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
66 /** dynamic range table. converts codes to scale factors. */
67 static float dynamic_range_tab[256];
68 static float heavy_dynamic_range_tab[256];
70 /** Adjustments in dB gain */
71 static const float gain_levels[9] = {
75 LEVEL_MINUS_1POINT5DB,
77 LEVEL_MINUS_4POINT5DB,
83 /** Adjustments in dB gain (LFE, +10 to -21 dB) */
84 static const float gain_levels_lfe[32] = {
85 3.162275, 2.818382, 2.511886, 2.238719, 1.995261, 1.778278, 1.584893,
86 1.412536, 1.258924, 1.122018, 1.000000, 0.891251, 0.794328, 0.707946,
87 0.630957, 0.562341, 0.501187, 0.446683, 0.398107, 0.354813, 0.316227,
88 0.281838, 0.251188, 0.223872, 0.199526, 0.177828, 0.158489, 0.141253,
89 0.125892, 0.112201, 0.100000, 0.089125
93 * Table for default stereo downmixing coefficients
94 * reference: Section 7.8.2 Downmixing Into Two Channels
96 static const uint8_t ac3_default_coeffs[8][5][2] = {
97 { { 2, 7 }, { 7, 2 }, },
99 { { 2, 7 }, { 7, 2 }, },
100 { { 2, 7 }, { 5, 5 }, { 7, 2 }, },
101 { { 2, 7 }, { 7, 2 }, { 6, 6 }, },
102 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 8, 8 }, },
103 { { 2, 7 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
104 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
108 * Symmetrical Dequantization
109 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
110 * Tables 7.19 to 7.23
113 symmetric_dequant(int code, int levels)
115 return ((code - (levels >> 1)) * (1 << 24)) / levels;
119 * Initialize tables at runtime.
121 static av_cold void ac3_tables_init(void)
125 /* generate table for ungrouping 3 values in 7 bits
126 reference: Section 7.1.3 Exponent Decoding */
127 for (i = 0; i < 128; i++) {
128 ungroup_3_in_7_bits_tab[i][0] = i / 25;
129 ungroup_3_in_7_bits_tab[i][1] = (i % 25) / 5;
130 ungroup_3_in_7_bits_tab[i][2] = (i % 25) % 5;
133 /* generate grouped mantissa tables
134 reference: Section 7.3.5 Ungrouping of Mantissas */
135 for (i = 0; i < 32; i++) {
136 /* bap=1 mantissas */
137 b1_mantissas[i][0] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][0], 3);
138 b1_mantissas[i][1] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][1], 3);
139 b1_mantissas[i][2] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][2], 3);
141 for (i = 0; i < 128; i++) {
142 /* bap=2 mantissas */
143 b2_mantissas[i][0] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][0], 5);
144 b2_mantissas[i][1] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][1], 5);
145 b2_mantissas[i][2] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][2], 5);
147 /* bap=4 mantissas */
148 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
149 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
151 /* generate ungrouped mantissa tables
152 reference: Tables 7.21 and 7.23 */
153 for (i = 0; i < 7; i++) {
154 /* bap=3 mantissas */
155 b3_mantissas[i] = symmetric_dequant(i, 7);
157 for (i = 0; i < 15; i++) {
158 /* bap=5 mantissas */
159 b5_mantissas[i] = symmetric_dequant(i, 15);
162 /* generate dynamic range table
163 reference: Section 7.7.1 Dynamic Range Control */
164 for (i = 0; i < 256; i++) {
165 int v = (i >> 5) - ((i >> 7) << 3) - 5;
166 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
169 /* generate compr dynamic range table
170 reference: Section 7.7.2 Heavy Compression */
171 for (i = 0; i < 256; i++) {
172 int v = (i >> 4) - ((i >> 7) << 4) - 4;
173 heavy_dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0xF) | 0x10);
179 * AVCodec initialization
181 static av_cold int ac3_decode_init(AVCodecContext *avctx)
183 AC3DecodeContext *s = avctx->priv_data;
189 ff_mdct_init(&s->imdct_256, 8, 1, 1.0);
190 ff_mdct_init(&s->imdct_512, 9, 1, 1.0);
191 AC3_RENAME(ff_kbd_window_init)(s->window, 5.0, 256);
192 ff_bswapdsp_init(&s->bdsp);
195 s->fdsp = avpriv_alloc_fixed_dsp(avctx->flags & AV_CODEC_FLAG_BITEXACT);
197 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
198 ff_fmt_convert_init(&s->fmt_conv, avctx);
201 ff_ac3dsp_init(&s->ac3dsp, avctx->flags & AV_CODEC_FLAG_BITEXACT);
202 av_lfg_init(&s->dith_state, 0);
205 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
207 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
209 /* allow downmixing to stereo or mono */
210 if (avctx->channels > 1 &&
211 avctx->request_channel_layout == AV_CH_LAYOUT_MONO)
213 else if (avctx->channels > 2 &&
214 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO)
218 for (i = 0; i < AC3_MAX_CHANNELS; i++) {
219 s->xcfptr[i] = s->transform_coeffs[i];
220 s->dlyptr[i] = s->delay[i];
227 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
228 * GetBitContext within AC3DecodeContext must point to
229 * the start of the synchronized AC-3 bitstream.
