);
393 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][0];
394 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][1];
395 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][2];
398 /* convert to absolute exps and expand groups */
400 for (i = 0, j = 0; i < ngrps * 3; i++) {
401 prevexp += dexp[i] - 2;
404 switch (group_size) {
405 case 4: dexps[j++] = prevexp;
406 dexps[j++] = prevexp;
407 case 2: dexps[j++] = prevexp;
408 case 1: dexps[j++] = prevexp;
415 * Generate transform coefficients for each coupled channel in the coupling
416 * range using the coupling coefficients and coupling coordinates.
417 * reference: Section 7.4.3 Coupling Coordinate Format
419 static void calc_transform_coeffs_cpl(AC3DecodeContext *s)
423 bin = s->start_freq[CPL_CH];
424 for (band = 0; band < s->num_cpl_bands; band++) {
425 int band_start = bin;
426 int band_end = bin + s->cpl_band_sizes[band];
427 for (ch = 1; ch <= s->fbw_channels; ch++) {
428 if (s->channel_in_cpl[ch]) {
429 int cpl_coord = s->cpl_coords[ch][band] << 5;
430 for (bin = band_start; bin < band_end; bin++) {
431 s->fixed_coeffs[ch][bin] =
432 MULH(s->fixed_coeffs[CPL_CH][bin] << 4, cpl_coord);
434 if (ch == 2 && s->phase_flags[band]) {
435 for (bin = band_start; bin < band_end; bin++)
436 s->fixed_coeffs[2][bin] = -s->fixed_coeffs[2][bin];
445 * Grouped mantissas for 3-level 5-level and 11-level quantization
447 typedef struct mant_groups {
457 * Decode the transform coefficients for a particular channel
458 * reference: Section 7.3 Quantization and Decoding of Mantissas
460 static void ac3_decode_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
462 int start_freq = s->start_freq[ch_index];
463 int end_freq = s->end_freq[ch_index];
464 uint8_t *baps = s->bap[ch_index];
465 int8_t *exps = s->dexps[ch_index];
466 int32_t *coeffs = s->fixed_coeffs[ch_index];
467 int dither = (ch_index == CPL_CH) || s->dither_flag[ch_index];
468 GetBitContext *gbc = &s->gbc;
471 for (freq = start_freq; freq < end_freq; freq++) {
472 int bap = baps[freq];
476 /* random noise with approximate range of -0.707 to 0.707 */
478 mantissa = (av_lfg_get(&s->dith_state) / 362) - 5932275;
485 mantissa = m->b1_mant[m->b1];
487 int bits = get_bits(gbc, 5);
488 mantissa = b1_mantissas[bits][0];
489 m->b1_mant[1] = b1_mantissas[bits][1];
490 m->b1_mant[0] = b1_mantissas[bits][2];
497 mantissa = m->b2_mant[m->b2];
499 int bits = get_bits(gbc, 7);
500 mantissa = b2_mantissas[bits][0];
501 m->b2_mant[1] = b2_mantissas[bits][1];
502 m->b2_mant[0] = b2_mantissas[bits][2];
507 mantissa = b3_mantissas[get_bits(gbc, 3)];
512 mantissa = m->b4_mant;
514 int bits = get_bits(gbc, 7);
515 mantissa = b4_mantissas[bits][0];
516 m->b4_mant = b4_mantissas[bits][1];
521 mantissa = b5_mantissas[get_bits(gbc, 4)];
523 default: /* 6 to 15 */
524 /* Shift mantissa and sign-extend it. */
525 mantissa = get_sbits(gbc, quantization_tab[bap]);
526 mantissa <<= 24 - quantization_tab[bap];
529 coeffs[freq] = mantissa >> exps[freq];
534 * Remove random dithering from coupling range coefficients with zero-bit
535 * mantissas for coupled channels which do not use dithering.
536 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
538 static void remove_dithering(AC3DecodeContext *s) {
541 for (ch = 1; ch <= s->fbw_channels; ch++) {
542 if (!s->dither_flag[ch] && s->channel_in_cpl[ch]) {
543 for (i = s->start_freq[CPL_CH]; i < s->end_freq[CPL_CH]; i++) {
544 if (!s->bap[CPL_CH][i])
545 s->fixed_coeffs[ch][i] = 0;
551 static void decode_transform_coeffs_ch(AC3DecodeContext *s, int blk, int ch,
554 if (!s->channel_uses_aht[ch]) {
555 ac3_decode_transform_coeffs_ch(s, ch, m);
557 /* if AHT is used, mantissas for all blocks are encoded in the first
558 block of the frame. */
560 if (!blk && CONFIG_EAC3_DECODER)
561 ff_eac3_decode_transform_coeffs_aht_ch(s, ch);
562 for (bin = s->start_freq[ch]; bin < s->end_freq[ch]; bin++) {
563 s->fixed_coeffs[ch][bin] = s->pre_mantissa[ch][bin][blk] >> s->dexps[ch][bin];
569 * Decode the transform coefficients.
