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 * Portions of this code are derived from liba52
11 * http://liba52.sourceforge.net
12 * Copyright (C) 2000-2003 Michel Lespinasse <walken@zoy.org>
13 * Copyright (C) 1999-2000 Aaron Holtzman <aholtzma@ess.engr.uvic.ca>
15 * This file is part of FFmpeg.
17 * FFmpeg is free software; you can redistribute it and/or
18 * modify it under the terms of the GNU General Public
19 * License as published by the Free Software Foundation; either
20 * version 2 of the License, or (at your option) any later version.
22 * FFmpeg is distributed in the hope that it will be useful,
23 * but WITHOUT ANY WARRANTY; without even the implied warranty of
24 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
25 * General Public License for more details.
27 * You should have received a copy of the GNU General Public
28 * License along with FFmpeg; if not, write to the Free Software
29 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
37 #include "libavutil/crc.h"
38 #include "ac3_parser.h"
40 #include "ac3dec_data.h"
42 /** Large enough for maximum possible frame size when the specification limit is ignored */
43 #define AC3_FRAME_BUFFER_SIZE 32768
46 * table for ungrouping 3 values in 7 bits.
47 * used for exponents and bap=2 mantissas
49 static uint8_t ungroup_3_in_7_bits_tab[128][3];
52 /** tables for ungrouping mantissas */
53 static int b1_mantissas[32][3];
54 static int b2_mantissas[128][3];
55 static int b3_mantissas[8];
56 static int b4_mantissas[128][2];
57 static int b5_mantissas[16];
60 * Quantization table: levels for symmetric. bits for asymmetric.
61 * reference: Table 7.18 Mapping of bap to Quantizer
63 static const uint8_t quantization_tab[16] = {
65 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
68 /** dynamic range table. converts codes to scale factors. */
69 static float dynamic_range_tab[256];
71 /** Adjustments in dB gain */
72 #define LEVEL_PLUS_3DB 1.4142135623730950
73 #define LEVEL_PLUS_1POINT5DB 1.1892071150027209
74 #define LEVEL_MINUS_1POINT5DB 0.8408964152537145
75 #define LEVEL_MINUS_3DB 0.7071067811865476
76 #define LEVEL_MINUS_4POINT5DB 0.5946035575013605
77 #define LEVEL_MINUS_6DB 0.5000000000000000
78 #define LEVEL_MINUS_9DB 0.3535533905932738
79 #define LEVEL_ZERO 0.0000000000000000
80 #define LEVEL_ONE 1.0000000000000000
82 static const float gain_levels[9] = {
86 LEVEL_MINUS_1POINT5DB,
88 LEVEL_MINUS_4POINT5DB,
95 * Table for center mix levels
96 * reference: Section 5.4.2.4 cmixlev
98 static const uint8_t center_levels[4] = { 4, 5, 6, 5 };
101 * Table for surround mix levels
102 * reference: Section 5.4.2.5 surmixlev
104 static const uint8_t surround_levels[4] = { 4, 6, 7, 6 };
107 * Table for default stereo downmixing coefficients
108 * reference: Section 7.8.2 Downmixing Into Two Channels
110 static const uint8_t ac3_default_coeffs[8][5][2] = {
111 { { 2, 7 }, { 7, 2 }, },
113 { { 2, 7 }, { 7, 2 }, },
114 { { 2, 7 }, { 5, 5 }, { 7, 2 }, },
115 { { 2, 7 }, { 7, 2 }, { 6, 6 }, },
116 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 8, 8 }, },
117 { { 2, 7 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
118 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
122 * Symmetrical Dequantization
123 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
124 * Tables 7.19 to 7.23
127 symmetric_dequant(int code, int levels)
129 return ((code - (levels >> 1)) << 24) / levels;
133 * Initialize tables at runtime.
