3 * This code was developed as part of Google Summer of Code 2006.
4 * E-AC-3 support was added as part of Google Summer of Code 2007.
6 * Copyright (c) 2006 Kartikey Mahendra BHATT (bhattkm at gmail dot com)
7 * Copyright (c) 2007-2008 Bartlomiej Wolowiec <bartek.wolowiec@gmail.com>
8 * Copyright (c) 2007 Justin Ruggles <justin.ruggles@gmail.com>
10 * This file is part of FFmpeg.
12 * FFmpeg is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public
14 * License as published by the Free Software Foundation; either
15 * version 2.1 of the License, or (at your option) any later version.
17 * FFmpeg is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with FFmpeg; if not, write to the Free Software
24 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
32 #include "libavutil/channel_layout.h"
33 #include "libavutil/crc.h"
34 #include "libavutil/downmix_info.h"
35 #include "libavutil/opt.h"
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
67 /** dynamic range table. converts codes to scale factors. */
68 static float dynamic_range_tab[256];
69 float ff_ac3_heavy_dynamic_range_tab[256];
72 /** Adjustments in dB gain */
73 static const float gain_levels[9] = {
77 LEVEL_MINUS_1POINT5DB,
79 LEVEL_MINUS_4POINT5DB,
85 /** Adjustments in dB gain (LFE, +10 to -21 dB) */
86 static const float gain_levels_lfe[32] = {
87 3.162275, 2.818382, 2.511886, 2.238719, 1.995261, 1.778278, 1.584893,
88 1.412536, 1.258924, 1.122018, 1.000000, 0.891251, 0.794328, 0.707946,
89 0.630957, 0.562341, 0.501187, 0.446683, 0.398107, 0.354813, 0.316227,
90 0.281838, 0.251188, 0.223872, 0.199526, 0.177828, 0.158489, 0.141253,
91 0.125892, 0.112201, 0.100000, 0.089125
95 * Table for default stereo downmixing coefficients
96 * reference: Section 7.8.2 Downmixing Into Two Channels
98 static const uint8_t ac3_default_coeffs[8][5][2] = {
99 { { 2, 7 }, { 7, 2 }, },
101 { { 2, 7 }, { 7, 2 }, },
102 { { 2, 7 }, { 5, 5 }, { 7, 2 }, },
103 { { 2, 7 }, { 7, 2 }, { 6, 6 }, },
104 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 8, 8 }, },
105 { { 2, 7 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
106 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
110 * Symmetrical Dequantization
111 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
112 * Tables 7.19 to 7.23
115 symmetric_dequant(int code, int levels)
117 return ((code - (levels >> 1)) * (1 << 24)) / levels;
121 * Initialize tables at runtime.
123 static av_cold void ac3_tables_init(void)
127 /* generate table for ungrouping 3 values in 7 bits
128 reference: Section 7.1.3 Exponent Decoding */
129 for (i = 0; i < 128; i++) {
130 ungroup_3_in_7_bits_tab[i][0] = i / 25;
131 ungroup_3_in_7_bits_tab[i][1] = (i % 25) / 5;
132 ungroup_3_in_7_bits_tab[i][2] = (i % 25) % 5;
135 /* generate grouped mantissa tables
136 reference: Section 7.3.5 Ungrouping of Mantissas */
137 for (i = 0; i < 32; i++) {
138 /* bap=1 mantissas */
139 b1_mantissas[i][0] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][0], 3);
140 b1_mantissas[i][1] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][1], 3);
141 b1_mantissas[i][2] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][2], 3);
143 for (i = 0; i < 128; i++) {
144 /* bap=2 mantissas */
145 b2_mantissas[i][0] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][0], 5);
146 b2_mantissas[i][1] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][1], 5);
147 b2_mantissas[i][2] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][2], 5);
149 /* bap=4 mantissas */
150 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
151 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
153 /* generate ungrouped mantissa tables
154 reference: Tables 7.21 and 7.23 */
155 for (i = 0; i < 7; i++) {
156 /* bap=3 mantissas */
157 b3_mantissas[i] = symmetric_dequant(i, 7);
159 for (i = 0; i < 15; i++) {
160 /* bap=5 mantissas */
161 b5_mantissas[i] = symmetric_dequant(i, 15);
165 /* generate dynamic range table
166 reference: Section 7.7.1 Dynamic Range Control */
167 for (i = 0; i < 256; i++) {
168 int v = (i >> 5) - ((i >> 7) << 3) - 5;
169 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
172 /* generate compr dynamic range table
173 reference: Section 7.7.2 Heavy Compression */
174 for (i = 0; i < 256; i++) {
175 int v = (i >> 4) - ((i >> 7) << 4) - 4;
176 ff_ac3_heavy_dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0xF) | 0x10);
182 * AVCodec initialization
184 static av_cold int ac3_decode_init(AVCodecContext *avctx)
186 AC3DecodeContext *s = avctx->priv_data;
192 ff_mdct_init(&s->imdct_256, 8, 1, 1.0);
193 ff_mdct_init(&s->imdct_512, 9, 1, 1.0);
194 AC3_RENAME(ff_kbd_window_init)(s->window, 5.0, 256);
195 ff_bswapdsp_init(&s->bdsp);
198 s->fdsp = avpriv_alloc_fixed_dsp(avctx->flags & AV_CODEC_FLAG_BITEXACT);
200 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
201 ff_fmt_convert_init(&s->fmt_conv, avctx);
204 ff_ac3dsp_init(&s->ac3dsp, avctx->flags & AV_CODEC_FLAG_BITEXACT);
205 av_lfg_init(&s->dith_state, 0);
208 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
210 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
212 /* allow downmixing to stereo or mono */
213 if (avctx->channels > 1 &&
214 avctx->request_channel_layout == AV_CH_LAYOUT_MONO)
216 else if (avctx->channels > 2 &&
217 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO)
221 for (i = 0; i < AC3_MAX_CHANNELS; i++) {
222 s->xcfptr[i] = s->transform_coeffs[i];
223 s->dlyptr[i] = s->delay[i];
230 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
231 * GetBitContext within AC3DecodeContext must point to
232 * the start of the synchronized AC-3 bitstream.