571 static void decode_transform_coeffs(AC3DecodeContext *s, int blk)
577 m.b1 = m.b2 = m.b4 = 0;
579 for (ch = 1; ch <= s->channels; ch++) {
580 /* transform coefficients for full-bandwidth channel */
581 decode_transform_coeffs_ch(s, blk, ch, &m);
582 /* transform coefficients for coupling channel come right after the
583 coefficients for the first coupled channel*/
584 if (s->channel_in_cpl[ch]) {
586 decode_transform_coeffs_ch(s, blk, CPL_CH, &m);
587 calc_transform_coeffs_cpl(s);
590 end = s->end_freq[CPL_CH];
592 end = s->end_freq[ch];
595 s->fixed_coeffs[ch][end] = 0;
599 /* zero the dithered coefficients for appropriate channels */
604 * Stereo rematrixing.
605 * reference: Section 7.5.4 Rematrixing : Decoding Technique
607 static void do_rematrixing(AC3DecodeContext *s)
612 end = FFMIN(s->end_freq[1], s->end_freq[2]);
614 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++) {
615 if (s->rematrixing_flags[bnd]) {
616 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd + 1]);
617 for (i = ff_ac3_rematrix_band_tab[bnd]; i < bndend; i++) {
618 int tmp0 = s->fixed_coeffs[1][i];
619 s->fixed_coeffs[1][i] += s->fixed_coeffs[2][i];
620 s->fixed_coeffs[2][i] = tmp0 - s->fixed_coeffs[2][i];
627 * Inverse MDCT Transform.
628 * Convert frequency domain coefficients to time-domain audio samples.
629 * reference: Section 7.9.4 Transformation Equations
631 static inline void do_imdct(AC3DecodeContext *s, int channels)
635 for (ch = 1; ch <= channels; ch++) {
636 if (s->block_switch[ch]) {
638 float *x = s->tmp_output + 128;
639 for (i = 0; i < 128; i++)
640 x[i] = s->transform_coeffs[ch][2 * i];
641 s->imdct_256.imdct_half(&s->imdct_256, s->tmp_output, x);
642 s->fdsp.vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1],
643 s->tmp_output, s->window, 128);
644 for (i = 0; i < 128; i++)
645 x[i] = s->transform_coeffs[ch][2 * i + 1];
646 s->imdct_256.imdct_half(&s->imdct_256, s->delay[ch - 1], x);
648 s->imdct_512.imdct_half(&s->imdct_512, s->tmp_output, s->transform_coeffs[ch]);
649 s->fdsp.vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1],
650 s->tmp_output, s->window, 128);
651 memcpy(s->delay[ch - 1], s->tmp_output + 128, 128 * sizeof(float));
657 * Upmix delay samples from stereo to original channel layout.
659 static void ac3_upmix_delay(AC3DecodeContext *s)
661 int channel_data_size = sizeof(s->delay[0]);
662 switch (s->channel_mode) {
663 case AC3_CHMODE_DUALMONO:
664 case AC3_CHMODE_STEREO:
665 /* upmix mono to stereo */
666 memcpy(s->delay[1], s->delay[0], channel_data_size);
668 case AC3_CHMODE_2F2R:
669 memset(s->delay[3], 0, channel_data_size);
670 case AC3_CHMODE_2F1R:
671 memset(s->delay[2], 0, channel_data_size);
673 case AC3_CHMODE_3F2R:
674 memset(s->delay[4], 0, channel_data_size);
675 case AC3_CHMODE_3F1R:
676 memset(s->delay[3], 0, channel_data_size);
678 memcpy(s->delay[2], s->delay[1], channel_data_size);
679 memset(s->delay[1], 0, channel_data_size);
685 * Decode band structure for coupling, spectral extension, or enhanced coupling.
686 * The band structure defines how many subbands are in each band. For each
687 * subband in the range, 1 means it is combined with the previous band, and 0
688 * means that it starts a new band.
690 * @param[in] gbc bit reader context
691 * @param[in] blk block number
692 * @param[in] eac3 flag to indicate E-AC-3
693 * @param[in] ecpl flag to indicate enhanced coupling
694 * @param[in] start_subband subband number for start of range
695 * @param[in] end_subband subband number for end of range
696 * @param[in] default_band_struct default band structure table
697 * @param[out] num_bands number of bands (optionally NULL)
698 * @param[out] band_sizes array containing the number of bins in each band (optionally NULL)
700 static void decode_band_structure(GetBitContext *gbc, int blk, int eac3,
701 int ecpl, int start_subband, int end_subband,
702 const uint8_t *default_band_struct,
703 int *num_bands, uint8_t *band_sizes)
705 int subbnd, bnd, n_subbands, n_bands=0;
707 uint8_t coded_band_struct[22];
708 const uint8_t *band_struct;
710 n_subbands = end_subband - start_subband;
712 /* decode band structure from bitstream or use default */
713 if (!eac3 || get_bits1(gbc)) {
714 for (subbnd = 0; subbnd < n_subbands - 1; subbnd++) {
715 coded_band_struct[subbnd] = get_bits1(gbc);
717 band_struct = coded_band_struct;
719 band_struct = &default_band_struct[start_subband+1];
721 /* no change in band structure */
725 /* calculate number of bands and band sizes based on band structure.
