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
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 "libavutil/random.h"
38 #include "ac3_parser.h"
39 #include "bitstream.h"
42 #include "ac3dec_data.h"
44 /** Maximum possible frame size when the specification limit is ignored */
45 #define AC3_MAX_FRAME_SIZE 21695
48 * table for ungrouping 3 values in 7 bits.
49 * used for exponents and bap=2 mantissas
51 static uint8_t ungroup_3_in_7_bits_tab[128][3];
54 /** tables for ungrouping mantissas */
55 static int b1_mantissas[32][3];
56 static int b2_mantissas[128][3];
57 static int b3_mantissas[8];
58 static int b4_mantissas[128][2];
59 static int b5_mantissas[16];
62 * Quantization table: levels for symmetric. bits for asymmetric.
63 * reference: Table 7.18 Mapping of bap to Quantizer
65 static const uint8_t quantization_tab[16] = {
67 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
70 /** dynamic range table. converts codes to scale factors. */
71 static float dynamic_range_tab[256];
73 /** Adjustments in dB gain */
74 #define LEVEL_PLUS_3DB 1.4142135623730950
75 #define LEVEL_PLUS_1POINT5DB 1.1892071150027209
76 #define LEVEL_MINUS_1POINT5DB 0.8408964152537145
77 #define LEVEL_MINUS_3DB 0.7071067811865476
78 #define LEVEL_MINUS_4POINT5DB 0.5946035575013605
79 #define LEVEL_MINUS_6DB 0.5000000000000000
80 #define LEVEL_MINUS_9DB 0.3535533905932738
81 #define LEVEL_ZERO 0.0000000000000000
82 #define LEVEL_ONE 1.0000000000000000
84 static const float gain_levels[9] = {
88 LEVEL_MINUS_1POINT5DB,
90 LEVEL_MINUS_4POINT5DB,
97 * Table for center mix levels
98 * reference: Section 5.4.2.4 cmixlev
100 static const uint8_t center_levels[4] = { 4, 5, 6, 5 };
103 * Table for surround mix levels
104 * reference: Section 5.4.2.5 surmixlev
106 static const uint8_t surround_levels[4] = { 4, 6, 7, 6 };
109 * Table for default stereo downmixing coefficients
110 * reference: Section 7.8.2 Downmixing Into Two Channels
112 static const uint8_t ac3_default_coeffs[8][5][2] = {
113 { { 2, 7 }, { 7, 2 }, },
115 { { 2, 7 }, { 7, 2 }, },
116 { { 2, 7 }, { 5, 5 }, { 7, 2 }, },
117 { { 2, 7 }, { 7, 2 }, { 6, 6 }, },
118 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 8, 8 }, },
119 { { 2, 7 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
120 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
124 * Symmetrical Dequantization
125 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
126 * Tables 7.19 to 7.23
129 symmetric_dequant(int code, int levels)
131 return ((code - (levels >> 1)) << 24) / levels;
135 * Initialize tables at runtime.