135 static av_cold void ac3_tables_init(void)
139 /* generate table for ungrouping 3 values in 7 bits
140 reference: Section 7.1.3 Exponent Decoding */
141 for(i=0; i<128; i++) {
142 ungroup_3_in_7_bits_tab[i][0] = i / 25;
143 ungroup_3_in_7_bits_tab[i][1] = (i % 25) / 5;
144 ungroup_3_in_7_bits_tab[i][2] = (i % 25) % 5;
147 /* generate grouped mantissa tables
148 reference: Section 7.3.5 Ungrouping of Mantissas */
149 for(i=0; i<32; i++) {
150 /* bap=1 mantissas */
151 b1_mantissas[i][0] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][0], 3);
152 b1_mantissas[i][1] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][1], 3);
153 b1_mantissas[i][2] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][2], 3);
155 for(i=0; i<128; i++) {
156 /* bap=2 mantissas */
157 b2_mantissas[i][0] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][0], 5);
158 b2_mantissas[i][1] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][1], 5);
159 b2_mantissas[i][2] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][2], 5);
161 /* bap=4 mantissas */
162 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
163 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
165 /* generate ungrouped mantissa tables
166 reference: Tables 7.21 and 7.23 */
168 /* bap=3 mantissas */
169 b3_mantissas[i] = symmetric_dequant(i, 7);
171 for(i=0; i<15; i++) {
172 /* bap=5 mantissas */
173 b5_mantissas[i] = symmetric_dequant(i, 15);
176 /* generate dynamic range table
177 reference: Section 7.7.1 Dynamic Range Control */
178 for(i=0; i<256; i++) {
179 int v = (i >> 5) - ((i >> 7) << 3) - 5;
180 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
186 * AVCodec initialization
188 static av_cold int ac3_decode_init(AVCodecContext *avctx)
190 AC3DecodeContext *s = avctx->priv_data;
195 ff_mdct_init(&s->imdct_256, 8, 1);
196 ff_mdct_init(&s->imdct_512, 9, 1);
197 ff_kbd_window_init(s->window, 5.0, 256);
198 dsputil_init(&s->dsp, avctx);
199 av_lfg_init(&s->dith_state, 0);
201 /* set bias values for float to int16 conversion */
202 if(s->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
203 s->add_bias = 385.0f;
207 s->mul_bias = 32767.0f;
210 /* allow downmixing to stereo or mono */
211 if (avctx->channels > 0 && avctx->request_channels > 0 &&
212 avctx->request_channels < avctx->channels &&
213 avctx->request_channels <= 2) {
214 avctx->channels = avctx->request_channels;
218 /* allocate context input buffer */
219 if (avctx->error_resilience >= FF_ER_CAREFUL) {
220 s->input_buffer = av_mallocz(AC3_FRAME_BUFFER_SIZE + FF_INPUT_BUFFER_PADDING_SIZE);
221 if (!s->input_buffer)
222 return AVERROR_NOMEM;
225 avctx->sample_fmt = SAMPLE_FMT_S16;
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 skip_bits(gbc, 5); // skip dialog normalization
244 skip_bits(gbc, 8); //skip compression
246 skip_bits(gbc, 8); //skip language code
248 skip_bits(gbc, 7); //skip audio production information
251 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
253 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
254 TODO: read & use the xbsi1 downmix levels */
256 skip_bits(gbc, 14); //skip timecode1 / xbsi1
258 skip_bits(gbc, 14); //skip timecode2 / xbsi2
260 /* skip additional bitstream info */
261 if (get_bits1(gbc)) {
262 i = get_bits(gbc, 6);
272 * Common function to parse AC-3 or E-AC-3 frame header
274 static int parse_frame_header(AC3DecodeContext *s)
279 err = ff_ac3_parse_header(&s->gbc, &hdr);
283 /* get decoding parameters from header info */
284 s->bit_alloc_params.sr_code = hdr.sr_code;
285 s->channel_mode = hdr.channel_mode;
286 s->lfe_on = hdr.lfe_on;
287 s->bit_alloc_params.sr_shift = hdr.sr_shift;
288 s->sample_rate = hdr.sample_rate;
289 s->bit_rate = hdr.bit_rate;
290 s->channels = hdr.channels;
291 s->fbw_channels = s->channels - s->lfe_on;
292 s->lfe_ch = s->fbw_channels + 1;
293 s->frame_size = hdr.frame_size;
294 s->center_mix_level = hdr.center_mix_level;
295 s->surround_mix_level = hdr.surround_mix_level;
296 s->num_blocks = hdr.num_blocks;
297 s->frame_type = hdr.frame_type;
298 s->substreamid = hdr.substreamid;
301 s->start_freq[s->lfe_ch] = 0;
302 s->end_freq[s->lfe_ch] = 7;
303 s->num_exp_groups[s->lfe_ch] = 2;
304 s->channel_in_cpl[s->lfe_ch] = 0;
307 if (hdr.bitstream_id <= 10) {
309 s->snr_offset_strategy = 2;
310 s->block_switch_syntax = 1;
311 s->dither_flag_syntax = 1;
312 s->bit_allocation_syntax = 1;
313 s->fast_gain_syntax = 0;
314 s->first_cpl_leak = 0;
317 memset(s->channel_uses_aht, 0, sizeof(s->channel_uses_aht));
318 return ac3_parse_header(s);
321 return ff_eac3_parse_header(s);*/
322 return AC3_PARSE_ERROR_BSID;
327 * Set stereo downmixing coefficients based on frame header info.