726 note that the first 4 subbands in enhanced coupling span only 6 bins
728 if (num_bands || band_sizes ) {
729 n_bands = n_subbands;
730 bnd_sz[0] = ecpl ? 6 : 12;
731 for (bnd = 0, subbnd = 1; subbnd < n_subbands; subbnd++) {
732 int subbnd_size = (ecpl && subbnd < 4) ? 6 : 12;
733 if (band_struct[subbnd - 1]) {
735 bnd_sz[bnd] += subbnd_size;
737 bnd_sz[++bnd] = subbnd_size;
742 /* set optional output params */
744 *num_bands = n_bands;
746 memcpy(band_sizes, bnd_sz, n_bands);
750 * Decode a single audio block from the AC-3 bitstream.
752 static int decode_audio_block(AC3DecodeContext *s, int blk)
754 int fbw_channels = s->fbw_channels;
755 int channel_mode = s->channel_mode;
757 int different_transforms;
760 GetBitContext *gbc = &s->gbc;
761 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS] = { 0 };
763 /* block switch flags */
764 different_transforms = 0;
765 if (s->block_switch_syntax) {
766 for (ch = 1; ch <= fbw_channels; ch++) {
767 s->block_switch[ch] = get_bits1(gbc);
768 if (ch > 1 && s->block_switch[ch] != s->block_switch[1])
769 different_transforms = 1;
773 /* dithering flags */
774 if (s->dither_flag_syntax) {
775 for (ch = 1; ch <= fbw_channels; ch++) {
776 s->dither_flag[ch] = get_bits1(gbc);
781 i = !s->channel_mode;
783 if (get_bits1(gbc)) {
784 /* Allow asymmetric application of DRC when drc_scale > 1.
785 Amplification of quiet sounds is enhanced */
786 float range = dynamic_range_tab[get_bits(gbc, 8)];
787 if (range > 1.0 || s->drc_scale <= 1.0)
788 s->dynamic_range[i] = powf(range, s->drc_scale);
790 s->dynamic_range[i] = range;
791 } else if (blk == 0) {
792 s->dynamic_range[i] = 1.0f;
796 /* spectral extension strategy */
797 if (s->eac3 && (!blk || get_bits1(gbc))) {
798 s->spx_in_use = get_bits1(gbc);
800 int dst_start_freq, dst_end_freq, src_start_freq,
801 start_subband, end_subband;
803 /* determine which channels use spx */
804 if (s->channel_mode == AC3_CHMODE_MONO) {
805 s->channel_uses_spx[1] = 1;
807 for (ch = 1; ch <= fbw_channels; ch++)
808 s->channel_uses_spx[ch] = get_bits1(gbc);
811 /* get the frequency bins of the spx copy region and the spx start
813 dst_start_freq = get_bits(gbc, 2);
814 start_subband = get_bits(gbc, 3) + 2;
815 if (start_subband > 7)
816 start_subband += start_subband - 7;
817 end_subband = get_bits(gbc, 3) + 5;
819 end_subband += end_subband - 7;
820 dst_start_freq = dst_start_freq * 12 + 25;
821 src_start_freq = start_subband * 12 + 25;
822 dst_end_freq = end_subband * 12 + 25;
824 /* check validity of spx ranges */
825 if (start_subband >= end_subband) {
826 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
827 "range (%d >= %d)\n", start_subband, end_subband);
828 return AVERROR_INVALIDDATA;
830 if (dst_start_freq >= src_start_freq) {
831 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
832 "copy start bin (%d >= %d)\n", dst_start_freq, src_start_freq);
833 return AVERROR_INVALIDDATA;
836 s->spx_dst_start_freq = dst_start_freq;
837 s->spx_src_start_freq = src_start_freq;
838 s->spx_dst_end_freq = dst_end_freq;
840 decode_band_structure(gbc, blk, s->eac3, 0,
841 start_subband, end_subband,
842 ff_eac3_default_spx_band_struct,
846 for (ch = 1; ch <= fbw_channels; ch++) {
847 s->channel_uses_spx[ch] = 0;
848 s->first_spx_coords[ch] = 1;
853 /* spectral extension coordinates */
855 for (ch = 1; ch <= fbw_channels; ch++) {
856 if (s->channel_uses_spx[ch]) {
857 if (s->first_spx_coords[ch] || get_bits1(gbc)) {
859 int bin, master_spx_coord;
861 s->first_spx_coords[ch] = 0;
862 spx_blend = get_bits(gbc, 5) * (1.0f/32);
863 master_spx_coord = get_bits(gbc, 2) * 3;
865 bin = s->spx_src_start_freq;
866 for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
868 int spx_coord_exp, spx_coord_mant;
869 float nratio, sblend, nblend, spx_coord;
871 /* calculate blending factors */
872 bandsize = s->spx_band_sizes[bnd];
873 nratio = ((float)((bin + (bandsize >> 1))) / s->spx_dst_end_freq) - spx_blend;
874 nratio = av_clipf(nratio, 0.0f, 1.0f);
875 nblend = sqrtf(3.0f * nratio); // noise is scaled by sqrt(3)
876 // to give unity variance
877 sblend = sqrtf(1.0f - nratio);
880 /* decode spx coordinates */
881 spx_coord_exp = get_bits(gbc, 4);
882 spx_coord_mant = get_bits(gbc, 2);
883 if (spx_coord_exp == 15) spx_coord_mant <<= 1;
884 else spx_coord_mant += 4;
885 spx_coord_mant <<= (25 - spx_coord_exp - master_spx_coord);
