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);
749 static inline int spx_strategy(AC3DecodeContext *s, int blk)
751 GetBitContext *bc = &s->gbc;
752 int fbw_channels = s->fbw_channels;
753 int dst_start_freq, dst_end_freq, src_start_freq,
754 start_subband, end_subband, ch;
756 /* determine which channels use spx */
757 if (s->channel_mode == AC3_CHMODE_MONO) {
758 s->channel_uses_spx[1] = 1;
760 for (ch = 1; ch <= fbw_channels; ch++)
761 s->channel_uses_spx[ch] = get_bits1(bc);
764 /* get the frequency bins of the spx copy region and the spx start
766 dst_start_freq = get_bits(bc, 2);
767 start_subband = get_bits(bc, 3) + 2;
768 if (start_subband > 7)
769 start_subband += start_subband - 7;
770 end_subband = get_bits(bc, 3) + 5;
772 end_subband += end_subband - 7;
773 dst_start_freq = dst_start_freq * 12 + 25;
774 src_start_freq = start_subband * 12 + 25;
775 dst_end_freq = end_subband * 12 + 25;
777 /* check validity of spx ranges */
778 if (start_subband >= end_subband) {
779 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
780 "range (%d >= %d)\n", start_subband, end_subband);
781 return AVERROR_INVALIDDATA;
783 if (dst_start_freq >= src_start_freq) {
784 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
785 "copy start bin (%d >= %d)\n", dst_start_freq, src_start_freq);
786 return AVERROR_INVALIDDATA;
789 s->spx_dst_start_freq = dst_start_freq;
790 s->spx_src_start_freq = src_start_freq;
791 s->spx_dst_end_freq = dst_end_freq;
793 decode_band_structure(bc, blk, s->eac3, 0,
794 start_subband, end_subband,
795 ff_eac3_default_spx_band_struct,
802 static inline void spx_coordinates(AC3DecodeContext *s)
804 GetBitContext *bc = &s->gbc;
805 int fbw_channels = s->fbw_channels;
808 for (ch = 1; ch <= fbw_channels; ch++) {
809 if (s->channel_uses_spx[ch]) {
810 if (s->first_spx_coords[ch] || get_bits1(bc)) {
812 int bin, master_spx_coord;
814 s->first_spx_coords[ch] = 0;
815 spx_blend = get_bits(bc, 5) * (1.0f / 32);
816 master_spx_coord = get_bits(bc, 2) * 3;
818 bin = s->spx_src_start_freq;
819 for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
821 int spx_coord_exp, spx_coord_mant;
822 float nratio, sblend, nblend, spx_coord;
824 /* calculate blending factors */
825 bandsize = s->spx_band_sizes[bnd];
826 nratio = ((float)((bin + (bandsize >> 1))) / s->spx_dst_end_freq) - spx_blend;
827 nratio = av_clipf(nratio, 0.0f, 1.0f);
828 nblend = sqrtf(3.0f * nratio); // noise is scaled by sqrt(3)
829 // to give unity variance
830 sblend = sqrtf(1.0f - nratio);
833 /* decode spx coordinates */
834 spx_coord_exp = get_bits(bc, 4);
835 spx_coord_mant = get_bits(bc, 2);
836 if (spx_coord_exp == 15) spx_coord_mant <<= 1;
837 else spx_coord_mant += 4;
838 spx_coord_mant <<= (25 - spx_coord_exp - master_spx_coord);
839 spx_coord = spx_coord_mant * (1.0f / (1 << 23));
841 /* multiply noise and signal blending factors by spx coordinate */
842 s->spx_noise_blend [ch][bnd] = nblend * spx_coord;
843 s->spx_signal_blend[ch][bnd] = sblend * spx_coord;
847 s->first_spx_coords[ch] = 1;
852 static inline int coupling_strategy(AC3DecodeContext *s, int blk,
853 uint8_t *bit_alloc_stages)
855 GetBitContext *bc = &s->gbc;
856 int fbw_channels = s->fbw_channels;
857 int channel_mode = s->channel_mode;
860 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
862 s->cpl_in_use[blk] = get_bits1(bc);
863 if (s->cpl_in_use[blk]) {
864 /* coupling in use */
865 int cpl_start_subband, cpl_end_subband;
867 if (channel_mode < AC3_CHMODE_STEREO) {
868 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
