3 * This code was developed as part of Google Summer of Code 2006.
4 * E-AC-3 support was added as part of Google Summer of Code 2007.
6 * Copyright (c) 2006 Kartikey Mahendra BHATT (bhattkm at gmail dot com)
7 * Copyright (c) 2007-2008 Bartlomiej Wolowiec <bartek.wolowiec@gmail.com>
8 * Copyright (c) 2007 Justin Ruggles <justin.ruggles@gmail.com>
10 * This file is part of FFmpeg.
12 * FFmpeg is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public
14 * License as published by the Free Software Foundation; either
15 * version 2.1 of the License, or (at your option) any later version.
17 * FFmpeg is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with FFmpeg; if not, write to the Free Software
24 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
32 #include "libavutil/channel_layout.h"
33 #include "libavutil/crc.h"
34 #include "libavutil/downmix_info.h"
35 #include "libavutil/opt.h"
38 #include "aac_ac3_parser.h"
39 #include "ac3_parser.h"
41 #include "ac3dec_data.h"
45 * table for ungrouping 3 values in 7 bits.
46 * used for exponents and bap=2 mantissas
48 static uint8_t ungroup_3_in_7_bits_tab[128][3];
50 /** tables for ungrouping mantissas */
51 static int b1_mantissas[32][3];
52 static int b2_mantissas[128][3];
53 static int b3_mantissas[8];
54 static int b4_mantissas[128][2];
55 static int b5_mantissas[16];
58 * Quantization table: levels for symmetric. bits for asymmetric.
59 * reference: Table 7.18 Mapping of bap to Quantizer
61 static const uint8_t quantization_tab[16] = {
63 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
67 /** dynamic range table. converts codes to scale factors. */
68 static float dynamic_range_tab[256];
69 float ff_ac3_heavy_dynamic_range_tab[256];
72 /** Adjustments in dB gain */
73 static const float gain_levels[9] = {
77 LEVEL_MINUS_1POINT5DB,
79 LEVEL_MINUS_4POINT5DB,
85 /** Adjustments in dB gain (LFE, +10 to -21 dB) */
86 static const float gain_levels_lfe[32] = {
87 3.162275, 2.818382, 2.511886, 2.238719, 1.995261, 1.778278, 1.584893,
88 1.412536, 1.258924, 1.122018, 1.000000, 0.891251, 0.794328, 0.707946,
89 0.630957, 0.562341, 0.501187, 0.446683, 0.398107, 0.354813, 0.316227,
90 0.281838, 0.251188, 0.223872, 0.199526, 0.177828, 0.158489, 0.141253,
91 0.125892, 0.112201, 0.100000, 0.089125
95 * Table for default stereo downmixing coefficients
96 * reference: Section 7.8.2 Downmixing Into Two Channels
98 static const uint8_t ac3_default_coeffs[8][5][2] = {
99 { { 2, 7 }, { 7, 2 }, },
101 { { 2, 7 }, { 7, 2 }, },
102 { { 2, 7 }, { 5, 5 }, { 7, 2 }, },
103 { { 2, 7 }, { 7, 2 }, { 6, 6 }, },
104 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 8, 8 }, },
105 { { 2, 7 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
106 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
110 * Symmetrical Dequantization
111 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
112 * Tables 7.19 to 7.23
115 symmetric_dequant(int code, int levels)
117 return ((code - (levels >> 1)) * (1 << 24)) / levels;
121 * Initialize tables at runtime.
123 static av_cold void ac3_tables_init(void)
127 /* generate table for ungrouping 3 values in 7 bits
128 reference: Section 7.1.3 Exponent Decoding */
129 for (i = 0; i < 128; i++) {
130 ungroup_3_in_7_bits_tab[i][0] = i / 25;
131 ungroup_3_in_7_bits_tab[i][1] = (i % 25) / 5;
132 ungroup_3_in_7_bits_tab[i][2] = (i % 25) % 5;
135 /* generate grouped mantissa tables
136 reference: Section 7.3.5 Ungrouping of Mantissas */
137 for (i = 0; i < 32; i++) {
138 /* bap=1 mantissas */
139 b1_mantissas[i][0] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][0], 3);
140 b1_mantissas[i][1] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][1], 3);
141 b1_mantissas[i][2] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][2], 3);
143 for (i = 0; i < 128; i++) {
144 /* bap=2 mantissas */
145 b2_mantissas[i][0] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][0], 5);
146 b2_mantissas[i][1] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][1], 5);
147 b2_mantissas[i][2] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][2], 5);
149 /* bap=4 mantissas */
150 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
151 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
153 /* generate ungrouped mantissa tables
154 reference: Tables 7.21 and 7.23 */
155 for (i = 0; i < 7; i++) {
156 /* bap=3 mantissas */
157 b3_mantissas[i] = symmetric_dequant(i, 7);
159 for (i = 0; i < 15; i++) {
160 /* bap=5 mantissas */
161 b5_mantissas[i] = symmetric_dequant(i, 15);
165 /* generate dynamic range table
166 reference: Section 7.7.1 Dynamic Range Control */
167 for (i = 0; i < 256; i++) {
168 int v = (i >> 5) - ((i >> 7) << 3) - 5;
169 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
172 /* generate compr dynamic range table
173 reference: Section 7.7.2 Heavy Compression */
174 for (i = 0; i < 256; i++) {
175 int v = (i >> 4) - ((i >> 7) << 4) - 4;
176 ff_ac3_heavy_dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0xF) | 0x10);
182 * AVCodec initialization
184 static av_cold int ac3_decode_init(AVCodecContext *avctx)
186 AC3DecodeContext *s = avctx->priv_data;
192 ff_mdct_init(&s->imdct_256, 8, 1, 1.0);
193 ff_mdct_init(&s->imdct_512, 9, 1, 1.0);
194 AC3_RENAME(ff_kbd_window_init)(s->window, 5.0, 256);
195 ff_bswapdsp_init(&s->bdsp);
198 s->fdsp = avpriv_alloc_fixed_dsp(avctx->flags & AV_CODEC_FLAG_BITEXACT);
200 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
201 ff_fmt_convert_init(&s->fmt_conv, avctx);
204 ff_ac3dsp_init(&s->ac3dsp, avctx->flags & AV_CODEC_FLAG_BITEXACT);
205 av_lfg_init(&s->dith_state, 0);
208 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
210 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
212 /* allow downmixing to stereo or mono */
213 if (avctx->channels > 1 &&
214 avctx->request_channel_layout == AV_CH_LAYOUT_MONO)
216 else if (avctx->channels > 2 &&
217 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO)
221 for (i = 0; i < AC3_MAX_CHANNELS; i++) {
222 s->xcfptr[i] = s->transform_coeffs[i];
223 s->dlyptr[i] = s->delay[i];
230 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
231 * GetBitContext within AC3DecodeContext must point to
232 * the start of the synchronized AC-3 bitstream.