886 spx_coord = spx_coord_mant * (1.0f / (1 << 23));
888 /* multiply noise and signal blending factors by spx coordinate */
889 s->spx_noise_blend [ch][bnd] = nblend * spx_coord;
890 s->spx_signal_blend[ch][bnd] = sblend * spx_coord;
894 s->first_spx_coords[ch] = 1;
899 /* coupling strategy */
900 if (s->eac3 ? s->cpl_strategy_exists[blk] : get_bits1(gbc)) {
901 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
903 s->cpl_in_use[blk] = get_bits1(gbc);
904 if (s->cpl_in_use[blk]) {
905 /* coupling in use */
906 int cpl_start_subband, cpl_end_subband;
908 if (channel_mode < AC3_CHMODE_STEREO) {
909 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
910 return AVERROR_INVALIDDATA;
913 /* check for enhanced coupling */
914 if (s->eac3 && get_bits1(gbc)) {
915 /* TODO: parse enhanced coupling strategy info */
916 avpriv_request_sample(s->avctx, "Enhanced coupling");
917 return AVERROR_PATCHWELCOME;
920 /* determine which channels are coupled */
921 if (s->eac3 && s->channel_mode == AC3_CHMODE_STEREO) {
922 s->channel_in_cpl[1] = 1;
923 s->channel_in_cpl[2] = 1;
925 for (ch = 1; ch <= fbw_channels; ch++)
926 s->channel_in_cpl[ch] = get_bits1(gbc);
929 /* phase flags in use */
930 if (channel_mode == AC3_CHMODE_STEREO)
931 s->phase_flags_in_use = get_bits1(gbc);
933 /* coupling frequency range */
934 cpl_start_subband = get_bits(gbc, 4);
935 cpl_end_subband = s->spx_in_use ? (s->spx_src_start_freq - 37) / 12 :
936 get_bits(gbc, 4) + 3;
937 if (cpl_start_subband >= cpl_end_subband) {
938 av_log(s->avctx, AV_LOG_ERROR, "invalid coupling range (%d >= %d)\n",
939 cpl_start_subband, cpl_end_subband);
940 return AVERROR_INVALIDDATA;
942 s->start_freq[CPL_CH] = cpl_start_subband * 12 + 37;
943 s->end_freq[CPL_CH] = cpl_end_subband * 12 + 37;
945 decode_band_structure(gbc, blk, s->eac3, 0, cpl_start_subband,
947 ff_eac3_default_cpl_band_struct,
948 &s->num_cpl_bands, s->cpl_band_sizes);
950 /* coupling not in use */
951 for (ch = 1; ch <= fbw_channels; ch++) {
952 s->channel_in_cpl[ch] = 0;
953 s->first_cpl_coords[ch] = 1;
955 s->first_cpl_leak = s->eac3;
956 s->phase_flags_in_use = 0;
958 } else if (!s->eac3) {
960 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must "
961 "be present in block 0\n");
962 return AVERROR_INVALIDDATA;
964 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
967 cpl_in_use = s->cpl_in_use[blk];
969 /* coupling coordinates */
971 int cpl_coords_exist = 0;
973 for (ch = 1; ch <= fbw_channels; ch++) {
974 if (s->channel_in_cpl[ch]) {
975 if ((s->eac3 && s->first_cpl_coords[ch]) || get_bits1(gbc)) {
976 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
977 s->first_cpl_coords[ch] = 0;
978 cpl_coords_exist = 1;
979 master_cpl_coord = 3 * get_bits(gbc, 2);
980 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
981 cpl_coord_exp = get_bits(gbc, 4);
982 cpl_coord_mant = get_bits(gbc, 4);
983 if (cpl_coord_exp == 15)
984 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
986 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
987 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
990 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must "
991 "be present in block 0\n");
992 return AVERROR_INVALIDDATA;
995 /* channel not in coupling */
996 s->first_cpl_coords[ch] = 1;
1000 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
1001 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
1002 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
1007 /* stereo rematrixing strategy and band structure */
1008 if (channel_mode == AC3_CHMODE_STEREO) {
1009 if ((s->eac3 && !blk) || get_bits1(gbc)) {
1010 s->num_rematrixing_bands = 4;
1011 if (cpl_in_use && s->start_freq[CPL_CH] <= 61) {
1012 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
1013 } else if (s->spx_in_use && s->spx_src_start_freq <= 61) {
1014 s->num_rematrixing_bands--;
1016 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++)
1017 s->rematrixing_flags[bnd] = get_bits1(gbc);
1019 av_log(s->avctx, AV_LOG_WARNING, "Warning: "
1020 "new rematrixing strategy not present in block 0\n");
1021 s->num_rematrixing_bands = 0;
1025 /* exponent strategies for each channel */
1026 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1028 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
1029 if (s->exp_strategy[blk][ch] != EXP_REUSE)
1030 bit_alloc_stages[ch] = 3;
1033 /* channel bandwidth */
