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,
899 for (ch = 1; ch <= fbw_channels; ch++) {
900 s->channel_uses_spx[ch] = 0;
901 s->first_spx_coords[ch] = 1;
906 /* spectral extension coordinates */
908 for (ch = 1; ch <= fbw_channels; ch++) {
909 if (s->channel_uses_spx[ch]) {
910 if (s->first_spx_coords[ch] || get_bits1(gbc)) {
912 int bin, master_spx_coord;
914 s->first_spx_coords[ch] = 0;
915 spx_blend = AC3_SPX_BLEND(get_bits(gbc, 5));
916 master_spx_coord = get_bits(gbc, 2) * 3;
918 bin = s->spx_src_start_freq;
919 for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
920 int bandsize = s->spx_band_sizes[bnd];
921 int spx_coord_exp, spx_coord_mant;
922 INTFLOAT nratio, sblend, nblend;
924 /* calculate blending factors */
925 int64_t accu = ((bin << 23) + (bandsize << 22))
926 * (int64_t)s->spx_dst_end_freq;
927 nratio = (int)(accu >> 32);
928 nratio -= spx_blend << 18;
933 } else if (nratio > 0x7fffff) {
934 nblend = 14529495; // sqrt(3) in FP.23
937 nblend = fixed_sqrt(nratio, 23);
938 accu = (int64_t)nblend * 1859775393;
939 nblend = (int)((accu + (1<<29)) >> 30);
940 sblend = fixed_sqrt(0x800000 - nratio, 23);
945 /* calculate blending factors */
946 nratio = ((float)((bin + (bandsize >> 1))) / s->spx_dst_end_freq) - spx_blend;
947 nratio = av_clipf(nratio, 0.0f, 1.0f);
948 nblend = sqrtf(3.0f * nratio); // noise is scaled by sqrt(3)
949 // to give unity variance
950 sblend = sqrtf(1.0f - nratio);
954 /* decode spx coordinates */
955 spx_coord_exp = get_bits(gbc, 4);
956 spx_coord_mant = get_bits(gbc, 2);
957 if (spx_coord_exp == 15) spx_coord_mant <<= 1;
958 else spx_coord_mant += 4;
959 spx_coord_mant <<= (25 - spx_coord_exp - master_spx_coord);
961 /* multiply noise and signal blending factors by spx coordinate */
963 accu = (int64_t)nblend * spx_coord_mant;
964 s->spx_noise_blend[ch][bnd] = (int)((accu + (1<<22)) >> 23);
965 accu = (int64_t)sblend * spx_coord_mant;
966 s->spx_signal_blend[ch][bnd] = (int)((accu + (1<<22)) >> 23);
968 spx_coord = spx_coord_mant * (1.0f / (1 << 23));
969 s->spx_noise_blend [ch][bnd] = nblend * spx_coord;
970 s->spx_signal_blend[ch][bnd] = sblend * spx_coord;
975 s->first_spx_coords[ch] = 1;
980 /* coupling strategy */
981 if (s->eac3 ? s->cpl_strategy_exists[blk] : get_bits1(gbc)) {
982 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
984 s->cpl_in_use[blk] = get_bits1(gbc);
985 if (s->cpl_in_use[blk]) {
986 /* coupling in use */
987 int cpl_start_subband, cpl_end_subband;
989 if (channel_mode < AC3_CHMODE_STEREO) {
990 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
991 return AVERROR_INVALIDDATA;
994 /* check for enhanced coupling */
995 if (s->eac3 && get_bits1(gbc)) {
996 /* TODO: parse enhanced coupling strategy info */
997 avpriv_request_sample(s->avctx, "Enhanced coupling");
998 return AVERROR_PATCHWELCOME;
1001 /* determine which channels are coupled */
1002 if (s->eac3 && s->channel_mode == AC3_CHMODE_STEREO) {
1003 s->channel_in_cpl[1] = 1;
1004 s->channel_in_cpl[2] = 1;
1006 for (ch = 1; ch <= fbw_channels; ch++)
1007 s->channel_in_cpl[ch] = get_bits1(gbc);
1010 /* phase flags in use */
1011 if (channel_mode == AC3_CHMODE_STEREO)
1012 s->phase_flags_in_use = get_bits1(gbc);
1014 /* coupling frequency range */
1015 cpl_start_subband = get_bits(gbc, 4);
1016 cpl_end_subband = s->spx_in_use ? (s->spx_src_start_freq - 37) / 12 :
1017 get_bits(gbc, 4) + 3;
1018 if (cpl_start_subband >= cpl_end_subband) {
1019 av_log(s->avctx, AV_LOG_ERROR, "invalid coupling range (%d >= %d)\n",
1020 cpl_start_subband, cpl_end_subband);
1021 return AVERROR_INVALIDDATA;
1023 s->start_freq[CPL_CH] = cpl_start_subband * 12 + 37;
1024 s->end_freq[CPL_CH] = cpl_end_subband * 12 + 37;
1026 decode_band_structure(gbc, blk, s->eac3, 0, cpl_start_subband,
1028 ff_eac3_default_cpl_band_struct,
1029 &s->num_cpl_bands, s->cpl_band_sizes);
1031 /* coupling not in use */
1032 for (ch = 1; ch <= fbw_channels; ch++) {
1033 s->channel_in_cpl[ch] = 0;
1034 s->first_cpl_coords[ch] = 1;
1036 s->first_cpl_leak = s->eac3;
1037 s->phase_flags_in_use = 0;
1039 } else if (!s->eac3) {
1041 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must "
1042 "be present in block 0\n");
1043 return AVERROR_INVALIDDATA;
1045 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