234 static int ac3_parse_header(AC3DecodeContext *s)
236 GetBitContext *gbc = &s->gbc;
239 /* read the rest of the bsi. read twice for dual mono mode. */
240 i = !s->channel_mode;
242 s->dialog_normalization[(!s->channel_mode)-i] = -get_bits(gbc, 5);
243 if (s->dialog_normalization[(!s->channel_mode)-i] == 0) {
244 s->dialog_normalization[(!s->channel_mode)-i] = -31;
246 if (s->target_level != 0) {
247 s->level_gain[(!s->channel_mode)-i] = powf(2.0f,
248 (float)(s->target_level -
249 s->dialog_normalization[(!s->channel_mode)-i])/6.0f);
251 if (s->compression_exists[(!s->channel_mode)-i] = get_bits1(gbc)) {
252 s->heavy_dynamic_range[(!s->channel_mode)-i] =
253 AC3_HEAVY_RANGE(get_bits(gbc, 8));
256 skip_bits(gbc, 8); //skip language code
258 skip_bits(gbc, 7); //skip audio production information
261 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
263 /* skip the timecodes or parse the Alternate Bit Stream Syntax */
264 if (s->bitstream_id != 6) {
266 skip_bits(gbc, 14); //skip timecode1
268 skip_bits(gbc, 14); //skip timecode2
270 if (get_bits1(gbc)) {
271 s->preferred_downmix = get_bits(gbc, 2);
272 s->center_mix_level_ltrt = get_bits(gbc, 3);
273 s->surround_mix_level_ltrt = av_clip(get_bits(gbc, 3), 3, 7);
274 s->center_mix_level = get_bits(gbc, 3);
275 s->surround_mix_level = av_clip(get_bits(gbc, 3), 3, 7);
277 if (get_bits1(gbc)) {
278 s->dolby_surround_ex_mode = get_bits(gbc, 2);
279 s->dolby_headphone_mode = get_bits(gbc, 2);
280 skip_bits(gbc, 10); // skip adconvtyp (1), xbsi2 (8), encinfo (1)
284 /* skip additional bitstream info */
285 if (get_bits1(gbc)) {
286 i = get_bits(gbc, 6);
296 * Common function to parse AC-3 or E-AC-3 frame header
298 static int parse_frame_header(AC3DecodeContext *s)
300 AC3HeaderInfo hdr, *phdr=&hdr;
303 err = avpriv_ac3_parse_header(&s->gbc, &phdr);
307 /* get decoding parameters from header info */
308 s->bit_alloc_params.sr_code = hdr.sr_code;
309 s->bitstream_id = hdr.bitstream_id;
310 s->bitstream_mode = hdr.bitstream_mode;
311 s->channel_mode = hdr.channel_mode;
312 s->lfe_on = hdr.lfe_on;
313 s->bit_alloc_params.sr_shift = hdr.sr_shift;
314 s->sample_rate = hdr.sample_rate;
315 s->bit_rate = hdr.bit_rate;
316 s->channels = hdr.channels;
317 s->fbw_channels = s->channels - s->lfe_on;
318 s->lfe_ch = s->fbw_channels + 1;
319 s->frame_size = hdr.frame_size;
320 s->preferred_downmix = AC3_DMIXMOD_NOTINDICATED;
321 s->center_mix_level = hdr.center_mix_level;
322 s->center_mix_level_ltrt = 4; // -3.0dB
323 s->surround_mix_level = hdr.surround_mix_level;
324 s->surround_mix_level_ltrt = 4; // -3.0dB
325 s->lfe_mix_level_exists = 0;
326 s->num_blocks = hdr.num_blocks;
327 s->frame_type = hdr.frame_type;
328 s->substreamid = hdr.substreamid;
329 s->dolby_surround_mode = hdr.dolby_surround_mode;
330 s->dolby_surround_ex_mode = AC3_DSUREXMOD_NOTINDICATED;
331 s->dolby_headphone_mode = AC3_DHEADPHONMOD_NOTINDICATED;
334 s->start_freq[s->lfe_ch] = 0;
335 s->end_freq[s->lfe_ch] = 7;
336 s->num_exp_groups[s->lfe_ch] = 2;
337 s->channel_in_cpl[s->lfe_ch] = 0;
340 if (s->bitstream_id <= 10) {
342 s->snr_offset_strategy = 2;
343 s->block_switch_syntax = 1;
344 s->dither_flag_syntax = 1;
345 s->bit_allocation_syntax = 1;
346 s->fast_gain_syntax = 0;
347 s->first_cpl_leak = 0;
350 memset(s->channel_uses_aht, 0, sizeof(s->channel_uses_aht));
351 return ac3_parse_header(s);
352 } else if (CONFIG_EAC3_DECODER) {
354 return ff_eac3_parse_header(s);
356 av_log(s->avctx, AV_LOG_ERROR, "E-AC-3 support not compiled in\n");
357 return AVERROR(ENOSYS);
362 * Set stereo downmixing coefficients based on frame header info.
363 * reference: Section 7.8.2 Downmixing Into Two Channels
365 static void set_downmix_coeffs(AC3DecodeContext *s)
368 float cmix = gain_levels[s-> center_mix_level];
369 float smix = gain_levels[s->surround_mix_level];
371 float downmix_coeffs[AC3_MAX_CHANNELS][2];
373 for (i = 0; i < s->fbw_channels; i++) {
374 downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
375 downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
377 if (s->channel_mode > 1 && s->channel_mode & 1) {
378 downmix_coeffs[1][0] = downmix_coeffs[1][1] = cmix;
380 if (s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
381 int nf = s->channel_mode - 2;
382 downmix_coeffs[nf][0] = downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
384 if (s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
385 int nf = s->channel_mode - 4;
386 downmix_coeffs[nf][0] = downmix_coeffs[nf+1][1] = smix;
391 for (i = 0; i < s->fbw_channels; i++) {
392 norm0 += downmix_coeffs[i][0];
393 norm1 += downmix_coeffs[i][1];
395 norm0 = 1.0f / norm0;
396 norm1 = 1.0f / norm1;
397 for (i = 0; i < s->fbw_channels; i++) {
398 downmix_coeffs[i][0] *= norm0;
399 downmix_coeffs[i][1] *= norm1;
402 if (s->output_mode == AC3_CHMODE_MONO) {
403 for (i = 0; i < s->fbw_channels; i++)
404 downmix_coeffs[i][0] = (downmix_coeffs[i][0] +
405 downmix_coeffs[i][1]) * LEVEL_MINUS_3DB;
407 for (i = 0; i < s->fbw_channels; i++) {
408 s->downmix_coeffs[i][0] = FIXR12(downmix_coeffs[i][0]);
409 s->downmix_coeffs[i][1] = FIXR12(downmix_coeffs[i][1]);
414 * Decode the grouped exponents according to exponent strategy.
415 * reference: Section 7.1.3 Exponent Decoding
417 static int decode_exponents(AC3DecodeContext *s,
418 GetBitContext *gbc, int exp_strategy, int ngrps,
419 uint8_t absexp, int8_t *dexps)
421 int i, j, grp, group_size;
426 group_size = exp_strategy + (exp_strategy == EXP_D45);
427 for (grp = 0, i = 0; grp < ngrps; grp++) {
428 expacc = get_bits(gbc, 7);
429 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][0];
430 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][1];
431 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][2];
434 /* convert to absolute exps and expand groups */
436 for (i = 0, j = 0; i < ngrps * 3; i++) {
437 prevexp += dexp[i] - 2;
439 av_log(s->avctx, AV_LOG_ERROR, "exponent %d is out-of-range\n", prevexp);
442 switch (group_size) {
443 case 4: dexps[j++] = prevexp;
444 dexps[j++] = prevexp;
445 case 2: dexps[j++] = prevexp;
446 case 1: dexps[j++] = prevexp;
453 * Generate transform coefficients for each coupled channel in the coupling
454 * range using the coupling coefficients and coupling coordinates.