231 static int ac3_parse_header(AC3DecodeContext *s)
233 GetBitContext *gbc = &s->gbc;
236 /* read the rest of the bsi. read twice for dual mono mode. */
237 i = !s->channel_mode;
239 s->dialog_normalization[(!s->channel_mode)-i] = -get_bits(gbc, 5);
240 if (s->dialog_normalization[(!s->channel_mode)-i] == 0) {
241 s->dialog_normalization[(!s->channel_mode)-i] = -31;
243 if (s->target_level != 0) {
244 s->level_gain[(!s->channel_mode)-i] = powf(2.0f,
245 (float)(s->target_level -
246 s->dialog_normalization[(!s->channel_mode)-i])/6.0f);
248 if (s->compression_exists[(!s->channel_mode)-i] = get_bits1(gbc)) {
249 s->heavy_dynamic_range[(!s->channel_mode)-i] =
250 AC3_HEAVY_RANGE(get_bits(gbc, 8));
253 skip_bits(gbc, 8); //skip language code
255 skip_bits(gbc, 7); //skip audio production information
258 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
260 /* skip the timecodes or parse the Alternate Bit Stream Syntax */
261 if (s->bitstream_id != 6) {
263 skip_bits(gbc, 14); //skip timecode1
265 skip_bits(gbc, 14); //skip timecode2
267 if (get_bits1(gbc)) {
268 s->preferred_downmix = get_bits(gbc, 2);
269 s->center_mix_level_ltrt = get_bits(gbc, 3);
270 s->surround_mix_level_ltrt = av_clip(get_bits(gbc, 3), 3, 7);
271 s->center_mix_level = get_bits(gbc, 3);
272 s->surround_mix_level = av_clip(get_bits(gbc, 3), 3, 7);
274 if (get_bits1(gbc)) {
275 s->dolby_surround_ex_mode = get_bits(gbc, 2);
276 s->dolby_headphone_mode = get_bits(gbc, 2);
277 skip_bits(gbc, 10); // skip adconvtyp (1), xbsi2 (8), encinfo (1)
281 /* skip additional bitstream info */
282 if (get_bits1(gbc)) {
283 i = get_bits(gbc, 6);
293 * Common function to parse AC-3 or E-AC-3 frame header
295 static int parse_frame_header(AC3DecodeContext *s)
297 AC3HeaderInfo hdr, *phdr=&hdr;
300 err = avpriv_ac3_parse_header(&s->gbc, &phdr);
304 /* get decoding parameters from header info */
305 s->bit_alloc_params.sr_code = hdr.sr_code;
306 s->bitstream_id = hdr.bitstream_id;
307 s->bitstream_mode = hdr.bitstream_mode;
308 s->channel_mode = hdr.channel_mode;
309 s->lfe_on = hdr.lfe_on;
310 s->bit_alloc_params.sr_shift = hdr.sr_shift;
311 s->sample_rate = hdr.sample_rate;
312 s->bit_rate = hdr.bit_rate;
313 s->channels = hdr.channels;
314 s->fbw_channels = s->channels - s->lfe_on;
315 s->lfe_ch = s->fbw_channels + 1;
316 s->frame_size = hdr.frame_size;
317 s->preferred_downmix = AC3_DMIXMOD_NOTINDICATED;
318 s->center_mix_level = hdr.center_mix_level;
319 s->center_mix_level_ltrt = 4; // -3.0dB
320 s->surround_mix_level = hdr.surround_mix_level;
321 s->surround_mix_level_ltrt = 4; // -3.0dB
322 s->lfe_mix_level_exists = 0;
323 s->num_blocks = hdr.num_blocks;
324 s->frame_type = hdr.frame_type;
325 s->substreamid = hdr.substreamid;
326 s->dolby_surround_mode = hdr.dolby_surround_mode;
327 s->dolby_surround_ex_mode = AC3_DSUREXMOD_NOTINDICATED;
328 s->dolby_headphone_mode = AC3_DHEADPHONMOD_NOTINDICATED;
331 s->start_freq[s->lfe_ch] = 0;
332 s->end_freq[s->lfe_ch] = 7;
333 s->num_exp_groups[s->lfe_ch] = 2;
334 s->channel_in_cpl[s->lfe_ch] = 0;
337 if (s->bitstream_id <= 10) {
339 s->snr_offset_strategy = 2;
340 s->block_switch_syntax = 1;
341 s->dither_flag_syntax = 1;
342 s->bit_allocation_syntax = 1;
343 s->fast_gain_syntax = 0;
344 s->first_cpl_leak = 0;
347 memset(s->channel_uses_aht, 0, sizeof(s->channel_uses_aht));
348 return ac3_parse_header(s);
349 } else if (CONFIG_EAC3_DECODER) {
351 return ff_eac3_parse_header(s);
353 av_log(s->avctx, AV_LOG_ERROR, "E-AC-3 support not compiled in\n");
354 return AVERROR(ENOSYS);
359 * Set stereo downmixing coefficients based on frame header info.
360 * reference: Section 7.8.2 Downmixing Into Two Channels
362 static void set_downmix_coeffs(AC3DecodeContext *s)
365 float cmix = gain_levels[s-> center_mix_level];
366 float smix = gain_levels[s->surround_mix_level];
368 float downmix_coeffs[AC3_MAX_CHANNELS][2];
370 for (i = 0; i < s->fbw_channels; i++) {
371 downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
372 downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
374 if (s->channel_mode > 1 && s->channel_mode & 1) {
375 downmix_coeffs[1][0] = downmix_coeffs[1][1] = cmix;
377 if (s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
378 int nf = s->channel_mode - 2;
379 downmix_coeffs[nf][0] = downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
381 if (s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
382 int nf = s->channel_mode - 4;
383 downmix_coeffs[nf][0] = downmix_coeffs[nf+1][1] = smix;
388 for (i = 0; i < s->fbw_channels; i++) {
389 norm0 += downmix_coeffs[i][0];
390 norm1 += downmix_coeffs[i][1];
392 norm0 = 1.0f / norm0;
393 norm1 = 1.0f / norm1;
394 for (i = 0; i < s->fbw_channels; i++) {
395 downmix_coeffs[i][0] *= norm0;
396 downmix_coeffs[i][1] *= norm1;
399 if (s->output_mode == AC3_CHMODE_MONO) {
400 for (i = 0; i < s->fbw_channels; i++)
401 downmix_coeffs[i][0] = (downmix_coeffs[i][0] +
402 downmix_coeffs[i][1]) * LEVEL_MINUS_3DB;
404 for (i = 0; i < s->fbw_channels; i++) {
405 s->downmix_coeffs[i][0] = FIXR12(downmix_coeffs[i][0]);
406 s->downmix_coeffs[i][1] = FIXR12(downmix_coeffs[i][1]);
411 * Decode the grouped exponents according to exponent strategy.
412 * reference: Section 7.1.3 Exponent Decoding
414 static int decode_exponents(AC3DecodeContext *s,
415 GetBitContext *gbc, int exp_strategy, int ngrps,
416 uint8_t absexp, int8_t *dexps)
418 int i, j, grp, group_size;
423 group_size = exp_strategy + (exp_strategy == EXP_D45);
424 for (grp = 0, i = 0; grp < ngrps; grp++) {
425 expacc = get_bits(gbc, 7);
426 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][0];
427 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][1];
428 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][2];
431 /* convert to absolute exps and expand groups */
433 for (i = 0, j = 0; i < ngrps * 3; i++) {
434 prevexp += dexp[i] - 2;
436 av_log(s->avctx, AV_LOG_ERROR, "exponent %d is out-of-range\n", prevexp);
439 switch (group_size) {
440 case 4: dexps[j++] = prevexp;
441 dexps[j++] = prevexp;
442 case 2: dexps[j++] = prevexp;
443 case 1: dexps[j++] = prevexp;
450 * Generate transform coefficients for each coupled channel in the coupling
451 * range using the coupling coefficients and coupling coordinates.