234 static int ac3_parse_header(AC3DecodeContext *s)
236 GetBitContext *gbc = &s->gbc;
239 /* read the rest of the bsi. read twice for dual mono mode. */
240 i = !s->channel_mode;
242 s->dialog_normalization[(!s->channel_mode)-i] = -get_bits(gbc, 5);
243 if (s->dialog_normalization[(!s->channel_mode)-i] == 0) {
244 s->dialog_normalization[(!s->channel_mode)-i] = -31;
246 if (s->target_level != 0) {
247 s->level_gain[(!s->channel_mode)-i] = powf(2.0f,
248 (float)(s->target_level -
249 s->dialog_normalization[(!s->channel_mode)-i])/6.0f);
251 if (s->compression_exists[(!s->channel_mode)-i] = get_bits1(gbc)) {
252 s->heavy_dynamic_range[(!s->channel_mode)-i] =
253 AC3_HEAVY_RANGE(get_bits(gbc, 8));
256 skip_bits(gbc, 8); //skip language code
258 skip_bits(gbc, 7); //skip audio production information
261 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
263 /* skip the timecodes or parse the Alternate Bit Stream Syntax */
264 if (s->bitstream_id != 6) {
266 skip_bits(gbc, 14); //skip timecode1
268 skip_bits(gbc, 14); //skip timecode2
270 if (get_bits1(gbc)) {
271 s->preferred_downmix = get_bits(gbc, 2);
272 s->center_mix_level_ltrt = get_bits(gbc, 3);
273 s->surround_mix_level_ltrt = av_clip(get_bits(gbc, 3), 3, 7);
274 s->center_mix_level = get_bits(gbc, 3);
275 s->surround_mix_level = av_clip(get_bits(gbc, 3), 3, 7);
277 if (get_bits1(gbc)) {
278 s->dolby_surround_ex_mode = get_bits(gbc, 2);
279 s->dolby_headphone_mode = get_bits(gbc, 2);
280 skip_bits(gbc, 10); // skip adconvtyp (1), xbsi2 (8), encinfo (1)
284 /* skip additional bitstream info */
285 if (get_bits1(gbc)) {
286 i = get_bits(gbc, 6);
296 * Common function to parse AC-3 or E-AC-3 frame header
298 static int parse_frame_header(AC3DecodeContext *s)
300 AC3HeaderInfo hdr, *phdr=&hdr;
303 err = avpriv_ac3_parse_header(&s->gbc, &phdr);
307 /* get decoding parameters from header info */
308 s->bit_alloc_params.sr_code = hdr.sr_code;
309 s->bitstream_id = hdr.bitstream_id;
310 s->bitstream_mode = hdr.bitstream_mode;
311 s->channel_mode = hdr.channel_mode;
312 s->lfe_on = hdr.lfe_on;
313 s->bit_alloc_params.sr_shift = hdr.sr_shift;
314 s->sample_rate = hdr.sample_rate;
315 s->bit_rate = hdr.bit_rate;
316 s->channels = hdr.channels;
317 s->fbw_channels = s->channels - s->lfe_on;
318 s->lfe_ch = s->fbw_channels + 1;
319 s->frame_size = hdr.frame_size;
320 s->preferred_downmix = AC3_DMIXMOD_NOTINDICATED;
321 s->center_mix_level = hdr.center_mix_level;
322 s->center_mix_level_ltrt = 4; // -3.0dB
323 s->surround_mix_level = hdr.surround_mix_level;
324 s->surround_mix_level_ltrt = 4; // -3.0dB
325 s->lfe_mix_level_exists = 0;
326 s->num_blocks = hdr.num_blocks;
327 s->frame_type = hdr.frame_type;
328 s->substreamid = hdr.substreamid;
329 s->dolby_surround_mode = hdr.dolby_surround_mode;
330 s->dolby_surround_ex_mode = AC3_DSUREXMOD_NOTINDICATED;
331 s->dolby_headphone_mode = AC3_DHEADPHONMOD_NOTINDICATED;
334 s->start_freq[s->lfe_ch] = 0;
335 s->end_freq[s->lfe_ch] = 7;
336 s->num_exp_groups[s->lfe_ch] = 2;
337 s->channel_in_cpl[s->lfe_ch] = 0;
340 if (s->bitstream_id <= 10) {
342 s->snr_offset_strategy = 2;
343 s->block_switch_syntax = 1;
344 s->dither_flag_syntax = 1;
345 s->bit_allocation_syntax = 1;
346 s->fast_gain_syntax = 0;
347 s->first_cpl_leak = 0;
350 memset(s->channel_uses_aht, 0, sizeof(s->channel_uses_aht));
351 return ac3_parse_header(s);
352 } else if (CONFIG_EAC3_DECODER) {
354 return ff_eac3_parse_header(s);
356 av_log(s->avctx, AV_LOG_ERROR, "E-AC-3 support not compiled in\n");
357 return AVERROR(ENOSYS);
362 * Set stereo downmixing coefficients based on frame header info.
363 * reference: Section 7.8.2 Downmixing Into Two Channels
365 static void set_downmix_coeffs(AC3DecodeContext *s)
368 float cmix = gain_levels[s-> center_mix_level];
369 float smix = gain_levels[s->surround_mix_level];
371 float downmix_coeffs[AC3_MAX_CHANNELS][2];
373 for (i = 0; i < s->fbw_channels; i++) {
374 downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
375 downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
377 if (s->channel_mode > 1 && s->channel_mode & 1) {
378 downmix_coeffs[1][0] = downmix_coeffs[1][1] = cmix;
380 if (s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
381 int nf = s->channel_mode - 2;
382 downmix_coeffs[nf][0] = downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
384 if (s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
385 int nf = s->channel_mode - 4;
386 downmix_coeffs[nf][0] = downmix_coeffs[nf+1][1] = smix;
391 for (i = 0; i < s->fbw_channels; i++) {
392 norm0 += downmix_coeffs[i][0];
393 norm1 += downmix_coeffs[i][1];
395 norm0 = 1.0f / norm0;
396 norm1 = 1.0f / norm1;
397 for (i = 0; i < s->fbw_channels; i++) {
398 downmix_coeffs[i][0] *= norm0;
399 downmix_coeffs[i][1] *= norm1;
402 if (s->output_mode == AC3_CHMODE_MONO) {
403 for (i = 0; i < s->fbw_channels; i++)
404 downmix_coeffs[i][0] = (downmix_coeffs[i][0] +
405 downmix_coeffs[i][1]) * LEVEL_MINUS_3DB;
407 for (i = 0; i < s->fbw_channels; i++) {
408 s->downmix_coeffs[i][0] = FIXR12(downmix_coeffs[i][0]);
409 s->downmix_coeffs[i][1] = FIXR12(downmix_coeffs[i][1]);
414 * Decode the grouped exponents according to exponent strategy.
