3 * This code is developed as part of Google Summer of Code 2006 Program.
5 * Copyright (c) 2006 Kartikey Mahendra BHATT (bhattkm at gmail dot com).
6 * Copyright (c) 2007 Justin Ruggles <justin.ruggles@gmail.com>
8 * Portions of this code are derived from liba52
9 * http://liba52.sourceforge.net
10 * Copyright (C) 2000-2003 Michel Lespinasse <walken@zoy.org>
11 * Copyright (C) 1999-2000 Aaron Holtzman <aholtzma@ess.engr.uvic.ca>
13 * This file is part of FFmpeg.
15 * FFmpeg is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public
17 * License as published by the Free Software Foundation; either
18 * version 2 of the License, or (at your option) any later version.
20 * FFmpeg is distributed in the hope that it will be useful,
21 * but WITHOUT ANY WARRANTY; without even the implied warranty of
22 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
23 * General Public License for more details.
25 * You should have received a copy of the GNU General Public
26 * License along with FFmpeg; if not, write to the Free Software
27 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
35 #include "libavutil/crc.h"
36 #include "ac3_parser.h"
38 #include "ac3dec_data.h"
40 /** Maximum possible frame size when the specification limit is ignored */
41 #define AC3_MAX_FRAME_SIZE 21695
44 * table for ungrouping 3 values in 7 bits.
45 * used for exponents and bap=2 mantissas
47 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];
69 /** Adjustments in dB gain */
70 #define LEVEL_PLUS_3DB 1.4142135623730950
71 #define LEVEL_PLUS_1POINT5DB 1.1892071150027209
72 #define LEVEL_MINUS_1POINT5DB 0.8408964152537145
73 #define LEVEL_MINUS_3DB 0.7071067811865476
74 #define LEVEL_MINUS_4POINT5DB 0.5946035575013605
75 #define LEVEL_MINUS_6DB 0.5000000000000000
76 #define LEVEL_MINUS_9DB 0.3535533905932738
77 #define LEVEL_ZERO 0.0000000000000000
78 #define LEVEL_ONE 1.0000000000000000
80 static const float gain_levels[9] = {
84 LEVEL_MINUS_1POINT5DB,
86 LEVEL_MINUS_4POINT5DB,
93 * Table for center mix levels
94 * reference: Section 5.4.2.4 cmixlev
96 static const uint8_t center_levels[4] = { 4, 5, 6, 5 };
99 * Table for surround mix levels
100 * reference: Section 5.4.2.5 surmixlev
102 static const uint8_t surround_levels[4] = { 4, 6, 7, 6 };
105 * Table for default stereo downmixing coefficients
106 * reference: Section 7.8.2 Downmixing Into Two Channels
108 static const uint8_t ac3_default_coeffs[8][5][2] = {
109 { { 2, 7 }, { 7, 2 }, },
111 { { 2, 7 }, { 7, 2 }, },
112 { { 2, 7 }, { 5, 5 }, { 7, 2 }, },
113 { { 2, 7 }, { 7, 2 }, { 6, 6 }, },
114 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 8, 8 }, },
115 { { 2, 7 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
116 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
120 * Symmetrical Dequantization
121 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
122 * Tables 7.19 to 7.23
125 symmetric_dequant(int code, int levels)
127 return ((code - (levels >> 1)) << 24) / levels;
131 * Initialize tables at runtime.