137 static av_cold void ac3_tables_init(void)
141 /* generate table for ungrouping 3 values in 7 bits
142 reference: Section 7.1.3 Exponent Decoding */
143 for(i=0; i<128; i++) {
144 ungroup_3_in_7_bits_tab[i][0] = i / 25;
145 ungroup_3_in_7_bits_tab[i][1] = (i % 25) / 5;
146 ungroup_3_in_7_bits_tab[i][2] = (i % 25) % 5;
149 /* generate grouped mantissa tables
150 reference: Section 7.3.5 Ungrouping of Mantissas */
151 for(i=0; i<32; i++) {
152 /* bap=1 mantissas */
153 b1_mantissas[i][0] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][0], 3);
154 b1_mantissas[i][1] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][1], 3);
155 b1_mantissas[i][2] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][2], 3);
157 for(i=0; i<128; i++) {
158 /* bap=2 mantissas */
159 b2_mantissas[i][0] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][0], 5);
160 b2_mantissas[i][1] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][1], 5);
161 b2_mantissas[i][2] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][2], 5);
163 /* bap=4 mantissas */
164 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
165 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
167 /* generate ungrouped mantissa tables
168 reference: Tables 7.21 and 7.23 */
170 /* bap=3 mantissas */
171 b3_mantissas[i] = symmetric_dequant(i, 7);
173 for(i=0; i<15; i++) {
174 /* bap=5 mantissas */
175 b5_mantissas[i] = symmetric_dequant(i, 15);
178 /* generate dynamic range table
179 reference: Section 7.7.1 Dynamic Range Control */
180 for(i=0; i<256; i++) {
181 int v = (i >> 5) - ((i >> 7) << 3) - 5;
182 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
188 * AVCodec initialization
190 static av_cold int ac3_decode_init(AVCodecContext *avctx)
192 AC3DecodeContext *s = avctx->priv_data;
197 ff_mdct_init(&s->imdct_256, 8, 1);
198 ff_mdct_init(&s->imdct_512, 9, 1);
199 ff_kbd_window_init(s->window, 5.0, 256);
200 dsputil_init(&s->dsp, avctx);
201 av_init_random(0, &s->dith_state);
203 /* set bias values for float to int16 conversion */
204 if(s->dsp.float_to_int16 == ff_float_to_int16_c) {
205 s->add_bias = 385.0f;
209 s->mul_bias = 32767.0f;
212 /* allow downmixing to stereo or mono */
213 if (avctx->channels > 0 && avctx->request_channels > 0 &&
214 avctx->request_channels < avctx->channels &&
215 avctx->request_channels <= 2) {
216 avctx->channels = avctx->request_channels;
220 /* allocate context input buffer */
221 if (avctx->error_resilience >= FF_ER_CAREFUL) {
222 s->input_buffer = av_mallocz(AC3_MAX_FRAME_SIZE + FF_INPUT_BUFFER_PADDING_SIZE);
223 if (!s->input_buffer)
224 return AVERROR_NOMEM;
227 avctx->sample_fmt = SAMPLE_FMT_S16;
232 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
233 * GetBitContext within AC3DecodeContext must point to
234 * the start of the synchronized AC-3 bitstream.
236 static int ac3_parse_header(AC3DecodeContext *s)
238 GetBitContext *gbc = &s->gbc;
241 /* read the rest of the bsi. read twice for dual mono mode. */
242 i = !(s->channel_mode);
244 skip_bits(gbc, 5); // skip dialog normalization
246 skip_bits(gbc, 8); //skip compression
248 skip_bits(gbc, 8); //skip language code
250 skip_bits(gbc, 7); //skip audio production information
253 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
255 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
256 TODO: read & use the xbsi1 downmix levels */
258 skip_bits(gbc, 14); //skip timecode1 / xbsi1
260 skip_bits(gbc, 14); //skip timecode2 / xbsi2
262 /* skip additional bitstream info */
263 if (get_bits1(gbc)) {
264 i = get_bits(gbc, 6);
274 * Common function to parse AC-3 or E-AC-3 frame header
276 static int parse_frame_header(AC3DecodeContext *s)
281 err = ff_ac3_parse_header(&s->gbc, &hdr);
285 /* get decoding parameters from header info */
286 s->bit_alloc_params.sr_code = hdr.sr_code;
287 s->channel_mode = hdr.channel_mode;
288 s->lfe_on = hdr.lfe_on;
289 s->bit_alloc_params.sr_shift = hdr.sr_shift;
290 s->sample_rate = hdr.sample_rate;
291 s->bit_rate = hdr.bit_rate;
292 s->channels = hdr.channels;
293 s->fbw_channels = s->channels - s->lfe_on;
294 s->lfe_ch = s->fbw_channels + 1;
295 s->frame_size = hdr.frame_size;
296 s->center_mix_level = hdr.center_mix_level;
297 s->surround_mix_level = hdr.surround_mix_level;
298 s->num_blocks = hdr.num_blocks;
299 s->frame_type = hdr.frame_type;
300 s->substreamid = hdr.substreamid;
303 s->start_freq[s->lfe_ch] = 0;
304 s->end_freq[s->lfe_ch] = 7;
305 s->num_exp_groups[s->lfe_ch] = 2;
306 s->channel_in_cpl[s->lfe_ch] = 0;
309 if(hdr.bitstream_id > 10)
310 return AC3_PARSE_ERROR_BSID;
312 return ac3_parse_header(s);
316 * Set stereo downmixing coefficients based on frame header info.