452 * reference: Section 7.4.3 Coupling Coordinate Format
454 static void calc_transform_coeffs_cpl(AC3DecodeContext *s)
458 bin = s->start_freq[CPL_CH];
459 for (band = 0; band < s->num_cpl_bands; band++) {
460 int band_start = bin;
461 int band_end = bin + s->cpl_band_sizes[band];
462 for (ch = 1; ch <= s->fbw_channels; ch++) {
463 if (s->channel_in_cpl[ch]) {
464 int cpl_coord = s->cpl_coords[ch][band] << 5;
465 for (bin = band_start; bin < band_end; bin++) {
466 s->fixed_coeffs[ch][bin] =
467 MULH(s->fixed_coeffs[CPL_CH][bin] * (1 << 4), cpl_coord);
469 if (ch == 2 && s->phase_flags[band]) {
470 for (bin = band_start; bin < band_end; bin++)
471 s->fixed_coeffs[2][bin] = -s->fixed_coeffs[2][bin];
480 * Grouped mantissas for 3-level 5-level and 11-level quantization
482 typedef struct mant_groups {
492 * Decode the transform coefficients for a particular channel
493 * reference: Section 7.3 Quantization and Decoding of Mantissas
495 static void ac3_decode_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
497 int start_freq = s->start_freq[ch_index];
498 int end_freq = s->end_freq[ch_index];
499 uint8_t *baps = s->bap[ch_index];
500 int8_t *exps = s->dexps[ch_index];
501 int32_t *coeffs = s->fixed_coeffs[ch_index];
502 int dither = (ch_index == CPL_CH) || s->dither_flag[ch_index];
503 GetBitContext *gbc = &s->gbc;
506 for (freq = start_freq; freq < end_freq; freq++) {
507 int bap = baps[freq];
511 /* random noise with approximate range of -0.707 to 0.707 */
513 mantissa = (((av_lfg_get(&s->dith_state)>>8)*181)>>8) - 5931008;
520 mantissa = m->b1_mant[m->b1];
522 int bits = get_bits(gbc, 5);
523 mantissa = b1_mantissas[bits][0];
524 m->b1_mant[1] = b1_mantissas[bits][1];
525 m->b1_mant[0] = b1_mantissas[bits][2];
532 mantissa = m->b2_mant[m->b2];
534 int bits = get_bits(gbc, 7);
535 mantissa = b2_mantissas[bits][0];
536 m->b2_mant[1] = b2_mantissas[bits][1];
537 m->b2_mant[0] = b2_mantissas[bits][2];
542 mantissa = b3_mantissas[get_bits(gbc, 3)];
547 mantissa = m->b4_mant;
549 int bits = get_bits(gbc, 7);
550 mantissa = b4_mantissas[bits][0];
551 m->b4_mant = b4_mantissas[bits][1];
556 mantissa = b5_mantissas[get_bits(gbc, 4)];
558 default: /* 6 to 15 */
559 /* Shift mantissa and sign-extend it. */
561 av_log(s->avctx, AV_LOG_ERROR, "bap %d is invalid in plain AC-3\n", bap);
564 mantissa = (unsigned)get_sbits(gbc, quantization_tab[bap]) << (24 - quantization_tab[bap]);
567 coeffs[freq] = mantissa >> exps[freq];
572 * Remove random dithering from coupling range coefficients with zero-bit
573 * mantissas for coupled channels which do not use dithering.
574 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
576 static void remove_dithering(AC3DecodeContext *s) {
579 for (ch = 1; ch <= s->fbw_channels; ch++) {
580 if (!s->dither_flag[ch] && s->channel_in_cpl[ch]) {
581 for (i = s->start_freq[CPL_CH]; i < s->end_freq[CPL_CH]; i++) {
582 if (!s->bap[CPL_CH][i])
583 s->fixed_coeffs[ch][i] = 0;
589 static void decode_transform_coeffs_ch(AC3DecodeContext *s, int blk, int ch,
592 if (!s->channel_uses_aht[ch]) {
593 ac3_decode_transform_coeffs_ch(s, ch, m);
595 /* if AHT is used, mantissas for all blocks are encoded in the first
596 block of the frame. */
598 if (CONFIG_EAC3_DECODER && !blk)
599 ff_eac3_decode_transform_coeffs_aht_ch(s, ch);
600 for (bin = s->start_freq[ch]; bin < s->end_freq[ch]; bin++) {
601 s->fixed_coeffs[ch][bin] = s->pre_mantissa[ch][bin][blk] >> s->dexps[ch][bin];
607 * Decode the transform coefficients.
609 static void decode_transform_coeffs(AC3DecodeContext *s, int blk)
615 m.b1 = m.b2 = m.b4 = 0;
617 for (ch = 1; ch <= s->channels; ch++) {
618 /* transform coefficients for full-bandwidth channel */
619 decode_transform_coeffs_ch(s, blk, ch, &m);
620 /* transform coefficients for coupling channel come right after the
621 coefficients for the first coupled channel*/
622 if (s->channel_in_cpl[ch]) {
624 decode_transform_coeffs_ch(s, blk, CPL_CH, &m);
625 calc_transform_coeffs_cpl(s);
628 end = s->end_freq[CPL_CH];
630 end = s->end_freq[ch];
633 s->fixed_coeffs[ch][end] = 0;
637 /* zero the dithered coefficients for appropriate channels */
642 * Stereo rematrixing.
643 * reference: Section 7.5.4 Rematrixing : Decoding Technique
645 static void do_rematrixing(AC3DecodeContext *s)
650 end = FFMIN(s->end_freq[1], s->end_freq[2]);
652 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++) {
653 if (s->rematrixing_flags[bnd]) {
654 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd + 1]);
655 for (i = ff_ac3_rematrix_band_tab[bnd]; i < bndend; i++) {
656 int tmp0 = s->fixed_coeffs[1][i];
657 s->fixed_coeffs[1][i] += s->fixed_coeffs[2][i];
658 s->fixed_coeffs[2][i] = tmp0 - s->fixed_coeffs[2][i];
665 * Inverse MDCT Transform.
666 * Convert frequency domain coefficients to time-domain audio samples.
667 * reference: Section 7.9.4 Transformation Equations
669 static inline void do_imdct(AC3DecodeContext *s, int channels)
673 for (ch = 1; ch <= channels; ch++) {
674 if (s->block_switch[ch]) {
676 FFTSample *x = s->tmp_output + 128;
677 for (i = 0; i < 128; i++)
678 x[i] = s->transform_coeffs[ch][2 * i];
679 s->imdct_256.imdct_half(&s->imdct_256, s->tmp_output, x);
681 s->fdsp->vector_fmul_window_scaled(s->outptr[ch - 1], s->delay[ch - 1],
682 s->tmp_output, s->window, 128, 8);
684 s->fdsp->vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1],
685 s->tmp_output, s->window, 128);
687 for (i = 0; i < 128; i++)
688 x[i] = s->transform_coeffs[ch][2 * i + 1];
689 s->imdct_256.imdct_half(&s->imdct_256, s->delay[ch - 1], x);
691 s->imdct_512.imdct_half(&s->imdct_512, s->tmp_output, s->transform_coeffs[ch]);
693 s->fdsp->vector_fmul_window_scaled(s->outptr[ch - 1], s->delay[ch - 1],
694 s->tmp_output, s->window, 128, 8);
696 s->fdsp->vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1],
697 s->tmp_output, s->window, 128);
699 memcpy(s->delay[ch - 1], s->tmp_output + 128, 128 * sizeof(FFTSample));
705 * Upmix delay samples from stereo to original channel layout.
