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;
188 ff_ac3_common_init();
190 ff_mdct_init(&s->imdct_256, 8, 1, 1.0);
191 ff_mdct_init(&s->imdct_512, 9, 1, 1.0);
192 AC3_RENAME(ff_kbd_window_init)(s->window, 5.0, 256);
193 ff_bswapdsp_init(&s->bdsp);
196 s->fdsp = avpriv_alloc_fixed_dsp(avctx->flags & AV_CODEC_FLAG_BITEXACT);
198 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
199 ff_fmt_convert_init(&s->fmt_conv, avctx);
202 ff_ac3dsp_init(&s->ac3dsp, avctx->flags & AV_CODEC_FLAG_BITEXACT);
203 av_lfg_init(&s->dith_state, 0);
206 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
208 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
210 /* allow downmixing to stereo or mono */
211 if (avctx->channels > 1 &&
212 avctx->request_channel_layout == AV_CH_LAYOUT_MONO)
214 else if (avctx->channels > 2 &&
215 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO)
219 for (i = 0; i < AC3_MAX_CHANNELS; i++) {
220 s->xcfptr[i] = s->transform_coeffs[i];
221 s->dlyptr[i] = s->delay[i];
228 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
229 * GetBitContext within AC3DecodeContext must point to
230 * the start of the synchronized AC-3 bitstream.
232 static int ac3_parse_header(AC3DecodeContext *s)
234 GetBitContext *gbc = &s->gbc;
237 /* read the rest of the bsi. read twice for dual mono mode. */
238 i = !s->channel_mode;
240 s->dialog_normalization[(!s->channel_mode)-i] = -get_bits(gbc, 5);
241 if (s->dialog_normalization[(!s->channel_mode)-i] == 0) {
242 s->dialog_normalization[(!s->channel_mode)-i] = -31;
244 if (s->target_level != 0) {
245 s->level_gain[(!s->channel_mode)-i] = powf(2.0f,
246 (float)(s->target_level -
247 s->dialog_normalization[(!s->channel_mode)-i])/6.0f);
249 if (s->compression_exists[(!s->channel_mode)-i] = get_bits1(gbc)) {
250 s->heavy_dynamic_range[(!s->channel_mode)-i] =
251 AC3_HEAVY_RANGE(get_bits(gbc, 8));
254 skip_bits(gbc, 8); //skip language code
256 skip_bits(gbc, 7); //skip audio production information
259 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
261 /* skip the timecodes or parse the Alternate Bit Stream Syntax */
262 if (s->bitstream_id != 6) {
264 skip_bits(gbc, 14); //skip timecode1
266 skip_bits(gbc, 14); //skip timecode2
268 if (get_bits1(gbc)) {
269 s->preferred_downmix = get_bits(gbc, 2);
270 s->center_mix_level_ltrt = get_bits(gbc, 3);
271 s->surround_mix_level_ltrt = av_clip(get_bits(gbc, 3), 3, 7);
272 s->center_mix_level = get_bits(gbc, 3);
273 s->surround_mix_level = av_clip(get_bits(gbc, 3), 3, 7);
275 if (get_bits1(gbc)) {
276 s->dolby_surround_ex_mode = get_bits(gbc, 2);
277 s->dolby_headphone_mode = get_bits(gbc, 2);
278 skip_bits(gbc, 10); // skip adconvtyp (1), xbsi2 (8), encinfo (1)
282 /* skip additional bitstream info */
283 if (get_bits1(gbc)) {
284 i = get_bits(gbc, 6);
294 * Common function to parse AC-3 or E-AC-3 frame header
296 static int parse_frame_header(AC3DecodeContext *s)
298 AC3HeaderInfo hdr, *phdr=&hdr;
301 err = avpriv_ac3_parse_header2(&s->gbc, &phdr);
305 /* get decoding parameters from header info */
306 s->bit_alloc_params.sr_code = hdr.sr_code;
307 s->bitstream_id = hdr.bitstream_id;
308 s->bitstream_mode = hdr.bitstream_mode;
309 s->channel_mode = hdr.channel_mode;
310 s->lfe_on = hdr.lfe_on;
311 s->bit_alloc_params.sr_shift = hdr.sr_shift;
312 s->sample_rate = hdr.sample_rate;
313 s->bit_rate = hdr.bit_rate;
314 s->channels = hdr.channels;
315 s->fbw_channels = s->channels - s->lfe_on;
316 s->lfe_ch = s->fbw_channels + 1;
317 s->frame_size = hdr.frame_size;
318 s->preferred_downmix = AC3_DMIXMOD_NOTINDICATED;
319 s->center_mix_level = hdr.center_mix_level;
320 s->center_mix_level_ltrt = 4; // -3.0dB
321 s->surround_mix_level = hdr.surround_mix_level;
322 s->surround_mix_level_ltrt = 4; // -3.0dB
323 s->lfe_mix_level_exists = 0;
324 s->num_blocks = hdr.num_blocks;
325 s->frame_type = hdr.frame_type;
326 s->substreamid = hdr.substreamid;
327 s->dolby_surround_mode = hdr.dolby_surround_mode;
328 s->dolby_surround_ex_mode = AC3_DSUREXMOD_NOTINDICATED;
329 s->dolby_headphone_mode = AC3_DHEADPHONMOD_NOTINDICATED;
332 s->start_freq[s->lfe_ch] = 0;
333 s->end_freq[s->lfe_ch] = 7;
334 s->num_exp_groups[s->lfe_ch] = 2;
335 s->channel_in_cpl[s->lfe_ch] = 0;
338 if (s->bitstream_id <= 10) {
340 s->snr_offset_strategy = 2;
341 s->block_switch_syntax = 1;
342 s->dither_flag_syntax = 1;
343 s->bit_allocation_syntax = 1;
344 s->fast_gain_syntax = 0;
345 s->first_cpl_leak = 0;
348 memset(s->channel_uses_aht, 0, sizeof(s->channel_uses_aht));
349 return ac3_parse_header(s);
350 } else if (CONFIG_EAC3_DECODER) {
352 return ff_eac3_parse_header(s);
354 av_log(s->avctx, AV_LOG_ERROR, "E-AC-3 support not compiled in\n");
355 return AVERROR(ENOSYS);
360 * Set stereo downmixing coefficients based on frame header info.