317 * reference: Section 7.8.2 Downmixing Into Two Channels
319 static void set_downmix_coeffs(AC3DecodeContext *s)
322 float cmix = gain_levels[center_levels[s->center_mix_level]];
323 float smix = gain_levels[surround_levels[s->surround_mix_level]];
325 for(i=0; i<s->fbw_channels; i++) {
326 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
327 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
329 if(s->channel_mode > 1 && s->channel_mode & 1) {
330 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
332 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
333 int nf = s->channel_mode - 2;
334 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
336 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
337 int nf = s->channel_mode - 4;
338 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
341 /* calculate adjustment needed for each channel to avoid clipping */
342 s->downmix_coeff_adjust[0] = s->downmix_coeff_adjust[1] = 0.0f;
343 for(i=0; i<s->fbw_channels; i++) {
344 s->downmix_coeff_adjust[0] += s->downmix_coeffs[i][0];
345 s->downmix_coeff_adjust[1] += s->downmix_coeffs[i][1];
347 s->downmix_coeff_adjust[0] = 1.0f / s->downmix_coeff_adjust[0];
348 s->downmix_coeff_adjust[1] = 1.0f / s->downmix_coeff_adjust[1];
352 * Decode the grouped exponents according to exponent strategy.
353 * reference: Section 7.1.3 Exponent Decoding
355 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
356 uint8_t absexp, int8_t *dexps)
358 int i, j, grp, group_size;
363 group_size = exp_strategy + (exp_strategy == EXP_D45);
364 for(grp=0,i=0; grp<ngrps; grp++) {
365 expacc = get_bits(gbc, 7);
366 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][0];
367 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][1];
368 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][2];
371 /* convert to absolute exps and expand groups */
373 for(i=0; i<ngrps*3; i++) {
374 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
375 for(j=0; j<group_size; j++) {
376 dexps[(i*group_size)+j] = prevexp;
382 * Generate transform coefficients for each coupled channel in the coupling
383 * range using the coupling coefficients and coupling coordinates.
384 * reference: Section 7.4.3 Coupling Coordinate Format
386 static void calc_transform_coeffs_cpl(AC3DecodeContext *s)
388 int i, j, ch, bnd, subbnd;
391 i = s->start_freq[CPL_CH];
392 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
395 for(j=0; j<12; j++) {
396 for(ch=1; ch<=s->fbw_channels; ch++) {
397 if(s->channel_in_cpl[ch]) {
398 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
399 if (ch == 2 && s->phase_flags[bnd])
400 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
405 } while(s->cpl_band_struct[subbnd]);
410 * Grouped mantissas for 3-level 5-level and 11-level quantization
422 * Get the transform coefficients for a particular channel
423 * reference: Section 7.3 Quantization and Decoding of Mantissas
425 static void get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
427 GetBitContext *gbc = &s->gbc;
428 int i, gcode, tbap, start, end;
433 exps = s->dexps[ch_index];
434 bap = s->bap[ch_index];
435 coeffs = s->fixed_coeffs[ch_index];
436 start = s->start_freq[ch_index];
437 end = s->end_freq[ch_index];
439 for (i = start; i < end; i++) {
443 coeffs[i] = (av_random(&s->dith_state) & 0x7FFFFF) - 0x400000;
448 gcode = get_bits(gbc, 5);