707 static void ac3_upmix_delay(AC3DecodeContext *s)
709 int channel_data_size = sizeof(s->delay[0]);
710 switch (s->channel_mode) {
711 case AC3_CHMODE_DUALMONO:
712 case AC3_CHMODE_STEREO:
713 /* upmix mono to stereo */
714 memcpy(s->delay[1], s->delay[0], channel_data_size);
716 case AC3_CHMODE_2F2R:
717 memset(s->delay[3], 0, channel_data_size);
718 case AC3_CHMODE_2F1R:
719 memset(s->delay[2], 0, channel_data_size);
721 case AC3_CHMODE_3F2R:
722 memset(s->delay[4], 0, channel_data_size);
723 case AC3_CHMODE_3F1R:
724 memset(s->delay[3], 0, channel_data_size);
726 memcpy(s->delay[2], s->delay[1], channel_data_size);
727 memset(s->delay[1], 0, channel_data_size);
733 * Decode band structure for coupling, spectral extension, or enhanced coupling.
734 * The band structure defines how many subbands are in each band. For each
735 * subband in the range, 1 means it is combined with the previous band, and 0
736 * means that it starts a new band.
738 * @param[in] gbc bit reader context
739 * @param[in] blk block number
740 * @param[in] eac3 flag to indicate E-AC-3
741 * @param[in] ecpl flag to indicate enhanced coupling
742 * @param[in] start_subband subband number for start of range
743 * @param[in] end_subband subband number for end of range
744 * @param[in] default_band_struct default band structure table
745 * @param[out] num_bands number of bands (optionally NULL)
746 * @param[out] band_sizes array containing the number of bins in each band (optionally NULL)
748 static void decode_band_structure(GetBitContext *gbc, int blk, int eac3,
749 int ecpl, int start_subband, int end_subband,
750 const uint8_t *default_band_struct,
751 int *num_bands, uint8_t *band_sizes)
753 int subbnd, bnd, n_subbands, n_bands=0;
755 uint8_t coded_band_struct[22];
756 const uint8_t *band_struct;
758 n_subbands = end_subband - start_subband;
760 /* decode band structure from bitstream or use default */
761 if (!eac3 || get_bits1(gbc)) {
762 for (subbnd = 0; subbnd < n_subbands - 1; subbnd++) {
763 coded_band_struct[subbnd] = get_bits1(gbc);
765 band_struct = coded_band_struct;
767 band_struct = &default_band_struct[start_subband+1];
769 /* no change in band structure */
773 /* calculate number of bands and band sizes based on band structure.
774 note that the first 4 subbands in enhanced coupling span only 6 bins
776 if (num_bands || band_sizes ) {
777 n_bands = n_subbands;
778 bnd_sz[0] = ecpl ? 6 : 12;
779 for (bnd = 0, subbnd = 1; subbnd < n_subbands; subbnd++) {
780 int subbnd_size = (ecpl && subbnd < 4) ? 6 : 12;
781 if (band_struct[subbnd - 1]) {
783 bnd_sz[bnd] += subbnd_size;
785 bnd_sz[++bnd] = subbnd_size;
790 /* set optional output params */
792 *num_bands = n_bands;
794 memcpy(band_sizes, bnd_sz, n_bands);
798 * Decode a single audio block from the AC-3 bitstream.
800 static int decode_audio_block(AC3DecodeContext *s, int blk)
802 int fbw_channels = s->fbw_channels;
803 int channel_mode = s->channel_mode;
805 int different_transforms;
808 GetBitContext *gbc = &s->gbc;
809 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS] = { 0 };
811 /* block switch flags */
812 different_transforms = 0;
813 if (s->block_switch_syntax) {
814 for (ch = 1; ch <= fbw_channels; ch++) {
815 s->block_switch[ch] = get_bits1(gbc);
816 if (ch > 1 && s->block_switch[ch] != s->block_switch[1])
817 different_transforms = 1;
821 /* dithering flags */
822 if (s->dither_flag_syntax) {
823 for (ch = 1; ch <= fbw_channels; ch++) {
824 s->dither_flag[ch] = get_bits1(gbc);
829 i = !s->channel_mode;
831 if (get_bits1(gbc)) {
832 /* Allow asymmetric application of DRC when drc_scale > 1.