328 * reference: Section 7.8.2 Downmixing Into Two Channels
330 static void set_downmix_coeffs(AC3DecodeContext *s)
333 float cmix = gain_levels[center_levels[s->center_mix_level]];
334 float smix = gain_levels[surround_levels[s->surround_mix_level]];
337 for(i=0; i<s->fbw_channels; i++) {
338 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
339 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
341 if(s->channel_mode > 1 && s->channel_mode & 1) {
342 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
344 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
345 int nf = s->channel_mode - 2;
346 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
348 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
349 int nf = s->channel_mode - 4;
350 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
355 for(i=0; i<s->fbw_channels; i++) {
356 norm0 += s->downmix_coeffs[i][0];
357 norm1 += s->downmix_coeffs[i][1];
359 norm0 = 1.0f / norm0;
360 norm1 = 1.0f / norm1;
361 for(i=0; i<s->fbw_channels; i++) {
362 s->downmix_coeffs[i][0] *= norm0;
363 s->downmix_coeffs[i][1] *= norm1;
366 if(s->output_mode == AC3_CHMODE_MONO) {
367 for(i=0; i<s->fbw_channels; i++)
368 s->downmix_coeffs[i][0] = (s->downmix_coeffs[i][0] + s->downmix_coeffs[i][1]) * LEVEL_MINUS_3DB;
373 * Decode the grouped exponents according to exponent strategy.
374 * reference: Section 7.1.3 Exponent Decoding
376 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
377 uint8_t absexp, int8_t *dexps)
379 int i, j, grp, group_size;
384 group_size = exp_strategy + (exp_strategy == EXP_D45);
385 for(grp=0,i=0; grp<ngrps; grp++) {
386 expacc = get_bits(gbc, 7);
387 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][0];
388 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][1];
389 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][2];
392 /* convert to absolute exps and expand groups */
394 for(i=0; i<ngrps*3; i++) {
395 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
396 for(j=0; j<group_size; j++) {
397 dexps[(i*group_size)+j] = prevexp;
403 * Generate transform coefficients for each coupled channel in the coupling
404 * range using the coupling coefficients and coupling coordinates.
405 * reference: Section 7.4.3 Coupling Coordinate Format
407 static void calc_transform_coeffs_cpl(AC3DecodeContext *s)
409 int i, j, ch, bnd, subbnd;
412 i = s->start_freq[CPL_CH];
413 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
416 for(j=0; j<12; j++) {
417 for(ch=1; ch<=s->fbw_channels; ch++) {
418 if(s->channel_in_cpl[ch]) {
419 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
420 if (ch == 2 && s->phase_flags[bnd])
421 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
426 } while(s->cpl_band_struct[subbnd]);
431 * Grouped mantissas for 3-level 5-level and 11-level quantization
443 * Decode the transform coefficients for a particular channel
444 * reference: Section 7.3 Quantization and Decoding of Mantissas
446 static void ac3_decode_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
448 GetBitContext *gbc = &s->gbc;
449 int i, gcode, tbap, start, end;
454 exps = s->dexps[ch_index];
455 bap = s->bap[ch_index];
456 coeffs = s->fixed_coeffs[ch_index];
457 start = s->start_freq[ch_index];
458 end = s->end_freq[ch_index];
460 for (i = start; i < end; i++) {
464 coeffs[i] = (av_lfg_get(&s->dith_state) & 0x7FFFFF) - 0x400000;
469 gcode = get_bits(gbc, 5);
470 m->b1_mant[0] = b1_mantissas[gcode][0];
471 m->b1_mant[1] = b1_mantissas[gcode][1];
472 m->b1_mant[2] = b1_mantissas[gcode][2];
475 coeffs[i] = m->b1_mant[m->b1ptr++];
480 gcode = get_bits(gbc, 7);
481 m->b2_mant[0] = b2_mantissas[gcode][0];
482 m->b2_mant[1] = b2_mantissas[gcode][1];
483 m->b2_mant[2] = b2_mantissas[gcode][2];