361 * reference: Section 7.8.2 Downmixing Into Two Channels
363 static void set_downmix_coeffs(AC3DecodeContext *s)
366 float cmix = gain_levels[s-> center_mix_level];
367 float smix = gain_levels[s->surround_mix_level];
369 float downmix_coeffs[AC3_MAX_CHANNELS][2];
371 for (i = 0; i < s->fbw_channels; i++) {
372 downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
373 downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
375 if (s->channel_mode > 1 && s->channel_mode & 1) {
376 downmix_coeffs[1][0] = downmix_coeffs[1][1] = cmix;
378 if (s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
379 int nf = s->channel_mode - 2;
380 downmix_coeffs[nf][0] = downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
382 if (s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
383 int nf = s->channel_mode - 4;
384 downmix_coeffs[nf][0] = downmix_coeffs[nf+1][1] = smix;
389 for (i = 0; i < s->fbw_channels; i++) {
390 norm0 += downmix_coeffs[i][0];
391 norm1 += downmix_coeffs[i][1];
393 norm0 = 1.0f / norm0;
394 norm1 = 1.0f / norm1;
395 for (i = 0; i < s->fbw_channels; i++) {
396 downmix_coeffs[i][0] *= norm0;
397 downmix_coeffs[i][1] *= norm1;
400 if (s->output_mode == AC3_CHMODE_MONO) {
401 for (i = 0; i < s->fbw_channels; i++)
402 downmix_coeffs[i][0] = (downmix_coeffs[i][0] +
403 downmix_coeffs[i][1]) * LEVEL_MINUS_3DB;
405 for (i = 0; i < s->fbw_channels; i++) {
406 s->downmix_coeffs[i][0] = FIXR12(downmix_coeffs[i][0]);
407 s->downmix_coeffs[i][1] = FIXR12(downmix_coeffs[i][1]);
412 * Decode the grouped exponents according to exponent strategy.
413 * reference: Section 7.1.3 Exponent Decoding
415 static int decode_exponents(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;
437 switch (group_size) {
438 case 4: dexps[j++] = prevexp;
439 dexps[j++] = prevexp;
440 case 2: dexps[j++] = prevexp;
441 case 1: dexps[j++] = prevexp;
448 * Generate transform coefficients for each coupled channel in the coupling
449 * range using the coupling coefficients and coupling coordinates.
450 * reference: Section 7.4.3 Coupling Coordinate Format
452 static void calc_transform_coeffs_cpl(AC3DecodeContext *s)
456 bin = s->start_freq[CPL_CH];
457 for (band = 0; band < s->num_cpl_bands; band++) {
458 int band_start = bin;
459 int band_end = bin + s->cpl_band_sizes[band];
460 for (ch = 1; ch <= s->fbw_channels; ch++) {
461 if (s->channel_in_cpl[ch]) {
462 int cpl_coord = s->cpl_coords[ch][band] << 5;
463 for (bin = band_start; bin < band_end; bin++) {
464 s->fixed_coeffs[ch][bin] =
465 MULH(s->fixed_coeffs[CPL_CH][bin] * (1 << 4), cpl_coord);
467 if (ch == 2 && s->phase_flags[band]) {
468 for (bin = band_start; bin < band_end; bin++)
469 s->fixed_coeffs[2][bin] = -s->fixed_coeffs[2][bin];
478 * Grouped mantissas for 3-level 5-level and 11-level quantization
480 typedef struct mant_groups {
490 * Decode the transform coefficients for a particular channel
491 * reference: Section 7.3 Quantization and Decoding of Mantissas
493 static void ac3_decode_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
495 int start_freq = s->start_freq[ch_index];
496 int end_freq = s->end_freq[ch_index];
497 uint8_t *baps = s->bap[ch_index];
498 int8_t *exps = s->dexps[ch_index];
499 int32_t *coeffs = s->fixed_coeffs[ch_index];
500 int dither = (ch_index == CPL_CH) || s->dither_flag[ch_index];
501 GetBitContext *gbc = &s->gbc;
504 for (freq = start_freq; freq < end_freq; freq++) {
505 int bap = baps[freq];
509 /* random noise with approximate range of -0.707 to 0.707 */
511 mantissa = (((av_lfg_get(&s->dith_state)>>8)*181)>>8) - 5931008;
518 mantissa = m->b1_mant[m->b1];
520 int bits = get_bits(gbc, 5);
521 mantissa = b1_mantissas[bits][0];
522 m->b1_mant[1] = b1_mantissas[bits][1];
523 m->b1_mant[0] = b1_mantissas[bits][2];
530 mantissa = m->b2_mant[m->b2];
532 int bits = get_bits(gbc, 7);
533 mantissa = b2_mantissas[bits][0];
534 m->b2_mant[1] = b2_mantissas[bits][1];
535 m->b2_mant[0] = b2_mantissas[bits][2];
540 mantissa = b3_mantissas[get_bits(gbc, 3)];
545 mantissa = m->b4_mant;
547 int bits = get_bits(gbc, 7);
548 mantissa = b4_mantissas[bits][0];
549 m->b4_mant = b4_mantissas[bits][1];
554 mantissa = b5_mantissas[get_bits(gbc, 4)];
556 default: /* 6 to 15 */
557 /* Shift mantissa and sign-extend it. */
559 av_log(s->avctx, AV_LOG_ERROR, "bap %d is invalid in plain AC-3\n", bap);
562 mantissa = (unsigned)get_sbits(gbc, quantization_tab[bap]) << (24 - quantization_tab[bap]);
565 coeffs[freq] = mantissa >> exps[freq];
570 * Remove random dithering from coupling range coefficients with zero-bit
571 * mantissas for coupled channels which do not use dithering.