415 * reference: Section 7.1.3 Exponent Decoding
417 static int decode_exponents(AC3DecodeContext *s,
418 GetBitContext *gbc, int exp_strategy, int ngrps,
419 uint8_t absexp, int8_t *dexps)
421 int i, j, grp, group_size;
426 group_size = exp_strategy + (exp_strategy == EXP_D45);
427 for (grp = 0, i = 0; grp < ngrps; grp++) {
428 expacc = get_bits(gbc, 7);
430 av_log(s->avctx, AV_LOG_ERROR, "expacc %d is out-of-range\n", expacc);
431 return AVERROR_INVALIDDATA;
433 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][0];
434 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][1];
435 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][2];
438 /* convert to absolute exps and expand groups */
440 for (i = 0, j = 0; i < ngrps * 3; i++) {
441 prevexp += dexp[i] - 2;
443 av_log(s->avctx, AV_LOG_ERROR, "exponent %d is out-of-range\n", prevexp);
446 switch (group_size) {
447 case 4: dexps[j++] = prevexp;
448 dexps[j++] = prevexp;
449 case 2: dexps[j++] = prevexp;
450 case 1: dexps[j++] = prevexp;
457 * Generate transform coefficients for each coupled channel in the coupling
458 * range using the coupling coefficients and coupling coordinates.
459 * reference: Section 7.4.3 Coupling Coordinate Format
461 static void calc_transform_coeffs_cpl(AC3DecodeContext *s)
465 bin = s->start_freq[CPL_CH];
466 for (band = 0; band < s->num_cpl_bands; band++) {
467 int band_start = bin;
468 int band_end = bin + s->cpl_band_sizes[band];
469 for (ch = 1; ch <= s->fbw_channels; ch++) {
470 if (s->channel_in_cpl[ch]) {
471 int cpl_coord = s->cpl_coords[ch][band] << 5;
472 for (bin = band_start; bin < band_end; bin++) {
473 s->fixed_coeffs[ch][bin] =
474 MULH(s->fixed_coeffs[CPL_CH][bin] * (1 << 4), cpl_coord);
476 if (ch == 2 && s->phase_flags[band]) {
477 for (bin = band_start; bin < band_end; bin++)
478 s->fixed_coeffs[2][bin] = -s->fixed_coeffs[2][bin];
487 * Grouped mantissas for 3-level 5-level and 11-level quantization
489 typedef struct mant_groups {
499 * Decode the transform coefficients for a particular channel
500 * reference: Section 7.3 Quantization and Decoding of Mantissas
502 static void ac3_decode_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
504 int start_freq = s->start_freq[ch_index];
505 int end_freq = s->end_freq[ch_index];
506 uint8_t *baps = s->bap[ch_index];
507 int8_t *exps = s->dexps[ch_index];
508 int32_t *coeffs = s->fixed_coeffs[ch_index];
509 int dither = (ch_index == CPL_CH) || s->dither_flag[ch_index];
510 GetBitContext *gbc = &s->gbc;
513 for (freq = start_freq; freq < end_freq; freq++) {
514 int bap = baps[freq];
518 /* random noise with approximate range of -0.707 to 0.707 */
520 mantissa = (((av_lfg_get(&s->dith_state)>>8)*181)>>8) - 5931008;
527 mantissa = m->b1_mant[m->b1];
529 int bits = get_bits(gbc, 5);
530 mantissa = b1_mantissas[bits][0];
531 m->b1_mant[1] = b1_mantissas[bits][1];
532 m->b1_mant[0] = b1_mantissas[bits][2];
539 mantissa = m->b2_mant[m->b2];
541 int bits = get_bits(gbc, 7);
542 mantissa = b2_mantissas[bits][0];
543 m->b2_mant[1] = b2_mantissas[bits][1];
544 m->b2_mant[0] = b2_mantissas[bits][2];
549 mantissa = b3_mantissas[get_bits(gbc, 3)];
554 mantissa = m->b4_mant;
556 int bits = get_bits(gbc, 7);
557 mantissa = b4_mantissas[bits][0];
558 m->b4_mant = b4_mantissas[bits][1];
563 mantissa = b5_mantissas[get_bits(gbc, 4)];
565 default: /* 6 to 15 */
566 /* Shift mantissa and sign-extend it. */
568 av_log(s->avctx, AV_LOG_ERROR, "bap %d is invalid in plain AC-3\n", bap);
571 mantissa = (unsigned)get_sbits(gbc, quantization_tab[bap]) << (24 - quantization_tab[bap]);
574 coeffs[freq] = mantissa >> exps[freq];
579 * Remove random dithering from coupling range coefficients with zero-bit
580 * mantissas for coupled channels which do not use dithering.
581 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
583 static void remove_dithering(AC3DecodeContext *s) {
586 for (ch = 1; ch <= s->fbw_channels; ch++) {
587 if (!s->dither_flag[ch] && s->channel_in_cpl[ch]) {
588 for (i = s->start_freq[CPL_CH]; i < s->end_freq[CPL_CH]; i++) {
589 if (!s->bap[CPL_CH][i])
590 s->fixed_coeffs[ch][i] = 0;
596 static void decode_transform_coeffs_ch(AC3DecodeContext *s, int blk, int ch,
599 if (!s->channel_uses_aht[ch]) {
600 ac3_decode_transform_coeffs_ch(s, ch, m);
602 /* if AHT is used, mantissas for all blocks are encoded in the first
603 block of the frame. */
605 if (CONFIG_EAC3_DECODER && !blk)
606 ff_eac3_decode_transform_coeffs_aht_ch(s, ch);
607 for (bin = s->start_freq[ch]; bin < s->end_freq[ch]; bin++) {
608 s->fixed_coeffs[ch][bin] = s->pre_mantissa[ch][bin][blk] >> s->dexps[ch][bin];
614 * Decode the transform coefficients.
616 static void decode_transform_coeffs(AC3DecodeContext *s, int blk)
622 m.b1 = m.b2 = m.b4 = 0;
624 for (ch = 1; ch <= s->channels; ch++) {
625 /* transform coefficients for full-bandwidth channel */
626 decode_transform_coeffs_ch(s, blk, ch, &m);
627 /* transform coefficients for coupling channel come right after the
628 coefficients for the first coupled channel*/
629 if (s->channel_in_cpl[ch]) {
631 decode_transform_coeffs_ch(s, blk, CPL_CH, &m);
632 calc_transform_coeffs_cpl(s);
635 end = s->end_freq[CPL_CH];
637 end = s->end_freq[ch];
640 s->fixed_coeffs[ch][end] = 0;
644 /* zero the dithered coefficients for appropriate channels */
649 * Stereo rematrixing.
650 * reference: Section 7.5.4 Rematrixing : Decoding Technique
652 static void do_rematrixing(AC3DecodeContext *s)
657 end = FFMIN(s->end_freq[1], s->end_freq[2]);
659 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++) {
660 if (s->rematrixing_flags[bnd]) {
661 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd + 1]);
662 for (i = ff_ac3_rematrix_band_tab[bnd]; i < bndend; i++) {
663 int tmp0 = s->fixed_coeffs[1][i];
664 s->fixed_coeffs[1][i] += s->fixed_coeffs[2][i];
665 s->fixed_coeffs[2][i] = tmp0 - s->fixed_coeffs[2][i];
672 * Inverse MDCT Transform.