869 return AVERROR_INVALIDDATA;
872 /* check for enhanced coupling */
873 if (s->eac3 && get_bits1(bc)) {
874 /* TODO: parse enhanced coupling strategy info */
875 avpriv_request_sample(s->avctx, "Enhanced coupling");
876 return AVERROR_PATCHWELCOME;
879 /* determine which channels are coupled */
880 if (s->eac3 && s->channel_mode == AC3_CHMODE_STEREO) {
881 s->channel_in_cpl[1] = 1;
882 s->channel_in_cpl[2] = 1;
884 for (ch = 1; ch <= fbw_channels; ch++)
885 s->channel_in_cpl[ch] = get_bits1(bc);
888 /* phase flags in use */
889 if (channel_mode == AC3_CHMODE_STEREO)
890 s->phase_flags_in_use = get_bits1(bc);
892 /* coupling frequency range */
893 cpl_start_subband = get_bits(bc, 4);
894 cpl_end_subband = s->spx_in_use ? (s->spx_src_start_freq - 37) / 12 :
896 if (cpl_start_subband >= cpl_end_subband) {
897 av_log(s->avctx, AV_LOG_ERROR, "invalid coupling range (%d >= %d)\n",
898 cpl_start_subband, cpl_end_subband);
899 return AVERROR_INVALIDDATA;
901 s->start_freq[CPL_CH] = cpl_start_subband * 12 + 37;
902 s->end_freq[CPL_CH] = cpl_end_subband * 12 + 37;
904 decode_band_structure(bc, blk, s->eac3, 0, cpl_start_subband,
906 ff_eac3_default_cpl_band_struct,
907 &s->num_cpl_bands, s->cpl_band_sizes);
909 /* coupling not in use */
910 for (ch = 1; ch <= fbw_channels; ch++) {
911 s->channel_in_cpl[ch] = 0;
912 s->first_cpl_coords[ch] = 1;
914 s->first_cpl_leak = s->eac3;
915 s->phase_flags_in_use = 0;
921 static inline int coupling_coordinates(AC3DecodeContext *s, int blk)
923 GetBitContext *bc = &s->gbc;
924 int fbw_channels = s->fbw_channels;
926 int cpl_coords_exist = 0;
928 for (ch = 1; ch <= fbw_channels; ch++) {
929 if (s->channel_in_cpl[ch]) {
930 if ((s->eac3 && s->first_cpl_coords[ch]) || get_bits1(bc)) {
931 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
932 s->first_cpl_coords[ch] = 0;
933 cpl_coords_exist = 1;
934 master_cpl_coord = 3 * get_bits(bc, 2);
935 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
936 cpl_coord_exp = get_bits(bc, 4);
937 cpl_coord_mant = get_bits(bc, 4);
938 if (cpl_coord_exp == 15)
939 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
941 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
942 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
945 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must "
946 "be present in block 0\n");
947 return AVERROR_INVALIDDATA;
950 /* channel not in coupling */
951 s->first_cpl_coords[ch] = 1;
955 if (s->channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
956 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
957 s->phase_flags[bnd] = s->phase_flags_in_use ? get_bits1(bc) : 0;
965 * Decode a single audio block from the AC-3 bitstream.
967 static int decode_audio_block(AC3DecodeContext *s, int blk)
969 int fbw_channels = s->fbw_channels;
970 int channel_mode = s->channel_mode;
971 int i, bnd, seg, ch, ret;
972 int different_transforms;
975 GetBitContext *gbc = &s->gbc;
976 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS] = { 0 };
978 /* block switch flags */
979 different_transforms = 0;
980 if (s->block_switch_syntax) {
981 for (ch = 1; ch <= fbw_channels; ch++) {
982 s->block_switch[ch] = get_bits1(gbc);
983 if (ch > 1 && s->block_switch[ch] != s->block_switch[1])
984 different_transforms = 1;
988 /* dithering flags */
989 if (s->dither_flag_syntax) {
990 for (ch = 1; ch <= fbw_channels; ch++) {
991 s->dither_flag[ch] = get_bits1(gbc);
996 i = !s->channel_mode;
998 if (get_bits1(gbc)) {
999 /* Allow asymmetric application of DRC when drc_scale > 1.