486 coeffs[i] = m->b2_mant[m->b2ptr++];
490 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
495 gcode = get_bits(gbc, 7);
496 m->b4_mant[0] = b4_mantissas[gcode][0];
497 m->b4_mant[1] = b4_mantissas[gcode][1];
500 coeffs[i] = m->b4_mant[m->b4ptr++];
504 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
508 /* asymmetric dequantization */
509 int qlevel = quantization_tab[tbap];
510 coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
514 coeffs[i] >>= exps[i];
519 * Remove random dithering from coefficients with zero-bit mantissas
520 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
522 static void remove_dithering(AC3DecodeContext *s) {
528 for(ch=1; ch<=s->fbw_channels; ch++) {
529 if(!s->dither_flag[ch]) {
530 coeffs = s->fixed_coeffs[ch];
532 if(s->channel_in_cpl[ch])
533 end = s->start_freq[CPL_CH];
535 end = s->end_freq[ch];
536 for(i=0; i<end; i++) {
540 if(s->channel_in_cpl[ch]) {
541 bap = s->bap[CPL_CH];
542 for(; i<s->end_freq[CPL_CH]; i++) {
551 static void decode_transform_coeffs_ch(AC3DecodeContext *s, int blk, int ch,
554 if (!s->channel_uses_aht[ch]) {
555 ac3_decode_transform_coeffs_ch(s, ch, m);
557 /* if AHT is used, mantissas for all blocks are encoded in the first
558 block of the frame. */
562 ff_eac3_decode_transform_coeffs_aht_ch(s, ch);
564 for (bin = s->start_freq[ch]; bin < s->end_freq[ch]; bin++) {
565 s->fixed_coeffs[ch][bin] = s->pre_mantissa[ch][bin][blk] >> s->dexps[ch][bin];
571 * Decode the transform coefficients.
573 static void decode_transform_coeffs(AC3DecodeContext *s, int blk)
579 m.b1ptr = m.b2ptr = m.b4ptr = 3;
581 for (ch = 1; ch <= s->channels; ch++) {
582 /* transform coefficients for full-bandwidth channel */
583 decode_transform_coeffs_ch(s, blk, ch, &m);
584 /* tranform coefficients for coupling channel come right after the
585 coefficients for the first coupled channel*/
586 if (s->channel_in_cpl[ch]) {
588 decode_transform_coeffs_ch(s, blk, CPL_CH, &m);
589 calc_transform_coeffs_cpl(s);
592 end = s->end_freq[CPL_CH];
594 end = s->end_freq[ch];
597 s->fixed_coeffs[ch][end] = 0;
601 /* if any channel doesn't use dithering, zero appropriate coefficients */
606 * Stereo rematrixing.
607 * reference: Section 7.5.4 Rematrixing : Decoding Technique
609 static void do_rematrixing(AC3DecodeContext *s)
615 end = FFMIN(s->end_freq[1], s->end_freq[2]);
617 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
618 if(s->rematrixing_flags[bnd]) {
619 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd+1]);
620 for(i=ff_ac3_rematrix_band_tab[bnd]; i<bndend; i++) {
621 tmp0 = s->fixed_coeffs[1][i];
622 tmp1 = s->fixed_coeffs[2][i];
623 s->fixed_coeffs[1][i] = tmp0 + tmp1;
624 s->fixed_coeffs[2][i] = tmp0 - tmp1;
631 * Inverse MDCT Transform.
632 * Convert frequency domain coefficients to time-domain audio samples.
633 * reference: Section 7.9.4 Transformation Equations
635 static inline void do_imdct(AC3DecodeContext *s, int channels)
638 float add_bias = s->add_bias;
639 if(s->out_channels==1 && channels>1)
640 add_bias *= LEVEL_MINUS_3DB; // compensate for the gain in downmix
642 for (ch=1; ch<=channels; ch++) {
643 if (s->block_switch[ch]) {
645 float *x = s->tmp_output+128;
647 x[i] = s->transform_coeffs[ch][2*i];
648 ff_imdct_half(&s->imdct_256, s->tmp_output, x);
649 s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, add_bias, 128);
651 x[i] = s->transform_coeffs[ch][2*i+1];
652 ff_imdct_half(&s->imdct_256, s->delay[ch-1], x);
654 ff_imdct_half(&s->imdct_512, s->tmp_output, s->transform_coeffs[ch]);
655 s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, add_bias, 128);
656 memcpy(s->delay[ch-1], s->tmp_output+128, 128*sizeof(float));
662 * Downmix the output to mono or stereo.