1048 cpl_in_use = s->cpl_in_use[blk];
1050 /* coupling coordinates */
1052 int cpl_coords_exist = 0;
1054 for (ch = 1; ch <= fbw_channels; ch++) {
1055 if (s->channel_in_cpl[ch]) {
1056 if ((s->eac3 && s->first_cpl_coords[ch]) || get_bits1(gbc)) {
1057 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
1058 s->first_cpl_coords[ch] = 0;
1059 cpl_coords_exist = 1;
1060 master_cpl_coord = 3 * get_bits(gbc, 2);
1061 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
1062 cpl_coord_exp = get_bits(gbc, 4);
1063 cpl_coord_mant = get_bits(gbc, 4);
1064 if (cpl_coord_exp == 15)
1065 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
1067 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
1068 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
1071 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must "
1072 "be present in block 0\n");
1073 return AVERROR_INVALIDDATA;
1076 /* channel not in coupling */
1077 s->first_cpl_coords[ch] = 1;
1081 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
1082 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
1083 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
1088 /* stereo rematrixing strategy and band structure */
1089 if (channel_mode == AC3_CHMODE_STEREO) {
1090 if ((s->eac3 && !blk) || get_bits1(gbc)) {
1091 s->num_rematrixing_bands = 4;
1092 if (cpl_in_use && s->start_freq[CPL_CH] <= 61) {
1093 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
1094 } else if (s->spx_in_use && s->spx_src_start_freq <= 61) {
1095 s->num_rematrixing_bands--;
1097 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++)
1098 s->rematrixing_flags[bnd] = get_bits1(gbc);
1100 av_log(s->avctx, AV_LOG_WARNING, "Warning: "
1101 "new rematrixing strategy not present in block 0\n");
1102 s->num_rematrixing_bands = 0;
1106 /* exponent strategies for each channel */
1107 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1109 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
1110 if (s->exp_strategy[blk][ch] != EXP_REUSE)
1111 bit_alloc_stages[ch] = 3;
1114 /* channel bandwidth */
1115 for (ch = 1; ch <= fbw_channels; ch++) {
1116 s->start_freq[ch] = 0;
1117 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1119 int prev = s->end_freq[ch];
1120 if (s->channel_in_cpl[ch])
1121 s->end_freq[ch] = s->start_freq[CPL_CH];
1122 else if (s->channel_uses_spx[ch])
1123 s->end_freq[ch] = s->spx_src_start_freq;
1125 int bandwidth_code = get_bits(gbc, 6);
1126 if (bandwidth_code > 60) {
1127 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60\n", bandwidth_code);
1128 return AVERROR_INVALIDDATA;
1130 s->end_freq[ch] = bandwidth_code * 3 + 73;
1132 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
1133 s->num_exp_groups[ch] = (s->end_freq[ch] + group_size-4) / group_size;
1134 if (blk > 0 && s->end_freq[ch] != prev)
1135 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
1138 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
1139 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
1140 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
1143 /* decode exponents for each channel */
1144 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1145 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1146 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
1147 if (decode_exponents(s, gbc, s->exp_strategy[blk][ch],
1148 s->num_exp_groups[ch], s->dexps[ch][0],
1149 &s->dexps[ch][s->start_freq[ch]+!!ch])) {
1150 return AVERROR_INVALIDDATA;
1152 if (ch != CPL_CH && ch != s->lfe_ch)
1153 skip_bits(gbc, 2); /* skip gainrng */
1157 /* bit allocation information */
1158 if (s->bit_allocation_syntax) {
1159 if (get_bits1(gbc)) {
1160 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1161 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1162 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
1163 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
1164 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
1165 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1166 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1168 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must "
1169 "be present in block 0\n");
1170 return AVERROR_INVALIDDATA;
1174 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
1175 if (!s->eac3 || !blk) {
1176 if (s->snr_offset_strategy && get_bits1(gbc)) {