1000 Amplification of quiet sounds is enhanced */
1001 float range = dynamic_range_tab[get_bits(gbc, 8)];
1002 if (range > 1.0 || s->drc_scale <= 1.0)
1003 s->dynamic_range[i] = powf(range, s->drc_scale);
1005 s->dynamic_range[i] = range;
1006 } else if (blk == 0) {
1007 s->dynamic_range[i] = 1.0f;
1011 /* spectral extension strategy */
1012 if (s->eac3 && (!blk || get_bits1(gbc))) {
1013 s->spx_in_use = get_bits1(gbc);
1014 if (s->spx_in_use) {
1015 if ((ret = spx_strategy(s, blk)) < 0)
1018 for (ch = 1; ch <= fbw_channels; ch++) {
1019 s->channel_uses_spx[ch] = 0;
1020 s->first_spx_coords[ch] = 1;
1025 /* spectral extension coordinates */
1029 /* coupling strategy */
1030 if (s->eac3 ? s->cpl_strategy_exists[blk] : get_bits1(gbc)) {
1031 if ((ret = coupling_strategy(s, blk, bit_alloc_stages)) < 0)
1033 } else if (!s->eac3) {
1035 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must "
1036 "be present in block 0\n");
1037 return AVERROR_INVALIDDATA;
1039 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
1042 cpl_in_use = s->cpl_in_use[blk];
1044 /* coupling coordinates */
1046 if ((ret = coupling_coordinates(s, blk)) < 0)
1050 /* stereo rematrixing strategy and band structure */
1051 if (channel_mode == AC3_CHMODE_STEREO) {
1052 if ((s->eac3 && !blk) || get_bits1(gbc)) {
1053 s->num_rematrixing_bands = 4;
1054 if (cpl_in_use && s->start_freq[CPL_CH] <= 61) {
1055 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
1056 } else if (s->spx_in_use && s->spx_src_start_freq <= 61) {
1057 s->num_rematrixing_bands--;
1059 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++)
1060 s->rematrixing_flags[bnd] = get_bits1(gbc);
1062 av_log(s->avctx, AV_LOG_WARNING, "Warning: "
1063 "new rematrixing strategy not present in block 0\n");
1064 s->num_rematrixing_bands = 0;
1068 /* exponent strategies for each channel */
1069 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1071 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
1072 if (s->exp_strategy[blk][ch] != EXP_REUSE)
1073 bit_alloc_stages[ch] = 3;
1076 /* channel bandwidth */
1077 for (ch = 1; ch <= fbw_channels; ch++) {
1078 s->start_freq[ch] = 0;
1079 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1081 int prev = s->end_freq[ch];
1082 if (s->channel_in_cpl[ch])
1083 s->end_freq[ch] = s->start_freq[CPL_CH];
1084 else if (s->channel_uses_spx[ch])
1085 s->end_freq[ch] = s->spx_src_start_freq;
1087 int bandwidth_code = get_bits(gbc, 6);
1088 if (bandwidth_code > 60) {
1089 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60\n", bandwidth_code);
1090 return AVERROR_INVALIDDATA;
1092 s->end_freq[ch] = bandwidth_code * 3 + 73;
1094 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
1095 s->num_exp_groups[ch] = (s->end_freq[ch] + group_size-4) / group_size;
1096 if (blk > 0 && s->end_freq[ch] != prev)
1097 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
1100 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
1101 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
1102 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
1105 /* decode exponents for each channel */
1106 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1107 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1108 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
1109 if (decode_exponents(gbc, s->exp_strategy[blk][ch],
1110 s->num_exp_groups[ch], s->dexps[ch][0],
1111 &s->dexps[ch][s->start_freq[ch]+!!ch])) {
1112 av_log(s->avctx, AV_LOG_ERROR, "exponent out-of-range\n");
1113 return AVERROR_INVALIDDATA;
1115 if (ch != CPL_CH && ch != s->lfe_ch)
1116 skip_bits(gbc, 2); /* skip gainrng */
1120 /* bit allocation information */
1121 if (s->bit_allocation_syntax) {
1122 if (get_bits1(gbc)) {
1123 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1124 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1125 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
1126 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
1127 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
1128 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1129 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1131 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must "
1132 "be present in block 0\n");
1133 return AVERROR_INVALIDDATA;
1137 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
1138 if (!s->eac3 || !blk) {