664 void ff_ac3_downmix_c(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len)
669 for(i=0; i<len; i++) {
671 for(j=0; j<in_ch; j++) {
672 v0 += samples[j][i] * matrix[j][0];
673 v1 += samples[j][i] * matrix[j][1];
678 } else if(out_ch == 1) {
679 for(i=0; i<len; i++) {
681 for(j=0; j<in_ch; j++)
682 v0 += samples[j][i] * matrix[j][0];
689 * Upmix delay samples from stereo to original channel layout.
691 static void ac3_upmix_delay(AC3DecodeContext *s)
693 int channel_data_size = sizeof(s->delay[0]);
694 switch(s->channel_mode) {
695 case AC3_CHMODE_DUALMONO:
696 case AC3_CHMODE_STEREO:
697 /* upmix mono to stereo */
698 memcpy(s->delay[1], s->delay[0], channel_data_size);
700 case AC3_CHMODE_2F2R:
701 memset(s->delay[3], 0, channel_data_size);
702 case AC3_CHMODE_2F1R:
703 memset(s->delay[2], 0, channel_data_size);
705 case AC3_CHMODE_3F2R:
706 memset(s->delay[4], 0, channel_data_size);
707 case AC3_CHMODE_3F1R:
708 memset(s->delay[3], 0, channel_data_size);
710 memcpy(s->delay[2], s->delay[1], channel_data_size);
711 memset(s->delay[1], 0, channel_data_size);
717 * Decode a single audio block from the AC-3 bitstream.
719 static int decode_audio_block(AC3DecodeContext *s, int blk)
721 int fbw_channels = s->fbw_channels;
722 int channel_mode = s->channel_mode;
724 int different_transforms;
727 GetBitContext *gbc = &s->gbc;
728 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
730 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
732 /* block switch flags */
733 different_transforms = 0;
734 if (s->block_switch_syntax) {
735 for (ch = 1; ch <= fbw_channels; ch++) {
736 s->block_switch[ch] = get_bits1(gbc);
737 if(ch > 1 && s->block_switch[ch] != s->block_switch[1])
738 different_transforms = 1;
742 /* dithering flags */
743 if (s->dither_flag_syntax) {
744 for (ch = 1; ch <= fbw_channels; ch++) {
745 s->dither_flag[ch] = get_bits1(gbc);
750 i = !(s->channel_mode);
753 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
754 s->avctx->drc_scale)+1.0;
755 } else if(blk == 0) {
756 s->dynamic_range[i] = 1.0f;
760 /* spectral extension strategy */
761 if (s->eac3 && (!blk || get_bits1(gbc))) {
762 if (get_bits1(gbc)) {
763 av_log_missing_feature(s->avctx, "Spectral extension", 1);
766 /* TODO: parse spectral extension strategy info */
769 /* TODO: spectral extension coordinates */
771 /* coupling strategy */
772 if (get_bits1(gbc)) {
773 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
775 s->cpl_in_use[blk] = get_bits1(gbc);
776 if (s->cpl_in_use[blk]) {
777 /* coupling in use */
778 int cpl_begin_freq, cpl_end_freq;
780 if (channel_mode < AC3_CHMODE_STEREO) {
781 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
785 /* check for enhanced coupling */
786 if (s->eac3 && get_bits1(gbc)) {
787 /* TODO: parse enhanced coupling strategy info */
788 av_log_missing_feature(s->avctx, "Enhanced coupling", 1);
792 /* determine which channels are coupled */
793 if (s->eac3 && s->channel_mode == AC3_CHMODE_STEREO) {
794 s->channel_in_cpl[1] = 1;
795 s->channel_in_cpl[2] = 1;
797 for (ch = 1; ch <= fbw_channels; ch++)
798 s->channel_in_cpl[ch] = get_bits1(gbc);
801 /* phase flags in use */
802 if (channel_mode == AC3_CHMODE_STEREO)
803 s->phase_flags_in_use = get_bits1(gbc);
805 /* coupling frequency range */
806 /* TODO: modify coupling end freq if spectral extension is used */
807 cpl_begin_freq = get_bits(gbc, 4);
808 cpl_end_freq = get_bits(gbc, 4);
809 if (3 + cpl_end_freq - cpl_begin_freq < 0) {
810 av_log(s->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
813 s->num_cpl_bands = s->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
814 s->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
815 s->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
817 /* coupling band structure */
818 if (!s->eac3 || get_bits1(gbc)) {
819 for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
820 s->cpl_band_struct[bnd] = get_bits1(gbc);
823 memcpy(s->cpl_band_struct,
824 &ff_eac3_default_cpl_band_struct[cpl_begin_freq+1],
825 s->num_cpl_subbands-1);
827 s->cpl_band_struct[s->num_cpl_subbands-1] = 0;
829 /* calculate number of coupling bands based on band structure */
830 for (bnd = 0; bnd < s->num_cpl_subbands-1; bnd++) {
831 s->num_cpl_bands -= s->cpl_band_struct[bnd];
834 /* coupling not in use */
835 for (ch = 1; ch <= fbw_channels; ch++) {
836 s->channel_in_cpl[ch] = 0;
837 s->first_cpl_coords[ch] = 1;
839 s->first_cpl_leak = s->eac3;
840 s->phase_flags_in_use = 0;
842 } else if (!s->eac3) {