455 * reference: Section 7.4.3 Coupling Coordinate Format
457 static void calc_transform_coeffs_cpl(AC3DecodeContext *s)
461 bin = s->start_freq[CPL_CH];
462 for (band = 0; band < s->num_cpl_bands; band++) {
463 int band_start = bin;
464 int band_end = bin + s->cpl_band_sizes[band];
465 for (ch = 1; ch <= s->fbw_channels; ch++) {
466 if (s->channel_in_cpl[ch]) {
467 int cpl_coord = s->cpl_coords[ch][band] << 5;
468 for (bin = band_start; bin < band_end; bin++) {
469 s->fixed_coeffs[ch][bin] =
470 MULH(s->fixed_coeffs[CPL_CH][bin] * (1 << 4), cpl_coord);
472 if (ch == 2 && s->phase_flags[band]) {
473 for (bin = band_start; bin < band_end; bin++)
474 s->fixed_coeffs[2][bin] = -s->fixed_coeffs[2][bin];
483 * Grouped mantissas for 3-level 5-level and 11-level quantization
485 typedef struct mant_groups {
495 * Decode the transform coefficients for a particular channel
496 * reference: Section 7.3 Quantization and Decoding of Mantissas
498 static void ac3_decode_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
500 int start_freq = s->start_freq[ch_index];
501 int end_freq = s->end_freq[ch_index];
502 uint8_t *baps = s->bap[ch_index];
503 int8_t *exps = s->dexps[ch_index];
504 int32_t *coeffs = s->fixed_coeffs[ch_index];
505 int dither = (ch_index == CPL_CH) || s->dither_flag[ch_index];
506 GetBitContext *gbc = &s->gbc;
509 for (freq = start_freq; freq < end_freq; freq++) {
510 int bap = baps[freq];
514 /* random noise with approximate range of -0.707 to 0.707 */
516 mantissa = (((av_lfg_get(&s->dith_state)>>8)*181)>>8) - 5931008;
523 mantissa = m->b1_mant[m->b1];
525 int bits = get_bits(gbc, 5);
526 mantissa = b1_mantissas[bits][0];
527 m->b1_mant[1] = b1_mantissas[bits][1];
528 m->b1_mant[0] = b1_mantissas[bits][2];
535 mantissa = m->b2_mant[m->b2];
537 int bits = get_bits(gbc, 7);
538 mantissa = b2_mantissas[bits][0];
539 m->b2_mant[1] = b2_mantissas[bits][1];
540 m->b2_mant[0] = b2_mantissas[bits][2];
545 mantissa = b3_mantissas[get_bits(gbc, 3)];
550 mantissa = m->b4_mant;
552 int bits = get_bits(gbc, 7);
553 mantissa = b4_mantissas[bits][0];
554 m->b4_mant = b4_mantissas[bits][1];
559 mantissa = b5_mantissas[get_bits(gbc, 4)];
561 default: /* 6 to 15 */
562 /* Shift mantissa and sign-extend it. */
564 av_log(s->avctx, AV_LOG_ERROR, "bap %d is invalid in plain AC-3\n", bap);
567 mantissa = (unsigned)get_sbits(gbc, quantization_tab[bap]) << (24 - quantization_tab[bap]);
570 coeffs[freq] = mantissa >> exps[freq];
575 * Remove random dithering from coupling range coefficients with zero-bit
576 * mantissas for coupled channels which do not use dithering.
577 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
579 static void remove_dithering(AC3DecodeContext *s) {
582 for (ch = 1; ch <= s->fbw_channels; ch++) {
583 if (!s->dither_flag[ch] && s->channel_in_cpl[ch]) {
584 for (i = s->start_freq[CPL_CH]; i < s->end_freq[CPL_CH]; i++) {
585 if (!s->bap[CPL_CH][i])
586 s->fixed_coeffs[ch][i] = 0;
592 static void decode_transform_coeffs_ch(AC3DecodeContext *s, int blk, int ch,
595 if (!s->channel_uses_aht[ch]) {
596 ac3_decode_transform_coeffs_ch(s, ch, m);
598 /* if AHT is used, mantissas for all blocks are encoded in the first
599 block of the frame. */
601 if (CONFIG_EAC3_DECODER && !blk)
602 ff_eac3_decode_transform_coeffs_aht_ch(s, ch);
603 for (bin = s->start_freq[ch]; bin < s->end_freq[ch]; bin++) {
604 s->fixed_coeffs[ch][bin] = s->pre_mantissa[ch][bin][blk] >> s->dexps[ch][bin];
610 * Decode the transform coefficients.
612 static void decode_transform_coeffs(AC3DecodeContext *s, int blk)
618 m.b1 = m.b2 = m.b4 = 0;
620 for (ch = 1; ch <= s->channels; ch++) {
621 /* transform coefficients for full-bandwidth channel */
622 decode_transform_coeffs_ch(s, blk, ch, &m);
623 /* transform coefficients for coupling channel come right after the
624 coefficients for the first coupled channel*/
625 if (s->channel_in_cpl[ch]) {
627 decode_transform_coeffs_ch(s, blk, CPL_CH, &m);
628 calc_transform_coeffs_cpl(s);
631 end = s->end_freq[CPL_CH];
633 end = s->end_freq[ch];
636 s->fixed_coeffs[ch][end] = 0;
640 /* zero the dithered coefficients for appropriate channels */
645 * Stereo rematrixing.
646 * reference: Section 7.5.4 Rematrixing : Decoding Technique
648 static void do_rematrixing(AC3DecodeContext *s)
653 end = FFMIN(s->end_freq[1], s->end_freq[2]);
655 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++) {
656 if (s->rematrixing_flags[bnd]) {
657 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd + 1]);
658 for (i = ff_ac3_rematrix_band_tab[bnd]; i < bndend; i++) {
659 int tmp0 = s->fixed_coeffs[1][i];
660 s->fixed_coeffs[1][i] += s->fixed_coeffs[2][i];
661 s->fixed_coeffs[2][i] = tmp0 - s->fixed_coeffs[2][i];
668 * Inverse MDCT Transform.
669 * Convert frequency domain coefficients to time-domain audio samples.
