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 /** Maximum possible frame size when the specification limit is ignored */
43 #define AC3_MAX_FRAME_SIZE 21695
46 * Table of bin locations for rematrixing bands
47 * reference: Section 7.5.2 Rematrixing : Frequency Band Definitions
49 static const uint8_t rematrix_band_tab[5] = { 13, 25, 37, 61, 253 };
51 /** table for grouping exponents */
52 static uint8_t exp_ungroup_tab[128][3];
55 /** tables for ungrouping mantissas */
56 static int b1_mantissas[32][3];
57 static int b2_mantissas[128][3];
58 static int b3_mantissas[8];
59 static int b4_mantissas[128][2];
60 static int b5_mantissas[16];
63 * Quantization table: levels for symmetric. bits for asymmetric.
64 * reference: Table 7.18 Mapping of bap to Quantizer
66 static const uint8_t quantization_tab[16] = {
68 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
71 /** dynamic range table. converts codes to scale factors. */
72 static float dynamic_range_tab[256];
74 /** Adjustments in dB gain */
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[6] = {
86 LEVEL_MINUS_4POINT5DB,
92 * Table for default stereo downmixing coefficients
93 * reference: Section 7.8.2 Downmixing Into Two Channels
95 static const uint8_t ac3_default_coeffs[8][5][2] = {
96 { { 1, 0 }, { 0, 1 }, },
98 { { 1, 0 }, { 0, 1 }, },
99 { { 1, 0 }, { 3, 3 }, { 0, 1 }, },
100 { { 1, 0 }, { 0, 1 }, { 4, 4 }, },
101 { { 1, 0 }, { 3, 3 }, { 0, 1 }, { 5, 5 }, },
102 { { 1, 0 }, { 0, 1 }, { 4, 0 }, { 0, 4 }, },
103 { { 1, 0 }, { 3, 3 }, { 0, 1 }, { 4, 0 }, { 0, 4 }, },
106 /* override ac3.h to include coupling channel */
107 #undef AC3_MAX_CHANNELS
108 #define AC3_MAX_CHANNELS 7
111 #define AC3_OUTPUT_LFEON 8
114 int num_blocks; ///< number of audio blocks
115 int channel_mode; ///< channel mode (acmod)
116 int block_switch[AC3_MAX_CHANNELS]; ///< block switch flags
117 int dither_flag[AC3_MAX_CHANNELS]; ///< dither flags
118 int dither_all; ///< true if all channels are dithered
119 int cpl_in_use; ///< coupling in use
120 int channel_in_cpl[AC3_MAX_CHANNELS]; ///< channel in coupling
121 int phase_flags_in_use; ///< phase flags in use
122 int phase_flags[18]; ///< phase flags
123 int cpl_band_struct[18]; ///< coupling band structure
124 int num_rematrixing_bands; ///< number of rematrixing bands
125 int rematrixing_flags[4]; ///< rematrixing flags
126 int exp_strategy[AC3_MAX_CHANNELS]; ///< exponent strategies
127 int snr_offset[AC3_MAX_CHANNELS]; ///< signal-to-noise ratio offsets
128 int fast_gain[AC3_MAX_CHANNELS]; ///< fast gain values (signal-to-mask ratio)
129 int dba_mode[AC3_MAX_CHANNELS]; ///< delta bit allocation mode
130 int dba_nsegs[AC3_MAX_CHANNELS]; ///< number of delta segments
131 uint8_t dba_offsets[AC3_MAX_CHANNELS][8]; ///< delta segment offsets
132 uint8_t dba_lengths[AC3_MAX_CHANNELS][8]; ///< delta segment lengths
133 uint8_t dba_values[AC3_MAX_CHANNELS][8]; ///< delta values for each segment
135 int sample_rate; ///< sample frequency, in Hz
136 int bit_rate; ///< stream bit rate, in bits-per-second
137 int frame_type; ///< frame type (strmtyp)