1179 csnr = (get_bits(gbc, 6) - 15) << 4;
1180 for (i = ch = !cpl_in_use; ch <= s->channels; ch++) {
1182 if (ch == i || s->snr_offset_strategy == 2)
1183 snr = (csnr + get_bits(gbc, 4)) << 2;
1184 /* run at least last bit allocation stage if snr offset changes */
1185 if (blk && s->snr_offset[ch] != snr) {
1186 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 1);
1188 s->snr_offset[ch] = snr;
1190 /* fast gain (normal AC-3 only) */
1192 int prev = s->fast_gain[ch];
1193 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1194 /* run last 2 bit allocation stages if fast gain changes */
1195 if (blk && prev != s->fast_gain[ch])
1196 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1199 } else if (!s->eac3 && !blk) {
1200 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
1201 return AVERROR_INVALIDDATA;
1205 /* fast gain (E-AC-3 only) */
1206 if (s->fast_gain_syntax && get_bits1(gbc)) {
1207 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1208 int prev = s->fast_gain[ch];
1209 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1210 /* run last 2 bit allocation stages if fast gain changes */
1211 if (blk && prev != s->fast_gain[ch])
1212 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1214 } else if (s->eac3 && !blk) {
1215 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1216 s->fast_gain[ch] = ff_ac3_fast_gain_tab[4];
1219 /* E-AC-3 to AC-3 converter SNR offset */
1220 if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT && get_bits1(gbc)) {
1221 skip_bits(gbc, 10); // skip converter snr offset
1224 /* coupling leak information */
1226 if (s->first_cpl_leak || get_bits1(gbc)) {
1227 int fl = get_bits(gbc, 3);
1228 int sl = get_bits(gbc, 3);
1229 /* run last 2 bit allocation stages for coupling channel if
1230 coupling leak changes */
1231 if (blk && (fl != s->bit_alloc_params.cpl_fast_leak ||
1232 sl != s->bit_alloc_params.cpl_slow_leak)) {
1233 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
1235 s->bit_alloc_params.cpl_fast_leak = fl;
1236 s->bit_alloc_params.cpl_slow_leak = sl;
1237 } else if (!s->eac3 && !blk) {
1238 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must "
1239 "be present in block 0\n");
1240 return AVERROR_INVALIDDATA;
1242 s->first_cpl_leak = 0;
1245 /* delta bit allocation information */
1246 if (s->dba_syntax && get_bits1(gbc)) {
1247 /* delta bit allocation exists (strategy) */
1248 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1249 s->dba_mode[ch] = get_bits(gbc, 2);
1250 if (s->dba_mode[ch] == DBA_RESERVED) {
1251 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1252 return AVERROR_INVALIDDATA;
1254 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1256 /* channel delta offset, len and bit allocation */
1257 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1258 if (s->dba_mode[ch] == DBA_NEW) {
1259 s->dba_nsegs[ch] = get_bits(gbc, 3) + 1;
1260 for (seg = 0; seg < s->dba_nsegs[ch]; seg++) {
1261 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1262 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1263 s->dba_values[ch][seg] = get_bits(gbc, 3);
1265 /* run last 2 bit allocation stages if new dba values */
1266 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1269 } else if (blk == 0) {
1270 for (ch = 0; ch <= s->channels; ch++) {
1271 s->dba_mode[ch] = DBA_NONE;
1275 /* Bit allocation */
1276 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1277 if (bit_alloc_stages[ch] > 2) {
1278 /* Exponent mapping into PSD and PSD integration */
1279 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1280 s->start_freq[ch], s->end_freq[ch],
1281 s->psd[ch], s->band_psd[ch]);
1283 if (bit_alloc_stages[ch] > 1) {
1284 /* Compute excitation function, Compute masking curve, and
1285 Apply delta bit allocation */
1286 if (ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1287 s->start_freq[ch], s->end_freq[ch],
1288 s->fast_gain[ch], (ch == s->lfe_ch),
1289 s->dba_mode[ch], s->dba_nsegs[ch],
1290 s->dba_offsets[ch], s->dba_lengths[ch],
1291 s->dba_values[ch], s->mask[ch])) {
1292 av_log(s->avctx, AV_LOG_ERROR, "error in bit allocation\n");
1293 return AVERROR_INVALIDDATA;
1296 if (bit_alloc_stages[ch] > 0) {
1297 /* Compute bit allocation */
1298 const uint8_t *bap_tab = s->channel_uses_aht[ch] ?