449 m->b1_mant[0] = b1_mantissas[gcode][0];
450 m->b1_mant[1] = b1_mantissas[gcode][1];
451 m->b1_mant[2] = b1_mantissas[gcode][2];
454 coeffs[i] = m->b1_mant[m->b1ptr++];
459 gcode = get_bits(gbc, 7);
460 m->b2_mant[0] = b2_mantissas[gcode][0];
461 m->b2_mant[1] = b2_mantissas[gcode][1];
462 m->b2_mant[2] = b2_mantissas[gcode][2];
465 coeffs[i] = m->b2_mant[m->b2ptr++];
469 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
474 gcode = get_bits(gbc, 7);
475 m->b4_mant[0] = b4_mantissas[gcode][0];
476 m->b4_mant[1] = b4_mantissas[gcode][1];
479 coeffs[i] = m->b4_mant[m->b4ptr++];
483 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
487 /* asymmetric dequantization */
488 int qlevel = quantization_tab[tbap];
489 coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
493 coeffs[i] >>= exps[i];
498 * Remove random dithering from coefficients with zero-bit mantissas
499 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
501 static void remove_dithering(AC3DecodeContext *s) {
507 for(ch=1; ch<=s->fbw_channels; ch++) {
508 if(!s->dither_flag[ch]) {
509 coeffs = s->fixed_coeffs[ch];
511 if(s->channel_in_cpl[ch])
512 end = s->start_freq[CPL_CH];
514 end = s->end_freq[ch];
515 for(i=0; i<end; i++) {
519 if(s->channel_in_cpl[ch]) {
520 bap = s->bap[CPL_CH];
521 for(; i<s->end_freq[CPL_CH]; i++) {
531 * Get the transform coefficients.
533 static void get_transform_coeffs(AC3DecodeContext *s)
539 m.b1ptr = m.b2ptr = m.b4ptr = 3;
541 for (ch = 1; ch <= s->channels; ch++) {
542 /* transform coefficients for full-bandwidth channel */
543 get_transform_coeffs_ch(s, ch, &m);
544 /* tranform coefficients for coupling channel come right after the
545 coefficients for the first coupled channel*/
546 if (s->channel_in_cpl[ch]) {
548 get_transform_coeffs_ch(s, CPL_CH, &m);
549 calc_transform_coeffs_cpl(s);
552 end = s->end_freq[CPL_CH];
554 end = s->end_freq[ch];
557 s->fixed_coeffs[ch][end] = 0;
561 /* if any channel doesn't use dithering, zero appropriate coefficients */
567 * Stereo rematrixing.
568 * reference: Section 7.5.4 Rematrixing : Decoding Technique
570 static void do_rematrixing(AC3DecodeContext *s)
576 end = FFMIN(s->end_freq[1], s->end_freq[2]);
578 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
579 if(s->rematrixing_flags[bnd]) {
580 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd+1]);
581 for(i=ff_ac3_rematrix_band_tab[bnd]; i<bndend; i++) {
582 tmp0 = s->fixed_coeffs[1][i];
583 tmp1 = s->fixed_coeffs[2][i];
584 s->fixed_coeffs[1][i] = tmp0 + tmp1;
585 s->fixed_coeffs[2][i] = tmp0 - tmp1;
592 * Inverse MDCT Transform.
593 * Convert frequency domain coefficients to time-domain audio samples.
594 * reference: Section 7.9.4 Transformation Equations
596 static inline void do_imdct(AC3DecodeContext *s, int channels)
600 for (ch=1; ch<=channels; ch++) {
601 if (s->block_switch[ch]) {
603 float *x = s->tmp_output+128;
605 x[i] = s->transform_coeffs[ch][2*i];
606 ff_imdct_half(&s->imdct_256, s->tmp_output, x);
607 s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, 0, 128);
609 x[i] = s->transform_coeffs[ch][2*i+1];
610 ff_imdct_half(&s->imdct_256, s->delay[ch-1], x);
612 ff_imdct_half(&s->imdct_512, s->tmp_output, s->transform_coeffs[ch]);
613 s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, 0, 128);
614 memcpy(s->delay[ch-1], s->tmp_output+128, 128*sizeof(float));
620 * Downmix the output to mono or stereo.