844 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must be present in block 0\n");
847 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
850 cpl_in_use = s->cpl_in_use[blk];
852 /* coupling coordinates */
854 int cpl_coords_exist = 0;
856 for (ch = 1; ch <= fbw_channels; ch++) {
857 if (s->channel_in_cpl[ch]) {
858 if (get_bits1(gbc)) {
859 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
860 cpl_coords_exist = 1;
861 master_cpl_coord = 3 * get_bits(gbc, 2);
862 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
863 cpl_coord_exp = get_bits(gbc, 4);
864 cpl_coord_mant = get_bits(gbc, 4);
865 if (cpl_coord_exp == 15)
866 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
868 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
869 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
872 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must be present in block 0\n");
878 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
879 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
880 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
885 /* stereo rematrixing strategy and band structure */
886 if (channel_mode == AC3_CHMODE_STEREO) {
887 if ((s->eac3 && !blk) || get_bits1(gbc)) {
888 s->num_rematrixing_bands = 4;
889 if(cpl_in_use && s->start_freq[CPL_CH] <= 61)
890 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
891 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
892 s->rematrixing_flags[bnd] = get_bits1(gbc);
894 av_log(s->avctx, AV_LOG_ERROR, "new rematrixing strategy must be present in block 0\n");
899 /* exponent strategies for each channel */
900 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
902 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
903 if(s->exp_strategy[blk][ch] != EXP_REUSE)
904 bit_alloc_stages[ch] = 3;
907 /* channel bandwidth */
908 for (ch = 1; ch <= fbw_channels; ch++) {
909 s->start_freq[ch] = 0;
910 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
912 int prev = s->end_freq[ch];
913 if (s->channel_in_cpl[ch])
914 s->end_freq[ch] = s->start_freq[CPL_CH];
916 int bandwidth_code = get_bits(gbc, 6);
917 if (bandwidth_code > 60) {
918 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
921 s->end_freq[ch] = bandwidth_code * 3 + 73;
923 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
924 s->num_exp_groups[ch] = (s->end_freq[ch]+group_size-4) / group_size;
925 if(blk > 0 && s->end_freq[ch] != prev)
926 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
929 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
930 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
931 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
934 /* decode exponents for each channel */
935 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
936 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
937 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
938 decode_exponents(gbc, s->exp_strategy[blk][ch],
939 s->num_exp_groups[ch], s->dexps[ch][0],
940 &s->dexps[ch][s->start_freq[ch]+!!ch]);
941 if(ch != CPL_CH && ch != s->lfe_ch)
942 skip_bits(gbc, 2); /* skip gainrng */
946 /* bit allocation information */
947 if (s->bit_allocation_syntax) {
948 if (get_bits1(gbc)) {
949 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
950 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
951 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
952 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
953 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
954 for(ch=!cpl_in_use; ch<=s->channels; ch++)
955 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
957 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must be present in block 0\n");
962 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
963 if (get_bits1(gbc)) {
965 csnr = (get_bits(gbc, 6) - 15) << 4;
966 for (ch = !cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
967 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
968 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
970 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
972 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
976 /* fast gain (E-AC-3 only) */
977 if (s->fast_gain_syntax && get_bits1(gbc)) {
978 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
979 int prev = s->fast_gain[ch];
980 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
981 /* run last 2 bit allocation stages if fast gain changes */
982 if(blk && prev != s->fast_gain[ch])
983 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
985 } else if (s->eac3 && !blk) {