1299 ff_eac3_hebap_tab : ff_ac3_bap_tab;
1300 s->ac3dsp.bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1301 s->start_freq[ch], s->end_freq[ch],
1303 s->bit_alloc_params.floor,
1304 bap_tab, s->bap[ch]);
1308 /* unused dummy data */
1309 if (s->skip_syntax && get_bits1(gbc)) {
1310 int skipl = get_bits(gbc, 9);
1315 /* unpack the transform coefficients
1316 this also uncouples channels if coupling is in use. */
1317 decode_transform_coeffs(s, blk);
1319 /* TODO: generate enhanced coupling coordinates and uncouple */
1321 /* recover coefficients if rematrixing is in use */
1322 if (s->channel_mode == AC3_CHMODE_STEREO)
1325 /* apply scaling to coefficients (headroom, dynrng) */
1326 for (ch = 1; ch <= s->channels; ch++) {
1327 int audio_channel = 0;
1329 if (s->channel_mode == AC3_CHMODE_DUALMONO)
1330 audio_channel = 2-ch;
1331 if (s->heavy_compression && s->compression_exists[audio_channel])
1332 gain = s->heavy_dynamic_range[audio_channel];
1334 gain = s->dynamic_range[audio_channel];
1337 scale_coefs(s->transform_coeffs[ch], s->fixed_coeffs[ch], gain, 256);
1339 if (s->target_level != 0)
1340 gain = gain * s->level_gain[audio_channel];
1341 gain *= 1.0 / 4194304.0f;
1342 s->fmt_conv.int32_to_float_fmul_scalar(s->transform_coeffs[ch],
1343 s->fixed_coeffs[ch], gain, 256);
1347 /* apply spectral extension to high frequency bins */
1348 if (CONFIG_EAC3_DECODER && s->spx_in_use) {
1349 ff_eac3_apply_spectral_extension(s);
1352 /* downmix and MDCT. order depends on whether block switching is used for
1353 any channel in this block. this is because coefficients for the long
1354 and short transforms cannot be mixed. */
1355 downmix_output = s->channels != s->out_channels &&
1356 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1357 s->fbw_channels == s->out_channels);
1358 if (different_transforms) {
1359 /* the delay samples have already been downmixed, so we upmix the delay
1360 samples in order to reconstruct all channels before downmixing. */
1366 do_imdct(s, s->channels);
1368 if (downmix_output) {
1370 ac3_downmix_c_fixed16(s->outptr, s->downmix_coeffs,
1371 s->out_channels, s->fbw_channels, 256);
1373 s->ac3dsp.downmix(s->outptr, s->downmix_coeffs,
1374 s->out_channels, s->fbw_channels, 256);
1378 if (downmix_output) {
1379 s->ac3dsp.AC3_RENAME(downmix)(s->xcfptr + 1, s->downmix_coeffs,
1380 s->out_channels, s->fbw_channels, 256);
1383 if (downmix_output && !s->downmixed) {
1385 s->ac3dsp.AC3_RENAME(downmix)(s->dlyptr, s->downmix_coeffs,
1386 s->out_channels, s->fbw_channels, 128);
1389 do_imdct(s, s->out_channels);
1396 * Decode a single AC-3 frame.