673 * Convert frequency domain coefficients to time-domain audio samples.
674 * reference: Section 7.9.4 Transformation Equations
676 static inline void do_imdct(AC3DecodeContext *s, int channels)
680 for (ch = 1; ch <= channels; ch++) {
681 if (s->block_switch[ch]) {
683 FFTSample *x = s->tmp_output + 128;
684 for (i = 0; i < 128; i++)
685 x[i] = s->transform_coeffs[ch][2 * i];
686 s->imdct_256.imdct_half(&s->imdct_256, s->tmp_output, x);
688 s->fdsp->vector_fmul_window_scaled(s->outptr[ch - 1], s->delay[ch - 1],
689 s->tmp_output, s->window, 128, 8);
691 s->fdsp->vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1],
692 s->tmp_output, s->window, 128);
694 for (i = 0; i < 128; i++)
695 x[i] = s->transform_coeffs[ch][2 * i + 1];
696 s->imdct_256.imdct_half(&s->imdct_256, s->delay[ch - 1], x);
698 s->imdct_512.imdct_half(&s->imdct_512, s->tmp_output, s->transform_coeffs[ch]);
700 s->fdsp->vector_fmul_window_scaled(s->outptr[ch - 1], s->delay[ch - 1],
701 s->tmp_output, s->window, 128, 8);
703 s->fdsp->vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1],
704 s->tmp_output, s->window, 128);
706 memcpy(s->delay[ch - 1], s->tmp_output + 128, 128 * sizeof(FFTSample));
712 * Upmix delay samples from stereo to original channel layout.
714 static void ac3_upmix_delay(AC3DecodeContext *s)
716 int channel_data_size = sizeof(s->delay[0]);
717 switch (s->channel_mode) {
718 case AC3_CHMODE_DUALMONO:
719 case AC3_CHMODE_STEREO:
720 /* upmix mono to stereo */
721 memcpy(s->delay[1], s->delay[0], channel_data_size);
723 case AC3_CHMODE_2F2R:
724 memset(s->delay[3], 0, channel_data_size);
725 case AC3_CHMODE_2F1R:
726 memset(s->delay[2], 0, channel_data_size);
728 case AC3_CHMODE_3F2R:
729 memset(s->delay[4], 0, channel_data_size);
730 case AC3_CHMODE_3F1R:
731 memset(s->delay[3], 0, channel_data_size);
733 memcpy(s->delay[2], s->delay[1], channel_data_size);
734 memset(s->delay[1], 0, channel_data_size);
740 * Decode band structure for coupling, spectral extension, or enhanced coupling.
741 * The band structure defines how many subbands are in each band. For each
742 * subband in the range, 1 means it is combined with the previous band, and 0
743 * means that it starts a new band.
745 * @param[in] gbc bit reader context
746 * @param[in] blk block number
747 * @param[in] eac3 flag to indicate E-AC-3
748 * @param[in] ecpl flag to indicate enhanced coupling
749 * @param[in] start_subband subband number for start of range
750 * @param[in] end_subband subband number for end of range
751 * @param[in] default_band_struct default band structure table
752 * @param[out] num_bands number of bands (optionally NULL)
753 * @param[out] band_sizes array containing the number of bins in each band (optionally NULL)
755 static void decode_band_structure(GetBitContext *gbc, int blk, int eac3,
756 int ecpl, int start_subband, int end_subband,
757 const uint8_t *default_band_struct,
758 int *num_bands, uint8_t *band_sizes)
760 int subbnd, bnd, n_subbands, n_bands=0;
762 uint8_t coded_band_struct[22];
763 const uint8_t *band_struct;
765 n_subbands = end_subband - start_subband;
767 /* decode band structure from bitstream or use default */
768 if (!eac3 || get_bits1(gbc)) {
769 for (subbnd = 0; subbnd < n_subbands - 1; subbnd++) {
770 coded_band_struct[subbnd] = get_bits1(gbc);
772 band_struct = coded_band_struct;
774 band_struct = &default_band_struct[start_subband+1];
776 /* no change in band structure */
780 /* calculate number of bands and band sizes based on band structure.
781 note that the first 4 subbands in enhanced coupling span only 6 bins
783 if (num_bands || band_sizes ) {
784 n_bands = n_subbands;
785 bnd_sz[0] = ecpl ? 6 : 12;
786 for (bnd = 0, subbnd = 1; subbnd < n_subbands; subbnd++) {
787 int subbnd_size = (ecpl && subbnd < 4) ? 6 : 12;
788 if (band_struct[subbnd - 1]) {
790 bnd_sz[bnd] += subbnd_size;
792 bnd_sz[++bnd] = subbnd_size;
797 /* set optional output params */
799 *num_bands = n_bands;
801 memcpy(band_sizes, bnd_sz, n_bands);
805 * Decode a single audio block from the AC-3 bitstream.
807 static int decode_audio_block(AC3DecodeContext *s, int blk)
809 int fbw_channels = s->fbw_channels;
810 int channel_mode = s->channel_mode;
812 int different_transforms;
815 GetBitContext *gbc = &s->gbc;
816 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS] = { 0 };
818 /* block switch flags */
819 different_transforms = 0;
820 if (s->block_switch_syntax) {
821 for (ch = 1; ch <= fbw_channels; ch++) {
822 s->block_switch[ch] = get_bits1(gbc);
823 if (ch > 1 && s->block_switch[ch] != s->block_switch[1])
824 different_transforms = 1;
828 /* dithering flags */
829 if (s->dither_flag_syntax) {
830 for (ch = 1; ch <= fbw_channels; ch++) {
831 s->dither_flag[ch] = get_bits1(gbc);
836 i = !s->channel_mode;
838 if (get_bits1(gbc)) {
839 /* Allow asymmetric application of DRC when drc_scale > 1.