622 static void ac3_downmix(AC3DecodeContext *s,
623 float samples[AC3_MAX_CHANNELS][256], int ch_offset)
628 for(i=0; i<256; i++) {
630 for(j=0; j<s->fbw_channels; j++) {
631 v0 += samples[j+ch_offset][i] * s->downmix_coeffs[j][0];
632 v1 += samples[j+ch_offset][i] * s->downmix_coeffs[j][1];
634 v0 *= s->downmix_coeff_adjust[0];
635 v1 *= s->downmix_coeff_adjust[1];
636 if(s->output_mode == AC3_CHMODE_MONO) {
637 samples[ch_offset][i] = (v0 + v1) * LEVEL_MINUS_3DB;
638 } else if(s->output_mode == AC3_CHMODE_STEREO) {
639 samples[ ch_offset][i] = v0;
640 samples[1+ch_offset][i] = v1;
646 * Upmix delay samples from stereo to original channel layout.
648 static void ac3_upmix_delay(AC3DecodeContext *s)
650 int channel_data_size = 128*sizeof(float);
651 switch(s->channel_mode) {
652 case AC3_CHMODE_DUALMONO:
653 case AC3_CHMODE_STEREO:
654 /* upmix mono to stereo */
655 memcpy(s->delay[1], s->delay[0], channel_data_size);
657 case AC3_CHMODE_2F2R:
658 memset(s->delay[3], 0, channel_data_size);
659 case AC3_CHMODE_2F1R:
660 memset(s->delay[2], 0, channel_data_size);
662 case AC3_CHMODE_3F2R:
663 memset(s->delay[4], 0, channel_data_size);
664 case AC3_CHMODE_3F1R:
665 memset(s->delay[3], 0, channel_data_size);
667 memcpy(s->delay[2], s->delay[1], channel_data_size);
668 memset(s->delay[1], 0, channel_data_size);
674 * Decode a single audio block from the AC-3 bitstream.
676 static int decode_audio_block(AC3DecodeContext *s, int blk)
678 int fbw_channels = s->fbw_channels;
679 int channel_mode = s->channel_mode;
681 int different_transforms;
684 GetBitContext *gbc = &s->gbc;
685 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
687 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
689 /* block switch flags */
690 different_transforms = 0;
691 for (ch = 1; ch <= fbw_channels; ch++) {
692 s->block_switch[ch] = get_bits1(gbc);
693 if(ch > 1 && s->block_switch[ch] != s->block_switch[1])
694 different_transforms = 1;
697 /* dithering flags */
699 for (ch = 1; ch <= fbw_channels; ch++) {
700 s->dither_flag[ch] = get_bits1(gbc);
701 if(!s->dither_flag[ch])
706 i = !(s->channel_mode);
709 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
710 s->avctx->drc_scale)+1.0;
711 } else if(blk == 0) {
712 s->dynamic_range[i] = 1.0f;
716 /* coupling strategy */
717 if (get_bits1(gbc)) {
718 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
719 s->cpl_in_use[blk] = get_bits1(gbc);
720 if (s->cpl_in_use[blk]) {
721 /* coupling in use */
722 int cpl_begin_freq, cpl_end_freq;
724 if (channel_mode < AC3_CHMODE_STEREO) {
725 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
729 /* determine which channels are coupled */
730 for (ch = 1; ch <= fbw_channels; ch++)
731 s->channel_in_cpl[ch] = get_bits1(gbc);
733 /* phase flags in use */
734 if (channel_mode == AC3_CHMODE_STEREO)
735 s->phase_flags_in_use = get_bits1(gbc);
737 /* coupling frequency range and band structure */
738 cpl_begin_freq = get_bits(gbc, 4);
739 cpl_end_freq = get_bits(gbc, 4);
740 if (3 + cpl_end_freq - cpl_begin_freq < 0) {
741 av_log(s->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
744 s->num_cpl_bands = s->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
745 s->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
746 s->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
747 for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
748 if (get_bits1(gbc)) {
749 s->cpl_band_struct[bnd] = 1;
753 s->cpl_band_struct[s->num_cpl_subbands-1] = 0;
755 /* coupling not in use */
756 for (ch = 1; ch <= fbw_channels; ch++)
757 s->channel_in_cpl[ch] = 0;
760 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must be present in block 0\n");
763 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
765 cpl_in_use = s->cpl_in_use[blk];
767 /* coupling coordinates */
769 int cpl_coords_exist = 0;
771 for (ch = 1; ch <= fbw_channels; ch++) {
772 if (s->channel_in_cpl[ch]) {