833 Amplification of quiet sounds is enhanced */
834 int range_bits = get_bits(gbc, 8);
835 INTFLOAT range = AC3_RANGE(range_bits);
836 if (range_bits <= 127 || s->drc_scale <= 1.0)
837 s->dynamic_range[i] = AC3_DYNAMIC_RANGE(range);
839 s->dynamic_range[i] = range;
840 } else if (blk == 0) {
841 s->dynamic_range[i] = AC3_DYNAMIC_RANGE1;
845 /* spectral extension strategy */
846 if (s->eac3 && (!blk || get_bits1(gbc))) {
847 s->spx_in_use = get_bits1(gbc);
849 int dst_start_freq, dst_end_freq, src_start_freq,
850 start_subband, end_subband;
852 /* determine which channels use spx */
853 if (s->channel_mode == AC3_CHMODE_MONO) {
854 s->channel_uses_spx[1] = 1;
856 for (ch = 1; ch <= fbw_channels; ch++)
857 s->channel_uses_spx[ch] = get_bits1(gbc);
860 /* get the frequency bins of the spx copy region and the spx start
862 dst_start_freq = get_bits(gbc, 2);
863 start_subband = get_bits(gbc, 3) + 2;
864 if (start_subband > 7)
865 start_subband += start_subband - 7;
866 end_subband = get_bits(gbc, 3) + 5;
868 s->spx_dst_end_freq = end_freq_inv_tab[end_subband-5];
871 end_subband += end_subband - 7;
872 dst_start_freq = dst_start_freq * 12 + 25;
873 src_start_freq = start_subband * 12 + 25;
874 dst_end_freq = end_subband * 12 + 25;
876 /* check validity of spx ranges */
877 if (start_subband >= end_subband) {
878 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
879 "range (%d >= %d)\n", start_subband, end_subband);
880 return AVERROR_INVALIDDATA;
882 if (dst_start_freq >= src_start_freq) {
883 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
884 "copy start bin (%d >= %d)\n", dst_start_freq, src_start_freq);
885 return AVERROR_INVALIDDATA;
888 s->spx_dst_start_freq = dst_start_freq;
889 s->spx_src_start_freq = src_start_freq;
891 s->spx_dst_end_freq = dst_end_freq;
893 decode_band_structure(gbc, blk, s->eac3, 0,
894 start_subband, end_subband,
895 ff_eac3_default_spx_band_struct,
900 if (!s->eac3 || !s->spx_in_use) {
902 for (ch = 1; ch <= fbw_channels; ch++) {
903 s->channel_uses_spx[ch] = 0;
904 s->first_spx_coords[ch] = 1;
908 /* spectral extension coordinates */
910 for (ch = 1; ch <= fbw_channels; ch++) {
911 if (s->channel_uses_spx[ch]) {
912 if (s->first_spx_coords[ch] || get_bits1(gbc)) {
914 int bin, master_spx_coord;
916 s->first_spx_coords[ch] = 0;
917 spx_blend = AC3_SPX_BLEND(get_bits(gbc, 5));
918 master_spx_coord = get_bits(gbc, 2) * 3;
920 bin = s->spx_src_start_freq;
921 for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
922 int bandsize = s->spx_band_sizes[bnd];
923 int spx_coord_exp, spx_coord_mant;
924 INTFLOAT nratio, sblend, nblend;
926 /* calculate blending factors */
927 int64_t accu = ((bin << 23) + (bandsize << 22))
928 * (int64_t)s->spx_dst_end_freq;
929 nratio = (int)(accu >> 32);
930 nratio -= spx_blend << 18;
935 } else if (nratio > 0x7fffff) {
936 nblend = 14529495; // sqrt(3) in FP.23
939 nblend = fixed_sqrt(nratio, 23);
940 accu = (int64_t)nblend * 1859775393;
941 nblend = (int)((accu + (1<<29)) >> 30);
942 sblend = fixed_sqrt(0x800000 - nratio, 23);
947 /* calculate blending factors */
948 nratio = ((float)((bin + (bandsize >> 1))) / s->spx_dst_end_freq) - spx_blend;
949 nratio = av_clipf(nratio, 0.0f, 1.0f);
950 nblend = sqrtf(3.0f * nratio); // noise is scaled by sqrt(3)
951 // to give unity variance
952 sblend = sqrtf(1.0f - nratio);
956 /* decode spx coordinates */
957 spx_coord_exp = get_bits(gbc, 4);
958 spx_coord_mant = get_bits(gbc, 2);
959 if (spx_coord_exp == 15) spx_coord_mant <<= 1;
960 else spx_coord_mant += 4;
961 spx_coord_mant <<= (25 - spx_coord_exp - master_spx_coord);
963 /* multiply noise and signal blending factors by spx coordinate */
965 accu = (int64_t)nblend * spx_coord_mant;
966 s->spx_noise_blend[ch][bnd] = (int)((accu + (1<<22)) >> 23);
967 accu = (int64_t)sblend * spx_coord_mant;
968 s->spx_signal_blend[ch][bnd] = (int)((accu + (1<<22)) >> 23);
970 spx_coord = spx_coord_mant * (1.0f / (1 << 23));
971 s->spx_noise_blend [ch][bnd] = nblend * spx_coord;
972 s->spx_signal_blend[ch][bnd] = sblend * spx_coord;
977 s->first_spx_coords[ch] = 1;
982 /* coupling strategy */
983 if (s->eac3 ? s->cpl_strategy_exists[blk] : get_bits1(gbc)) {
984 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
986 s->cpl_in_use[blk] = get_bits1(gbc);
987 if (s->cpl_in_use[blk]) {
988 /* coupling in use */
989 int cpl_start_subband, cpl_end_subband;
991 if (channel_mode < AC3_CHMODE_STEREO) {
992 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
993 return AVERROR_INVALIDDATA;
996 /* check for enhanced coupling */
997 if (s->eac3 && get_bits1(gbc)) {
998 /* TODO: parse enhanced coupling strategy info */
999 avpriv_request_sample(s->avctx, "Enhanced coupling");
1000 return AVERROR_PATCHWELCOME;
1003 /* determine which channels are coupled */
1004 if (s->eac3 && s->channel_mode == AC3_CHMODE_STEREO) {
1005 s->channel_in_cpl[1] = 1;
1006 s->channel_in_cpl[2] = 1;
1008 for (ch = 1; ch <= fbw_channels; ch++)
1009 s->channel_in_cpl[ch] = get_bits1(gbc);
1012 /* phase flags in use */
1013 if (channel_mode == AC3_CHMODE_STEREO)
1014 s->phase_flags_in_use = get_bits1(gbc);
1016 /* coupling frequency range */
1017 cpl_start_subband = get_bits(gbc, 4);
1018 cpl_end_subband = s->spx_in_use ? (s->spx_src_start_freq - 37) / 12 :
1019 get_bits(gbc, 4) + 3;
1020 if (cpl_start_subband >= cpl_end_subband) {
1021 av_log(s->avctx, AV_LOG_ERROR, "invalid coupling range (%d >= %d)\n",
1022 cpl_start_subband, cpl_end_subband);
1023 return AVERROR_INVALIDDATA;
1025 s->start_freq[CPL_CH] = cpl_start_subband * 12 + 37;
1026 s->end_freq[CPL_CH] = cpl_end_subband * 12 + 37;
1028 decode_band_structure(gbc, blk, s->eac3, 0, cpl_start_subband,
1030 ff_eac3_default_cpl_band_struct,
1031 &s->num_cpl_bands, s->cpl_band_sizes);
1033 /* coupling not in use */
1034 for (ch = 1; ch <= fbw_channels; ch++) {
1035 s->channel_in_cpl[ch] = 0;
1036 s->first_cpl_coords[ch] = 1;
1038 s->first_cpl_leak = s->eac3;
1039 s->phase_flags_in_use = 0;
1041 } else if (!s->eac3) {
1043 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must "
1044 "be present in block 0\n");
1045 return AVERROR_INVALIDDATA;
1047 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
1050 cpl_in_use = s->cpl_in_use[blk];
1052 /* coupling coordinates */
1054 int cpl_coords_exist = 0;
1056 for (ch = 1; ch <= fbw_channels; ch++) {
1057 if (s->channel_in_cpl[ch]) {
1058 if ((s->eac3 && s->first_cpl_coords[ch]) || get_bits1(gbc)) {
1059 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
1060 s->first_cpl_coords[ch] = 0;
1061 cpl_coords_exist = 1;
1062 master_cpl_coord = 3 * get_bits(gbc, 2);
1063 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
1064 cpl_coord_exp = get_bits(gbc, 4);
1065 cpl_coord_mant = get_bits(gbc, 4);
1066 if (cpl_coord_exp == 15)
1067 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
1069 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
1070 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
1073 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must "
1074 "be present in block 0\n");
1075 return AVERROR_INVALIDDATA;
1078 /* channel not in coupling */
1079 s->first_cpl_coords[ch] = 1;
1083 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
1084 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
1085 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
1090 /* stereo rematrixing strategy and band structure */