670 * reference: Section 7.9.4 Transformation Equations
672 static inline void do_imdct(AC3DecodeContext *s, int channels)
676 for (ch = 1; ch <= channels; ch++) {
677 if (s->block_switch[ch]) {
679 FFTSample *x = s->tmp_output + 128;
680 for (i = 0; i < 128; i++)
681 x[i] = s->transform_coeffs[ch][2 * i];
682 s->imdct_256.imdct_half(&s->imdct_256, s->tmp_output, x);
684 s->fdsp->vector_fmul_window_scaled(s->outptr[ch - 1], s->delay[ch - 1],
685 s->tmp_output, s->window, 128, 8);
687 s->fdsp->vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1],
688 s->tmp_output, s->window, 128);
690 for (i = 0; i < 128; i++)
691 x[i] = s->transform_coeffs[ch][2 * i + 1];
692 s->imdct_256.imdct_half(&s->imdct_256, s->delay[ch - 1], x);
694 s->imdct_512.imdct_half(&s->imdct_512, s->tmp_output, s->transform_coeffs[ch]);
696 s->fdsp->vector_fmul_window_scaled(s->outptr[ch - 1], s->delay[ch - 1],
697 s->tmp_output, s->window, 128, 8);
699 s->fdsp->vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1],
700 s->tmp_output, s->window, 128);
702 memcpy(s->delay[ch - 1], s->tmp_output + 128, 128 * sizeof(FFTSample));
708 * Upmix delay samples from stereo to original channel layout.
710 static void ac3_upmix_delay(AC3DecodeContext *s)
712 int channel_data_size = sizeof(s->delay[0]);
713 switch (s->channel_mode) {
714 case AC3_CHMODE_DUALMONO:
715 case AC3_CHMODE_STEREO:
716 /* upmix mono to stereo */
717 memcpy(s->delay[1], s->delay[0], channel_data_size);
719 case AC3_CHMODE_2F2R:
720 memset(s->delay[3], 0, channel_data_size);
721 case AC3_CHMODE_2F1R:
722 memset(s->delay[2], 0, channel_data_size);
724 case AC3_CHMODE_3F2R:
725 memset(s->delay[4], 0, channel_data_size);
726 case AC3_CHMODE_3F1R:
727 memset(s->delay[3], 0, channel_data_size);
729 memcpy(s->delay[2], s->delay[1], channel_data_size);
730 memset(s->delay[1], 0, channel_data_size);
736 * Decode band structure for coupling, spectral extension, or enhanced coupling.
737 * The band structure defines how many subbands are in each band. For each
738 * subband in the range, 1 means it is combined with the previous band, and 0
739 * means that it starts a new band.
741 * @param[in] gbc bit reader context
742 * @param[in] blk block number
743 * @param[in] eac3 flag to indicate E-AC-3
744 * @param[in] ecpl flag to indicate enhanced coupling
745 * @param[in] start_subband subband number for start of range
746 * @param[in] end_subband subband number for end of range
747 * @param[in] default_band_struct default band structure table
748 * @param[out] num_bands number of bands (optionally NULL)
749 * @param[out] band_sizes array containing the number of bins in each band (optionally NULL)
751 static void decode_band_structure(GetBitContext *gbc, int blk, int eac3,
752 int ecpl, int start_subband, int end_subband,
753 const uint8_t *default_band_struct,
754 int *num_bands, uint8_t *band_sizes)
756 int subbnd, bnd, n_subbands, n_bands=0;
758 uint8_t coded_band_struct[22];
759 const uint8_t *band_struct;
761 n_subbands = end_subband - start_subband;
763 /* decode band structure from bitstream or use default */
764 if (!eac3 || get_bits1(gbc)) {
765 for (subbnd = 0; subbnd < n_subbands - 1; subbnd++) {
766 coded_band_struct[subbnd] = get_bits1(gbc);
768 band_struct = coded_band_struct;
770 band_struct = &default_band_struct[start_subband+1];
772 /* no change in band structure */
776 /* calculate number of bands and band sizes based on band structure.
777 note that the first 4 subbands in enhanced coupling span only 6 bins
779 if (num_bands || band_sizes ) {
780 n_bands = n_subbands;
781 bnd_sz[0] = ecpl ? 6 : 12;
782 for (bnd = 0, subbnd = 1; subbnd < n_subbands; subbnd++) {
783 int subbnd_size = (ecpl && subbnd < 4) ? 6 : 12;
784 if (band_struct[subbnd - 1]) {
786 bnd_sz[bnd] += subbnd_size;
788 bnd_sz[++bnd] = subbnd_size;
793 /* set optional output params */
795 *num_bands = n_bands;
797 memcpy(band_sizes, bnd_sz, n_bands);
801 * Decode a single audio block from the AC-3 bitstream.
803 static int decode_audio_block(AC3DecodeContext *s, int blk)
805 int fbw_channels = s->fbw_channels;
806 int channel_mode = s->channel_mode;
808 int different_transforms;
811 GetBitContext *gbc = &s->gbc;
812 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS] = { 0 };
814 /* block switch flags */
815 different_transforms = 0;
816 if (s->block_switch_syntax) {
817 for (ch = 1; ch <= fbw_channels; ch++) {
818 s->block_switch[ch] = get_bits1(gbc);
819 if (ch > 1 && s->block_switch[ch] != s->block_switch[1])
820 different_transforms = 1;
824 /* dithering flags */
825 if (s->dither_flag_syntax) {
826 for (ch = 1; ch <= fbw_channels; ch++) {
827 s->dither_flag[ch] = get_bits1(gbc);
832 i = !s->channel_mode;
834 if (get_bits1(gbc)) {
835 /* Allow asymmetric application of DRC when drc_scale > 1.