138 int substreamid; ///< substream identification
139 int frame_size; ///< current frame size, in bytes
141 int channels; ///< number of total channels
142 int fbw_channels; ///< number of full-bandwidth channels
143 int lfe_on; ///< lfe channel in use
144 int lfe_ch; ///< index of LFE channel
145 int output_mode; ///< output channel configuration
146 int out_channels; ///< number of output channels
148 int center_mix_level; ///< Center mix level index
149 int surround_mix_level; ///< Surround mix level index
150 float downmix_coeffs[AC3_MAX_CHANNELS][2]; ///< stereo downmix coefficients
151 float downmix_coeff_adjust[2]; ///< adjustment needed for each output channel when downmixing
152 float dynamic_range[2]; ///< dynamic range
153 int cpl_coords[AC3_MAX_CHANNELS][18]; ///< coupling coordinates
154 int num_cpl_bands; ///< number of coupling bands
155 int num_cpl_subbands; ///< number of coupling sub bands
156 int start_freq[AC3_MAX_CHANNELS]; ///< start frequency bin
157 int end_freq[AC3_MAX_CHANNELS]; ///< end frequency bin
158 AC3BitAllocParameters bit_alloc_params; ///< bit allocation parameters
160 int num_exp_groups[AC3_MAX_CHANNELS]; ///< Number of exponent groups
161 int8_t dexps[AC3_MAX_CHANNELS][256]; ///< decoded exponents
162 uint8_t bap[AC3_MAX_CHANNELS][256]; ///< bit allocation pointers
163 int16_t psd[AC3_MAX_CHANNELS][256]; ///< scaled exponents
164 int16_t band_psd[AC3_MAX_CHANNELS][50]; ///< interpolated exponents
165 int16_t mask[AC3_MAX_CHANNELS][50]; ///< masking curve values
167 int fixed_coeffs[AC3_MAX_CHANNELS][256]; ///> fixed-point transform coefficients
168 DECLARE_ALIGNED_16(float, transform_coeffs[AC3_MAX_CHANNELS][256]); ///< transform coefficients
169 int downmixed; ///< indicates if coeffs are currently downmixed
172 MDCTContext imdct_512; ///< for 512 sample IMDCT
173 MDCTContext imdct_256; ///< for 256 sample IMDCT
174 DSPContext dsp; ///< for optimization
175 float add_bias; ///< offset for float_to_int16 conversion
176 float mul_bias; ///< scaling for float_to_int16 conversion
178 DECLARE_ALIGNED_16(float, output[AC3_MAX_CHANNELS][256]); ///< output after imdct transform and windowing
179 DECLARE_ALIGNED_16(short, int_output[AC3_MAX_CHANNELS-1][256]); ///< final 16-bit integer output
180 DECLARE_ALIGNED_16(float, delay[AC3_MAX_CHANNELS][256]); ///< delay - added to the next block
181 DECLARE_ALIGNED_16(float, tmp_imdct[256]); ///< temporary storage for imdct transform
182 DECLARE_ALIGNED_16(float, tmp_output[512]); ///< temporary storage for output before windowing
183 DECLARE_ALIGNED_16(float, window[256]); ///< window coefficients
186 GetBitContext gbc; ///< bitstream reader
187 AVRandomState dith_state; ///< for dither generation
188 AVCodecContext *avctx; ///< parent context
189 uint8_t *input_buffer; ///< temp buffer to prevent overread
193 * Symmetrical Dequantization
194 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
195 * Tables 7.19 to 7.23
198 symmetric_dequant(int code, int levels)
200 return ((code - (levels >> 1)) << 24) / levels;
204 * Initialize tables at runtime.