986 for (ch = !cpl_in_use; ch <= s->channels; ch++)
987 s->fast_gain[ch] = ff_ac3_fast_gain_tab[4];
990 /* E-AC-3 to AC-3 converter SNR offset */
991 if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT && get_bits1(gbc)) {
992 skip_bits(gbc, 10); // skip converter snr offset
995 /* coupling leak information */
997 if (get_bits1(gbc)) {
998 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
999 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
1000 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
1002 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must be present in block 0\n");
1007 /* delta bit allocation information */
1008 if (s->dba_syntax && get_bits1(gbc)) {
1009 /* delta bit allocation exists (strategy) */
1010 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1011 s->dba_mode[ch] = get_bits(gbc, 2);
1012 if (s->dba_mode[ch] == DBA_RESERVED) {
1013 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1016 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1018 /* channel delta offset, len and bit allocation */
1019 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1020 if (s->dba_mode[ch] == DBA_NEW) {
1021 s->dba_nsegs[ch] = get_bits(gbc, 3);
1022 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
1023 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1024 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1025 s->dba_values[ch][seg] = get_bits(gbc, 3);
1027 /* run last 2 bit allocation stages if new dba values */
1028 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1031 } else if(blk == 0) {
1032 for(ch=0; ch<=s->channels; ch++) {
1033 s->dba_mode[ch] = DBA_NONE;
1037 /* Bit allocation */
1038 for(ch=!cpl_in_use; ch<=s->channels; ch++) {
1039 if(bit_alloc_stages[ch] > 2) {
1040 /* Exponent mapping into PSD and PSD integration */
1041 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1042 s->start_freq[ch], s->end_freq[ch],
1043 s->psd[ch], s->band_psd[ch]);
1045 if(bit_alloc_stages[ch] > 1) {
1046 /* Compute excitation function, Compute masking curve, and
1047 Apply delta bit allocation */
1048 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1049 s->start_freq[ch], s->end_freq[ch],
1050 s->fast_gain[ch], (ch == s->lfe_ch),
1051 s->dba_mode[ch], s->dba_nsegs[ch],
1052 s->dba_offsets[ch], s->dba_lengths[ch],
1053 s->dba_values[ch], s->mask[ch]);
1055 if(bit_alloc_stages[ch] > 0) {
1056 /* Compute bit allocation */
1057 const uint8_t *bap_tab = s->channel_uses_aht[ch] ?
1058 ff_eac3_hebap_tab : ff_ac3_bap_tab;
1059 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1060 s->start_freq[ch], s->end_freq[ch],
1062 s->bit_alloc_params.floor,
1063 bap_tab, s->bap[ch]);
1067 /* unused dummy data */
1068 if (s->skip_syntax && get_bits1(gbc)) {
1069 int skipl = get_bits(gbc, 9);
1074 /* unpack the transform coefficients
1075 this also uncouples channels if coupling is in use. */
1076 decode_transform_coeffs(s, blk);
1078 /* TODO: generate enhanced coupling coordinates and uncouple */
1080 /* TODO: apply spectral extension */
1082 /* recover coefficients if rematrixing is in use */
1083 if(s->channel_mode == AC3_CHMODE_STEREO)
1086 /* apply scaling to coefficients (headroom, dynrng) */
1087 for(ch=1; ch<=s->channels; ch++) {
1088 float gain = s->mul_bias / 4194304.0f;
1089 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1090 gain *= s->dynamic_range[ch-1];
1092 gain *= s->dynamic_range[0];
1094 s->dsp.int32_to_float_fmul_scalar(s->transform_coeffs[ch], s->fixed_coeffs[ch], gain, 256);
1097 /* downmix and MDCT. order depends on whether block switching is used for
1098 any channel in this block. this is because coefficients for the long
1099 and short transforms cannot be mixed. */
1100 downmix_output = s->channels != s->out_channels &&
1101 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1102 s->fbw_channels == s->out_channels);
1103 if(different_transforms) {
1104 /* the delay samples have already been downmixed, so we upmix the delay
1105 samples in order to reconstruct all channels before downmixing. */
1111 do_imdct(s, s->channels);
1113 if(downmix_output) {
1114 s->dsp.ac3_downmix(s->output, s->downmix_coeffs, s->out_channels, s->fbw_channels, 256);
1117 if(downmix_output) {
1118 s->dsp.ac3_downmix(s->transform_coeffs+1, s->downmix_coeffs, s->out_channels, s->fbw_channels, 256);
1121 if(downmix_output && !s->downmixed) {
1123 s->dsp.ac3_downmix(s->delay, s->downmix_coeffs, s->out_channels, s->fbw_channels, 128);
1126 do_imdct(s, s->out_channels);
1133 * Decode a single AC-3 frame.