1034 for (ch = 1; ch <= fbw_channels; ch++) {
1035 s->start_freq[ch] = 0;
1036 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1038 int prev = s->end_freq[ch];
1039 if (s->channel_in_cpl[ch])
1040 s->end_freq[ch] = s->start_freq[CPL_CH];
1041 else if (s->channel_uses_spx[ch])
1042 s->end_freq[ch] = s->spx_src_start_freq;
1044 int bandwidth_code = get_bits(gbc, 6);
1045 if (bandwidth_code > 60) {
1046 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60\n", bandwidth_code);
1047 return AVERROR_INVALIDDATA;
1049 s->end_freq[ch] = bandwidth_code * 3 + 73;
1051 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
1052 s->num_exp_groups[ch] = (s->end_freq[ch] + group_size-4) / group_size;
1053 if (blk > 0 && s->end_freq[ch] != prev)
1054 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
1057 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
1058 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
1059 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
1062 /* decode exponents for each channel */
1063 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1064 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1065 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
1066 if (decode_exponents(gbc, s->exp_strategy[blk][ch],
1067 s->num_exp_groups[ch], s->dexps[ch][0],
1068 &s->dexps[ch][s->start_freq[ch]+!!ch])) {
1069 av_log(s->avctx, AV_LOG_ERROR, "exponent out-of-range\n");
1070 return AVERROR_INVALIDDATA;
1072 if (ch != CPL_CH && ch != s->lfe_ch)
1073 skip_bits(gbc, 2); /* skip gainrng */
1077 /* bit allocation information */
1078 if (s->bit_allocation_syntax) {
1079 if (get_bits1(gbc)) {
1080 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1081 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1082 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
1083 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
1084 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
1085 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1086 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1088 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must "
1089 "be present in block 0\n");
1090 return AVERROR_INVALIDDATA;
1094 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
1095 if (!s->eac3 || !blk) {
1096 if (s->snr_offset_strategy && get_bits1(gbc)) {
1099 csnr = (get_bits(gbc, 6) - 15) << 4;
1100 for (i = ch = !cpl_in_use; ch <= s->channels; ch++) {
1102 if (ch == i || s->snr_offset_strategy == 2)
1103 snr = (csnr + get_bits(gbc, 4)) << 2;
1104 /* run at least last bit allocation stage if snr offset changes */
1105 if (blk && s->snr_offset[ch] != snr) {
1106 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 1);
1108 s->snr_offset[ch] = snr;
1110 /* fast gain (normal AC-3 only) */
1112 int prev = s->fast_gain[ch];
1113 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1114 /* run last 2 bit allocation stages if fast gain changes */
1115 if (blk && prev != s->fast_gain[ch])
1116 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1119 } else if (!s->eac3 && !blk) {
1120 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
1121 return AVERROR_INVALIDDATA;
1125 /* fast gain (E-AC-3 only) */
1126 if (s->fast_gain_syntax && get_bits1(gbc)) {
1127 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1128 int prev = s->fast_gain[ch];
1129 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1130 /* run last 2 bit allocation stages if fast gain changes */
1131 if (blk && prev != s->fast_gain[ch])
1132 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1134 } else if (s->eac3 && !blk) {
1135 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1136 s->fast_gain[ch] = ff_ac3_fast_gain_tab[4];
1139 /* E-AC-3 to AC-3 converter SNR offset */
1140 if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT && get_bits1(gbc)) {
1141 skip_bits(gbc, 10); // skip converter snr offset
1144 /* coupling leak information */
1146 if (s->first_cpl_leak || get_bits1(gbc)) {
1147 int fl = get_bits(gbc, 3);
1148 int sl = get_bits(gbc, 3);
1149 /* run last 2 bit allocation stages for coupling channel if
1150 coupling leak changes */
1151 if (blk && (fl != s->bit_alloc_params.cpl_fast_leak ||
1152 sl != s->bit_alloc_params.cpl_slow_leak)) {
1153 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
1155 s->bit_alloc_params.cpl_fast_leak = fl;
1156 s->bit_alloc_params.cpl_slow_leak = sl;
1157 } else if (!s->eac3 && !blk) {