836 Amplification of quiet sounds is enhanced */
837 int range_bits = get_bits(gbc, 8);
838 INTFLOAT range = AC3_RANGE(range_bits);
839 if (range_bits <= 127 || s->drc_scale <= 1.0)
840 s->dynamic_range[i] = AC3_DYNAMIC_RANGE(range);
842 s->dynamic_range[i] = range;
843 } else if (blk == 0) {
844 s->dynamic_range[i] = AC3_DYNAMIC_RANGE1;
848 /* spectral extension strategy */
849 if (s->eac3 && (!blk || get_bits1(gbc))) {
850 s->spx_in_use = get_bits1(gbc);
852 int dst_start_freq, dst_end_freq, src_start_freq,
853 start_subband, end_subband;
855 /* determine which channels use spx */
856 if (s->channel_mode == AC3_CHMODE_MONO) {
857 s->channel_uses_spx[1] = 1;
859 for (ch = 1; ch <= fbw_channels; ch++)
860 s->channel_uses_spx[ch] = get_bits1(gbc);
863 /* get the frequency bins of the spx copy region and the spx start
865 dst_start_freq = get_bits(gbc, 2);
866 start_subband = get_bits(gbc, 3) + 2;
867 if (start_subband > 7)
868 start_subband += start_subband - 7;
869 end_subband = get_bits(gbc, 3) + 5;
871 s->spx_dst_end_freq = end_freq_inv_tab[end_subband-5];
874 end_subband += end_subband - 7;
875 dst_start_freq = dst_start_freq * 12 + 25;
876 src_start_freq = start_subband * 12 + 25;
877 dst_end_freq = end_subband * 12 + 25;
879 /* check validity of spx ranges */
880 if (start_subband >= end_subband) {
881 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
882 "range (%d >= %d)\n", start_subband, end_subband);
883 return AVERROR_INVALIDDATA;
885 if (dst_start_freq >= src_start_freq) {
886 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
887 "copy start bin (%d >= %d)\n", dst_start_freq, src_start_freq);
888 return AVERROR_INVALIDDATA;
891 s->spx_dst_start_freq = dst_start_freq;
892 s->spx_src_start_freq = src_start_freq;
894 s->spx_dst_end_freq = dst_end_freq;
896 decode_band_structure(gbc, blk, s->eac3, 0,
897 start_subband, end_subband,
898 ff_eac3_default_spx_band_struct,
903 if (!s->eac3 || !s->spx_in_use) {
905 for (ch = 1; ch <= fbw_channels; ch++) {
906 s->channel_uses_spx[ch] = 0;
907 s->first_spx_coords[ch] = 1;
911 /* spectral extension coordinates */
913 for (ch = 1; ch <= fbw_channels; ch++) {
914 if (s->channel_uses_spx[ch]) {
915 if (s->first_spx_coords[ch] || get_bits1(gbc)) {
917 int bin, master_spx_coord;
919 s->first_spx_coords[ch] = 0;
920 spx_blend = AC3_SPX_BLEND(get_bits(gbc, 5));
921 master_spx_coord = get_bits(gbc, 2) * 3;
923 bin = s->spx_src_start_freq;
924 for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
925 int bandsize = s->spx_band_sizes[bnd];
926 int spx_coord_exp, spx_coord_mant;
927 INTFLOAT nratio, sblend, nblend;
929 /* calculate blending factors */
930 int64_t accu = ((bin << 23) + (bandsize << 22))
931 * (int64_t)s->spx_dst_end_freq;
932 nratio = (int)(accu >> 32);
933 nratio -= spx_blend << 18;
938 } else if (nratio > 0x7fffff) {
939 nblend = 14529495; // sqrt(3) in FP.23
942 nblend = fixed_sqrt(nratio, 23);
943 accu = (int64_t)nblend * 1859775393;
944 nblend = (int)((accu + (1<<29)) >> 30);
945 sblend = fixed_sqrt(0x800000 - nratio, 23);
950 /* calculate blending factors */
951 nratio = ((float)((bin + (bandsize >> 1))) / s->spx_dst_end_freq) - spx_blend;
952 nratio = av_clipf(nratio, 0.0f, 1.0f);
953 nblend = sqrtf(3.0f * nratio); // noise is scaled by sqrt(3)
954 // to give unity variance
955 sblend = sqrtf(1.0f - nratio);
959 /* decode spx coordinates */
960 spx_coord_exp = get_bits(gbc, 4);
961 spx_coord_mant = get_bits(gbc, 2);
962 if (spx_coord_exp == 15) spx_coord_mant <<= 1;
963 else spx_coord_mant += 4;
964 spx_coord_mant <<= (25 - spx_coord_exp - master_spx_coord);
966 /* multiply noise and signal blending factors by spx coordinate */
968 accu = (int64_t)nblend * spx_coord_mant;
969 s->spx_noise_blend[ch][bnd] = (int)((accu + (1<<22)) >> 23);
970 accu = (int64_t)sblend * spx_coord_mant;
971 s->spx_signal_blend[ch][bnd] = (int)((accu + (1<<22)) >> 23);
973 spx_coord = spx_coord_mant * (1.0f / (1 << 23));
974 s->spx_noise_blend [ch][bnd] = nblend * spx_coord;
975 s->spx_signal_blend[ch][bnd] = sblend * spx_coord;
980 s->first_spx_coords[ch] = 1;
985 /* coupling strategy */
986 if (s->eac3 ? s->cpl_strategy_exists[blk] : get_bits1(gbc)) {
987 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
989 s->cpl_in_use[blk] = get_bits1(gbc);
990 if (s->cpl_in_use[blk]) {
991 /* coupling in use */
992 int cpl_start_subband, cpl_end_subband;
994 if (channel_mode < AC3_CHMODE_STEREO) {
995 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
996 return AVERROR_INVALIDDATA;
999 /* check for enhanced coupling */
1000 if (s->eac3 && get_bits1(gbc)) {
1001 /* TODO: parse enhanced coupling strategy info */
1002 avpriv_request_sample(s->avctx, "Enhanced coupling");
1003 return AVERROR_PATCHWELCOME;
1006 /* determine which channels are coupled */
1007 if (s->eac3 && s->channel_mode == AC3_CHMODE_STEREO) {
1008 s->channel_in_cpl[1] = 1;
1009 s->channel_in_cpl[2] = 1;
1011 for (ch = 1; ch <= fbw_channels; ch++)
1012 s->channel_in_cpl[ch] = get_bits1(gbc);
1015 /* phase flags in use */
1016 if (channel_mode == AC3_CHMODE_STEREO)
1017 s->phase_flags_in_use = get_bits1(gbc);
1019 /* coupling frequency range */
1020 cpl_start_subband = get_bits(gbc, 4);
1021 cpl_end_subband = s->spx_in_use ? (s->spx_src_start_freq - 37) / 12 :
1022 get_bits(gbc, 4) + 3;
1023 if (cpl_start_subband >= cpl_end_subband) {
1024 av_log(s->avctx, AV_LOG_ERROR, "invalid coupling range (%d >= %d)\n",
1025 cpl_start_subband, cpl_end_subband);
1026 return AVERROR_INVALIDDATA;
1028 s->start_freq[CPL_CH] = cpl_start_subband * 12 + 37;
1029 s->end_freq[CPL_CH] = cpl_end_subband * 12 + 37;
1031 decode_band_structure(gbc, blk, s->eac3, 0, cpl_start_subband,
1033 ff_eac3_default_cpl_band_struct,
1034 &s->num_cpl_bands, s->cpl_band_sizes);
1036 /* coupling not in use */
1037 for (ch = 1; ch <= fbw_channels; ch++) {
1038 s->channel_in_cpl[ch] = 0;
1039 s->first_cpl_coords[ch] = 1;
1041 s->first_cpl_leak = s->eac3;
1042 s->phase_flags_in_use = 0;
1044 } else if (!s->eac3) {
1046 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must "
1047 "be present in block 0\n");
1048 return AVERROR_INVALIDDATA;
1050 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
1053 cpl_in_use = s->cpl_in_use[blk];
1055 /* coupling coordinates */
1057 int cpl_coords_exist = 0;
1059 for (ch = 1; ch <= fbw_channels; ch++) {
1060 if (s->channel_in_cpl[ch]) {
1061 if ((s->eac3 && s->first_cpl_coords[ch]) || get_bits1(gbc)) {
1062 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
1063 s->first_cpl_coords[ch] = 0;
1064 cpl_coords_exist = 1;
1065 master_cpl_coord = 3 * get_bits(gbc, 2);
1066 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
1067 cpl_coord_exp = get_bits(gbc, 4);
1068 cpl_coord_mant = get_bits(gbc, 4);
1069 if (cpl_coord_exp == 15)
1070 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
1072 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
1073 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
1076 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must "
1077 "be present in block 0\n");
1078 return AVERROR_INVALIDDATA;
1081 /* channel not in coupling */
1082 s->first_cpl_coords[ch] = 1;
1086 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
1087 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
1088 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
1093 /* stereo rematrixing strategy and band structure */