1139 if (s->snr_offset_strategy && get_bits1(gbc)) {
1142 csnr = (get_bits(gbc, 6) - 15) << 4;
1143 for (i = ch = !cpl_in_use; ch <= s->channels; ch++) {
1145 if (ch == i || s->snr_offset_strategy == 2)
1146 snr = (csnr + get_bits(gbc, 4)) << 2;
1147 /* run at least last bit allocation stage if snr offset changes */
1148 if (blk && s->snr_offset[ch] != snr) {
1149 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 1);
1151 s->snr_offset[ch] = snr;
1153 /* fast gain (normal AC-3 only) */
1155 int prev = s->fast_gain[ch];
1156 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1157 /* run last 2 bit allocation stages if fast gain changes */
1158 if (blk && prev != s->fast_gain[ch])
1159 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1162 } else if (!s->eac3 && !blk) {
1163 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
1164 return AVERROR_INVALIDDATA;
1168 /* fast gain (E-AC-3 only) */
1169 if (s->fast_gain_syntax && get_bits1(gbc)) {
1170 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1171 int prev = s->fast_gain[ch];
1172 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1173 /* run last 2 bit allocation stages if fast gain changes */
1174 if (blk && prev != s->fast_gain[ch])
1175 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1177 } else if (s->eac3 && !blk) {
1178 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1179 s->fast_gain[ch] = ff_ac3_fast_gain_tab[4];
1182 /* E-AC-3 to AC-3 converter SNR offset */
1183 if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT && get_bits1(gbc)) {
1184 skip_bits(gbc, 10); // skip converter snr offset
1187 /* coupling leak information */
1189 if (s->first_cpl_leak || get_bits1(gbc)) {
1190 int fl = get_bits(gbc, 3);
1191 int sl = get_bits(gbc, 3);
1192 /* run last 2 bit allocation stages for coupling channel if
1193 coupling leak changes */
1194 if (blk && (fl != s->bit_alloc_params.cpl_fast_leak ||
1195 sl != s->bit_alloc_params.cpl_slow_leak)) {
1196 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
1198 s->bit_alloc_params.cpl_fast_leak = fl;
1199 s->bit_alloc_params.cpl_slow_leak = sl;
1200 } else if (!s->eac3 && !blk) {
1201 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must "
1202 "be present in block 0\n");
1203 return AVERROR_INVALIDDATA;
1205 s->first_cpl_leak = 0;
1208 /* delta bit allocation information */
1209 if (s->dba_syntax && get_bits1(gbc)) {
1210 /* delta bit allocation exists (strategy) */
1211 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1212 s->dba_mode[ch] = get_bits(gbc, 2);
1213 if (s->dba_mode[ch] == DBA_RESERVED) {
1214 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1215 return AVERROR_INVALIDDATA;
1217 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1219 /* channel delta offset, len and bit allocation */
1220 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1221 if (s->dba_mode[ch] == DBA_NEW) {
1222 s->dba_nsegs[ch] = get_bits(gbc, 3) + 1;
1223 for (seg = 0; seg < s->dba_nsegs[ch]; seg++) {
1224 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1225 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1226 s->dba_values[ch][seg] = get_bits(gbc, 3);
1228 /* run last 2 bit allocation stages if new dba values */
1229 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1232 } else if (blk == 0) {
1233 for (ch = 0; ch <= s->channels; ch++) {
1234 s->dba_mode[ch] = DBA_NONE;
1238 /* Bit allocation */
1239 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1240 if (bit_alloc_stages[ch] > 2) {
1241 /* Exponent mapping into PSD and PSD integration */
1242 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1243 s->start_freq[ch], s->end_freq[ch],
1244 s->psd[ch], s->band_psd[ch]);
1246 if (bit_alloc_stages[ch] > 1) {
1247 /* Compute excitation function, Compute masking curve, and
1248 Apply delta bit allocation */
1249 if (ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1250 s->start_freq[ch], s->end_freq[ch],
1251 s->fast_gain[ch], (ch == s->lfe_ch),
1252 s->dba_mode[ch], s->dba_nsegs[ch],
1253 s->dba_offsets[ch], s->dba_lengths[ch],
1254 s->dba_values[ch], s->mask[ch])) {
1255 av_log(s->avctx, AV_LOG_ERROR, "error in bit allocation\n");
1256 return AVERROR_INVALIDDATA;
1259 if (bit_alloc_stages[ch] > 0) {
1260 /* Compute bit allocation */
1261 const uint8_t *bap_tab = s->channel_uses_aht[ch] ?