773 if (get_bits1(gbc)) {
774 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
775 cpl_coords_exist = 1;
776 master_cpl_coord = 3 * get_bits(gbc, 2);
777 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
778 cpl_coord_exp = get_bits(gbc, 4);
779 cpl_coord_mant = get_bits(gbc, 4);
780 if (cpl_coord_exp == 15)
781 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
783 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
784 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
787 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must be present in block 0\n");
793 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
794 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
795 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
800 /* stereo rematrixing strategy and band structure */
801 if (channel_mode == AC3_CHMODE_STEREO) {
802 if (get_bits1(gbc)) {
803 s->num_rematrixing_bands = 4;
804 if(cpl_in_use && s->start_freq[CPL_CH] <= 61)
805 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
806 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
807 s->rematrixing_flags[bnd] = get_bits1(gbc);
809 av_log(s->avctx, AV_LOG_ERROR, "new rematrixing strategy must be present in block 0\n");
814 /* exponent strategies for each channel */
815 s->exp_strategy[blk][CPL_CH] = EXP_REUSE;
816 s->exp_strategy[blk][s->lfe_ch] = EXP_REUSE;
817 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
818 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
819 if(s->exp_strategy[blk][ch] != EXP_REUSE)
820 bit_alloc_stages[ch] = 3;
823 /* channel bandwidth */
824 for (ch = 1; ch <= fbw_channels; ch++) {
825 s->start_freq[ch] = 0;
826 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
828 int prev = s->end_freq[ch];
829 if (s->channel_in_cpl[ch])
830 s->end_freq[ch] = s->start_freq[CPL_CH];
832 int bandwidth_code = get_bits(gbc, 6);
833 if (bandwidth_code > 60) {
834 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
837 s->end_freq[ch] = bandwidth_code * 3 + 73;
839 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
840 s->num_exp_groups[ch] = (s->end_freq[ch]+group_size-4) / group_size;
841 if(blk > 0 && s->end_freq[ch] != prev)
842 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
845 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
846 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
847 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
850 /* decode exponents for each channel */
851 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
852 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
853 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
854 decode_exponents(gbc, s->exp_strategy[blk][ch],
855 s->num_exp_groups[ch], s->dexps[ch][0],
856 &s->dexps[ch][s->start_freq[ch]+!!ch]);
857 if(ch != CPL_CH && ch != s->lfe_ch)
858 skip_bits(gbc, 2); /* skip gainrng */
862 /* bit allocation information */
863 if (get_bits1(gbc)) {
864 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
865 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
866 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
867 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
868 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
869 for(ch=!cpl_in_use; ch<=s->channels; ch++)
870 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
872 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must be present in block 0\n");
876 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
877 if (get_bits1(gbc)) {
879 csnr = (get_bits(gbc, 6) - 15) << 4;
880 for (ch = !cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
881 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
882 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
884 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
886 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
890 /* coupling leak information */
892 if (get_bits1(gbc)) {