1398 static int ac3_decode_frame(AVCodecContext * avctx, void *data,
1399 int *got_frame_ptr, AVPacket *avpkt)
1401 AVFrame *frame = data;
1402 const uint8_t *buf = avpkt->data;
1403 int buf_size = avpkt->size;
1404 AC3DecodeContext *s = avctx->priv_data;
1405 int blk, ch, err, ret;
1406 const uint8_t *channel_map;
1407 const SHORTFLOAT *output[AC3_MAX_CHANNELS];
1408 enum AVMatrixEncoding matrix_encoding;
1409 AVDownmixInfo *downmix_info;
1411 /* copy input buffer to decoder context to avoid reading past the end
1412 of the buffer, which can be caused by a damaged input stream. */
1413 if (buf_size >= 2 && AV_RB16(buf) == 0x770B) {
1414 // seems to be byte-swapped AC-3
1415 int cnt = FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE) >> 1;
1416 s->bdsp.bswap16_buf((uint16_t *) s->input_buffer,
1417 (const uint16_t *) buf, cnt);
1419 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE));
1420 buf = s->input_buffer;
1421 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1422 if ((ret = init_get_bits8(&s->gbc, buf, buf_size)) < 0)
1425 /* parse the syncinfo */
1426 err = parse_frame_header(s);
1430 case AAC_AC3_PARSE_ERROR_SYNC:
1431 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1432 return AVERROR_INVALIDDATA;
1433 case AAC_AC3_PARSE_ERROR_BSID:
1434 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1436 case AAC_AC3_PARSE_ERROR_SAMPLE_RATE:
1437 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1439 case AAC_AC3_PARSE_ERROR_FRAME_SIZE:
1440 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1442 case AAC_AC3_PARSE_ERROR_FRAME_TYPE:
1443 /* skip frame if CRC is ok. otherwise use error concealment. */
1444 /* TODO: add support for substreams and dependent frames */
1445 if (s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1446 av_log(avctx, AV_LOG_WARNING, "unsupported frame type : "
1447 "skipping frame\n");
1451 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1454 case AAC_AC3_PARSE_ERROR_CRC:
1455 case AAC_AC3_PARSE_ERROR_CHANNEL_CFG:
1457 default: // Normal AVERROR do not try to recover.
1462 /* check that reported frame size fits in input buffer */
1463 if (s->frame_size > buf_size) {
1464 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1465 err = AAC_AC3_PARSE_ERROR_FRAME_SIZE;
1466 } else if (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL)) {
1467 /* check for crc mismatch */
1468 if (av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2],
1469 s->frame_size - 2)) {
1470 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1471 if (avctx->err_recognition & AV_EF_EXPLODE)
1472 return AVERROR_INVALIDDATA;
1473 err = AAC_AC3_PARSE_ERROR_CRC;
1478 /* if frame is ok, set audio parameters */
1480 avctx->sample_rate = s->sample_rate;
1481 avctx->bit_rate = s->bit_rate;
1484 /* channel config */
1485 if (!err || (s->channels && s->out_channels != s->channels)) {
1486 s->out_channels = s->channels;
1487 s->output_mode = s->channel_mode;
1489 s->output_mode |= AC3_OUTPUT_LFEON;
1490 if (s->channels > 1 &&
1491 avctx->request_channel_layout == AV_CH_LAYOUT_MONO) {
1492 s->out_channels = 1;
1493 s->output_mode = AC3_CHMODE_MONO;
1494 } else if (s->channels > 2 &&
1495 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
1496 s->out_channels = 2;
1497 s->output_mode = AC3_CHMODE_STEREO;
1500 s->loro_center_mix_level = gain_levels[s-> center_mix_level];
1501 s->loro_surround_mix_level = gain_levels[s->surround_mix_level];
1502 s->ltrt_center_mix_level = LEVEL_MINUS_3DB;
1503 s->ltrt_surround_mix_level = LEVEL_MINUS_3DB;
1504 /* set downmixing coefficients if needed */
1505 if (s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1506 s->fbw_channels == s->out_channels)) {
1507 set_downmix_coeffs(s);
1509 } else if (!s->channels) {
1510 av_log(avctx, AV_LOG_ERROR, "unable to determine channel mode\n");
1511 return AVERROR_INVALIDDATA;
1513 avctx->channels = s->out_channels;
1514 avctx->channel_layout = avpriv_ac3_channel_layout_tab[s->output_mode & ~AC3_OUTPUT_LFEON];