1262 ff_eac3_hebap_tab : ff_ac3_bap_tab;
1263 s->ac3dsp.bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1264 s->start_freq[ch], s->end_freq[ch],
1266 s->bit_alloc_params.floor,
1267 bap_tab, s->bap[ch]);
1271 /* unused dummy data */
1272 if (s->skip_syntax && get_bits1(gbc)) {
1273 int skipl = get_bits(gbc, 9);
1274 skip_bits_long(gbc, 8 * skipl);
1277 /* unpack the transform coefficients
1278 this also uncouples channels if coupling is in use. */
1279 decode_transform_coeffs(s, blk);
1281 /* TODO: generate enhanced coupling coordinates and uncouple */
1283 /* recover coefficients if rematrixing is in use */
1284 if (s->channel_mode == AC3_CHMODE_STEREO)
1287 /* apply scaling to coefficients (headroom, dynrng) */
1288 for (ch = 1; ch <= s->channels; ch++) {
1289 float gain = 1.0 / 4194304.0f;
1290 if (s->channel_mode == AC3_CHMODE_DUALMONO) {
1291 gain *= s->dynamic_range[2 - ch];
1293 gain *= s->dynamic_range[0];
1295 s->fmt_conv.int32_to_float_fmul_scalar(s->transform_coeffs[ch],
1296 s->fixed_coeffs[ch], gain, 256);
1299 /* apply spectral extension to high frequency bins */
1300 if (s->spx_in_use && CONFIG_EAC3_DECODER) {
1301 ff_eac3_apply_spectral_extension(s);
1304 /* downmix and MDCT. order depends on whether block switching is used for
1305 any channel in this block. this is because coefficients for the long
1306 and short transforms cannot be mixed. */
1307 downmix_output = s->channels != s->out_channels &&
1308 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1309 s->fbw_channels == s->out_channels);
1310 if (different_transforms) {
1311 /* the delay samples have already been downmixed, so we upmix the delay
1312 samples in order to reconstruct all channels before downmixing. */
1318 do_imdct(s, s->channels);
1320 if (downmix_output) {
1321 s->ac3dsp.downmix(s->outptr, s->downmix_coeffs,
1322 s->out_channels, s->fbw_channels, 256);
1325 if (downmix_output) {
1326 s->ac3dsp.downmix(s->xcfptr + 1, s->downmix_coeffs,
1327 s->out_channels, s->fbw_channels, 256);
1330 if (downmix_output && !s->downmixed) {
1332 s->ac3dsp.downmix(s->dlyptr, s->downmix_coeffs, s->out_channels,
1333 s->fbw_channels, 128);
1336 do_imdct(s, s->out_channels);
1343 * Decode a single AC-3 frame.
1345 static int ac3_decode_frame(AVCodecContext * avctx, void *data,
1346 int *got_frame_ptr, AVPacket *avpkt)
1348 AVFrame *frame = data;
1349 const uint8_t *buf = avpkt->data;
1350 int buf_size = avpkt->size;
1351 AC3DecodeContext *s = avctx->priv_data;
1352 int blk, ch, err, ret;
1353 const uint8_t *channel_map;
1354 const float *output[AC3_MAX_CHANNELS];
1355 enum AVMatrixEncoding matrix_encoding;
1356 AVDownmixInfo *downmix_info;
1358 /* copy input buffer to decoder context to avoid reading past the end
1359 of the buffer, which can be caused by a damaged input stream. */
1360 if (buf_size >= 2 && AV_RB16(buf) == 0x770B) {
1361 // seems to be byte-swapped AC-3
1362 int cnt = FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE) >> 1;
1363 s->bdsp.bswap16_buf((uint16_t *) s->input_buffer,
1364 (const uint16_t *) buf, cnt);
1366 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE));
1367 buf = s->input_buffer;
1368 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1369 init_get_bits(&s->gbc, buf, buf_size * 8);
1371 /* parse the syncinfo */
1372 err = parse_frame_header(s);
1376 case AAC_AC3_PARSE_ERROR_SYNC:
1377 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1378 return AVERROR_INVALIDDATA;
1379 case AAC_AC3_PARSE_ERROR_BSID:
1380 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1382 case AAC_AC3_PARSE_ERROR_SAMPLE_RATE:
1383 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1385 case AAC_AC3_PARSE_ERROR_FRAME_SIZE:
1386 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1388 case AAC_AC3_PARSE_ERROR_FRAME_TYPE:
1389 /* skip frame if CRC is ok. otherwise use error concealment. */
1390 /* TODO: add support for substreams and dependent frames */
1391 if (s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1392 av_log(avctx, AV_LOG_DEBUG,
1393 "unsupported frame type %d: skipping frame\n",
1398 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1401 case AAC_AC3_PARSE_ERROR_CRC:
1402 case AAC_AC3_PARSE_ERROR_CHANNEL_CFG:
1404 default: // Normal AVERROR do not try to recover.