133 static av_cold void ac3_tables_init(void)
137 /* generate table for ungrouping 3 values in 7 bits
138 reference: Section 7.1.3 Exponent Decoding */
139 for(i=0; i<128; i++) {
140 ungroup_3_in_7_bits_tab[i][0] = i / 25;
141 ungroup_3_in_7_bits_tab[i][1] = (i % 25) / 5;
142 ungroup_3_in_7_bits_tab[i][2] = (i % 25) % 5;
145 /* generate grouped mantissa tables
146 reference: Section 7.3.5 Ungrouping of Mantissas */
147 for(i=0; i<32; i++) {
148 /* bap=1 mantissas */
149 b1_mantissas[i][0] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][0], 3);
150 b1_mantissas[i][1] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][1], 3);
151 b1_mantissas[i][2] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][2], 3);
153 for(i=0; i<128; i++) {
154 /* bap=2 mantissas */
155 b2_mantissas[i][0] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][0], 5);
156 b2_mantissas[i][1] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][1], 5);
157 b2_mantissas[i][2] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][2], 5);
159 /* bap=4 mantissas */
160 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
161 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
163 /* generate ungrouped mantissa tables
164 reference: Tables 7.21 and 7.23 */
166 /* bap=3 mantissas */
167 b3_mantissas[i] = symmetric_dequant(i, 7);
169 for(i=0; i<15; i++) {
170 /* bap=5 mantissas */
171 b5_mantissas[i] = symmetric_dequant(i, 15);
174 /* generate dynamic range table
175 reference: Section 7.7.1 Dynamic Range Control */
176 for(i=0; i<256; i++) {
177 int v = (i >> 5) - ((i >> 7) << 3) - 5;
178 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
184 * AVCodec initialization
186 static av_cold int ac3_decode_init(AVCodecContext *avctx)
188 AC3DecodeContext *s = avctx->priv_data;
193 ff_mdct_init(&s->imdct_256, 8, 1);
194 ff_mdct_init(&s->imdct_512, 9, 1);
195 ff_kbd_window_init(s->window, 5.0, 256);
196 dsputil_init(&s->dsp, avctx);
197 av_lfg_init(&s->dith_state, 0);
199 /* set bias values for float to int16 conversion */
200 if(s->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
201 s->add_bias = 385.0f;
205 s->mul_bias = 32767.0f;
208 /* allow downmixing to stereo or mono */
209 if (avctx->channels > 0 && avctx->request_channels > 0 &&
210 avctx->request_channels < avctx->channels &&
211 avctx->request_channels <= 2) {
212 avctx->channels = avctx->request_channels;
216 /* allocate context input buffer */
217 if (avctx->error_resilience >= FF_ER_CAREFUL) {
218 s->input_buffer = av_mallocz(AC3_MAX_FRAME_SIZE + FF_INPUT_BUFFER_PADDING_SIZE);
219 if (!s->input_buffer)
220 return AVERROR_NOMEM;
223 avctx->sample_fmt = SAMPLE_FMT_S16;
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 skip_bits(gbc, 5); // skip dialog normalization
242 skip_bits(gbc, 8); //skip compression
244 skip_bits(gbc, 8); //skip language code
246 skip_bits(gbc, 7); //skip audio production information
249 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
251 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
252 TODO: read & use the xbsi1 downmix levels */
254 skip_bits(gbc, 14); //skip timecode1 / xbsi1
256 skip_bits(gbc, 14); //skip timecode2 / xbsi2
258 /* skip additional bitstream info */
259 if (get_bits1(gbc)) {
260 i = get_bits(gbc, 6);
270 * Common function to parse AC-3 or E-AC-3 frame header
272 static int parse_frame_header(AC3DecodeContext *s)
277 err = ff_ac3_parse_header(&s->gbc, &hdr);
281 /* get decoding parameters from header info */
282 s->bit_alloc_params.sr_code = hdr.sr_code;
283 s->channel_mode = hdr.channel_mode;
284 s->lfe_on = hdr.lfe_on;
285 s->bit_alloc_params.sr_shift = hdr.sr_shift;
286 s->sample_rate = hdr.sample_rate;
287 s->bit_rate = hdr.bit_rate;
288 s->channels = hdr.channels;
289 s->fbw_channels = s->channels - s->lfe_on;
290 s->lfe_ch = s->fbw_channels + 1;
291 s->frame_size = hdr.frame_size;
292 s->center_mix_level = hdr.center_mix_level;
293 s->surround_mix_level = hdr.surround_mix_level;
294 s->num_blocks = hdr.num_blocks;
295 s->frame_type = hdr.frame_type;
296 s->substreamid = hdr.substreamid;
299 s->start_freq[s->lfe_ch] = 0;
300 s->end_freq[s->lfe_ch] = 7;
301 s->num_exp_groups[s->lfe_ch] = 2;
302 s->channel_in_cpl[s->lfe_ch] = 0;
305 if(hdr.bitstream_id > 10)
306 return AC3_PARSE_ERROR_BSID;
308 return ac3_parse_header(s);
312 * Set stereo downmixing coefficients based on frame header info.