1158 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must "
1159 "be present in block 0\n");
1160 return AVERROR_INVALIDDATA;
1162 s->first_cpl_leak = 0;
1165 /* delta bit allocation information */
1166 if (s->dba_syntax && get_bits1(gbc)) {
1167 /* delta bit allocation exists (strategy) */
1168 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1169 s->dba_mode[ch] = get_bits(gbc, 2);
1170 if (s->dba_mode[ch] == DBA_RESERVED) {
1171 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1172 return AVERROR_INVALIDDATA;
1174 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1176 /* channel delta offset, len and bit allocation */
1177 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1178 if (s->dba_mode[ch] == DBA_NEW) {
1179 s->dba_nsegs[ch] = get_bits(gbc, 3) + 1;
1180 for (seg = 0; seg < s->dba_nsegs[ch]; seg++) {
1181 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1182 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1183 s->dba_values[ch][seg] = get_bits(gbc, 3);
1185 /* run last 2 bit allocation stages if new dba values */
1186 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1189 } else if (blk == 0) {
1190 for (ch = 0; ch <= s->channels; ch++) {
1191 s->dba_mode[ch] = DBA_NONE;
1195 /* Bit allocation */
1196 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1197 if (bit_alloc_stages[ch] > 2) {
1198 /* Exponent mapping into PSD and PSD integration */
1199 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1200 s->start_freq[ch], s->end_freq[ch],
1201 s->psd[ch], s->band_psd[ch]);
1203 if (bit_alloc_stages[ch] > 1) {
1204 /* Compute excitation function, Compute masking curve, and
1205 Apply delta bit allocation */
1206 if (ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1207 s->start_freq[ch], s->end_freq[ch],
1208 s->fast_gain[ch], (ch == s->lfe_ch),
1209 s->dba_mode[ch], s->dba_nsegs[ch],
1210 s->dba_offsets[ch], s->dba_lengths[ch],
1211 s->dba_values[ch], s->mask[ch])) {
1212 av_log(s->avctx, AV_LOG_ERROR, "error in bit allocation\n");
1213 return AVERROR_INVALIDDATA;
1216 if (bit_alloc_stages[ch] > 0) {
1217 /* Compute bit allocation */
1218 const uint8_t *bap_tab = s->channel_uses_aht[ch] ?
1219 ff_eac3_hebap_tab : ff_ac3_bap_tab;
1220 s->ac3dsp.bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1221 s->start_freq[ch], s->end_freq[ch],
1223 s->bit_alloc_params.floor,
1224 bap_tab, s->bap[ch]);
1228 /* unused dummy data */
1229 if (s->skip_syntax && get_bits1(gbc)) {
1230 int skipl = get_bits(gbc, 9);
1235 /* unpack the transform coefficients
1236 this also uncouples channels if coupling is in use. */
1237 decode_transform_coeffs(s, blk);
1239 /* TODO: generate enhanced coupling coordinates and uncouple */
1241 /* recover coefficients if rematrixing is in use */
1242 if (s->channel_mode == AC3_CHMODE_STEREO)
1245 /* apply scaling to coefficients (headroom, dynrng) */
1246 for (ch = 1; ch <= s->channels; ch++) {
1247 float gain = 1.0 / 4194304.0f;
1248 if (s->channel_mode == AC3_CHMODE_DUALMONO) {
1249 gain *= s->dynamic_range[2 - ch];
1251 gain *= s->dynamic_range[0];
1253 s->fmt_conv.int32_to_float_fmul_scalar(s->transform_coeffs[ch],
1254 s->fixed_coeffs[ch], gain, 256);
1257 /* apply spectral extension to high frequency bins */
1258 if (s->spx_in_use && CONFIG_EAC3_DECODER) {
1259 ff_eac3_apply_spectral_extension(s);
1262 /* downmix and MDCT. order depends on whether block switching is used for
1263 any channel in this block. this is because coefficients for the long
1264 and short transforms cannot be mixed. */
1265 downmix_output = s->channels != s->out_channels &&
1266 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1267 s->fbw_channels == s->out_channels);
1268 if (different_transforms) {
1269 /* the delay samples have already been downmixed, so we upmix the delay
1270 samples in order to reconstruct all channels before downmixing. */
1276 do_imdct(s, s->channels);
1278 if (downmix_output) {
1279 s->ac3dsp.downmix(s->outptr, s->downmix_coeffs,
1280 s->out_channels, s->fbw_channels, 256);
1283 if (downmix_output) {
1284 s->ac3dsp.downmix(s->xcfptr + 1, s->downmix_coeffs,
1285 s->out_channels, s->fbw_channels, 256);
1288 if (downmix_output && !s->downmixed) {
1290 s->ac3dsp.downmix(s->dlyptr, s->downmix_coeffs, s->out_channels,
1291 s->fbw_channels, 128);
1294 do_imdct(s, s->out_channels);
1301 * Decode a single AC-3 frame.