1409 /* check that reported frame size fits in input buffer */
1410 if (s->frame_size > buf_size) {
1411 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1412 err = AAC_AC3_PARSE_ERROR_FRAME_SIZE;
1413 } else if (avctx->err_recognition & AV_EF_CRCCHECK) {
1414 /* check for crc mismatch */
1415 if (av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2],
1416 s->frame_size - 2)) {
1417 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1418 if (avctx->err_recognition & AV_EF_EXPLODE)
1419 return AVERROR_INVALIDDATA;
1420 err = AAC_AC3_PARSE_ERROR_CRC;
1425 /* if frame is ok, set audio parameters */
1427 avctx->sample_rate = s->sample_rate;
1428 avctx->bit_rate = s->bit_rate;
1431 /* channel config */
1432 if (!err || (s->channels && s->out_channels != s->channels)) {
1433 s->out_channels = s->channels;
1434 s->output_mode = s->channel_mode;
1436 s->output_mode |= AC3_OUTPUT_LFEON;
1437 if (s->channels > 1 &&
1438 avctx->request_channel_layout == AV_CH_LAYOUT_MONO) {
1439 s->out_channels = 1;
1440 s->output_mode = AC3_CHMODE_MONO;
1441 } else if (s->channels > 2 &&
1442 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
1443 s->out_channels = 2;
1444 s->output_mode = AC3_CHMODE_STEREO;
1447 /* set downmixing coefficients if needed */
1448 if (s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1449 s->fbw_channels == s->out_channels)) {
1450 set_downmix_coeffs(s);
1452 } else if (!s->channels) {
1453 av_log(avctx, AV_LOG_ERROR, "unable to determine channel mode\n");
1454 return AVERROR_INVALIDDATA;
1456 avctx->channels = s->out_channels;
1457 avctx->channel_layout = avpriv_ac3_channel_layout_tab[s->output_mode & ~AC3_OUTPUT_LFEON];
1458 if (s->output_mode & AC3_OUTPUT_LFEON)
1459 avctx->channel_layout |= AV_CH_LOW_FREQUENCY;
1461 /* set audio service type based on bitstream mode for AC-3 */
1462 avctx->audio_service_type = s->bitstream_mode;
1463 if (s->bitstream_mode == 0x7 && s->channels > 1)
1464 avctx->audio_service_type = AV_AUDIO_SERVICE_TYPE_KARAOKE;
1466 /* get output buffer */
1467 frame->nb_samples = s->num_blocks * AC3_BLOCK_SIZE;
1468 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1469 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1473 /* decode the audio blocks */
1474 channel_map = ff_ac3_dec_channel_map[s->output_mode & ~AC3_OUTPUT_LFEON][s->lfe_on];
1475 for (ch = 0; ch < s->channels; ch++) {
1476 if (ch < s->out_channels)
1477 s->outptr[channel_map[ch]] = (float *)frame->data[ch];
1479 s->outptr[ch] = s->output[ch];
1480 output[ch] = s->output[ch];
1482 for (blk = 0; blk < s->num_blocks; blk++) {
1483 if (!err && decode_audio_block(s, blk)) {
1484 av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n");
1488 for (ch = 0; ch < s->out_channels; ch++)
1489 memcpy(s->outptr[channel_map[ch]], output[ch], sizeof(**output) * AC3_BLOCK_SIZE);
1490 for (ch = 0; ch < s->out_channels; ch++)
1491 output[ch] = s->outptr[channel_map[ch]];
1492 for (ch = 0; ch < s->out_channels; ch++)
1493 s->outptr[ch] += AC3_BLOCK_SIZE;
1496 /* keep last block for error concealment in next frame */
1497 for (ch = 0; ch < s->out_channels; ch++)
1498 memcpy(s->output[ch], output[ch], sizeof(**output) * AC3_BLOCK_SIZE);
1503 * Check whether the input layout is compatible, and make sure we're not
1504 * downmixing (else the matrix encoding is no longer applicable).