1094 if (channel_mode == AC3_CHMODE_STEREO) {
1095 if ((s->eac3 && !blk) || get_bits1(gbc)) {
1096 s->num_rematrixing_bands = 4;
1097 if (cpl_in_use && s->start_freq[CPL_CH] <= 61) {
1098 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
1099 } else if (s->spx_in_use && s->spx_src_start_freq <= 61) {
1100 s->num_rematrixing_bands--;
1102 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++)
1103 s->rematrixing_flags[bnd] = get_bits1(gbc);
1105 av_log(s->avctx, AV_LOG_WARNING, "Warning: "
1106 "new rematrixing strategy not present in block 0\n");
1107 s->num_rematrixing_bands = 0;
1111 /* exponent strategies for each channel */
1112 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1114 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
1115 if (s->exp_strategy[blk][ch] != EXP_REUSE)
1116 bit_alloc_stages[ch] = 3;
1119 /* channel bandwidth */
1120 for (ch = 1; ch <= fbw_channels; ch++) {
1121 s->start_freq[ch] = 0;
1122 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1124 int prev = s->end_freq[ch];
1125 if (s->channel_in_cpl[ch])
1126 s->end_freq[ch] = s->start_freq[CPL_CH];
1127 else if (s->channel_uses_spx[ch])
1128 s->end_freq[ch] = s->spx_src_start_freq;
1130 int bandwidth_code = get_bits(gbc, 6);
1131 if (bandwidth_code > 60) {
1132 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60\n", bandwidth_code);
1133 return AVERROR_INVALIDDATA;
1135 s->end_freq[ch] = bandwidth_code * 3 + 73;
1137 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
1138 s->num_exp_groups[ch] = (s->end_freq[ch] + group_size-4) / group_size;
1139 if (blk > 0 && s->end_freq[ch] != prev)
1140 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
1143 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
1144 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
1145 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
1148 /* decode exponents for each channel */
1149 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1150 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1151 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
1152 if (decode_exponents(s, gbc, s->exp_strategy[blk][ch],
1153 s->num_exp_groups[ch], s->dexps[ch][0],
1154 &s->dexps[ch][s->start_freq[ch]+!!ch])) {
1155 return AVERROR_INVALIDDATA;
1157 if (ch != CPL_CH && ch != s->lfe_ch)
1158 skip_bits(gbc, 2); /* skip gainrng */
1162 /* bit allocation information */
1163 if (s->bit_allocation_syntax) {
1164 if (get_bits1(gbc)) {
1165 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1166 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1167 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
1168 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
1169 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
1170 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1171 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1173 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must "
1174 "be present in block 0\n");
1175 return AVERROR_INVALIDDATA;
1179 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
1180 if (!s->eac3 || !blk) {
1181 if (s->snr_offset_strategy && get_bits1(gbc)) {
1184 csnr = (get_bits(gbc, 6) - 15) << 4;
1185 for (i = ch = !cpl_in_use; ch <= s->channels; ch++) {
1187 if (ch == i || s->snr_offset_strategy == 2)
1188 snr = (csnr + get_bits(gbc, 4)) << 2;
1189 /* run at least last bit allocation stage if snr offset changes */
1190 if (blk && s->snr_offset[ch] != snr) {
1191 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 1);
1193 s->snr_offset[ch] = snr;
1195 /* fast gain (normal AC-3 only) */
1197 int prev = s->fast_gain[ch];
1198 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1199 /* run last 2 bit allocation stages if fast gain changes */
1200 if (blk && prev != s->fast_gain[ch])
1201 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1204 } else if (!s->eac3 && !blk) {
1205 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
1206 return AVERROR_INVALIDDATA;
1210 /* fast gain (E-AC-3 only) */
1211 if (s->fast_gain_syntax && get_bits1(gbc)) {
1212 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1213 int prev = s->fast_gain[ch];
1214 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1215 /* run last 2 bit allocation stages if fast gain changes */
1216 if (blk && prev != s->fast_gain[ch])
1217 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1219 } else if (s->eac3 && !blk) {
1220 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1221 s->fast_gain[ch] = ff_ac3_fast_gain_tab[4];
1224 /* E-AC-3 to AC-3 converter SNR offset */
1225 if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT && get_bits1(gbc)) {
1226 skip_bits(gbc, 10); // skip converter snr offset
1229 /* coupling leak information */
1231 if (s->first_cpl_leak || get_bits1(gbc)) {
1232 int fl = get_bits(gbc, 3);
1233 int sl = get_bits(gbc, 3);
1234 /* run last 2 bit allocation stages for coupling channel if
1235 coupling leak changes */
1236 if (blk && (fl != s->bit_alloc_params.cpl_fast_leak ||
1237 sl != s->bit_alloc_params.cpl_slow_leak)) {
1238 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
1240 s->bit_alloc_params.cpl_fast_leak = fl;
1241 s->bit_alloc_params.cpl_slow_leak = sl;
1242 } else if (!s->eac3 && !blk) {
1243 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must "
1244 "be present in block 0\n");
1245 return AVERROR_INVALIDDATA;
1247 s->first_cpl_leak = 0;
1250 /* delta bit allocation information */
1251 if (s->dba_syntax && get_bits1(gbc)) {
1252 /* delta bit allocation exists (strategy) */
1253 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1254 s->dba_mode[ch] = get_bits(gbc, 2);
1255 if (s->dba_mode[ch] == DBA_RESERVED) {
1256 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1257 return AVERROR_INVALIDDATA;
1259 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1261 /* channel delta offset, len and bit allocation */
1262 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1263 if (s->dba_mode[ch] == DBA_NEW) {
1264 s->dba_nsegs[ch] = get_bits(gbc, 3) + 1;
1265 for (seg = 0; seg < s->dba_nsegs[ch]; seg++) {
1266 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1267 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1268 s->dba_values[ch][seg] = get_bits(gbc, 3);
1270 /* run last 2 bit allocation stages if new dba values */
1271 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1274 } else if (blk == 0) {
1275 for (ch = 0; ch <= s->channels; ch++) {
1276 s->dba_mode[ch] = DBA_NONE;
1280 /* Bit allocation */
1281 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1282 if (bit_alloc_stages[ch] > 2) {
1283 /* Exponent mapping into PSD and PSD integration */
1284 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1285 s->start_freq[ch], s->end_freq[ch],
1286 s->psd[ch], s->band_psd[ch]);
1288 if (bit_alloc_stages[ch] > 1) {
1289 /* Compute excitation function, Compute masking curve, and
1290 Apply delta bit allocation */
1291 if (ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1292 s->start_freq[ch], s->end_freq[ch],
1293 s->fast_gain[ch], (ch == s->lfe_ch),
1294 s->dba_mode[ch], s->dba_nsegs[ch],
1295 s->dba_offsets[ch], s->dba_lengths[ch],
1296 s->dba_values[ch], s->mask[ch])) {
1297 av_log(s->avctx, AV_LOG_ERROR, "error in bit allocation\n");
1298 return AVERROR_INVALIDDATA;
1301 if (bit_alloc_stages[ch] > 0) {
1302 /* Compute bit allocation */
1303 const uint8_t *bap_tab = s->channel_uses_aht[ch] ?