840 Amplification of quiet sounds is enhanced */
841 int range_bits = get_bits(gbc, 8);
842 INTFLOAT range = AC3_RANGE(range_bits);
843 if (range_bits <= 127 || s->drc_scale <= 1.0)
844 s->dynamic_range[i] = AC3_DYNAMIC_RANGE(range);
846 s->dynamic_range[i] = range;
847 } else if (blk == 0) {
848 s->dynamic_range[i] = AC3_DYNAMIC_RANGE1;
852 /* spectral extension strategy */
853 if (s->eac3 && (!blk || get_bits1(gbc))) {
854 s->spx_in_use = get_bits1(gbc);
856 int dst_start_freq, dst_end_freq, src_start_freq,
857 start_subband, end_subband;
859 /* determine which channels use spx */
860 if (s->channel_mode == AC3_CHMODE_MONO) {
861 s->channel_uses_spx[1] = 1;
863 for (ch = 1; ch <= fbw_channels; ch++)
864 s->channel_uses_spx[ch] = get_bits1(gbc);
867 /* get the frequency bins of the spx copy region and the spx start
869 dst_start_freq = get_bits(gbc, 2);
870 start_subband = get_bits(gbc, 3) + 2;
871 if (start_subband > 7)
872 start_subband += start_subband - 7;
873 end_subband = get_bits(gbc, 3) + 5;
875 s->spx_dst_end_freq = end_freq_inv_tab[end_subband-5];
878 end_subband += end_subband - 7;
879 dst_start_freq = dst_start_freq * 12 + 25;
880 src_start_freq = start_subband * 12 + 25;
881 dst_end_freq = end_subband * 12 + 25;
883 /* check validity of spx ranges */
884 if (start_subband >= end_subband) {
885 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
886 "range (%d >= %d)\n", start_subband, end_subband);
887 return AVERROR_INVALIDDATA;
889 if (dst_start_freq >= src_start_freq) {
890 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
891 "copy start bin (%d >= %d)\n", dst_start_freq, src_start_freq);
892 return AVERROR_INVALIDDATA;
895 s->spx_dst_start_freq = dst_start_freq;
896 s->spx_src_start_freq = src_start_freq;
898 s->spx_dst_end_freq = dst_end_freq;
900 decode_band_structure(gbc, blk, s->eac3, 0,
901 start_subband, end_subband,
902 ff_eac3_default_spx_band_struct,
907 if (!s->eac3 || !s->spx_in_use) {
909 for (ch = 1; ch <= fbw_channels; ch++) {
910 s->channel_uses_spx[ch] = 0;
911 s->first_spx_coords[ch] = 1;
915 /* spectral extension coordinates */
917 for (ch = 1; ch <= fbw_channels; ch++) {
918 if (s->channel_uses_spx[ch]) {
919 if (s->first_spx_coords[ch] || get_bits1(gbc)) {
921 int bin, master_spx_coord;
923 s->first_spx_coords[ch] = 0;
924 spx_blend = AC3_SPX_BLEND(get_bits(gbc, 5));
925 master_spx_coord = get_bits(gbc, 2) * 3;
927 bin = s->spx_src_start_freq;
928 for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
929 int bandsize = s->spx_band_sizes[bnd];
930 int spx_coord_exp, spx_coord_mant;
931 INTFLOAT nratio, sblend, nblend;
933 /* calculate blending factors */
934 int64_t accu = ((bin << 23) + (bandsize << 22))
935 * (int64_t)s->spx_dst_end_freq;
936 nratio = (int)(accu >> 32);
937 nratio -= spx_blend << 18;
942 } else if (nratio > 0x7fffff) {
943 nblend = 14529495; // sqrt(3) in FP.23
946 nblend = fixed_sqrt(nratio, 23);
947 accu = (int64_t)nblend * 1859775393;
948 nblend = (int)((accu + (1<<29)) >> 30);
949 sblend = fixed_sqrt(0x800000 - nratio, 23);
954 /* calculate blending factors */
955 nratio = ((float)((bin + (bandsize >> 1))) / s->spx_dst_end_freq) - spx_blend;
956 nratio = av_clipf(nratio, 0.0f, 1.0f);
957 nblend = sqrtf(3.0f * nratio); // noise is scaled by sqrt(3)
958 // to give unity variance
959 sblend = sqrtf(1.0f - nratio);
963 /* decode spx coordinates */
964 spx_coord_exp = get_bits(gbc, 4);
965 spx_coord_mant = get_bits(gbc, 2);
966 if (spx_coord_exp == 15) spx_coord_mant <<= 1;
967 else spx_coord_mant += 4;
968 spx_coord_mant <<= (25 - spx_coord_exp - master_spx_coord);
970 /* multiply noise and signal blending factors by spx coordinate */
972 accu = (int64_t)nblend * spx_coord_mant;
973 s->spx_noise_blend[ch][bnd] = (int)((accu + (1<<22)) >> 23);
974 accu = (int64_t)sblend * spx_coord_mant;
975 s->spx_signal_blend[ch][bnd] = (int)((accu + (1<<22)) >> 23);
977 spx_coord = spx_coord_mant * (1.0f / (1 << 23));
978 s->spx_noise_blend [ch][bnd] = nblend * spx_coord;
979 s->spx_signal_blend[ch][bnd] = sblend * spx_coord;
984 s->first_spx_coords[ch] = 1;
989 /* coupling strategy */
990 if (s->eac3 ? s->cpl_strategy_exists[blk] : get_bits1(gbc)) {
991 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
993 s->cpl_in_use[blk] = get_bits1(gbc);
994 if (s->cpl_in_use[blk]) {
995 /* coupling in use */
996 int cpl_start_subband, cpl_end_subband;
998 if (channel_mode < AC3_CHMODE_STEREO) {
999 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
1000 return AVERROR_INVALIDDATA;
1003 /* check for enhanced coupling */
1004 if (s->eac3 && get_bits1(gbc)) {
1005 /* TODO: parse enhanced coupling strategy info */
1006 avpriv_request_sample(s->avctx, "Enhanced coupling");
1007 return AVERROR_PATCHWELCOME;
1010 /* determine which channels are coupled */
1011 if (s->eac3 && s->channel_mode == AC3_CHMODE_STEREO) {
1012 s->channel_in_cpl[1] = 1;
1013 s->channel_in_cpl[2] = 1;
1015 for (ch = 1; ch <= fbw_channels; ch++)
1016 s->channel_in_cpl[ch] = get_bits1(gbc);
1019 /* phase flags in use */
1020 if (channel_mode == AC3_CHMODE_STEREO)
1021 s->phase_flags_in_use = get_bits1(gbc);
1023 /* coupling frequency range */
1024 cpl_start_subband = get_bits(gbc, 4);
1025 cpl_end_subband = s->spx_in_use ? (s->spx_src_start_freq - 37) / 12 :
1026 get_bits(gbc, 4) + 3;
1027 if (cpl_start_subband >= cpl_end_subband) {
1028 av_log(s->avctx, AV_LOG_ERROR, "invalid coupling range (%d >= %d)\n",
1029 cpl_start_subband, cpl_end_subband);
1030 return AVERROR_INVALIDDATA;
1032 s->start_freq[CPL_CH] = cpl_start_subband * 12 + 37;
1033 s->end_freq[CPL_CH] = cpl_end_subband * 12 + 37;
1035 decode_band_structure(gbc, blk, s->eac3, 0, cpl_start_subband,
1037 ff_eac3_default_cpl_band_struct,
1038 &s->num_cpl_bands, s->cpl_band_sizes);
1040 /* coupling not in use */
1041 for (ch = 1; ch <= fbw_channels; ch++) {
1042 s->channel_in_cpl[ch] = 0;
1043 s->first_cpl_coords[ch] = 1;
1045 s->first_cpl_leak = s->eac3;
1046 s->phase_flags_in_use = 0;
1048 } else if (!s->eac3) {
1050 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must "
1051 "be present in block 0\n");
1052 return AVERROR_INVALIDDATA;
1054 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
1057 cpl_in_use = s->cpl_in_use[blk];
1059 /* coupling coordinates */
1061 int cpl_coords_exist = 0;
1063 for (ch = 1; ch <= fbw_channels; ch++) {
1064 if (s->channel_in_cpl[ch]) {
1065 if ((s->eac3 && s->first_cpl_coords[ch]) || get_bits1(gbc)) {
1066 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
1067 s->first_cpl_coords[ch] = 0;
1068 cpl_coords_exist = 1;
1069 master_cpl_coord = 3 * get_bits(gbc, 2);
1070 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
1071 cpl_coord_exp = get_bits(gbc, 4);
1072 cpl_coord_mant = get_bits(gbc, 4);
1073 if (cpl_coord_exp == 15)
1074 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
1076 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
1077 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
1080 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must "
1081 "be present in block 0\n");
1082 return AVERROR_INVALIDDATA;
1085 /* channel not in coupling */
1086 s->first_cpl_coords[ch] = 1;
1090 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
1091 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
1092 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
1097 /* stereo rematrixing strategy and band structure */