313 * reference: Section 7.8.2 Downmixing Into Two Channels
315 static void set_downmix_coeffs(AC3DecodeContext *s)
318 float cmix = gain_levels[center_levels[s->center_mix_level]];
319 float smix = gain_levels[surround_levels[s->surround_mix_level]];
322 for(i=0; i<s->fbw_channels; i++) {
323 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
324 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
326 if(s->channel_mode > 1 && s->channel_mode & 1) {
327 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
329 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
330 int nf = s->channel_mode - 2;
331 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
333 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
334 int nf = s->channel_mode - 4;
335 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
340 for(i=0; i<s->fbw_channels; i++) {
341 norm0 += s->downmix_coeffs[i][0];
342 norm1 += s->downmix_coeffs[i][1];
344 norm0 = 1.0f / norm0;
345 norm1 = 1.0f / norm1;
346 for(i=0; i<s->fbw_channels; i++) {
347 s->downmix_coeffs[i][0] *= norm0;
348 s->downmix_coeffs[i][1] *= norm1;
351 if(s->output_mode == AC3_CHMODE_MONO) {
352 for(i=0; i<s->fbw_channels; i++)
353 s->downmix_coeffs[i][0] = (s->downmix_coeffs[i][0] + s->downmix_coeffs[i][1]) * LEVEL_MINUS_3DB;
358 * Decode the grouped exponents according to exponent strategy.
359 * reference: Section 7.1.3 Exponent Decoding
361 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
362 uint8_t absexp, int8_t *dexps)
364 int i, j, grp, group_size;
369 group_size = exp_strategy + (exp_strategy == EXP_D45);
370 for(grp=0,i=0; grp<ngrps; grp++) {
371 expacc = get_bits(gbc, 7);
372 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][0];
373 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][1];
374 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][2];
377 /* convert to absolute exps and expand groups */
379 for(i=0; i<ngrps*3; i++) {
380 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
381 for(j=0; j<group_size; j++) {
382 dexps[(i*group_size)+j] = prevexp;
388 * Generate transform coefficients for each coupled channel in the coupling
389 * range using the coupling coefficients and coupling coordinates.