572 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
574 static void remove_dithering(AC3DecodeContext *s) {
577 for (ch = 1; ch <= s->fbw_channels; ch++) {
578 if (!s->dither_flag[ch] && s->channel_in_cpl[ch]) {
579 for (i = s->start_freq[CPL_CH]; i < s->end_freq[CPL_CH]; i++) {
580 if (!s->bap[CPL_CH][i])
581 s->fixed_coeffs[ch][i] = 0;
587 static void decode_transform_coeffs_ch(AC3DecodeContext *s, int blk, int ch,
590 if (!s->channel_uses_aht[ch]) {
591 ac3_decode_transform_coeffs_ch(s, ch, m);
593 /* if AHT is used, mantissas for all blocks are encoded in the first
594 block of the frame. */
596 if (CONFIG_EAC3_DECODER && !blk)
597 ff_eac3_decode_transform_coeffs_aht_ch(s, ch);
598 for (bin = s->start_freq[ch]; bin < s->end_freq[ch]; bin++) {
599 s->fixed_coeffs[ch][bin] = s->pre_mantissa[ch][bin][blk] >> s->dexps[ch][bin];
605 * Decode the transform coefficients.
607 static void decode_transform_coeffs(AC3DecodeContext *s, int blk)
613 m.b1 = m.b2 = m.b4 = 0;
615 for (ch = 1; ch <= s->channels; ch++) {
616 /* transform coefficients for full-bandwidth channel */
617 decode_transform_coeffs_ch(s, blk, ch, &m);
618 /* transform coefficients for coupling channel come right after the
619 coefficients for the first coupled channel*/
620 if (s->channel_in_cpl[ch]) {
622 decode_transform_coeffs_ch(s, blk, CPL_CH, &m);
623 calc_transform_coeffs_cpl(s);
626 end = s->end_freq[CPL_CH];
628 end = s->end_freq[ch];
631 s->fixed_coeffs[ch][end] = 0;
635 /* zero the dithered coefficients for appropriate channels */
640 * Stereo rematrixing.
641 * reference: Section 7.5.4 Rematrixing : Decoding Technique
643 static void do_rematrixing(AC3DecodeContext *s)
648 end = FFMIN(s->end_freq[1], s->end_freq[2]);
650 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++) {
651 if (s->rematrixing_flags[bnd]) {
652 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd + 1]);
653 for (i = ff_ac3_rematrix_band_tab[bnd]; i < bndend; i++) {
654 int tmp0 = s->fixed_coeffs[1][i];
655 s->fixed_coeffs[1][i] += s->fixed_coeffs[2][i];
656 s->fixed_coeffs[2][i] = tmp0 - s->fixed_coeffs[2][i];
663 * Inverse MDCT Transform.
664 * Convert frequency domain coefficients to time-domain audio samples.
665 * reference: Section 7.9.4 Transformation Equations
667 static inline void do_imdct(AC3DecodeContext *s, int channels)
671 for (ch = 1; ch <= channels; ch++) {
672 if (s->block_switch[ch]) {
674 FFTSample *x = s->tmp_output + 128;
675 for (i = 0; i < 128; i++)
676 x[i] = s->transform_coeffs[ch][2 * i];
677 s->imdct_256.imdct_half(&s->imdct_256, s->tmp_output, x);
679 s->fdsp->vector_fmul_window_scaled(s->outptr[ch - 1], s->delay[ch - 1],
680 s->tmp_output, s->window, 128, 8);
682 s->fdsp->vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1],
683 s->tmp_output, s->window, 128);
685 for (i = 0; i < 128; i++)
686 x[i] = s->transform_coeffs[ch][2 * i + 1];
687 s->imdct_256.imdct_half(&s->imdct_256, s->delay[ch - 1], x);
689 s->imdct_512.imdct_half(&s->imdct_512, s->tmp_output, s->transform_coeffs[ch]);
691 s->fdsp->vector_fmul_window_scaled(s->outptr[ch - 1], s->delay[ch - 1],
692 s->tmp_output, s->window, 128, 8);
694 s->fdsp->vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1],
695 s->tmp_output, s->window, 128);
697 memcpy(s->delay[ch - 1], s->tmp_output + 128, 128 * sizeof(FFTSample));
703 * Upmix delay samples from stereo to original channel layout.
705 static void ac3_upmix_delay(AC3DecodeContext *s)
707 int channel_data_size = sizeof(s->delay[0]);
708 switch (s->channel_mode) {
709 case AC3_CHMODE_DUALMONO:
710 case AC3_CHMODE_STEREO:
711 /* upmix mono to stereo */
712 memcpy(s->delay[1], s->delay[0], channel_data_size);
714 case AC3_CHMODE_2F2R:
715 memset(s->delay[3], 0, channel_data_size);
716 case AC3_CHMODE_2F1R:
717 memset(s->delay[2], 0, channel_data_size);
719 case AC3_CHMODE_3F2R:
720 memset(s->delay[4], 0, channel_data_size);
721 case AC3_CHMODE_3F1R:
722 memset(s->delay[3], 0, channel_data_size);
724 memcpy(s->delay[2], s->delay[1], channel_data_size);
725 memset(s->delay[1], 0, channel_data_size);
731 * Decode band structure for coupling, spectral extension, or enhanced coupling.
732 * The band structure defines how many subbands are in each band. For each
733 * subband in the range, 1 means it is combined with the previous band, and 0
734 * means that it starts a new band.
736 * @param[in] gbc bit reader context
737 * @param[in] blk block number
738 * @param[in] eac3 flag to indicate E-AC-3
739 * @param[in] ecpl flag to indicate enhanced coupling
740 * @param[in] start_subband subband number for start of range
741 * @param[in] end_subband subband number for end of range
742 * @param[in] default_band_struct default band structure table
743 * @param[out] num_bands number of bands (optionally NULL)
744 * @param[out] band_sizes array containing the number of bins in each band (optionally NULL)
746 static void decode_band_structure(GetBitContext *gbc, int blk, int eac3,
747 int ecpl, int start_subband, int end_subband,
748 const uint8_t *default_band_struct,
749 int *num_bands, uint8_t *band_sizes)
751 int subbnd, bnd, n_subbands, n_bands=0;
753 uint8_t coded_band_struct[22];
754 const uint8_t *band_struct;
756 n_subbands = end_subband - start_subband;
758 /* decode band structure from bitstream or use default */
759 if (!eac3 || get_bits1(gbc)) {
760 for (subbnd = 0; subbnd < n_subbands - 1; subbnd++) {
761 coded_band_struct[subbnd] = get_bits1(gbc);
763 band_struct = coded_band_struct;
765 band_struct = &default_band_struct[start_subband+1];
767 /* no change in band structure */
771 /* calculate number of bands and band sizes based on band structure.