1515 if (s->output_mode & AC3_OUTPUT_LFEON)
1516 avctx->channel_layout |= AV_CH_LOW_FREQUENCY;
1518 /* set audio service type based on bitstream mode for AC-3 */
1519 avctx->audio_service_type = s->bitstream_mode;
1520 if (s->bitstream_mode == 0x7 && s->channels > 1)
1521 avctx->audio_service_type = AV_AUDIO_SERVICE_TYPE_KARAOKE;
1523 /* get output buffer */
1524 frame->nb_samples = s->num_blocks * AC3_BLOCK_SIZE;
1525 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1528 /* decode the audio blocks */
1529 channel_map = ff_ac3_dec_channel_map[s->output_mode & ~AC3_OUTPUT_LFEON][s->lfe_on];
1530 for (ch = 0; ch < AC3_MAX_CHANNELS; ch++) {
1531 output[ch] = s->output[ch];
1532 s->outptr[ch] = s->output[ch];
1534 for (ch = 0; ch < s->channels; ch++) {
1535 if (ch < s->out_channels)
1536 s->outptr[channel_map[ch]] = (SHORTFLOAT *)frame->data[ch];
1538 for (blk = 0; blk < s->num_blocks; blk++) {
1539 if (!err && decode_audio_block(s, blk)) {
1540 av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n");
1544 for (ch = 0; ch < s->out_channels; ch++)
1545 memcpy(((SHORTFLOAT*)frame->data[ch]) + AC3_BLOCK_SIZE*blk, output[ch], AC3_BLOCK_SIZE*sizeof(SHORTFLOAT));
1546 for (ch = 0; ch < s->out_channels; ch++)
1547 output[ch] = s->outptr[channel_map[ch]];
1548 for (ch = 0; ch < s->out_channels; ch++) {
1549 if (!ch || channel_map[ch])
1550 s->outptr[channel_map[ch]] += AC3_BLOCK_SIZE;
1554 av_frame_set_decode_error_flags(frame, err ? FF_DECODE_ERROR_INVALID_BITSTREAM : 0);
1556 /* keep last block for error concealment in next frame */
1557 for (ch = 0; ch < s->out_channels; ch++)
1558 memcpy(s->output[ch], output[ch], AC3_BLOCK_SIZE*sizeof(SHORTFLOAT));
1563 * Check whether the input layout is compatible, and make sure we're not
1564 * downmixing (else the matrix encoding is no longer applicable).
1566 matrix_encoding = AV_MATRIX_ENCODING_NONE;
1567 if (s->channel_mode == AC3_CHMODE_STEREO &&
1568 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1569 if (s->dolby_surround_mode == AC3_DSURMOD_ON)
1570 matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
1571 else if (s->dolby_headphone_mode == AC3_DHEADPHONMOD_ON)
1572 matrix_encoding = AV_MATRIX_ENCODING_DOLBYHEADPHONE;
1573 } else if (s->channel_mode >= AC3_CHMODE_2F2R &&
1574 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1575 switch (s->dolby_surround_ex_mode) {
1576 case AC3_DSUREXMOD_ON: // EX or PLIIx
1577 matrix_encoding = AV_MATRIX_ENCODING_DOLBYEX;
1579 case AC3_DSUREXMOD_PLIIZ:
1580 matrix_encoding = AV_MATRIX_ENCODING_DPLIIZ;
1582 default: // not indicated or off
1586 if ((ret = ff_side_data_update_matrix_encoding(frame, matrix_encoding)) < 0)
1590 if ((downmix_info = av_downmix_info_update_side_data(frame))) {
1591 switch (s->preferred_downmix) {
1592 case AC3_DMIXMOD_LTRT:
1593 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LTRT;
1595 case AC3_DMIXMOD_LORO:
1596 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LORO;
1598 case AC3_DMIXMOD_DPLII:
1599 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_DPLII;
1602 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_UNKNOWN;
1605 downmix_info->center_mix_level = gain_levels[s-> center_mix_level];
1606 downmix_info->center_mix_level_ltrt = gain_levels[s-> center_mix_level_ltrt];
1607 downmix_info->surround_mix_level = gain_levels[s-> surround_mix_level];
1608 downmix_info->surround_mix_level_ltrt = gain_levels[s->surround_mix_level_ltrt];
1609 if (s->lfe_mix_level_exists)
1610 downmix_info->lfe_mix_level = gain_levels_lfe[s->lfe_mix_level];
1612 downmix_info->lfe_mix_level = 0.0; // -inf dB
1614 return AVERROR(ENOMEM);
1618 return FFMIN(buf_size, s->frame_size);
1622 * Uninitialize the AC-3 decoder.
1624 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1626 AC3DecodeContext *s = avctx->priv_data;
1627 ff_mdct_end(&s->imdct_512);
1628 ff_mdct_end(&s->imdct_256);
1634 #define OFFSET(x) offsetof(AC3DecodeContext, x)
1635 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)