390 * reference: Section 7.4.3 Coupling Coordinate Format
392 static void calc_transform_coeffs_cpl(AC3DecodeContext *s)
394 int i, j, ch, bnd, subbnd;
397 i = s->start_freq[CPL_CH];
398 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
401 for(j=0; j<12; j++) {
402 for(ch=1; ch<=s->fbw_channels; ch++) {
403 if(s->channel_in_cpl[ch]) {
404 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
405 if (ch == 2 && s->phase_flags[bnd])
406 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
411 } while(s->cpl_band_struct[subbnd]);
416 * Grouped mantissas for 3-level 5-level and 11-level quantization
428 * Get the transform coefficients for a particular channel
429 * reference: Section 7.3 Quantization and Decoding of Mantissas
431 static void get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
433 GetBitContext *gbc = &s->gbc;
434 int i, gcode, tbap, start, end;
439 exps = s->dexps[ch_index];
440 bap = s->bap[ch_index];
441 coeffs = s->fixed_coeffs[ch_index];
442 start = s->start_freq[ch_index];
443 end = s->end_freq[ch_index];
445 for (i = start; i < end; i++) {
449 coeffs[i] = (av_lfg_get(&s->dith_state) & 0x7FFFFF) - 0x400000;
454 gcode = get_bits(gbc, 5);
455 m->b1_mant[0] = b1_mantissas[gcode][0];
456 m->b1_mant[1] = b1_mantissas[gcode][1];
457 m->b1_mant[2] = b1_mantissas[gcode][2];
460 coeffs[i] = m->b1_mant[m->b1ptr++];
465 gcode = get_bits(gbc, 7);
466 m->b2_mant[0] = b2_mantissas[gcode][0];
467 m->b2_mant[1] = b2_mantissas[gcode][1];
468 m->b2_mant[2] = b2_mantissas[gcode][2];
471 coeffs[i] = m->b2_mant[m->b2ptr++];
475 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
480 gcode = get_bits(gbc, 7);
481 m->b4_mant[0] = b4_mantissas[gcode][0];
482 m->b4_mant[1] = b4_mantissas[gcode][1];
485 coeffs[i] = m->b4_mant[m->b4ptr++];
489 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
493 /* asymmetric dequantization */
494 int qlevel = quantization_tab[tbap];
495 coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
499 coeffs[i] >>= exps[i];
504 * Remove random dithering from coefficients with zero-bit mantissas
505 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
507 static void remove_dithering(AC3DecodeContext *s) {
513 for(ch=1; ch<=s->fbw_channels; ch++) {
514 if(!s->dither_flag[ch]) {
515 coeffs = s->fixed_coeffs[ch];
517 if(s->channel_in_cpl[ch])
518 end = s->start_freq[CPL_CH];
520 end = s->end_freq[ch];
521 for(i=0; i<end; i++) {
525 if(s->channel_in_cpl[ch]) {
526 bap = s->bap[CPL_CH];
527 for(; i<s->end_freq[CPL_CH]; i++) {
537 * Get the transform coefficients.
539 static void get_transform_coeffs(AC3DecodeContext *s)
545 m.b1ptr = m.b2ptr = m.b4ptr = 3;
547 for (ch = 1; ch <= s->channels; ch++) {
548 /* transform coefficients for full-bandwidth channel */
549 get_transform_coeffs_ch(s, ch, &m);
550 /* tranform coefficients for coupling channel come right after the
551 coefficients for the first coupled channel*/
552 if (s->channel_in_cpl[ch]) {
554 get_transform_coeffs_ch(s, CPL_CH, &m);
555 calc_transform_coeffs_cpl(s);
558 end = s->end_freq[CPL_CH];
560 end = s->end_freq[ch];
563 s->fixed_coeffs[ch][end] = 0;
567 /* if any channel doesn't use dithering, zero appropriate coefficients */
573 * Stereo rematrixing.
574 * reference: Section 7.5.4 Rematrixing : Decoding Technique
576 static void do_rematrixing(AC3DecodeContext *s)
582 end = FFMIN(s->end_freq[1], s->end_freq[2]);
584 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
585 if(s->rematrixing_flags[bnd]) {
586 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd+1]);
587 for(i=ff_ac3_rematrix_band_tab[bnd]; i<bndend; i++) {
588 tmp0 = s->fixed_coeffs[1][i];
589 tmp1 = s->fixed_coeffs[2][i];
590 s->fixed_coeffs[1][i] = tmp0 + tmp1;
591 s->fixed_coeffs[2][i] = tmp0 - tmp1;
598 * Inverse MDCT Transform.
599 * Convert frequency domain coefficients to time-domain audio samples.