772 note that the first 4 subbands in enhanced coupling span only 6 bins
774 if (num_bands || band_sizes ) {
775 n_bands = n_subbands;
776 bnd_sz[0] = ecpl ? 6 : 12;
777 for (bnd = 0, subbnd = 1; subbnd < n_subbands; subbnd++) {
778 int subbnd_size = (ecpl && subbnd < 4) ? 6 : 12;
779 if (band_struct[subbnd - 1]) {
781 bnd_sz[bnd] += subbnd_size;
783 bnd_sz[++bnd] = subbnd_size;
788 /* set optional output params */
790 *num_bands = n_bands;
792 memcpy(band_sizes, bnd_sz, n_bands);
796 * Decode a single audio block from the AC-3 bitstream.
798 static int decode_audio_block(AC3DecodeContext *s, int blk)
800 int fbw_channels = s->fbw_channels;
801 int channel_mode = s->channel_mode;
803 int different_transforms;
806 GetBitContext *gbc = &s->gbc;
807 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS] = { 0 };
809 /* block switch flags */
810 different_transforms = 0;
811 if (s->block_switch_syntax) {
812 for (ch = 1; ch <= fbw_channels; ch++) {
813 s->block_switch[ch] = get_bits1(gbc);
814 if (ch > 1 && s->block_switch[ch] != s->block_switch[1])
815 different_transforms = 1;
819 /* dithering flags */
820 if (s->dither_flag_syntax) {
821 for (ch = 1; ch <= fbw_channels; ch++) {
822 s->dither_flag[ch] = get_bits1(gbc);
827 i = !s->channel_mode;
829 if (get_bits1(gbc)) {
830 /* Allow asymmetric application of DRC when drc_scale > 1.
831 Amplification of quiet sounds is enhanced */
832 int range_bits = get_bits(gbc, 8);
833 INTFLOAT range = AC3_RANGE(range_bits);
834 if (range_bits <= 127 || s->drc_scale <= 1.0)
835 s->dynamic_range[i] = AC3_DYNAMIC_RANGE(range);
837 s->dynamic_range[i] = range;
838 } else if (blk == 0) {
839 s->dynamic_range[i] = AC3_DYNAMIC_RANGE1;
843 /* spectral extension strategy */
844 if (s->eac3 && (!blk || get_bits1(gbc))) {
845 s->spx_in_use = get_bits1(gbc);
847 int dst_start_freq, dst_end_freq, src_start_freq,
848 start_subband, end_subband;
850 /* determine which channels use spx */
851 if (s->channel_mode == AC3_CHMODE_MONO) {
852 s->channel_uses_spx[1] = 1;
854 for (ch = 1; ch <= fbw_channels; ch++)
855 s->channel_uses_spx[ch] = get_bits1(gbc);
858 /* get the frequency bins of the spx copy region and the spx start
860 dst_start_freq = get_bits(gbc, 2);
861 start_subband = get_bits(gbc, 3) + 2;
862 if (start_subband > 7)
863 start_subband += start_subband - 7;
864 end_subband = get_bits(gbc, 3) + 5;
866 s->spx_dst_end_freq = end_freq_inv_tab[end_subband-5];
869 end_subband += end_subband - 7;
870 dst_start_freq = dst_start_freq * 12 + 25;
871 src_start_freq = start_subband * 12 + 25;
872 dst_end_freq = end_subband * 12 + 25;
874 /* check validity of spx ranges */
875 if (start_subband >= end_subband) {
876 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
877 "range (%d >= %d)\n", start_subband, end_subband);
878 return AVERROR_INVALIDDATA;
880 if (dst_start_freq >= src_start_freq) {
881 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
882 "copy start bin (%d >= %d)\n", dst_start_freq, src_start_freq);
883 return AVERROR_INVALIDDATA;
886 s->spx_dst_start_freq = dst_start_freq;
887 s->spx_src_start_freq = src_start_freq;
889 s->spx_dst_end_freq = dst_end_freq;
891 decode_band_structure(gbc, blk, s->eac3, 0,
892 start_subband, end_subband,
893 ff_eac3_default_spx_band_struct,
897 for (ch = 1; ch <= fbw_channels; ch++) {
898 s->channel_uses_spx[ch] = 0;
899 s->first_spx_coords[ch] = 1;
904 /* spectral extension coordinates */
906 for (ch = 1; ch <= fbw_channels; ch++) {
907 if (s->channel_uses_spx[ch]) {
908 if (s->first_spx_coords[ch] || get_bits1(gbc)) {
910 int bin, master_spx_coord;
912 s->first_spx_coords[ch] = 0;
913 spx_blend = AC3_SPX_BLEND(get_bits(gbc, 5));
914 master_spx_coord = get_bits(gbc, 2) * 3;
916 bin = s->spx_src_start_freq;
917 for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
918 int bandsize = s->spx_band_sizes[bnd];
919 int spx_coord_exp, spx_coord_mant;
920 INTFLOAT nratio, sblend, nblend;
922 /* calculate blending factors */
923 int64_t accu = ((bin << 23) + (bandsize << 22))
924 * (int64_t)s->spx_dst_end_freq;
925 nratio = (int)(accu >> 32);
926 nratio -= spx_blend << 18;
931 } else if (nratio > 0x7fffff) {
932 nblend = 14529495; // sqrt(3) in FP.23
935 nblend = fixed_sqrt(nratio, 23);
936 accu = (int64_t)nblend * 1859775393;
937 nblend = (int)((accu + (1<<29)) >> 30);
938 sblend = fixed_sqrt(0x800000 - nratio, 23);
943 /* calculate blending factors */
944 nratio = ((float)((bin + (bandsize >> 1))) / s->spx_dst_end_freq) - spx_blend;
945 nratio = av_clipf(nratio, 0.0f, 1.0f);
946 nblend = sqrtf(3.0f * nratio); // noise is scaled by sqrt(3)
947 // to give unity variance
948 sblend = sqrtf(1.0f - nratio);
952 /* decode spx coordinates */
953 spx_coord_exp = get_bits(gbc, 4);
954 spx_coord_mant = get_bits(gbc, 2);
955 if (spx_coord_exp == 15) spx_coord_mant <<= 1;
956 else spx_coord_mant += 4;
957 spx_coord_mant <<= (25 - spx_coord_exp - master_spx_coord);
959 /* multiply noise and signal blending factors by spx coordinate */
961 accu = (int64_t)nblend * spx_coord_mant;
962 s->spx_noise_blend[ch][bnd] = (int)((accu + (1<<22)) >> 23);
963 accu = (int64_t)sblend * spx_coord_mant;
964 s->spx_signal_blend[ch][bnd] = (int)((accu + (1<<22)) >> 23);
966 spx_coord = spx_coord_mant * (1.0f / (1 << 23));
967 s->spx_noise_blend [ch][bnd] = nblend * spx_coord;
968 s->spx_signal_blend[ch][bnd] = sblend * spx_coord;