1304 ff_eac3_hebap_tab : ff_ac3_bap_tab;
1305 s->ac3dsp.bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1306 s->start_freq[ch], s->end_freq[ch],
1308 s->bit_alloc_params.floor,
1309 bap_tab, s->bap[ch]);
1313 /* unused dummy data */
1314 if (s->skip_syntax && get_bits1(gbc)) {
1315 int skipl = get_bits(gbc, 9);
1320 /* unpack the transform coefficients
1321 this also uncouples channels if coupling is in use. */
1322 decode_transform_coeffs(s, blk);
1324 /* TODO: generate enhanced coupling coordinates and uncouple */
1326 /* recover coefficients if rematrixing is in use */
1327 if (s->channel_mode == AC3_CHMODE_STEREO)
1330 /* apply scaling to coefficients (headroom, dynrng) */
1331 for (ch = 1; ch <= s->channels; ch++) {
1332 int audio_channel = 0;
1334 if (s->channel_mode == AC3_CHMODE_DUALMONO)
1335 audio_channel = 2-ch;
1336 if (s->heavy_compression && s->compression_exists[audio_channel])
1337 gain = s->heavy_dynamic_range[audio_channel];
1339 gain = s->dynamic_range[audio_channel];
1342 scale_coefs(s->transform_coeffs[ch], s->fixed_coeffs[ch], gain, 256);
1344 if (s->target_level != 0)
1345 gain = gain * s->level_gain[audio_channel];
1346 gain *= 1.0 / 4194304.0f;
1347 s->fmt_conv.int32_to_float_fmul_scalar(s->transform_coeffs[ch],
1348 s->fixed_coeffs[ch], gain, 256);
1352 /* apply spectral extension to high frequency bins */
1353 if (CONFIG_EAC3_DECODER && s->spx_in_use) {
1354 ff_eac3_apply_spectral_extension(s);
1357 /* downmix and MDCT. order depends on whether block switching is used for
1358 any channel in this block. this is because coefficients for the long
1359 and short transforms cannot be mixed. */
1360 downmix_output = s->channels != s->out_channels &&
1361 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1362 s->fbw_channels == s->out_channels);
1363 if (different_transforms) {
1364 /* the delay samples have already been downmixed, so we upmix the delay
1365 samples in order to reconstruct all channels before downmixing. */
1371 do_imdct(s, s->channels);
1373 if (downmix_output) {
1375 ac3_downmix_c_fixed16(s->outptr, s->downmix_coeffs,
1376 s->out_channels, s->fbw_channels, 256);
1378 s->ac3dsp.downmix(s->outptr, s->downmix_coeffs,
1379 s->out_channels, s->fbw_channels, 256);
1383 if (downmix_output) {
1384 s->ac3dsp.AC3_RENAME(downmix)(s->xcfptr + 1, s->downmix_coeffs,
1385 s->out_channels, s->fbw_channels, 256);
1388 if (downmix_output && !s->downmixed) {
1390 s->ac3dsp.AC3_RENAME(downmix)(s->dlyptr, s->downmix_coeffs,
1391 s->out_channels, s->fbw_channels, 128);
1394 do_imdct(s, s->out_channels);
1401 * Decode a single AC-3 frame.
1403 static int ac3_decode_frame(AVCodecContext * avctx, void *data,
1404 int *got_frame_ptr, AVPacket *avpkt)
1406 AVFrame *frame = data;
1407 const uint8_t *buf = avpkt->data;
1408 int buf_size = avpkt->size;
1409 AC3DecodeContext *s = avctx->priv_data;
1410 int blk, ch, err, ret;
1411 const uint8_t *channel_map;
1412 const SHORTFLOAT *output[AC3_MAX_CHANNELS];
1413 enum AVMatrixEncoding matrix_encoding;
1414 AVDownmixInfo *downmix_info;
1416 /* copy input buffer to decoder context to avoid reading past the end
1417 of the buffer, which can be caused by a damaged input stream. */
1418 if (buf_size >= 2 && AV_RB16(buf) == 0x770B) {
1419 // seems to be byte-swapped AC-3
1420 int cnt = FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE) >> 1;
1421 s->bdsp.bswap16_buf((uint16_t *) s->input_buffer,
1422 (const uint16_t *) buf, cnt);
1424 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE));
1425 buf = s->input_buffer;
1426 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1427 if ((ret = init_get_bits8(&s->gbc, buf, buf_size)) < 0)
1430 /* parse the syncinfo */
1431 err = parse_frame_header(s);
1435 case AAC_AC3_PARSE_ERROR_SYNC:
1436 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1437 return AVERROR_INVALIDDATA;
1438 case AAC_AC3_PARSE_ERROR_BSID:
1439 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1441 case AAC_AC3_PARSE_ERROR_SAMPLE_RATE:
1442 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1444 case AAC_AC3_PARSE_ERROR_FRAME_SIZE:
1445 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1447 case AAC_AC3_PARSE_ERROR_FRAME_TYPE:
1448 /* skip frame if CRC is ok. otherwise use error concealment. */
1449 /* TODO: add support for substreams and dependent frames */
1450 if (s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1451 av_log(avctx, AV_LOG_DEBUG,
1452 "unsupported frame type %d: skipping frame\n",
1457 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1460 case AAC_AC3_PARSE_ERROR_CRC:
1461 case AAC_AC3_PARSE_ERROR_CHANNEL_CFG:
1463 default: // Normal AVERROR do not try to recover.