600 * reference: Section 7.9.4 Transformation Equations
602 static inline void do_imdct(AC3DecodeContext *s, int channels)
605 float add_bias = s->add_bias;
606 if(s->out_channels==1 && channels>1)
607 add_bias *= LEVEL_MINUS_3DB; // compensate for the gain in downmix
609 for (ch=1; ch<=channels; ch++) {
610 if (s->block_switch[ch]) {
612 float *x = s->tmp_output+128;
614 x[i] = s->transform_coeffs[ch][2*i];
615 ff_imdct_half(&s->imdct_256, s->tmp_output, x);
616 s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, add_bias, 128);
618 x[i] = s->transform_coeffs[ch][2*i+1];
619 ff_imdct_half(&s->imdct_256, s->delay[ch-1], x);
621 ff_imdct_half(&s->imdct_512, s->tmp_output, s->transform_coeffs[ch]);
622 s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, add_bias, 128);
623 memcpy(s->delay[ch-1], s->tmp_output+128, 128*sizeof(float));
629 * Downmix the output to mono or stereo.
631 void ff_ac3_downmix_c(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len)
636 for(i=0; i<len; i++) {
638 for(j=0; j<in_ch; j++) {
639 v0 += samples[j][i] * matrix[j][0];
640 v1 += samples[j][i] * matrix[j][1];
645 } else if(out_ch == 1) {
646 for(i=0; i<len; i++) {
648 for(j=0; j<in_ch; j++)
649 v0 += samples[j][i] * matrix[j][0];
656 * Upmix delay samples from stereo to original channel layout.
658 static void ac3_upmix_delay(AC3DecodeContext *s)
660 int channel_data_size = 128*sizeof(float);
661 switch(s->channel_mode) {
662 case AC3_CHMODE_DUALMONO:
663 case AC3_CHMODE_STEREO:
664 /* upmix mono to stereo */
665 memcpy(s->delay[1], s->delay[0], channel_data_size);
667 case AC3_CHMODE_2F2R:
668 memset(s->delay[3], 0, channel_data_size);
669 case AC3_CHMODE_2F1R:
670 memset(s->delay[2], 0, channel_data_size);
672 case AC3_CHMODE_3F2R:
673 memset(s->delay[4], 0, channel_data_size);
674 case AC3_CHMODE_3F1R:
675 memset(s->delay[3], 0, channel_data_size);
677 memcpy(s->delay[2], s->delay[1], channel_data_size);
678 memset(s->delay[1], 0, channel_data_size);
684 * Decode a single audio block from the AC-3 bitstream.
686 static int decode_audio_block(AC3DecodeContext *s, int blk)
688 int fbw_channels = s->fbw_channels;
689 int channel_mode = s->channel_mode;
691 int different_transforms;
694 GetBitContext *gbc = &s->gbc;
695 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
697 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
699 /* block switch flags */
700 different_transforms = 0;
701 for (ch = 1; ch <= fbw_channels; ch++) {
702 s->block_switch[ch] = get_bits1(gbc);
703 if(ch > 1 && s->block_switch[ch] != s->block_switch[1])
704 different_transforms = 1;
707 /* dithering flags */
709 for (ch = 1; ch <= fbw_channels; ch++) {
710 s->dither_flag[ch] = get_bits1(gbc);
711 if(!s->dither_flag[ch])
716 i = !(s->channel_mode);
719 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
720 s->avctx->drc_scale)+1.0;
721 } else if(blk == 0) {
722 s->dynamic_range[i] = 1.0f;
726 /* coupling strategy */
727 if (get_bits1(gbc)) {
728 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
729 s->cpl_in_use[blk] = get_bits1(gbc);
730 if (s->cpl_in_use[blk]) {
731 /* coupling in use */
732 int cpl_begin_freq, cpl_end_freq;
734 if (channel_mode < AC3_CHMODE_STEREO) {
735 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
739 /* determine which channels are coupled */
740 for (ch = 1; ch <= fbw_channels; ch++)
741 s->channel_in_cpl[ch] = get_bits1(gbc);
743 /* phase flags in use */
744 if (channel_mode == AC3_CHMODE_STEREO)
745 s->phase_flags_in_use = get_bits1(gbc);
747 /* coupling frequency range and band structure */
748 cpl_begin_freq = get_bits(gbc, 4);
749 cpl_end_freq = get_bits(gbc, 4);
750 if (3 + cpl_end_freq - cpl_begin_freq < 0) {
751 av_log(s->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
754 s->num_cpl_bands = s->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
755 s->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
756 s->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
757 for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
758 if (get_bits1(gbc)) {
759 s->cpl_band_struct[bnd] = 1;