973 s->first_spx_coords[ch] = 1;
978 /* coupling strategy */
979 if (s->eac3 ? s->cpl_strategy_exists[blk] : get_bits1(gbc)) {
980 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
982 s->cpl_in_use[blk] = get_bits1(gbc);
983 if (s->cpl_in_use[blk]) {
984 /* coupling in use */
985 int cpl_start_subband, cpl_end_subband;
987 if (channel_mode < AC3_CHMODE_STEREO) {
988 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
989 return AVERROR_INVALIDDATA;
992 /* check for enhanced coupling */
993 if (s->eac3 && get_bits1(gbc)) {
994 /* TODO: parse enhanced coupling strategy info */
995 avpriv_request_sample(s->avctx, "Enhanced coupling");
996 return AVERROR_PATCHWELCOME;
999 /* determine which channels are coupled */
1000 if (s->eac3 && s->channel_mode == AC3_CHMODE_STEREO) {
1001 s->channel_in_cpl[1] = 1;
1002 s->channel_in_cpl[2] = 1;
1004 for (ch = 1; ch <= fbw_channels; ch++)
1005 s->channel_in_cpl[ch] = get_bits1(gbc);
1008 /* phase flags in use */
1009 if (channel_mode == AC3_CHMODE_STEREO)
1010 s->phase_flags_in_use = get_bits1(gbc);
1012 /* coupling frequency range */
1013 cpl_start_subband = get_bits(gbc, 4);
1014 cpl_end_subband = s->spx_in_use ? (s->spx_src_start_freq - 37) / 12 :
1015 get_bits(gbc, 4) + 3;
1016 if (cpl_start_subband >= cpl_end_subband) {
1017 av_log(s->avctx, AV_LOG_ERROR, "invalid coupling range (%d >= %d)\n",
1018 cpl_start_subband, cpl_end_subband);
1019 return AVERROR_INVALIDDATA;
1021 s->start_freq[CPL_CH] = cpl_start_subband * 12 + 37;
1022 s->end_freq[CPL_CH] = cpl_end_subband * 12 + 37;
1024 decode_band_structure(gbc, blk, s->eac3, 0, cpl_start_subband,
1026 ff_eac3_default_cpl_band_struct,
1027 &s->num_cpl_bands, s->cpl_band_sizes);
1029 /* coupling not in use */
1030 for (ch = 1; ch <= fbw_channels; ch++) {
1031 s->channel_in_cpl[ch] = 0;
1032 s->first_cpl_coords[ch] = 1;
1034 s->first_cpl_leak = s->eac3;
1035 s->phase_flags_in_use = 0;
1037 } else if (!s->eac3) {
1039 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must "
1040 "be present in block 0\n");
1041 return AVERROR_INVALIDDATA;
1043 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
1046 cpl_in_use = s->cpl_in_use[blk];
1048 /* coupling coordinates */
1050 int cpl_coords_exist = 0;
1052 for (ch = 1; ch <= fbw_channels; ch++) {
1053 if (s->channel_in_cpl[ch]) {
1054 if ((s->eac3 && s->first_cpl_coords[ch]) || get_bits1(gbc)) {
1055 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
1056 s->first_cpl_coords[ch] = 0;
1057 cpl_coords_exist = 1;
1058 master_cpl_coord = 3 * get_bits(gbc, 2);
1059 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
1060 cpl_coord_exp = get_bits(gbc, 4);
1061 cpl_coord_mant = get_bits(gbc, 4);
1062 if (cpl_coord_exp == 15)
1063 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
1065 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
1066 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
1069 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must "
1070 "be present in block 0\n");
1071 return AVERROR_INVALIDDATA;
1074 /* channel not in coupling */
1075 s->first_cpl_coords[ch] = 1;
1079 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
1080 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
1081 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
1086 /* stereo rematrixing strategy and band structure */
1087 if (channel_mode == AC3_CHMODE_STEREO) {
1088 if ((s->eac3 && !blk) || get_bits1(gbc)) {
1089 s->num_rematrixing_bands = 4;
1090 if (cpl_in_use && s->start_freq[CPL_CH] <= 61) {
1091 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
1092 } else if (s->spx_in_use && s->spx_src_start_freq <= 61) {
1093 s->num_rematrixing_bands--;
1095 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++)
1096 s->rematrixing_flags[bnd] = get_bits1(gbc);
1098 av_log(s->avctx, AV_LOG_WARNING, "Warning: "
1099 "new rematrixing strategy not present in block 0\n");
1100 s->num_rematrixing_bands = 0;
1104 /* exponent strategies for each channel */
1105 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1107 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
1108 if (s->exp_strategy[blk][ch] != EXP_REUSE)
1109 bit_alloc_stages[ch] = 3;
1112 /* channel bandwidth */
1113 for (ch = 1; ch <= fbw_channels; ch++) {
1114 s->start_freq[ch] = 0;
1115 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1117 int prev = s->end_freq[ch];
1118 if (s->channel_in_cpl[ch])
1119 s->end_freq[ch] = s->start_freq[CPL_CH];
1120 else if (s->channel_uses_spx[ch])
1121 s->end_freq[ch] = s->spx_src_start_freq;
1123 int bandwidth_code = get_bits(gbc, 6);
1124 if (bandwidth_code > 60) {
1125 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60\n", bandwidth_code);
1126 return AVERROR_INVALIDDATA;
1128 s->end_freq[ch] = bandwidth_code * 3 + 73;
1130 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
1131 s->num_exp_groups[ch] = (s->end_freq[ch] + group_size-4) / group_size;
1132 if (blk > 0 && s->end_freq[ch] != prev)
1133 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
1136 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
1137 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
1138 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
1141 /* decode exponents for each channel */
1142 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1143 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1144 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