206 static av_cold void ac3_tables_init(void)
210 /* generate grouped mantissa tables
211 reference: Section 7.3.5 Ungrouping of Mantissas */
212 for(i=0; i<32; i++) {
213 /* bap=1 mantissas */
214 b1_mantissas[i][0] = symmetric_dequant( i / 9 , 3);
215 b1_mantissas[i][1] = symmetric_dequant((i % 9) / 3, 3);
216 b1_mantissas[i][2] = symmetric_dequant((i % 9) % 3, 3);
218 for(i=0; i<128; i++) {
219 /* bap=2 mantissas */
220 b2_mantissas[i][0] = symmetric_dequant( i / 25 , 5);
221 b2_mantissas[i][1] = symmetric_dequant((i % 25) / 5, 5);
222 b2_mantissas[i][2] = symmetric_dequant((i % 25) % 5, 5);
224 /* bap=4 mantissas */
225 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
226 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
228 /* generate ungrouped mantissa tables
229 reference: Tables 7.21 and 7.23 */
231 /* bap=3 mantissas */
232 b3_mantissas[i] = symmetric_dequant(i, 7);
234 for(i=0; i<15; i++) {
235 /* bap=5 mantissas */
236 b5_mantissas[i] = symmetric_dequant(i, 15);
239 /* generate dynamic range table
240 reference: Section 7.7.1 Dynamic Range Control */
241 for(i=0; i<256; i++) {
242 int v = (i >> 5) - ((i >> 7) << 3) - 5;
243 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
246 /* generate exponent tables
247 reference: Section 7.1.3 Exponent Decoding */
248 for(i=0; i<128; i++) {
249 exp_ungroup_tab[i][0] = i / 25;
250 exp_ungroup_tab[i][1] = (i % 25) / 5;
251 exp_ungroup_tab[i][2] = (i % 25) % 5;
257 * AVCodec initialization
259 static av_cold int ac3_decode_init(AVCodecContext *avctx)
261 AC3DecodeContext *s = avctx->priv_data;
266 ff_mdct_init(&s->imdct_256, 8, 1);
267 ff_mdct_init(&s->imdct_512, 9, 1);
268 ff_kbd_window_init(s->window, 5.0, 256);
269 dsputil_init(&s->dsp, avctx);
270 av_init_random(0, &s->dith_state);
272 /* set bias values for float to int16 conversion */
273 if(s->dsp.float_to_int16 == ff_float_to_int16_c) {
274 s->add_bias = 385.0f;
278 s->mul_bias = 32767.0f;
281 /* allow downmixing to stereo or mono */
282 if (avctx->channels > 0 && avctx->request_channels > 0 &&
283 avctx->request_channels < avctx->channels &&
284 avctx->request_channels <= 2) {
285 avctx->channels = avctx->request_channels;
289 /* allocate context input buffer */
290 if (avctx->error_resilience >= FF_ER_CAREFUL) {
291 s->input_buffer = av_mallocz(AC3_MAX_FRAME_SIZE + FF_INPUT_BUFFER_PADDING_SIZE);
292 if (!s->input_buffer)
293 return AVERROR_NOMEM;
300 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
301 * GetBitContext within AC3DecodeContext must point to
302 * start of the synchronized ac3 bitstream.
304 static int ac3_parse_header(AC3DecodeContext *s)
307 GetBitContext *gbc = &s->gbc;
310 err = ff_ac3_parse_header(gbc, &hdr);
314 if(hdr.bitstream_id > 10)
315 return AC3_PARSE_ERROR_BSID;
317 /* get decoding parameters from header info */
318 s->bit_alloc_params.sr_code = hdr.sr_code;
319 s->channel_mode = hdr.channel_mode;
320 s->lfe_on = hdr.lfe_on;
321 s->bit_alloc_params.sr_shift = hdr.sr_shift;
322 s->sample_rate = hdr.sample_rate;
323 s->bit_rate = hdr.bit_rate;
324 s->channels = hdr.channels;
325 s->fbw_channels = s->channels - s->lfe_on;
326 s->lfe_ch = s->fbw_channels + 1;
327 s->frame_size = hdr.frame_size;
328 s->center_mix_level = hdr.center_mix_level;
329 s->surround_mix_level = hdr.surround_mix_level;
330 s->num_blocks = hdr.num_blocks;
331 s->frame_type = hdr.frame_type;
332 s->substreamid = hdr.substreamid;
335 s->start_freq[s->lfe_ch] = 0;
336 s->end_freq[s->lfe_ch] = 7;
337 s->num_exp_groups[s->lfe_ch] = 2;
338 s->channel_in_cpl[s->lfe_ch] = 0;
341 /* read the rest of the bsi. read twice for dual mono mode. */
342 i = !(s->channel_mode);
344 skip_bits(gbc, 5); // skip dialog normalization
346 skip_bits(gbc, 8); //skip compression
348 skip_bits(gbc, 8); //skip language code
350 skip_bits(gbc, 7); //skip audio production information
353 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
355 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
356 TODO: read & use the xbsi1 downmix levels */
358 skip_bits(gbc, 14); //skip timecode1 / xbsi1
360 skip_bits(gbc, 14); //skip timecode2 / xbsi2
362 /* skip additional bitstream info */
363 if (get_bits1(gbc)) {
364 i = get_bits(gbc, 6);
374 * Set stereo downmixing coefficients based on frame header info.