1091 if (channel_mode == AC3_CHMODE_STEREO) {
1092 if ((s->eac3 && !blk) || get_bits1(gbc)) {
1093 s->num_rematrixing_bands = 4;
1094 if (cpl_in_use && s->start_freq[CPL_CH] <= 61) {
1095 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
1096 } else if (s->spx_in_use && s->spx_src_start_freq <= 61) {
1097 s->num_rematrixing_bands--;
1099 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++)
1100 s->rematrixing_flags[bnd] = get_bits1(gbc);
1102 av_log(s->avctx, AV_LOG_WARNING, "Warning: "
1103 "new rematrixing strategy not present in block 0\n");
1104 s->num_rematrixing_bands = 0;
1108 /* exponent strategies for each channel */
1109 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1111 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
1112 if (s->exp_strategy[blk][ch] != EXP_REUSE)
1113 bit_alloc_stages[ch] = 3;
1116 /* channel bandwidth */
1117 for (ch = 1; ch <= fbw_channels; ch++) {
1118 s->start_freq[ch] = 0;
1119 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1121 int prev = s->end_freq[ch];
1122 if (s->channel_in_cpl[ch])
1123 s->end_freq[ch] = s->start_freq[CPL_CH];
1124 else if (s->channel_uses_spx[ch])
1125 s->end_freq[ch] = s->spx_src_start_freq;
1127 int bandwidth_code = get_bits(gbc, 6);
1128 if (bandwidth_code > 60) {
1129 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60\n", bandwidth_code);
1130 return AVERROR_INVALIDDATA;
1132 s->end_freq[ch] = bandwidth_code * 3 + 73;
1134 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
1135 s->num_exp_groups[ch] = (s->end_freq[ch] + group_size-4) / group_size;
1136 if (blk > 0 && s->end_freq[ch] != prev)
1137 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
1140 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
1141 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
1142 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
1145 /* decode exponents for each channel */
1146 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1147 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1148 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
1149 if (decode_exponents(s, gbc, s->exp_strategy[blk][ch],
1150 s->num_exp_groups[ch], s->dexps[ch][0],
1151 &s->dexps[ch][s->start_freq[ch]+!!ch])) {
1152 return AVERROR_INVALIDDATA;
1154 if (ch != CPL_CH && ch != s->lfe_ch)
1155 skip_bits(gbc, 2); /* skip gainrng */
1159 /* bit allocation information */
1160 if (s->bit_allocation_syntax) {
1161 if (get_bits1(gbc)) {
1162 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1163 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1164 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
1165 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
1166 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
1167 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1168 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1170 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must "
1171 "be present in block 0\n");
1172 return AVERROR_INVALIDDATA;
1176 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
1177 if (!s->eac3 || !blk) {
1178 if (s->snr_offset_strategy && get_bits1(gbc)) {
1181 csnr = (get_bits(gbc, 6) - 15) << 4;
1182 for (i = ch = !cpl_in_use; ch <= s->channels; ch++) {
1184 if (ch == i || s->snr_offset_strategy == 2)
1185 snr = (csnr + get_bits(gbc, 4)) << 2;
1186 /* run at least last bit allocation stage if snr offset changes */
1187 if (blk && s->snr_offset[ch] != snr) {
1188 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 1);
1190 s->snr_offset[ch] = snr;
1192 /* fast gain (normal AC-3 only) */
1194 int prev = s->fast_gain[ch];
1195 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1196 /* run last 2 bit allocation stages if fast gain changes */
1197 if (blk && prev != s->fast_gain[ch])
1198 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1201 } else if (!s->eac3 && !blk) {
1202 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
1203 return AVERROR_INVALIDDATA;
1207 /* fast gain (E-AC-3 only) */
1208 if (s->fast_gain_syntax && get_bits1(gbc)) {
1209 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1210 int prev = s->fast_gain[ch];
1211 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1212 /* run last 2 bit allocation stages if fast gain changes */
1213 if (blk && prev != s->fast_gain[ch])
1214 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1216 } else if (s->eac3 && !blk) {
1217 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1218 s->fast_gain[ch] = ff_ac3_fast_gain_tab[4];
1221 /* E-AC-3 to AC-3 converter SNR offset */
1222 if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT && get_bits1(gbc)) {
1223 skip_bits(gbc, 10); // skip converter snr offset
1226 /* coupling leak information */
1228 if (s->first_cpl_leak || get_bits1(gbc)) {
1229 int fl = get_bits(gbc, 3);
1230 int sl = get_bits(gbc, 3);
1231 /* run last 2 bit allocation stages for coupling channel if
1232 coupling leak changes */
1233 if (blk && (fl != s->bit_alloc_params.cpl_fast_leak ||
1234 sl != s->bit_alloc_params.cpl_slow_leak)) {
1235 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
1237 s->bit_alloc_params.cpl_fast_leak = fl;
1238 s->bit_alloc_params.cpl_slow_leak = sl;
1239 } else if (!s->eac3 && !blk) {
1240 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must "
1241 "be present in block 0\n");
1242 return AVERROR_INVALIDDATA;
1244 s->first_cpl_leak = 0;
1247 /* delta bit allocation information */
1248 if (s->dba_syntax && get_bits1(gbc)) {
1249 /* delta bit allocation exists (strategy) */
1250 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1251 s->dba_mode[ch] = get_bits(gbc, 2);
1252 if (s->dba_mode[ch] == DBA_RESERVED) {
1253 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1254 return AVERROR_INVALIDDATA;
1256 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1258 /* channel delta offset, len and bit allocation */
1259 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1260 if (s->dba_mode[ch] == DBA_NEW) {
1261 s->dba_nsegs[ch] = get_bits(gbc, 3) + 1;
1262 for (seg = 0; seg < s->dba_nsegs[ch]; seg++) {
1263 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1264 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1265 s->dba_values[ch][seg] = get_bits(gbc, 3);
1267 /* run last 2 bit allocation stages if new dba values */
1268 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1271 } else if (blk == 0) {
1272 for (ch = 0; ch <= s->channels; ch++) {
1273 s->dba_mode[ch] = DBA_NONE;
1277 /* Bit allocation */
1278 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1279 if (bit_alloc_stages[ch] > 2) {
1280 /* Exponent mapping into PSD and PSD integration */
1281 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1282 s->start_freq[ch], s->end_freq[ch],
1283 s->psd[ch], s->band_psd[ch]);
1285 if (bit_alloc_stages[ch] > 1) {
1286 /* Compute excitation function, Compute masking curve, and
1287 Apply delta bit allocation */
1288 if (ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1289 s->start_freq[ch], s->end_freq[ch],
1290 s->fast_gain[ch], (ch == s->lfe_ch),
1291 s->dba_mode[ch], s->dba_nsegs[ch],
1292 s->dba_offsets[ch], s->dba_lengths[ch],
1293 s->dba_values[ch], s->mask[ch])) {
1294 av_log(s->avctx, AV_LOG_ERROR, "error in bit allocation\n");
1295 return AVERROR_INVALIDDATA;
1298 if (bit_alloc_stages[ch] > 0) {
1299 /* Compute bit allocation */
1300 const uint8_t *bap_tab = s->channel_uses_aht[ch] ?