1145 if (decode_exponents(gbc, s->exp_strategy[blk][ch],
1146 s->num_exp_groups[ch], s->dexps[ch][0],
1147 &s->dexps[ch][s->start_freq[ch]+!!ch])) {
1148 av_log(s->avctx, AV_LOG_ERROR, "exponent out-of-range\n");
1149 return AVERROR_INVALIDDATA;
1151 if (ch != CPL_CH && ch != s->lfe_ch)
1152 skip_bits(gbc, 2); /* skip gainrng */
1156 /* bit allocation information */
1157 if (s->bit_allocation_syntax) {
1158 if (get_bits1(gbc)) {
1159 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1160 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1161 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
1162 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
1163 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
1164 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1165 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1167 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must "
1168 "be present in block 0\n");
1169 return AVERROR_INVALIDDATA;
1173 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
1174 if (!s->eac3 || !blk) {
1175 if (s->snr_offset_strategy && get_bits1(gbc)) {
1178 csnr = (get_bits(gbc, 6) - 15) << 4;
1179 for (i = ch = !cpl_in_use; ch <= s->channels; ch++) {
1181 if (ch == i || s->snr_offset_strategy == 2)
1182 snr = (csnr + get_bits(gbc, 4)) << 2;
1183 /* run at least last bit allocation stage if snr offset changes */
1184 if (blk && s->snr_offset[ch] != snr) {
1185 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 1);
1187 s->snr_offset[ch] = snr;
1189 /* fast gain (normal AC-3 only) */
1191 int prev = s->fast_gain[ch];
1192 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1193 /* run last 2 bit allocation stages if fast gain changes */
1194 if (blk && prev != s->fast_gain[ch])
1195 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1198 } else if (!s->eac3 && !blk) {
1199 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
1200 return AVERROR_INVALIDDATA;
1204 /* fast gain (E-AC-3 only) */
1205 if (s->fast_gain_syntax && get_bits1(gbc)) {
1206 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1207 int prev = s->fast_gain[ch];
1208 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1209 /* run last 2 bit allocation stages if fast gain changes */
1210 if (blk && prev != s->fast_gain[ch])
1211 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1213 } else if (s->eac3 && !blk) {
1214 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1215 s->fast_gain[ch] = ff_ac3_fast_gain_tab[4];
1218 /* E-AC-3 to AC-3 converter SNR offset */
1219 if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT && get_bits1(gbc)) {
1220 skip_bits(gbc, 10); // skip converter snr offset
1223 /* coupling leak information */
1225 if (s->first_cpl_leak || get_bits1(gbc)) {
1226 int fl = get_bits(gbc, 3);
1227 int sl = get_bits(gbc, 3);
1228 /* run last 2 bit allocation stages for coupling channel if
1229 coupling leak changes */
1230 if (blk && (fl != s->bit_alloc_params.cpl_fast_leak ||
1231 sl != s->bit_alloc_params.cpl_slow_leak)) {
1232 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
1234 s->bit_alloc_params.cpl_fast_leak = fl;
1235 s->bit_alloc_params.cpl_slow_leak = sl;
1236 } else if (!s->eac3 && !blk) {
1237 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must "
1238 "be present in block 0\n");
1239 return AVERROR_INVALIDDATA;
1241 s->first_cpl_leak = 0;
1244 /* delta bit allocation information */
1245 if (s->dba_syntax && get_bits1(gbc)) {
1246 /* delta bit allocation exists (strategy) */
1247 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1248 s->dba_mode[ch] = get_bits(gbc, 2);
1249 if (s->dba_mode[ch] == DBA_RESERVED) {
1250 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1251 return AVERROR_INVALIDDATA;
1253 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1255 /* channel delta offset, len and bit allocation */
1256 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1257 if (s->dba_mode[ch] == DBA_NEW) {
1258 s->dba_nsegs[ch] = get_bits(gbc, 3) + 1;
1259 for (seg = 0; seg < s->dba_nsegs[ch]; seg++) {
1260 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1261 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1262 s->dba_values[ch][seg] = get_bits(gbc, 3);
1264 /* run last 2 bit allocation stages if new dba values */
1265 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1268 } else if (blk == 0) {
1269 for (ch = 0; ch <= s->channels; ch++) {
1270 s->dba_mode[ch] = DBA_NONE;
1274 /* Bit allocation */
1275 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1276 if (bit_alloc_stages[ch] > 2) {
1277 /* Exponent mapping into PSD and PSD integration */
1278 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1279 s->start_freq[ch], s->end_freq[ch],
1280 s->psd[ch], s->band_psd[ch]);
1282 if (bit_alloc_stages[ch] > 1) {
1283 /* Compute excitation function, Compute masking curve, and
1284 Apply delta bit allocation */
1285 if (ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1286 s->start_freq[ch], s->end_freq[ch],
1287 s->fast_gain[ch], (ch == s->lfe_ch),
1288 s->dba_mode[ch], s->dba_nsegs[ch],
1289 s->dba_offsets[ch], s->dba_lengths[ch],
1290 s->dba_values[ch], s->mask[ch])) {
1291 av_log(s->avctx, AV_LOG_ERROR, "error in bit allocation\n");
1292 return AVERROR_INVALIDDATA;
1295 if (bit_alloc_stages[ch] > 0) {
1296 /* Compute bit allocation */
1297 const uint8_t *bap_tab = s->channel_uses_aht[ch] ?