763 s->cpl_band_struct[s->num_cpl_subbands-1] = 0;
765 /* coupling not in use */
766 for (ch = 1; ch <= fbw_channels; ch++)
767 s->channel_in_cpl[ch] = 0;
770 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must be present in block 0\n");
773 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
775 cpl_in_use = s->cpl_in_use[blk];
777 /* coupling coordinates */
779 int cpl_coords_exist = 0;
781 for (ch = 1; ch <= fbw_channels; ch++) {
782 if (s->channel_in_cpl[ch]) {
783 if (get_bits1(gbc)) {
784 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
785 cpl_coords_exist = 1;
786 master_cpl_coord = 3 * get_bits(gbc, 2);
787 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
788 cpl_coord_exp = get_bits(gbc, 4);
789 cpl_coord_mant = get_bits(gbc, 4);
790 if (cpl_coord_exp == 15)
791 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
793 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
794 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
797 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must be present in block 0\n");
803 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
804 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
805 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
810 /* stereo rematrixing strategy and band structure */
811 if (channel_mode == AC3_CHMODE_STEREO) {
812 if (get_bits1(gbc)) {
813 s->num_rematrixing_bands = 4;
814 if(cpl_in_use && s->start_freq[CPL_CH] <= 61)
815 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
816 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
817 s->rematrixing_flags[bnd] = get_bits1(gbc);
819 av_log(s->avctx, AV_LOG_ERROR, "new rematrixing strategy must be present in block 0\n");
824 /* exponent strategies for each channel */
825 s->exp_strategy[blk][CPL_CH] = EXP_REUSE;
826 s->exp_strategy[blk][s->lfe_ch] = EXP_REUSE;
827 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
828 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
829 if(s->exp_strategy[blk][ch] != EXP_REUSE)
830 bit_alloc_stages[ch] = 3;
833 /* channel bandwidth */
834 for (ch = 1; ch <= fbw_channels; ch++) {
835 s->start_freq[ch] = 0;
836 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
838 int prev = s->end_freq[ch];
839 if (s->channel_in_cpl[ch])
840 s->end_freq[ch] = s->start_freq[CPL_CH];
842 int bandwidth_code = get_bits(gbc, 6);
843 if (bandwidth_code > 60) {
844 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
847 s->end_freq[ch] = bandwidth_code * 3 + 73;
849 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
850 s->num_exp_groups[ch] = (s->end_freq[ch]+group_size-4) / group_size;
851 if(blk > 0 && s->end_freq[ch] != prev)
852 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
855 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
856 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
857 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
860 /* decode exponents for each channel */
861 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
862 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
863 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
864 decode_exponents(gbc, s->exp_strategy[blk][ch],
865 s->num_exp_groups[ch], s->dexps[ch][0],
866 &s->dexps[ch][s->start_freq[ch]+!!ch]);
867 if(ch != CPL_CH && ch != s->lfe_ch)
868 skip_bits(gbc, 2); /* skip gainrng */
872 /* bit allocation information */
873 if (get_bits1(gbc)) {
874 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
875 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
876 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
877 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
878 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
879 for(ch=!cpl_in_use; ch<=s->channels; ch++)
880 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
882 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must be present in block 0\n");
886 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
887 if (get_bits1(gbc)) {
889 csnr = (get_bits(gbc, 6) - 15) << 4;
890 for (ch = !cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
891 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