1098 if (channel_mode == AC3_CHMODE_STEREO) {
1099 if ((s->eac3 && !blk) || get_bits1(gbc)) {
1100 s->num_rematrixing_bands = 4;
1101 if (cpl_in_use && s->start_freq[CPL_CH] <= 61) {
1102 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
1103 } else if (s->spx_in_use && s->spx_src_start_freq <= 61) {
1104 s->num_rematrixing_bands--;
1106 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++)
1107 s->rematrixing_flags[bnd] = get_bits1(gbc);
1109 av_log(s->avctx, AV_LOG_WARNING, "Warning: "
1110 "new rematrixing strategy not present in block 0\n");
1111 s->num_rematrixing_bands = 0;
1115 /* exponent strategies for each channel */
1116 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1118 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
1119 if (s->exp_strategy[blk][ch] != EXP_REUSE)
1120 bit_alloc_stages[ch] = 3;
1123 /* channel bandwidth */
1124 for (ch = 1; ch <= fbw_channels; ch++) {
1125 s->start_freq[ch] = 0;
1126 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1128 int prev = s->end_freq[ch];
1129 if (s->channel_in_cpl[ch])
1130 s->end_freq[ch] = s->start_freq[CPL_CH];
1131 else if (s->channel_uses_spx[ch])
1132 s->end_freq[ch] = s->spx_src_start_freq;
1134 int bandwidth_code = get_bits(gbc, 6);
1135 if (bandwidth_code > 60) {
1136 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60\n", bandwidth_code);
1137 return AVERROR_INVALIDDATA;
1139 s->end_freq[ch] = bandwidth_code * 3 + 73;
1141 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
1142 s->num_exp_groups[ch] = (s->end_freq[ch] + group_size-4) / group_size;
1143 if (blk > 0 && s->end_freq[ch] != prev)
1144 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
1147 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
1148 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
1149 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
1152 /* decode exponents for each channel */
1153 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1154 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1155 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
1156 if (decode_exponents(s, gbc, s->exp_strategy[blk][ch],
1157 s->num_exp_groups[ch], s->dexps[ch][0],
1158 &s->dexps[ch][s->start_freq[ch]+!!ch])) {
1159 return AVERROR_INVALIDDATA;
1161 if (ch != CPL_CH && ch != s->lfe_ch)
1162 skip_bits(gbc, 2); /* skip gainrng */
1166 /* bit allocation information */
1167 if (s->bit_allocation_syntax) {
1168 if (get_bits1(gbc)) {
1169 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1170 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1171 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
1172 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
1173 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
1174 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1175 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1177 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must "
1178 "be present in block 0\n");
1179 return AVERROR_INVALIDDATA;
1183 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
1184 if (!s->eac3 || !blk) {
1185 if (s->snr_offset_strategy && get_bits1(gbc)) {
1188 csnr = (get_bits(gbc, 6) - 15) << 4;
1189 for (i = ch = !cpl_in_use; ch <= s->channels; ch++) {
1191 if (ch == i || s->snr_offset_strategy == 2)
1192 snr = (csnr + get_bits(gbc, 4)) << 2;
1193 /* run at least last bit allocation stage if snr offset changes */
1194 if (blk && s->snr_offset[ch] != snr) {
1195 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 1);
1197 s->snr_offset[ch] = snr;
1199 /* fast gain (normal AC-3 only) */
1201 int prev = s->fast_gain[ch];
1202 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1203 /* run last 2 bit allocation stages if fast gain changes */
1204 if (blk && prev != s->fast_gain[ch])
1205 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1208 } else if (!s->eac3 && !blk) {
1209 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
1210 return AVERROR_INVALIDDATA;
1214 /* fast gain (E-AC-3 only) */
1215 if (s->fast_gain_syntax && get_bits1(gbc)) {
1216 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1217 int prev = s->fast_gain[ch];
1218 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1219 /* run last 2 bit allocation stages if fast gain changes */
1220 if (blk && prev != s->fast_gain[ch])
1221 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1223 } else if (s->eac3 && !blk) {
1224 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1225 s->fast_gain[ch] = ff_ac3_fast_gain_tab[4];
1228 /* E-AC-3 to AC-3 converter SNR offset */
1229 if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT && get_bits1(gbc)) {
1230 skip_bits(gbc, 10); // skip converter snr offset
1233 /* coupling leak information */
1235 if (s->first_cpl_leak || get_bits1(gbc)) {
1236 int fl = get_bits(gbc, 3);
1237 int sl = get_bits(gbc, 3);
1238 /* run last 2 bit allocation stages for coupling channel if
1239 coupling leak changes */
1240 if (blk && (fl != s->bit_alloc_params.cpl_fast_leak ||
1241 sl != s->bit_alloc_params.cpl_slow_leak)) {
1242 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
1244 s->bit_alloc_params.cpl_fast_leak = fl;
1245 s->bit_alloc_params.cpl_slow_leak = sl;
1246 } else if (!s->eac3 && !blk) {
1247 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must "
1248 "be present in block 0\n");
1249 return AVERROR_INVALIDDATA;
1251 s->first_cpl_leak = 0;
1254 /* delta bit allocation information */
1255 if (s->dba_syntax && get_bits1(gbc)) {
1256 /* delta bit allocation exists (strategy) */
1257 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1258 s->dba_mode[ch] = get_bits(gbc, 2);
1259 if (s->dba_mode[ch] == DBA_RESERVED) {
1260 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1261 return AVERROR_INVALIDDATA;
1263 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1265 /* channel delta offset, len and bit allocation */
1266 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1267 if (s->dba_mode[ch] == DBA_NEW) {
1268 s->dba_nsegs[ch] = get_bits(gbc, 3) + 1;
1269 for (seg = 0; seg < s->dba_nsegs[ch]; seg++) {
1270 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1271 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1272 s->dba_values[ch][seg] = get_bits(gbc, 3);
1274 /* run last 2 bit allocation stages if new dba values */
1275 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1278 } else if (blk == 0) {
1279 for (ch = 0; ch <= s->channels; ch++) {
1280 s->dba_mode[ch] = DBA_NONE;
1284 /* Bit allocation */
1285 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1286 if (bit_alloc_stages[ch] > 2) {
1287 /* Exponent mapping into PSD and PSD integration */
1288 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1289 s->start_freq[ch], s->end_freq[ch],
1290 s->psd[ch], s->band_psd[ch]);
1292 if (bit_alloc_stages[ch] > 1) {
1293 /* Compute excitation function, Compute masking curve, and
1294 Apply delta bit allocation */
1295 if (ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1296 s->start_freq[ch], s->end_freq[ch],
1297 s->fast_gain[ch], (ch == s->lfe_ch),
1298 s->dba_mode[ch], s->dba_nsegs[ch],
1299 s->dba_offsets[ch], s->dba_lengths[ch],
1300 s->dba_values[ch], s->mask[ch])) {
1301 av_log(s->avctx, AV_LOG_ERROR, "error in bit allocation\n");
1302 return AVERROR_INVALIDDATA;
1305 if (bit_alloc_stages[ch] > 0) {
1306 /* Compute bit allocation */
1307 const uint8_t *bap_tab = s->channel_uses_aht[ch] ?