1303 static int ac3_decode_frame(AVCodecContext * avctx, void *data,
1304 int *got_frame_ptr, AVPacket *avpkt)
1306 AVFrame *frame = data;
1307 const uint8_t *buf = avpkt->data;
1308 int buf_size = avpkt->size;
1309 AC3DecodeContext *s = avctx->priv_data;
1310 int blk, ch, err, ret;
1311 const uint8_t *channel_map;
1312 const float *output[AC3_MAX_CHANNELS];
1313 enum AVMatrixEncoding matrix_encoding;
1314 AVDownmixInfo *downmix_info;
1316 /* copy input buffer to decoder context to avoid reading past the end
1317 of the buffer, which can be caused by a damaged input stream. */
1318 if (buf_size >= 2 && AV_RB16(buf) == 0x770B) {
1319 // seems to be byte-swapped AC-3
1320 int cnt = FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE) >> 1;
1321 s->bdsp.bswap16_buf((uint16_t *) s->input_buffer,
1322 (const uint16_t *) buf, cnt);
1324 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE));
1325 buf = s->input_buffer;
1326 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1327 init_get_bits(&s->gbc, buf, buf_size * 8);
1329 /* parse the syncinfo */
1330 err = parse_frame_header(s);
1334 case AAC_AC3_PARSE_ERROR_SYNC:
1335 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1336 return AVERROR_INVALIDDATA;
1337 case AAC_AC3_PARSE_ERROR_BSID:
1338 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1340 case AAC_AC3_PARSE_ERROR_SAMPLE_RATE:
1341 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1343 case AAC_AC3_PARSE_ERROR_FRAME_SIZE:
1344 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1346 case AAC_AC3_PARSE_ERROR_FRAME_TYPE:
1347 /* skip frame if CRC is ok. otherwise use error concealment. */
1348 /* TODO: add support for substreams and dependent frames */
1349 if (s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1350 av_log(avctx, AV_LOG_WARNING, "unsupported frame type : "
1351 "skipping frame\n");
1355 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1358 case AAC_AC3_PARSE_ERROR_CRC:
1359 case AAC_AC3_PARSE_ERROR_CHANNEL_CFG:
1361 default: // Normal AVERROR do not try to recover.
1366 /* check that reported frame size fits in input buffer */
1367 if (s->frame_size > buf_size) {
1368 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1369 err = AAC_AC3_PARSE_ERROR_FRAME_SIZE;
1370 } else if (avctx->err_recognition & AV_EF_CRCCHECK) {
1371 /* check for crc mismatch */
1372 if (av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2],
1373 s->frame_size - 2)) {
1374 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1375 if (avctx->err_recognition & AV_EF_EXPLODE)
1376 return AVERROR_INVALIDDATA;
1377 err = AAC_AC3_PARSE_ERROR_CRC;
1382 /* if frame is ok, set audio parameters */
1384 avctx->sample_rate = s->sample_rate;
1385 avctx->bit_rate = s->bit_rate;
1388 /* channel config */
1389 if (!err || (s->channels && s->out_channels != s->channels)) {
1390 s->out_channels = s->channels;
1391 s->output_mode = s->channel_mode;
1393 s->output_mode |= AC3_OUTPUT_LFEON;
1394 if (s->channels > 1 &&
1395 avctx->request_channel_layout == AV_CH_LAYOUT_MONO) {
1396 s->out_channels = 1;
1397 s->output_mode = AC3_CHMODE_MONO;
1398 } else if (s->channels > 2 &&
1399 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
1400 s->out_channels = 2;
1401 s->output_mode = AC3_CHMODE_STEREO;
1404 /* set downmixing coefficients if needed */
1405 if (s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1406 s->fbw_channels == s->out_channels)) {
1407 set_downmix_coeffs(s);
1409 } else if (!s->channels) {
1410 av_log(avctx, AV_LOG_ERROR, "unable to determine channel mode\n");
1411 return AVERROR_INVALIDDATA;
1413 avctx->channels = s->out_channels;
1414 avctx->channel_layout = avpriv_ac3_channel_layout_tab[s->output_mode & ~AC3_OUTPUT_LFEON];
1415 if (s->output_mode & AC3_OUTPUT_LFEON)
1416 avctx->channel_layout |= AV_CH_LOW_FREQUENCY;
1418 /* set audio service type based on bitstream mode for AC-3 */
1419 avctx->audio_service_type = s->bitstream_mode;
1420 if (s->bitstream_mode == 0x7 && s->channels > 1)
1421 avctx->audio_service_type = AV_AUDIO_SERVICE_TYPE_KARAOKE;
1423 /* get output buffer */
1424 frame->nb_samples = s->num_blocks * AC3_BLOCK_SIZE;
1425 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1426 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1430 /* decode the audio blocks */
1431 channel_map = ff_ac3_dec_channel_map[s->output_mode & ~AC3_OUTPUT_LFEON][s->lfe_on];
1432 for (ch = 0; ch < s->channels; ch++) {
1433 if (ch < s->out_channels)
1434 s->outptr[channel_map[ch]] = (float *)frame->data[ch];
1436 s->outptr[ch] = s->output[ch];
1437 output[ch] = s->output[ch];
1439 for (blk = 0; blk < s->num_blocks; blk++) {
1440 if (!err && decode_audio_block(s, blk)) {
1441 av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n");
1445 for (ch = 0; ch < s->out_channels; ch++)
1446 memcpy(s->outptr[channel_map[ch]], output[ch], sizeof(**output) * AC3_BLOCK_SIZE);
1447 for (ch = 0; ch < s->out_channels; ch++)
1448 output[ch] = s->outptr[channel_map[ch]];
1449 for (ch = 0; ch < s->out_channels; ch++)
1450 s->outptr[ch] += AC3_BLOCK_SIZE;
1453 /* keep last block for error concealment in next frame */
1454 for (ch = 0; ch < s->out_channels; ch++)
1455 memcpy(s->output[ch], output[ch], sizeof(**output) * AC3_BLOCK_SIZE);
1460 * Check whether the input layout is compatible, and make sure we're not
1461 * downmixing (else the matrix encoding is no longer applicable).