893 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
894 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
895 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
897 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must be present in block 0\n");
902 /* delta bit allocation information */
903 if (get_bits1(gbc)) {
904 /* delta bit allocation exists (strategy) */
905 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
906 s->dba_mode[ch] = get_bits(gbc, 2);
907 if (s->dba_mode[ch] == DBA_RESERVED) {
908 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
911 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
913 /* channel delta offset, len and bit allocation */
914 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
915 if (s->dba_mode[ch] == DBA_NEW) {
916 s->dba_nsegs[ch] = get_bits(gbc, 3);
917 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
918 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
919 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
920 s->dba_values[ch][seg] = get_bits(gbc, 3);
922 /* run last 2 bit allocation stages if new dba values */
923 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
926 } else if(blk == 0) {
927 for(ch=0; ch<=s->channels; ch++) {
928 s->dba_mode[ch] = DBA_NONE;
933 for(ch=!cpl_in_use; ch<=s->channels; ch++) {
934 if(bit_alloc_stages[ch] > 2) {
935 /* Exponent mapping into PSD and PSD integration */
936 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
937 s->start_freq[ch], s->end_freq[ch],
938 s->psd[ch], s->band_psd[ch]);
940 if(bit_alloc_stages[ch] > 1) {
941 /* Compute excitation function, Compute masking curve, and
942 Apply delta bit allocation */
943 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
944 s->start_freq[ch], s->end_freq[ch],
945 s->fast_gain[ch], (ch == s->lfe_ch),
946 s->dba_mode[ch], s->dba_nsegs[ch],
947 s->dba_offsets[ch], s->dba_lengths[ch],
948 s->dba_values[ch], s->mask[ch]);
950 if(bit_alloc_stages[ch] > 0) {
951 /* Compute bit allocation */
952 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
953 s->start_freq[ch], s->end_freq[ch],
955 s->bit_alloc_params.floor,
956 ff_ac3_bap_tab, s->bap[ch]);
960 /* unused dummy data */
961 if (get_bits1(gbc)) {
962 int skipl = get_bits(gbc, 9);
967 /* unpack the transform coefficients
968 this also uncouples channels if coupling is in use. */
969 get_transform_coeffs(s);
971 /* recover coefficients if rematrixing is in use */
972 if(s->channel_mode == AC3_CHMODE_STEREO)
975 /* apply scaling to coefficients (headroom, dynrng) */
976 for(ch=1; ch<=s->channels; ch++) {
977 float gain = s->mul_bias / 4194304.0f;
978 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
979 gain *= s->dynamic_range[ch-1];
981 gain *= s->dynamic_range[0];
983 for(i=0; i<256; i++) {
984 s->transform_coeffs[ch][i] = s->fixed_coeffs[ch][i] * gain;
988 /* downmix and MDCT. order depends on whether block switching is used for
989 any channel in this block. this is because coefficients for the long
990 and short transforms cannot be mixed. */
991 downmix_output = s->channels != s->out_channels &&
992 !((s->output_mode & AC3_OUTPUT_LFEON) &&
993 s->fbw_channels == s->out_channels);
994 if(different_transforms) {
995 /* the delay samples have already been downmixed, so we upmix the delay
996 samples in order to reconstruct all channels before downmixing. */
1002 do_imdct(s, s->channels);
1004 if(downmix_output) {
1005 ac3_downmix(s, s->output, 0);
1008 if(downmix_output) {
1009 ac3_downmix(s, s->transform_coeffs, 1);
1014 // FIXME delay[] is half the size of the other downmixes
1015 ac3_downmix(s, s->delay, 0);
1018 do_imdct(s, s->out_channels);
1021 /* convert float to 16-bit integer */
1022 for(ch=0; ch<s->out_channels; ch++) {
1023 for(i=0; i<256; i++) {
1024 s->output[ch][i] += s->add_bias;
1026 s->dsp.float_to_int16(s->int_output[ch], s->output[ch], 256);
1033 * Decode a single AC-3 frame.