1298 ff_eac3_hebap_tab : ff_ac3_bap_tab;
1299 s->ac3dsp.bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1300 s->start_freq[ch], s->end_freq[ch],
1302 s->bit_alloc_params.floor,
1303 bap_tab, s->bap[ch]);
1307 /* unused dummy data */
1308 if (s->skip_syntax && get_bits1(gbc)) {
1309 int skipl = get_bits(gbc, 9);
1314 /* unpack the transform coefficients
1315 this also uncouples channels if coupling is in use. */
1316 decode_transform_coeffs(s, blk);
1318 /* TODO: generate enhanced coupling coordinates and uncouple */
1320 /* recover coefficients if rematrixing is in use */
1321 if (s->channel_mode == AC3_CHMODE_STEREO)
1324 /* apply scaling to coefficients (headroom, dynrng) */
1325 for (ch = 1; ch <= s->channels; ch++) {
1326 int audio_channel = 0;
1328 if (s->channel_mode == AC3_CHMODE_DUALMONO)
1329 audio_channel = 2-ch;
1330 if (s->heavy_compression && s->compression_exists[audio_channel])
1331 gain = s->heavy_dynamic_range[audio_channel];
1333 gain = s->dynamic_range[audio_channel];
1336 scale_coefs(s->transform_coeffs[ch], s->fixed_coeffs[ch], gain, 256);
1338 if (s->target_level != 0)
1339 gain = gain * s->level_gain[audio_channel];
1340 gain *= 1.0 / 4194304.0f;
1341 s->fmt_conv.int32_to_float_fmul_scalar(s->transform_coeffs[ch],
1342 s->fixed_coeffs[ch], gain, 256);
1346 /* apply spectral extension to high frequency bins */
1347 if (CONFIG_EAC3_DECODER && s->spx_in_use) {
1348 ff_eac3_apply_spectral_extension(s);
1351 /* downmix and MDCT. order depends on whether block switching is used for
1352 any channel in this block. this is because coefficients for the long
1353 and short transforms cannot be mixed. */
1354 downmix_output = s->channels != s->out_channels &&
1355 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1356 s->fbw_channels == s->out_channels);
1357 if (different_transforms) {
1358 /* the delay samples have already been downmixed, so we upmix the delay
1359 samples in order to reconstruct all channels before downmixing. */
1365 do_imdct(s, s->channels);
1367 if (downmix_output) {
1369 ac3_downmix_c_fixed16(s->outptr, s->downmix_coeffs,
1370 s->out_channels, s->fbw_channels, 256);
1372 s->ac3dsp.downmix(s->outptr, s->downmix_coeffs,
1373 s->out_channels, s->fbw_channels, 256);
1377 if (downmix_output) {
1378 s->ac3dsp.AC3_RENAME(downmix)(s->xcfptr + 1, s->downmix_coeffs,
1379 s->out_channels, s->fbw_channels, 256);
1382 if (downmix_output && !s->downmixed) {
1384 s->ac3dsp.AC3_RENAME(downmix)(s->dlyptr, s->downmix_coeffs,
1385 s->out_channels, s->fbw_channels, 128);
1388 do_imdct(s, s->out_channels);
1395 * Decode a single AC-3 frame.
1397 static int ac3_decode_frame(AVCodecContext * avctx, void *data,
1398 int *got_frame_ptr, AVPacket *avpkt)
1400 AVFrame *frame = data;
1401 const uint8_t *buf = avpkt->data;
1402 int buf_size = avpkt->size;
1403 AC3DecodeContext *s = avctx->priv_data;
1404 int blk, ch, err, ret;
1405 const uint8_t *channel_map;
1406 const SHORTFLOAT *output[AC3_MAX_CHANNELS];
1407 enum AVMatrixEncoding matrix_encoding;
1408 AVDownmixInfo *downmix_info;
1410 /* copy input buffer to decoder context to avoid reading past the end
1411 of the buffer, which can be caused by a damaged input stream. */
1412 if (buf_size >= 2 && AV_RB16(buf) == 0x770B) {
1413 // seems to be byte-swapped AC-3
1414 int cnt = FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE) >> 1;
1415 s->bdsp.bswap16_buf((uint16_t *) s->input_buffer,
1416 (const uint16_t *) buf, cnt);
1418 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE));
1419 buf = s->input_buffer;
1420 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1421 if ((ret = init_get_bits8(&s->gbc, buf, buf_size)) < 0)
1424 /* parse the syncinfo */
1425 err = parse_frame_header(s);
1429 case AAC_AC3_PARSE_ERROR_SYNC:
1430 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1431 return AVERROR_INVALIDDATA;
1432 case AAC_AC3_PARSE_ERROR_BSID:
1433 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1435 case AAC_AC3_PARSE_ERROR_SAMPLE_RATE:
1436 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1438 case AAC_AC3_PARSE_ERROR_FRAME_SIZE:
1439 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1441 case AAC_AC3_PARSE_ERROR_FRAME_TYPE:
1442 /* skip frame if CRC is ok. otherwise use error concealment. */
1443 /* TODO: add support for substreams and dependent frames */
1444 if (s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1445 av_log(avctx, AV_LOG_WARNING, "unsupported frame type : "
1446 "skipping frame\n");
1450 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1453 case AAC_AC3_PARSE_ERROR_CRC:
1454 case AAC_AC3_PARSE_ERROR_CHANNEL_CFG:
1456 default: // Normal AVERROR do not try to recover.