1463 matrix_encoding = AV_MATRIX_ENCODING_NONE;
1464 if (s->channel_mode == AC3_CHMODE_STEREO &&
1465 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1466 if (s->dolby_surround_mode == AC3_DSURMOD_ON)
1467 matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
1468 else if (s->dolby_headphone_mode == AC3_DHEADPHONMOD_ON)
1469 matrix_encoding = AV_MATRIX_ENCODING_DOLBYHEADPHONE;
1470 } else if (s->channel_mode >= AC3_CHMODE_2F2R &&
1471 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1472 switch (s->dolby_surround_ex_mode) {
1473 case AC3_DSUREXMOD_ON: // EX or PLIIx
1474 matrix_encoding = AV_MATRIX_ENCODING_DOLBYEX;
1476 case AC3_DSUREXMOD_PLIIZ:
1477 matrix_encoding = AV_MATRIX_ENCODING_DPLIIZ;
1479 default: // not indicated or off
1483 if ((ret = ff_side_data_update_matrix_encoding(frame, matrix_encoding)) < 0)
1487 if ((downmix_info = av_downmix_info_update_side_data(frame))) {
1488 switch (s->preferred_downmix) {
1489 case AC3_DMIXMOD_LTRT:
1490 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LTRT;
1492 case AC3_DMIXMOD_LORO:
1493 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LORO;
1495 case AC3_DMIXMOD_DPLII:
1496 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_DPLII;
1499 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_UNKNOWN;
1502 downmix_info->center_mix_level = gain_levels[s-> center_mix_level];
1503 downmix_info->center_mix_level_ltrt = gain_levels[s-> center_mix_level_ltrt];
1504 downmix_info->surround_mix_level = gain_levels[s-> surround_mix_level];
1505 downmix_info->surround_mix_level_ltrt = gain_levels[s->surround_mix_level_ltrt];
1506 if (s->lfe_mix_level_exists)
1507 downmix_info->lfe_mix_level = gain_levels_lfe[s->lfe_mix_level];
1509 downmix_info->lfe_mix_level = 0.0; // -inf dB
1511 return AVERROR(ENOMEM);
1515 return FFMIN(buf_size, s->frame_size);
1519 * Uninitialize the AC-3 decoder.
1521 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1523 AC3DecodeContext *s = avctx->priv_data;
1524 ff_mdct_end(&s->imdct_512);
1525 ff_mdct_end(&s->imdct_256);
1530 #define OFFSET(x) offsetof(AC3DecodeContext, x)
1531 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)
1532 static const AVOption options[] = {
1533 { "drc_scale", "percentage of dynamic range compression to apply", OFFSET(drc_scale), AV_OPT_TYPE_FLOAT, {.dbl = 1.0}, 0.0, 6.0, PAR },
1537 static const AVClass ac3_decoder_class = {
1538 .class_name = "AC3 decoder",
1539 .item_name = av_default_item_name,
1541 .version = LIBAVUTIL_VERSION_INT,
1544 AVCodec ff_ac3_decoder = {
1546 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52A (AC-3)"),
1547 .type = AVMEDIA_TYPE_AUDIO,
1548 .id = AV_CODEC_ID_AC3,
1549 .priv_data_size = sizeof (AC3DecodeContext),
1550 .init = ac3_decode_init,
1551 .close = ac3_decode_end,
1552 .decode = ac3_decode_frame,
1553 .capabilities = AV_CODEC_CAP_DR1,
1554 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1555 AV_SAMPLE_FMT_NONE },
1556 .priv_class = &ac3_decoder_class,
1559 #if CONFIG_EAC3_DECODER
1560 static const AVClass eac3_decoder_class = {
1561 .class_name = "E-AC3 decoder",
1562 .item_name = av_default_item_name,
1564 .version = LIBAVUTIL_VERSION_INT,
1567 AVCodec ff_eac3_decoder = {
1569 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52B (AC-3, E-AC-3)"),
1570 .type = AVMEDIA_TYPE_AUDIO,
1571 .id = AV_CODEC_ID_EAC3,
1572 .priv_data_size = sizeof (AC3DecodeContext),
1573 .init = ac3_decode_init,
1574 .close = ac3_decode_end,
1575 .decode = ac3_decode_frame,
1576 .capabilities = AV_CODEC_CAP_DR1,
1577 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1578 AV_SAMPLE_FMT_NONE },
1579 .priv_class = &eac3_decoder_class,