1301 ff_eac3_hebap_tab : ff_ac3_bap_tab;
1302 s->ac3dsp.bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1303 s->start_freq[ch], s->end_freq[ch],
1305 s->bit_alloc_params.floor,
1306 bap_tab, s->bap[ch]);
1310 /* unused dummy data */
1311 if (s->skip_syntax && get_bits1(gbc)) {
1312 int skipl = get_bits(gbc, 9);
1317 /* unpack the transform coefficients
1318 this also uncouples channels if coupling is in use. */
1319 decode_transform_coeffs(s, blk);
1321 /* TODO: generate enhanced coupling coordinates and uncouple */
1323 /* recover coefficients if rematrixing is in use */
1324 if (s->channel_mode == AC3_CHMODE_STEREO)
1327 /* apply scaling to coefficients (headroom, dynrng) */
1328 for (ch = 1; ch <= s->channels; ch++) {
1329 int audio_channel = 0;
1331 if (s->channel_mode == AC3_CHMODE_DUALMONO)
1332 audio_channel = 2-ch;
1333 if (s->heavy_compression && s->compression_exists[audio_channel])
1334 gain = s->heavy_dynamic_range[audio_channel];
1336 gain = s->dynamic_range[audio_channel];
1339 scale_coefs(s->transform_coeffs[ch], s->fixed_coeffs[ch], gain, 256);
1341 if (s->target_level != 0)
1342 gain = gain * s->level_gain[audio_channel];
1343 gain *= 1.0 / 4194304.0f;
1344 s->fmt_conv.int32_to_float_fmul_scalar(s->transform_coeffs[ch],
1345 s->fixed_coeffs[ch], gain, 256);
1349 /* apply spectral extension to high frequency bins */
1350 if (CONFIG_EAC3_DECODER && s->spx_in_use) {
1351 ff_eac3_apply_spectral_extension(s);
1354 /* downmix and MDCT. order depends on whether block switching is used for
1355 any channel in this block. this is because coefficients for the long
1356 and short transforms cannot be mixed. */
1357 downmix_output = s->channels != s->out_channels &&
1358 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1359 s->fbw_channels == s->out_channels);
1360 if (different_transforms) {
1361 /* the delay samples have already been downmixed, so we upmix the delay
1362 samples in order to reconstruct all channels before downmixing. */
1368 do_imdct(s, s->channels);
1370 if (downmix_output) {
1372 ac3_downmix_c_fixed16(s->outptr, s->downmix_coeffs,
1373 s->out_channels, s->fbw_channels, 256);
1375 s->ac3dsp.downmix(s->outptr, s->downmix_coeffs,
1376 s->out_channels, s->fbw_channels, 256);
1380 if (downmix_output) {
1381 s->ac3dsp.AC3_RENAME(downmix)(s->xcfptr + 1, s->downmix_coeffs,
1382 s->out_channels, s->fbw_channels, 256);
1385 if (downmix_output && !s->downmixed) {
1387 s->ac3dsp.AC3_RENAME(downmix)(s->dlyptr, s->downmix_coeffs,
1388 s->out_channels, s->fbw_channels, 128);
1391 do_imdct(s, s->out_channels);
1398 * Decode a single AC-3 frame.
1400 static int ac3_decode_frame(AVCodecContext * avctx, void *data,
1401 int *got_frame_ptr, AVPacket *avpkt)
1403 AVFrame *frame = data;
1404 const uint8_t *buf = avpkt->data;
1405 int buf_size = avpkt->size;
1406 AC3DecodeContext *s = avctx->priv_data;
1407 int blk, ch, err, ret;
1408 const uint8_t *channel_map;
1409 const SHORTFLOAT *output[AC3_MAX_CHANNELS];
1410 enum AVMatrixEncoding matrix_encoding;
1411 AVDownmixInfo *downmix_info;
1413 /* copy input buffer to decoder context to avoid reading past the end
1414 of the buffer, which can be caused by a damaged input stream. */
1415 if (buf_size >= 2 && AV_RB16(buf) == 0x770B) {
1416 // seems to be byte-swapped AC-3
1417 int cnt = FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE) >> 1;
1418 s->bdsp.bswap16_buf((uint16_t *) s->input_buffer,
1419 (const uint16_t *) buf, cnt);
1421 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE));
1422 buf = s->input_buffer;
1423 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1424 if ((ret = init_get_bits8(&s->gbc, buf, buf_size)) < 0)
1427 /* parse the syncinfo */
1428 err = parse_frame_header(s);
1432 case AAC_AC3_PARSE_ERROR_SYNC:
1433 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1434 return AVERROR_INVALIDDATA;
1435 case AAC_AC3_PARSE_ERROR_BSID:
1436 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1438 case AAC_AC3_PARSE_ERROR_SAMPLE_RATE:
1439 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1441 case AAC_AC3_PARSE_ERROR_FRAME_SIZE:
1442 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1444 case AAC_AC3_PARSE_ERROR_FRAME_TYPE:
1445 /* skip frame if CRC is ok. otherwise use error concealment. */
1446 /* TODO: add support for substreams and dependent frames */
1447 if (s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1448 av_log(avctx, AV_LOG_DEBUG,
1449 "unsupported frame type %d: skipping frame\n",
1454 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1457 case AAC_AC3_PARSE_ERROR_CRC:
1458 case AAC_AC3_PARSE_ERROR_CHANNEL_CFG:
1460 default: // Normal AVERROR do not try to recover.