1035 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
1036 const uint8_t *buf, int buf_size)
1038 AC3DecodeContext *s = avctx->priv_data;
1039 int16_t *out_samples = (int16_t *)data;
1040 int i, blk, ch, err;
1042 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1043 if (s->input_buffer) {
1044 /* copy input buffer to decoder context to avoid reading past the end
1045 of the buffer, which can be caused by a damaged input stream. */
1046 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_MAX_FRAME_SIZE));
1047 init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
1049 init_get_bits(&s->gbc, buf, buf_size * 8);
1052 /* parse the syncinfo */
1054 err = parse_frame_header(s);
1056 /* check that reported frame size fits in input buffer */
1057 if(s->frame_size > buf_size) {
1058 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1059 err = AC3_PARSE_ERROR_FRAME_SIZE;
1062 /* check for crc mismatch */
1063 if(err != AC3_PARSE_ERROR_FRAME_SIZE && avctx->error_resilience >= FF_ER_CAREFUL) {
1064 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1065 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1066 err = AC3_PARSE_ERROR_CRC;
1070 if(err && err != AC3_PARSE_ERROR_CRC) {
1072 case AC3_PARSE_ERROR_SYNC:
1073 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1075 case AC3_PARSE_ERROR_BSID:
1076 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1078 case AC3_PARSE_ERROR_SAMPLE_RATE:
1079 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1081 case AC3_PARSE_ERROR_FRAME_SIZE:
1082 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1084 case AC3_PARSE_ERROR_FRAME_TYPE:
1085 /* skip frame if CRC is ok. otherwise use error concealment. */
1086 /* TODO: add support for substreams and dependent frames */
1087 if(s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1088 av_log(avctx, AV_LOG_ERROR, "unsupported frame type : skipping frame\n");
1089 return s->frame_size;
1091 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1095 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1100 /* if frame is ok, set audio parameters */
1102 avctx->sample_rate = s->sample_rate;
1103 avctx->bit_rate = s->bit_rate;
1105 /* channel config */
1106 s->out_channels = s->channels;
1107 s->output_mode = s->channel_mode;
1109 s->output_mode |= AC3_OUTPUT_LFEON;
1110 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1111 avctx->request_channels < s->channels) {
1112 s->out_channels = avctx->request_channels;
1113 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1115 avctx->channels = s->out_channels;
1117 /* set downmixing coefficients if needed */
1118 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1119 s->fbw_channels == s->out_channels)) {
1120 set_downmix_coeffs(s);
1122 } else if (!s->out_channels) {
1123 s->out_channels = avctx->channels;
1124 if(s->out_channels < s->channels)
1125 s->output_mode = s->out_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1128 /* decode the audio blocks */
1129 for (blk = 0; blk < s->num_blocks; blk++) {
1130 if (!err && decode_audio_block(s, blk)) {
1131 av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n");
1134 /* interleave output samples */
1135 for (i = 0; i < 256; i++)
1136 for (ch = 0; ch < s->out_channels; ch++)
1137 *(out_samples++) = s->int_output[ch][i];
1139 *data_size = s->num_blocks * 256 * avctx->channels * sizeof (int16_t);
1140 return s->frame_size;
1144 * Uninitialize the AC-3 decoder.
1146 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1148 AC3DecodeContext *s = avctx->priv_data;
1149 ff_mdct_end(&s->imdct_512);
1150 ff_mdct_end(&s->imdct_256);
1152 av_freep(&s->input_buffer);
1157 AVCodec ac3_decoder = {
1159 .type = CODEC_TYPE_AUDIO,
1161 .priv_data_size = sizeof (AC3DecodeContext),
1162 .init = ac3_decode_init,
1163 .close = ac3_decode_end,
1164 .decode = ac3_decode_frame,
1165 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52 / AC-3"),