1308 ff_eac3_hebap_tab : ff_ac3_bap_tab;
1309 s->ac3dsp.bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1310 s->start_freq[ch], s->end_freq[ch],
1312 s->bit_alloc_params.floor,
1313 bap_tab, s->bap[ch]);
1317 /* unused dummy data */
1318 if (s->skip_syntax && get_bits1(gbc)) {
1319 int skipl = get_bits(gbc, 9);
1324 /* unpack the transform coefficients
1325 this also uncouples channels if coupling is in use. */
1326 decode_transform_coeffs(s, blk);
1328 /* TODO: generate enhanced coupling coordinates and uncouple */
1330 /* recover coefficients if rematrixing is in use */
1331 if (s->channel_mode == AC3_CHMODE_STEREO)
1334 /* apply scaling to coefficients (headroom, dynrng) */
1335 for (ch = 1; ch <= s->channels; ch++) {
1336 int audio_channel = 0;
1338 if (s->channel_mode == AC3_CHMODE_DUALMONO)
1339 audio_channel = 2-ch;
1340 if (s->heavy_compression && s->compression_exists[audio_channel])
1341 gain = s->heavy_dynamic_range[audio_channel];
1343 gain = s->dynamic_range[audio_channel];
1346 scale_coefs(s->transform_coeffs[ch], s->fixed_coeffs[ch], gain, 256);
1348 if (s->target_level != 0)
1349 gain = gain * s->level_gain[audio_channel];
1350 gain *= 1.0 / 4194304.0f;
1351 s->fmt_conv.int32_to_float_fmul_scalar(s->transform_coeffs[ch],
1352 s->fixed_coeffs[ch], gain, 256);
1356 /* apply spectral extension to high frequency bins */
1357 if (CONFIG_EAC3_DECODER && s->spx_in_use) {
1358 ff_eac3_apply_spectral_extension(s);
1361 /* downmix and MDCT. order depends on whether block switching is used for
1362 any channel in this block. this is because coefficients for the long
1363 and short transforms cannot be mixed. */
1364 downmix_output = s->channels != s->out_channels &&
1365 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1366 s->fbw_channels == s->out_channels);
1367 if (different_transforms) {
1368 /* the delay samples have already been downmixed, so we upmix the delay
1369 samples in order to reconstruct all channels before downmixing. */
1375 do_imdct(s, s->channels);
1377 if (downmix_output) {
1379 ac3_downmix_c_fixed16(s->outptr, s->downmix_coeffs,
1380 s->out_channels, s->fbw_channels, 256);
1382 s->ac3dsp.downmix(s->outptr, s->downmix_coeffs,
1383 s->out_channels, s->fbw_channels, 256);
1387 if (downmix_output) {
1388 s->ac3dsp.AC3_RENAME(downmix)(s->xcfptr + 1, s->downmix_coeffs,
1389 s->out_channels, s->fbw_channels, 256);
1392 if (downmix_output && !s->downmixed) {
1394 s->ac3dsp.AC3_RENAME(downmix)(s->dlyptr, s->downmix_coeffs,
1395 s->out_channels, s->fbw_channels, 128);
1398 do_imdct(s, s->out_channels);
1405 * Decode a single AC-3 frame.
1407 static int ac3_decode_frame(AVCodecContext * avctx, void *data,
1408 int *got_frame_ptr, AVPacket *avpkt)
1410 AVFrame *frame = data;
1411 const uint8_t *buf = avpkt->data;
1412 int buf_size = avpkt->size;
1413 AC3DecodeContext *s = avctx->priv_data;
1414 int blk, ch, err, ret;
1415 const uint8_t *channel_map;
1416 const SHORTFLOAT *output[AC3_MAX_CHANNELS];
1417 enum AVMatrixEncoding matrix_encoding;
1418 AVDownmixInfo *downmix_info;
1420 /* copy input buffer to decoder context to avoid reading past the end
1421 of the buffer, which can be caused by a damaged input stream. */
1422 if (buf_size >= 2 && AV_RB16(buf) == 0x770B) {
1423 // seems to be byte-swapped AC-3
1424 int cnt = FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE) >> 1;
1425 s->bdsp.bswap16_buf((uint16_t *) s->input_buffer,
1426 (const uint16_t *) buf, cnt);
1428 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE));
1430 /* if consistent noise generation is enabled, seed the linear feedback generator
1431 * with the contents of the AC-3 frame so that the noise is identical across
1432 * decodes given the same AC-3 frame data, for use with non-linear edititing software. */
1433 if (s->consistent_noise_generation)
1434 av_lfg_init_from_data(&s->dith_state, s->input_buffer, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE));
1436 buf = s->input_buffer;
1437 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1438 if ((ret = init_get_bits8(&s->gbc, buf, buf_size)) < 0)
1441 /* parse the syncinfo */
1442 err = parse_frame_header(s);
1446 case AAC_AC3_PARSE_ERROR_SYNC:
1447 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1448 return AVERROR_INVALIDDATA;
1449 case AAC_AC3_PARSE_ERROR_BSID:
1450 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1452 case AAC_AC3_PARSE_ERROR_SAMPLE_RATE:
1453 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1455 case AAC_AC3_PARSE_ERROR_FRAME_SIZE:
1456 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1458 case AAC_AC3_PARSE_ERROR_FRAME_TYPE:
1459 /* skip frame if CRC is ok. otherwise use error concealment. */
1460 /* TODO: add support for substreams and dependent frames */
1461 if (s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1462 av_log(avctx, AV_LOG_DEBUG,
1463 "unsupported frame type %d: skipping frame\n",
1468 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1471 case AAC_AC3_PARSE_ERROR_CRC:
1472 case AAC_AC3_PARSE_ERROR_CHANNEL_CFG:
1474 default: // Normal AVERROR do not try to recover.