1461 /* check that reported frame size fits in input buffer */
1462 if (s->frame_size > buf_size) {
1463 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1464 err = AAC_AC3_PARSE_ERROR_FRAME_SIZE;
1465 } else if (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL)) {
1466 /* check for crc mismatch */
1467 if (av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2],
1468 s->frame_size - 2)) {
1469 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1470 if (avctx->err_recognition & AV_EF_EXPLODE)
1471 return AVERROR_INVALIDDATA;
1472 err = AAC_AC3_PARSE_ERROR_CRC;
1477 /* if frame is ok, set audio parameters */
1479 avctx->sample_rate = s->sample_rate;
1480 avctx->bit_rate = s->bit_rate;
1483 /* channel config */
1484 if (!err || (s->channels && s->out_channels != s->channels)) {
1485 s->out_channels = s->channels;
1486 s->output_mode = s->channel_mode;
1488 s->output_mode |= AC3_OUTPUT_LFEON;
1489 if (s->channels > 1 &&
1490 avctx->request_channel_layout == AV_CH_LAYOUT_MONO) {
1491 s->out_channels = 1;
1492 s->output_mode = AC3_CHMODE_MONO;
1493 } else if (s->channels > 2 &&
1494 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
1495 s->out_channels = 2;
1496 s->output_mode = AC3_CHMODE_STEREO;
1499 s->loro_center_mix_level = gain_levels[s-> center_mix_level];
1500 s->loro_surround_mix_level = gain_levels[s->surround_mix_level];
1501 s->ltrt_center_mix_level = LEVEL_MINUS_3DB;
1502 s->ltrt_surround_mix_level = LEVEL_MINUS_3DB;
1503 /* set downmixing coefficients if needed */
1504 if (s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1505 s->fbw_channels == s->out_channels)) {
1506 set_downmix_coeffs(s);
1508 } else if (!s->channels) {
1509 av_log(avctx, AV_LOG_ERROR, "unable to determine channel mode\n");
1510 return AVERROR_INVALIDDATA;
1512 avctx->channels = s->out_channels;
1513 avctx->channel_layout = avpriv_ac3_channel_layout_tab[s->output_mode & ~AC3_OUTPUT_LFEON];
1514 if (s->output_mode & AC3_OUTPUT_LFEON)
1515 avctx->channel_layout |= AV_CH_LOW_FREQUENCY;
1517 /* set audio service type based on bitstream mode for AC-3 */
1518 avctx->audio_service_type = s->bitstream_mode;
1519 if (s->bitstream_mode == 0x7 && s->channels > 1)
1520 avctx->audio_service_type = AV_AUDIO_SERVICE_TYPE_KARAOKE;
1522 /* get output buffer */
1523 frame->nb_samples = s->num_blocks * AC3_BLOCK_SIZE;
1524 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1527 /* decode the audio blocks */
1528 channel_map = ff_ac3_dec_channel_map[s->output_mode & ~AC3_OUTPUT_LFEON][s->lfe_on];
1529 for (ch = 0; ch < AC3_MAX_CHANNELS; ch++) {
1530 output[ch] = s->output[ch];
1531 s->outptr[ch] = s->output[ch];
1533 for (ch = 0; ch < s->channels; ch++) {
1534 if (ch < s->out_channels)
1535 s->outptr[channel_map[ch]] = (SHORTFLOAT *)frame->data[ch];
1537 for (blk = 0; blk < s->num_blocks; blk++) {
1538 if (!err && decode_audio_block(s, blk)) {
1539 av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n");
1543 for (ch = 0; ch < s->out_channels; ch++)
1544 memcpy(((SHORTFLOAT*)frame->data[ch]) + AC3_BLOCK_SIZE*blk, output[ch], AC3_BLOCK_SIZE*sizeof(SHORTFLOAT));
1545 for (ch = 0; ch < s->out_channels; ch++)
1546 output[ch] = s->outptr[channel_map[ch]];
1547 for (ch = 0; ch < s->out_channels; ch++) {
1548 if (!ch || channel_map[ch])
1549 s->outptr[channel_map[ch]] += AC3_BLOCK_SIZE;
1553 av_frame_set_decode_error_flags(frame, err ? FF_DECODE_ERROR_INVALID_BITSTREAM : 0);
1555 /* keep last block for error concealment in next frame */
1556 for (ch = 0; ch < s->out_channels; ch++)
1557 memcpy(s->output[ch], output[ch], AC3_BLOCK_SIZE*sizeof(SHORTFLOAT));
1562 * Check whether the input layout is compatible, and make sure we're not
1563 * downmixing (else the matrix encoding is no longer applicable).
1565 matrix_encoding = AV_MATRIX_ENCODING_NONE;
1566 if (s->channel_mode == AC3_CHMODE_STEREO &&
1567 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1568 if (s->dolby_surround_mode == AC3_DSURMOD_ON)
1569 matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
1570 else if (s->dolby_headphone_mode == AC3_DHEADPHONMOD_ON)
1571 matrix_encoding = AV_MATRIX_ENCODING_DOLBYHEADPHONE;
1572 } else if (s->channel_mode >= AC3_CHMODE_2F2R &&
1573 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1574 switch (s->dolby_surround_ex_mode) {
1575 case AC3_DSUREXMOD_ON: // EX or PLIIx
1576 matrix_encoding = AV_MATRIX_ENCODING_DOLBYEX;
1578 case AC3_DSUREXMOD_PLIIZ:
1579 matrix_encoding = AV_MATRIX_ENCODING_DPLIIZ;
1581 default: // not indicated or off
1585 if ((ret = ff_side_data_update_matrix_encoding(frame, matrix_encoding)) < 0)
1589 if ((downmix_info = av_downmix_info_update_side_data(frame))) {
1590 switch (s->preferred_downmix) {
1591 case AC3_DMIXMOD_LTRT:
1592 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LTRT;
1594 case AC3_DMIXMOD_LORO:
1595 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LORO;
1597 case AC3_DMIXMOD_DPLII:
1598 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_DPLII;
1601 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_UNKNOWN;
1604 downmix_info->center_mix_level = gain_levels[s-> center_mix_level];
1605 downmix_info->center_mix_level_ltrt = gain_levels[s-> center_mix_level_ltrt];
1606 downmix_info->surround_mix_level = gain_levels[s-> surround_mix_level];
1607 downmix_info->surround_mix_level_ltrt = gain_levels[s->surround_mix_level_ltrt];
1608 if (s->lfe_mix_level_exists)
1609 downmix_info->lfe_mix_level = gain_levels_lfe[s->lfe_mix_level];
1611 downmix_info->lfe_mix_level = 0.0; // -inf dB
1613 return AVERROR(ENOMEM);
1617 return FFMIN(buf_size, s->frame_size);
1621 * Uninitialize the AC-3 decoder.
1623 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1625 AC3DecodeContext *s = avctx->priv_data;
1626 ff_mdct_end(&s->imdct_512);
1627 ff_mdct_end(&s->imdct_256);
1633 #define OFFSET(x) offsetof(AC3DecodeContext, x)
1634 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)