1135 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
1136 const uint8_t *buf, int buf_size)
1138 AC3DecodeContext *s = avctx->priv_data;
1139 int16_t *out_samples = (int16_t *)data;
1142 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1143 if (s->input_buffer) {
1144 /* copy input buffer to decoder context to avoid reading past the end
1145 of the buffer, which can be caused by a damaged input stream. */
1146 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE));
1147 init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
1149 init_get_bits(&s->gbc, buf, buf_size * 8);
1152 /* parse the syncinfo */
1154 err = parse_frame_header(s);
1156 /* check that reported frame size fits in input buffer */
1157 if(s->frame_size > buf_size) {
1158 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1159 err = AC3_PARSE_ERROR_FRAME_SIZE;
1162 /* check for crc mismatch */
1163 if(err != AC3_PARSE_ERROR_FRAME_SIZE && avctx->error_resilience >= FF_ER_CAREFUL) {
1164 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1165 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1166 err = AC3_PARSE_ERROR_CRC;
1170 if(err && err != AC3_PARSE_ERROR_CRC) {
1172 case AC3_PARSE_ERROR_SYNC:
1173 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1175 case AC3_PARSE_ERROR_BSID:
1176 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1178 case AC3_PARSE_ERROR_SAMPLE_RATE:
1179 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1181 case AC3_PARSE_ERROR_FRAME_SIZE:
1182 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1184 case AC3_PARSE_ERROR_FRAME_TYPE:
1185 /* skip frame if CRC is ok. otherwise use error concealment. */
1186 /* TODO: add support for substreams and dependent frames */
1187 if(s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1188 av_log(avctx, AV_LOG_ERROR, "unsupported frame type : skipping frame\n");
1189 return s->frame_size;
1191 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1195 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1200 /* if frame is ok, set audio parameters */
1202 avctx->sample_rate = s->sample_rate;
1203 avctx->bit_rate = s->bit_rate;
1205 /* channel config */
1206 s->out_channels = s->channels;
1207 s->output_mode = s->channel_mode;
1209 s->output_mode |= AC3_OUTPUT_LFEON;
1210 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1211 avctx->request_channels < s->channels) {
1212 s->out_channels = avctx->request_channels;
1213 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1215 avctx->channels = s->out_channels;
1217 /* set downmixing coefficients if needed */
1218 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1219 s->fbw_channels == s->out_channels)) {
1220 set_downmix_coeffs(s);
1222 } else if (!s->out_channels) {
1223 s->out_channels = avctx->channels;
1224 if(s->out_channels < s->channels)
1225 s->output_mode = s->out_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1228 /* decode the audio blocks */
1229 for (blk = 0; blk < s->num_blocks; blk++) {
1230 const float *output[s->out_channels];
1231 if (!err && decode_audio_block(s, blk)) {
1232 av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n");
1234 for (ch = 0; ch < s->out_channels; ch++)
1235 output[ch] = s->output[ch];
1236 s->dsp.float_to_int16_interleave(out_samples, output, 256, s->out_channels);
1237 out_samples += 256 * s->out_channels;
1239 *data_size = s->num_blocks * 256 * avctx->channels * sizeof (int16_t);
1240 return s->frame_size;
1244 * Uninitialize the AC-3 decoder.
1246 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1248 AC3DecodeContext *s = avctx->priv_data;
1249 ff_mdct_end(&s->imdct_512);
1250 ff_mdct_end(&s->imdct_256);
1252 av_freep(&s->input_buffer);
1257 AVCodec ac3_decoder = {
1259 .type = CODEC_TYPE_AUDIO,
1261 .priv_data_size = sizeof (AC3DecodeContext),
1262 .init = ac3_decode_init,
1263 .close = ac3_decode_end,
1264 .decode = ac3_decode_frame,
1265 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52 (AC-3, E-AC-3)"),