1479 /* check that reported frame size fits in input buffer */
1480 if (s->frame_size > buf_size) {
1481 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1482 err = AAC_AC3_PARSE_ERROR_FRAME_SIZE;
1483 } else if (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL)) {
1484 /* check for crc mismatch */
1485 if (av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2],
1486 s->frame_size - 2)) {
1487 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1488 if (avctx->err_recognition & AV_EF_EXPLODE)
1489 return AVERROR_INVALIDDATA;
1490 err = AAC_AC3_PARSE_ERROR_CRC;
1495 /* if frame is ok, set audio parameters */
1497 avctx->sample_rate = s->sample_rate;
1498 avctx->bit_rate = s->bit_rate;
1501 /* channel config */
1502 if (!err || (s->channels && s->out_channels != s->channels)) {
1503 s->out_channels = s->channels;
1504 s->output_mode = s->channel_mode;
1506 s->output_mode |= AC3_OUTPUT_LFEON;
1507 if (s->channels > 1 &&
1508 avctx->request_channel_layout == AV_CH_LAYOUT_MONO) {
1509 s->out_channels = 1;
1510 s->output_mode = AC3_CHMODE_MONO;
1511 } else if (s->channels > 2 &&
1512 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
1513 s->out_channels = 2;
1514 s->output_mode = AC3_CHMODE_STEREO;
1517 s->loro_center_mix_level = gain_levels[s-> center_mix_level];
1518 s->loro_surround_mix_level = gain_levels[s->surround_mix_level];
1519 s->ltrt_center_mix_level = LEVEL_MINUS_3DB;
1520 s->ltrt_surround_mix_level = LEVEL_MINUS_3DB;
1521 /* set downmixing coefficients if needed */
1522 if (s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1523 s->fbw_channels == s->out_channels)) {
1524 set_downmix_coeffs(s);
1526 } else if (!s->channels) {
1527 av_log(avctx, AV_LOG_ERROR, "unable to determine channel mode\n");
1528 return AVERROR_INVALIDDATA;
1530 avctx->channels = s->out_channels;
1531 avctx->channel_layout = avpriv_ac3_channel_layout_tab[s->output_mode & ~AC3_OUTPUT_LFEON];
1532 if (s->output_mode & AC3_OUTPUT_LFEON)
1533 avctx->channel_layout |= AV_CH_LOW_FREQUENCY;
1535 /* set audio service type based on bitstream mode for AC-3 */
1536 avctx->audio_service_type = s->bitstream_mode;
1537 if (s->bitstream_mode == 0x7 && s->channels > 1)
1538 avctx->audio_service_type = AV_AUDIO_SERVICE_TYPE_KARAOKE;
1540 /* get output buffer */
1541 frame->nb_samples = s->num_blocks * AC3_BLOCK_SIZE;
1542 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1545 /* decode the audio blocks */
1546 channel_map = ff_ac3_dec_channel_map[s->output_mode & ~AC3_OUTPUT_LFEON][s->lfe_on];
1547 for (ch = 0; ch < AC3_MAX_CHANNELS; ch++) {
1548 output[ch] = s->output[ch];
1549 s->outptr[ch] = s->output[ch];
1551 for (ch = 0; ch < s->channels; ch++) {
1552 if (ch < s->out_channels)
1553 s->outptr[channel_map[ch]] = (SHORTFLOAT *)frame->data[ch];
1555 for (blk = 0; blk < s->num_blocks; blk++) {
1556 if (!err && decode_audio_block(s, blk)) {
1557 av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n");
1561 for (ch = 0; ch < s->out_channels; ch++)
1562 memcpy(((SHORTFLOAT*)frame->data[ch]) + AC3_BLOCK_SIZE*blk, output[ch], AC3_BLOCK_SIZE*sizeof(SHORTFLOAT));
1563 for (ch = 0; ch < s->out_channels; ch++)
1564 output[ch] = s->outptr[channel_map[ch]];
1565 for (ch = 0; ch < s->out_channels; ch++) {
1566 if (!ch || channel_map[ch])
1567 s->outptr[channel_map[ch]] += AC3_BLOCK_SIZE;
1571 av_frame_set_decode_error_flags(frame, err ? FF_DECODE_ERROR_INVALID_BITSTREAM : 0);
1573 /* keep last block for error concealment in next frame */
1574 for (ch = 0; ch < s->out_channels; ch++)
1575 memcpy(s->output[ch], output[ch], AC3_BLOCK_SIZE*sizeof(SHORTFLOAT));
1580 * Check whether the input layout is compatible, and make sure we're not
1581 * downmixing (else the matrix encoding is no longer applicable).
1583 matrix_encoding = AV_MATRIX_ENCODING_NONE;
1584 if (s->channel_mode == AC3_CHMODE_STEREO &&
1585 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1586 if (s->dolby_surround_mode == AC3_DSURMOD_ON)
1587 matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
1588 else if (s->dolby_headphone_mode == AC3_DHEADPHONMOD_ON)
1589 matrix_encoding = AV_MATRIX_ENCODING_DOLBYHEADPHONE;
1590 } else if (s->channel_mode >= AC3_CHMODE_2F2R &&
1591 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1592 switch (s->dolby_surround_ex_mode) {
1593 case AC3_DSUREXMOD_ON: // EX or PLIIx
1594 matrix_encoding = AV_MATRIX_ENCODING_DOLBYEX;
1596 case AC3_DSUREXMOD_PLIIZ:
1597 matrix_encoding = AV_MATRIX_ENCODING_DPLIIZ;
1599 default: // not indicated or off
1603 if ((ret = ff_side_data_update_matrix_encoding(frame, matrix_encoding)) < 0)
1607 if ((downmix_info = av_downmix_info_update_side_data(frame))) {
1608 switch (s->preferred_downmix) {
1609 case AC3_DMIXMOD_LTRT:
1610 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LTRT;
1612 case AC3_DMIXMOD_LORO:
1613 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LORO;
1615 case AC3_DMIXMOD_DPLII:
1616 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_DPLII;
1619 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_UNKNOWN;
1622 downmix_info->center_mix_level = gain_levels[s-> center_mix_level];
1623 downmix_info->center_mix_level_ltrt = gain_levels[s-> center_mix_level_ltrt];
1624 downmix_info->surround_mix_level = gain_levels[s-> surround_mix_level];
1625 downmix_info->surround_mix_level_ltrt = gain_levels[s->surround_mix_level_ltrt];
1626 if (s->lfe_mix_level_exists)
1627 downmix_info->lfe_mix_level = gain_levels_lfe[s->lfe_mix_level];
1629 downmix_info->lfe_mix_level = 0.0; // -inf dB
1631 return AVERROR(ENOMEM);
1635 return FFMIN(buf_size, s->frame_size);
1639 * Uninitialize the AC-3 decoder.
1641 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1643 AC3DecodeContext *s = avctx->priv_data;
1644 ff_mdct_end(&s->imdct_512);
1645 ff_mdct_end(&s->imdct_256);
1651 #define OFFSET(x) offsetof(AC3DecodeContext, x)
1652 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)