892 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
894 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
896 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
900 /* coupling leak information */
902 if (get_bits1(gbc)) {
903 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
904 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
905 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
907 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must be present in block 0\n");
912 /* delta bit allocation information */
913 if (get_bits1(gbc)) {
914 /* delta bit allocation exists (strategy) */
915 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
916 s->dba_mode[ch] = get_bits(gbc, 2);
917 if (s->dba_mode[ch] == DBA_RESERVED) {
918 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
921 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
923 /* channel delta offset, len and bit allocation */
924 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
925 if (s->dba_mode[ch] == DBA_NEW) {
926 s->dba_nsegs[ch] = get_bits(gbc, 3);
927 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
928 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
929 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
930 s->dba_values[ch][seg] = get_bits(gbc, 3);
932 /* run last 2 bit allocation stages if new dba values */
933 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
936 } else if(blk == 0) {
937 for(ch=0; ch<=s->channels; ch++) {
938 s->dba_mode[ch] = DBA_NONE;
943 for(ch=!cpl_in_use; ch<=s->channels; ch++) {
944 if(bit_alloc_stages[ch] > 2) {
945 /* Exponent mapping into PSD and PSD integration */
946 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
947 s->start_freq[ch], s->end_freq[ch],
948 s->psd[ch], s->band_psd[ch]);
950 if(bit_alloc_stages[ch] > 1) {
951 /* Compute excitation function, Compute masking curve, and
952 Apply delta bit allocation */
953 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
954 s->start_freq[ch], s->end_freq[ch],
955 s->fast_gain[ch], (ch == s->lfe_ch),
956 s->dba_mode[ch], s->dba_nsegs[ch],
957 s->dba_offsets[ch], s->dba_lengths[ch],
958 s->dba_values[ch], s->mask[ch]);
960 if(bit_alloc_stages[ch] > 0) {
961 /* Compute bit allocation */
962 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
963 s->start_freq[ch], s->end_freq[ch],
965 s->bit_alloc_params.floor,
966 ff_ac3_bap_tab, s->bap[ch]);
970 /* unused dummy data */
971 if (get_bits1(gbc)) {
972 int skipl = get_bits(gbc, 9);
977 /* unpack the transform coefficients
978 this also uncouples channels if coupling is in use. */
979 get_transform_coeffs(s);
981 /* recover coefficients if rematrixing is in use */
982 if(s->channel_mode == AC3_CHMODE_STEREO)
985 /* apply scaling to coefficients (headroom, dynrng) */
986 for(ch=1; ch<=s->channels; ch++) {
987 float gain = s->mul_bias / 4194304.0f;
988 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
989 gain *= s->dynamic_range[ch-1];
991 gain *= s->dynamic_range[0];
993 s->dsp.int32_to_float_fmul_scalar(s->transform_coeffs[ch], s->fixed_coeffs[ch], gain, 256);
996 /* downmix and MDCT. order depends on whether block switching is used for
997 any channel in this block. this is because coefficients for the long
998 and short transforms cannot be mixed. */
999 downmix_output = s->channels != s->out_channels &&
1000 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1001 s->fbw_channels == s->out_channels);
1002 if(different_transforms) {
1003 /* the delay samples have already been downmixed, so we upmix the delay
1004 samples in order to reconstruct all channels before downmixing. */
1010 do_imdct(s, s->channels);
1012 if(downmix_output) {
1013 s->dsp.ac3_downmix(s->output, s->downmix_coeffs, s->out_channels, s->fbw_channels, 256);
1016 if(downmix_output) {
1017 s->dsp.ac3_downmix(s->transform_coeffs+1, s->downmix_coeffs, s->out_channels, s->fbw_channels, 256);
1020 if(downmix_output && !s->downmixed) {
1022 s->dsp.ac3_downmix(s->delay, s->downmix_coeffs, s->out_channels, s->fbw_channels, 128);
1025 do_imdct(s, s->out_channels);
1032 * Decode a single AC-3 frame.