1506 matrix_encoding = AV_MATRIX_ENCODING_NONE;
1507 if (s->channel_mode == AC3_CHMODE_STEREO &&
1508 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1509 if (s->dolby_surround_mode == AC3_DSURMOD_ON)
1510 matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
1511 else if (s->dolby_headphone_mode == AC3_DHEADPHONMOD_ON)
1512 matrix_encoding = AV_MATRIX_ENCODING_DOLBYHEADPHONE;
1513 } else if (s->channel_mode >= AC3_CHMODE_2F2R &&
1514 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1515 switch (s->dolby_surround_ex_mode) {
1516 case AC3_DSUREXMOD_ON: // EX or PLIIx
1517 matrix_encoding = AV_MATRIX_ENCODING_DOLBYEX;
1519 case AC3_DSUREXMOD_PLIIZ:
1520 matrix_encoding = AV_MATRIX_ENCODING_DPLIIZ;
1522 default: // not indicated or off
1526 if ((ret = ff_side_data_update_matrix_encoding(frame, matrix_encoding)) < 0)
1530 if ((downmix_info = av_downmix_info_update_side_data(frame))) {
1531 switch (s->preferred_downmix) {
1532 case AC3_DMIXMOD_LTRT:
1533 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LTRT;
1535 case AC3_DMIXMOD_LORO:
1536 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LORO;
1538 case AC3_DMIXMOD_DPLII:
1539 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_DPLII;
1542 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_UNKNOWN;
1545 downmix_info->center_mix_level = gain_levels[s-> center_mix_level];
1546 downmix_info->center_mix_level_ltrt = gain_levels[s-> center_mix_level_ltrt];
1547 downmix_info->surround_mix_level = gain_levels[s-> surround_mix_level];
1548 downmix_info->surround_mix_level_ltrt = gain_levels[s->surround_mix_level_ltrt];
1549 if (s->lfe_mix_level_exists)
1550 downmix_info->lfe_mix_level = gain_levels_lfe[s->lfe_mix_level];
1552 downmix_info->lfe_mix_level = 0.0; // -inf dB
1554 return AVERROR(ENOMEM);
1558 return FFMIN(buf_size, s->frame_size);
1562 * Uninitialize the AC-3 decoder.
1564 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1566 AC3DecodeContext *s = avctx->priv_data;
1567 ff_mdct_end(&s->imdct_512);
1568 ff_mdct_end(&s->imdct_256);
1573 #define OFFSET(x) offsetof(AC3DecodeContext, x)
1574 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)
1575 static const AVOption options[] = {
1576 { "drc_scale", "percentage of dynamic range compression to apply", OFFSET(drc_scale), AV_OPT_TYPE_FLOAT, {.dbl = 1.0}, 0.0, 6.0, PAR },
1580 static const AVClass ac3_decoder_class = {
1581 .class_name = "AC3 decoder",
1582 .item_name = av_default_item_name,
1584 .version = LIBAVUTIL_VERSION_INT,
1587 AVCodec ff_ac3_decoder = {
1589 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52A (AC-3)"),
1590 .type = AVMEDIA_TYPE_AUDIO,
1591 .id = AV_CODEC_ID_AC3,
1592 .priv_data_size = sizeof (AC3DecodeContext),
1593 .init = ac3_decode_init,
1594 .close = ac3_decode_end,
1595 .decode = ac3_decode_frame,
1596 .capabilities = AV_CODEC_CAP_DR1,
1597 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1598 AV_SAMPLE_FMT_NONE },
1599 .priv_class = &ac3_decoder_class,
1602 #if CONFIG_EAC3_DECODER
1603 static const AVClass eac3_decoder_class = {
1604 .class_name = "E-AC3 decoder",
1605 .item_name = av_default_item_name,
1607 .version = LIBAVUTIL_VERSION_INT,
1610 AVCodec ff_eac3_decoder = {
1612 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52B (AC-3, E-AC-3)"),
1613 .type = AVMEDIA_TYPE_AUDIO,
1614 .id = AV_CODEC_ID_EAC3,
1615 .priv_data_size = sizeof (AC3DecodeContext),
1616 .init = ac3_decode_init,
1617 .close = ac3_decode_end,
1618 .decode = ac3_decode_frame,
1619 .capabilities = AV_CODEC_CAP_DR1,
1620 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1621 AV_SAMPLE_FMT_NONE },
1622 .priv_class = &eac3_decoder_class,