1468 /* check that reported frame size fits in input buffer */
1469 if (s->frame_size > buf_size) {
1470 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1471 err = AAC_AC3_PARSE_ERROR_FRAME_SIZE;
1472 } else if (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL)) {
1473 /* check for crc mismatch */
1474 if (av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2],
1475 s->frame_size - 2)) {
1476 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1477 if (avctx->err_recognition & AV_EF_EXPLODE)
1478 return AVERROR_INVALIDDATA;
1479 err = AAC_AC3_PARSE_ERROR_CRC;
1484 /* if frame is ok, set audio parameters */
1486 avctx->sample_rate = s->sample_rate;
1487 avctx->bit_rate = s->bit_rate;
1490 /* channel config */
1491 if (!err || (s->channels && s->out_channels != s->channels)) {
1492 s->out_channels = s->channels;
1493 s->output_mode = s->channel_mode;
1495 s->output_mode |= AC3_OUTPUT_LFEON;
1496 if (s->channels > 1 &&
1497 avctx->request_channel_layout == AV_CH_LAYOUT_MONO) {
1498 s->out_channels = 1;
1499 s->output_mode = AC3_CHMODE_MONO;
1500 } else if (s->channels > 2 &&
1501 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
1502 s->out_channels = 2;
1503 s->output_mode = AC3_CHMODE_STEREO;
1506 s->loro_center_mix_level = gain_levels[s-> center_mix_level];
1507 s->loro_surround_mix_level = gain_levels[s->surround_mix_level];
1508 s->ltrt_center_mix_level = LEVEL_MINUS_3DB;
1509 s->ltrt_surround_mix_level = LEVEL_MINUS_3DB;
1510 /* set downmixing coefficients if needed */
1511 if (s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1512 s->fbw_channels == s->out_channels)) {
1513 set_downmix_coeffs(s);
1515 } else if (!s->channels) {
1516 av_log(avctx, AV_LOG_ERROR, "unable to determine channel mode\n");
1517 return AVERROR_INVALIDDATA;
1519 avctx->channels = s->out_channels;
1520 avctx->channel_layout = avpriv_ac3_channel_layout_tab[s->output_mode & ~AC3_OUTPUT_LFEON];
1521 if (s->output_mode & AC3_OUTPUT_LFEON)
1522 avctx->channel_layout |= AV_CH_LOW_FREQUENCY;
1524 /* set audio service type based on bitstream mode for AC-3 */
1525 avctx->audio_service_type = s->bitstream_mode;
1526 if (s->bitstream_mode == 0x7 && s->channels > 1)
1527 avctx->audio_service_type = AV_AUDIO_SERVICE_TYPE_KARAOKE;
1529 /* get output buffer */
1530 frame->nb_samples = s->num_blocks * AC3_BLOCK_SIZE;
1531 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1534 /* decode the audio blocks */
1535 channel_map = ff_ac3_dec_channel_map[s->output_mode & ~AC3_OUTPUT_LFEON][s->lfe_on];
1536 for (ch = 0; ch < AC3_MAX_CHANNELS; ch++) {
1537 output[ch] = s->output[ch];
1538 s->outptr[ch] = s->output[ch];
1540 for (ch = 0; ch < s->channels; ch++) {
1541 if (ch < s->out_channels)
1542 s->outptr[channel_map[ch]] = (SHORTFLOAT *)frame->data[ch];
1544 for (blk = 0; blk < s->num_blocks; blk++) {
1545 if (!err && decode_audio_block(s, blk)) {
1546 av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n");
1550 for (ch = 0; ch < s->out_channels; ch++)
1551 memcpy(((SHORTFLOAT*)frame->data[ch]) + AC3_BLOCK_SIZE*blk, output[ch], AC3_BLOCK_SIZE*sizeof(SHORTFLOAT));
1552 for (ch = 0; ch < s->out_channels; ch++)
1553 output[ch] = s->outptr[channel_map[ch]];
1554 for (ch = 0; ch < s->out_channels; ch++) {
1555 if (!ch || channel_map[ch])
1556 s->outptr[channel_map[ch]] += AC3_BLOCK_SIZE;
1560 av_frame_set_decode_error_flags(frame, err ? FF_DECODE_ERROR_INVALID_BITSTREAM : 0);
1562 /* keep last block for error concealment in next frame */
1563 for (ch = 0; ch < s->out_channels; ch++)
1564 memcpy(s->output[ch], output[ch], AC3_BLOCK_SIZE*sizeof(SHORTFLOAT));
1569 * Check whether the input layout is compatible, and make sure we're not
1570 * downmixing (else the matrix encoding is no longer applicable).
1572 matrix_encoding = AV_MATRIX_ENCODING_NONE;
1573 if (s->channel_mode == AC3_CHMODE_STEREO &&
1574 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1575 if (s->dolby_surround_mode == AC3_DSURMOD_ON)
1576 matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
1577 else if (s->dolby_headphone_mode == AC3_DHEADPHONMOD_ON)
1578 matrix_encoding = AV_MATRIX_ENCODING_DOLBYHEADPHONE;
1579 } else if (s->channel_mode >= AC3_CHMODE_2F2R &&
1580 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1581 switch (s->dolby_surround_ex_mode) {
1582 case AC3_DSUREXMOD_ON: // EX or PLIIx
1583 matrix_encoding = AV_MATRIX_ENCODING_DOLBYEX;
1585 case AC3_DSUREXMOD_PLIIZ:
1586 matrix_encoding = AV_MATRIX_ENCODING_DPLIIZ;
1588 default: // not indicated or off
1592 if ((ret = ff_side_data_update_matrix_encoding(frame, matrix_encoding)) < 0)
1596 if ((downmix_info = av_downmix_info_update_side_data(frame))) {
1597 switch (s->preferred_downmix) {
1598 case AC3_DMIXMOD_LTRT:
1599 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LTRT;
1601 case AC3_DMIXMOD_LORO:
1602 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LORO;
1604 case AC3_DMIXMOD_DPLII:
1605 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_DPLII;
1608 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_UNKNOWN;
1611 downmix_info->center_mix_level = gain_levels[s-> center_mix_level];
1612 downmix_info->center_mix_level_ltrt = gain_levels[s-> center_mix_level_ltrt];
1613 downmix_info->surround_mix_level = gain_levels[s-> surround_mix_level];
1614 downmix_info->surround_mix_level_ltrt = gain_levels[s->surround_mix_level_ltrt];
1615 if (s->lfe_mix_level_exists)
1616 downmix_info->lfe_mix_level = gain_levels_lfe[s->lfe_mix_level];
1618 downmix_info->lfe_mix_level = 0.0; // -inf dB
1620 return AVERROR(ENOMEM);
1624 return FFMIN(buf_size, s->frame_size);
1628 * Uninitialize the AC-3 decoder.
1630 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1632 AC3DecodeContext *s = avctx->priv_data;
1633 ff_mdct_end(&s->imdct_512);
1634 ff_mdct_end(&s->imdct_256);
1640 #define OFFSET(x) offsetof(AC3DecodeContext, x)
1641 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)