1034 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
1035 const uint8_t *buf, int buf_size)
1037 AC3DecodeContext *s = avctx->priv_data;
1038 int16_t *out_samples = (int16_t *)data;
1041 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1042 if (s->input_buffer) {
1043 /* copy input buffer to decoder context to avoid reading past the end
1044 of the buffer, which can be caused by a damaged input stream. */
1045 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_MAX_FRAME_SIZE));
1046 init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
1048 init_get_bits(&s->gbc, buf, buf_size * 8);
1051 /* parse the syncinfo */
1053 err = parse_frame_header(s);
1055 /* check that reported frame size fits in input buffer */
1056 if(s->frame_size > buf_size) {
1057 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1058 err = AC3_PARSE_ERROR_FRAME_SIZE;
1061 /* check for crc mismatch */
1062 if(err != AC3_PARSE_ERROR_FRAME_SIZE && avctx->error_resilience >= FF_ER_CAREFUL) {
1063 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1064 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1065 err = AC3_PARSE_ERROR_CRC;
1069 if(err && err != AC3_PARSE_ERROR_CRC) {
1071 case AC3_PARSE_ERROR_SYNC:
1072 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1074 case AC3_PARSE_ERROR_BSID:
1075 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1077 case AC3_PARSE_ERROR_SAMPLE_RATE:
1078 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1080 case AC3_PARSE_ERROR_FRAME_SIZE:
1081 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1083 case AC3_PARSE_ERROR_FRAME_TYPE:
1084 /* skip frame if CRC is ok. otherwise use error concealment. */
1085 /* TODO: add support for substreams and dependent frames */
1086 if(s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1087 av_log(avctx, AV_LOG_ERROR, "unsupported frame type : skipping frame\n");
1088 return s->frame_size;
1090 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1094 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1099 /* if frame is ok, set audio parameters */
1101 avctx->sample_rate = s->sample_rate;
1102 avctx->bit_rate = s->bit_rate;
1104 /* channel config */
1105 s->out_channels = s->channels;
1106 s->output_mode = s->channel_mode;
1108 s->output_mode |= AC3_OUTPUT_LFEON;
1109 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1110 avctx->request_channels < s->channels) {
1111 s->out_channels = avctx->request_channels;
1112 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1114 avctx->channels = s->out_channels;
1116 /* set downmixing coefficients if needed */
1117 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1118 s->fbw_channels == s->out_channels)) {
1119 set_downmix_coeffs(s);
1121 } else if (!s->out_channels) {
1122 s->out_channels = avctx->channels;
1123 if(s->out_channels < s->channels)
1124 s->output_mode = s->out_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1127 /* decode the audio blocks */
1128 for (blk = 0; blk < s->num_blocks; blk++) {
1129 const float *output[s->out_channels];
1130 if (!err && decode_audio_block(s, blk)) {
1131 av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n");
1133 for (ch = 0; ch < s->out_channels; ch++)
1134 output[ch] = s->output[ch];
1135 s->dsp.float_to_int16_interleave(out_samples, output, 256, s->out_channels);
1136 out_samples += 256 * s->out_channels;
1138 *data_size = s->num_blocks * 256 * avctx->channels * sizeof (int16_t);
1139 return s->frame_size;
1143 * Uninitialize the AC-3 decoder.
1145 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1147 AC3DecodeContext *s = avctx->priv_data;
1148 ff_mdct_end(&s->imdct_512);
1149 ff_mdct_end(&s->imdct_256);
1151 av_freep(&s->input_buffer);
1156 AVCodec ac3_decoder = {
1158 .type = CODEC_TYPE_AUDIO,
1160 .priv_data_size = sizeof (AC3DecodeContext),
1161 .init = ac3_decode_init,
1162 .close = ac3_decode_end,
1163 .decode = ac3_decode_frame,
1164 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52 / AC-3"),