1465 /* check that reported frame size fits in input buffer */
1466 if (s->frame_size > buf_size) {
1467 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1468 err = AAC_AC3_PARSE_ERROR_FRAME_SIZE;
1469 } else if (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL)) {
1470 /* check for crc mismatch */
1471 if (av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2],
1472 s->frame_size - 2)) {
1473 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1474 if (avctx->err_recognition & AV_EF_EXPLODE)
1475 return AVERROR_INVALIDDATA;
1476 err = AAC_AC3_PARSE_ERROR_CRC;
1481 /* if frame is ok, set audio parameters */
1483 avctx->sample_rate = s->sample_rate;
1484 avctx->bit_rate = s->bit_rate;
1487 /* channel config */
1488 if (!err || (s->channels && s->out_channels != s->channels)) {
1489 s->out_channels = s->channels;
1490 s->output_mode = s->channel_mode;
1492 s->output_mode |= AC3_OUTPUT_LFEON;
1493 if (s->channels > 1 &&
1494 avctx->request_channel_layout == AV_CH_LAYOUT_MONO) {
1495 s->out_channels = 1;
1496 s->output_mode = AC3_CHMODE_MONO;
1497 } else if (s->channels > 2 &&
1498 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
1499 s->out_channels = 2;
1500 s->output_mode = AC3_CHMODE_STEREO;
1503 s->loro_center_mix_level = gain_levels[s-> center_mix_level];
1504 s->loro_surround_mix_level = gain_levels[s->surround_mix_level];
1505 s->ltrt_center_mix_level = LEVEL_MINUS_3DB;
1506 s->ltrt_surround_mix_level = LEVEL_MINUS_3DB;
1507 /* set downmixing coefficients if needed */
1508 if (s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1509 s->fbw_channels == s->out_channels)) {
1510 set_downmix_coeffs(s);
1512 } else if (!s->channels) {
1513 av_log(avctx, AV_LOG_ERROR, "unable to determine channel mode\n");
1514 return AVERROR_INVALIDDATA;
1516 avctx->channels = s->out_channels;
1517 avctx->channel_layout = avpriv_ac3_channel_layout_tab[s->output_mode & ~AC3_OUTPUT_LFEON];
1518 if (s->output_mode & AC3_OUTPUT_LFEON)
1519 avctx->channel_layout |= AV_CH_LOW_FREQUENCY;
1521 /* set audio service type based on bitstream mode for AC-3 */
1522 avctx->audio_service_type = s->bitstream_mode;
1523 if (s->bitstream_mode == 0x7 && s->channels > 1)
1524 avctx->audio_service_type = AV_AUDIO_SERVICE_TYPE_KARAOKE;
1526 /* get output buffer */
1527 frame->nb_samples = s->num_blocks * AC3_BLOCK_SIZE;
1528 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1531 /* decode the audio blocks */
1532 channel_map = ff_ac3_dec_channel_map[s->output_mode & ~AC3_OUTPUT_LFEON][s->lfe_on];
1533 for (ch = 0; ch < AC3_MAX_CHANNELS; ch++) {
1534 output[ch] = s->output[ch];
1535 s->outptr[ch] = s->output[ch];
1537 for (ch = 0; ch < s->channels; ch++) {
1538 if (ch < s->out_channels)
1539 s->outptr[channel_map[ch]] = (SHORTFLOAT *)frame->data[ch];
1541 for (blk = 0; blk < s->num_blocks; blk++) {
1542 if (!err && decode_audio_block(s, blk)) {
1543 av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n");
1547 for (ch = 0; ch < s->out_channels; ch++)
1548 memcpy(((SHORTFLOAT*)frame->data[ch]) + AC3_BLOCK_SIZE*blk, output[ch], AC3_BLOCK_SIZE*sizeof(SHORTFLOAT));
1549 for (ch = 0; ch < s->out_channels; ch++)
1550 output[ch] = s->outptr[channel_map[ch]];
1551 for (ch = 0; ch < s->out_channels; ch++) {
1552 if (!ch || channel_map[ch])
1553 s->outptr[channel_map[ch]] += AC3_BLOCK_SIZE;
1557 av_frame_set_decode_error_flags(frame, err ? FF_DECODE_ERROR_INVALID_BITSTREAM : 0);
1559 /* keep last block for error concealment in next frame */
1560 for (ch = 0; ch < s->out_channels; ch++)
1561 memcpy(s->output[ch], output[ch], AC3_BLOCK_SIZE*sizeof(SHORTFLOAT));
1566 * Check whether the input layout is compatible, and make sure we're not
1567 * downmixing (else the matrix encoding is no longer applicable).
1569 matrix_encoding = AV_MATRIX_ENCODING_NONE;
1570 if (s->channel_mode == AC3_CHMODE_STEREO &&
1571 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1572 if (s->dolby_surround_mode == AC3_DSURMOD_ON)
1573 matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
1574 else if (s->dolby_headphone_mode == AC3_DHEADPHONMOD_ON)
1575 matrix_encoding = AV_MATRIX_ENCODING_DOLBYHEADPHONE;
1576 } else if (s->channel_mode >= AC3_CHMODE_2F2R &&
1577 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1578 switch (s->dolby_surround_ex_mode) {
1579 case AC3_DSUREXMOD_ON: // EX or PLIIx
1580 matrix_encoding = AV_MATRIX_ENCODING_DOLBYEX;
1582 case AC3_DSUREXMOD_PLIIZ:
1583 matrix_encoding = AV_MATRIX_ENCODING_DPLIIZ;
1585 default: // not indicated or off
1589 if ((ret = ff_side_data_update_matrix_encoding(frame, matrix_encoding)) < 0)
1593 if ((downmix_info = av_downmix_info_update_side_data(frame))) {
1594 switch (s->preferred_downmix) {
1595 case AC3_DMIXMOD_LTRT:
1596 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LTRT;
1598 case AC3_DMIXMOD_LORO:
1599 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LORO;
1601 case AC3_DMIXMOD_DPLII:
1602 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_DPLII;
1605 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_UNKNOWN;
1608 downmix_info->center_mix_level = gain_levels[s-> center_mix_level];
1609 downmix_info->center_mix_level_ltrt = gain_levels[s-> center_mix_level_ltrt];
1610 downmix_info->surround_mix_level = gain_levels[s-> surround_mix_level];
1611 downmix_info->surround_mix_level_ltrt = gain_levels[s->surround_mix_level_ltrt];
1612 if (s->lfe_mix_level_exists)
1613 downmix_info->lfe_mix_level = gain_levels_lfe[s->lfe_mix_level];
1615 downmix_info->lfe_mix_level = 0.0; // -inf dB
1617 return AVERROR(ENOMEM);
1621 return FFMIN(buf_size, s->frame_size);
1625 * Uninitialize the AC-3 decoder.
1627 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1629 AC3DecodeContext *s = avctx->priv_data;
1630 ff_mdct_end(&s->imdct_512);
1631 ff_mdct_end(&s->imdct_256);
1637 #define OFFSET(x) offsetof(AC3DecodeContext, x)
1638 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)