375 * reference: Section 7.8.2 Downmixing Into Two Channels
377 static void set_downmix_coeffs(AC3DecodeContext *s)
380 float cmix = gain_levels[s->center_mix_level];
381 float smix = gain_levels[s->surround_mix_level];
383 for(i=0; i<s->fbw_channels; i++) {
384 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
385 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
387 if(s->channel_mode > 1 && s->channel_mode & 1) {
388 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
390 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
391 int nf = s->channel_mode - 2;
392 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
394 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
395 int nf = s->channel_mode - 4;
396 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
399 /* calculate adjustment needed for each channel to avoid clipping */
400 s->downmix_coeff_adjust[0] = s->downmix_coeff_adjust[1] = 0.0f;
401 for(i=0; i<s->fbw_channels; i++) {
402 s->downmix_coeff_adjust[0] += s->downmix_coeffs[i][0];
403 s->downmix_coeff_adjust[1] += s->downmix_coeffs[i][1];
405 s->downmix_coeff_adjust[0] = 1.0f / s->downmix_coeff_adjust[0];
406 s->downmix_coeff_adjust[1] = 1.0f / s->downmix_coeff_adjust[1];
410 * Decode the grouped exponents according to exponent strategy.
411 * reference: Section 7.1.3 Exponent Decoding
413 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
414 uint8_t absexp, int8_t *dexps)
416 int i, j, grp, group_size;
421 group_size = exp_strategy + (exp_strategy == EXP_D45);
422 for(grp=0,i=0; grp<ngrps; grp++) {
423 expacc = get_bits(gbc, 7);
424 dexp[i++] = exp_ungroup_tab[expacc][0];
425 dexp[i++] = exp_ungroup_tab[expacc][1];
426 dexp[i++] = exp_ungroup_tab[expacc][2];
429 /* convert to absolute exps and expand groups */
431 for(i=0; i<ngrps*3; i++) {
432 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
433 for(j=0; j<group_size; j++) {
434 dexps[(i*group_size)+j] = prevexp;
440 * Generate transform coefficients for each coupled channel in the coupling
441 * range using the coupling coefficients and coupling coordinates.
442 * reference: Section 7.4.3 Coupling Coordinate Format
444 static void uncouple_channels(AC3DecodeContext *s)
446 int i, j, ch, bnd, subbnd;
449 i = s->start_freq[CPL_CH];
450 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
453 for(j=0; j<12; j++) {
454 for(ch=1; ch<=s->fbw_channels; ch++) {
455 if(s->channel_in_cpl[ch]) {
456 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
457 if (ch == 2 && s->phase_flags[bnd])
458 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
463 } while(s->cpl_band_struct[subbnd]);
468 * Grouped mantissas for 3-level 5-level and 11-level quantization
480 * Get the transform coefficients for a particular channel
481 * reference: Section 7.3 Quantization and Decoding of Mantissas
483 static void get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
485 GetBitContext *gbc = &s->gbc;
486 int i, gcode, tbap, start, end;
491 exps = s->dexps[ch_index];
492 bap = s->bap[ch_index];
493 coeffs = s->fixed_coeffs[ch_index];
494 start = s->start_freq[ch_index];
495 end = s->end_freq[ch_index];
497 for (i = start; i < end; i++) {
501 coeffs[i] = (av_random(&s->dith_state) & 0x7FFFFF) - 4194304;
506 gcode = get_bits(gbc, 5);
507 m->b1_mant[0] = b1_mantissas[gcode][0];
508 m->b1_mant[1] = b1_mantissas[gcode][1];
509 m->b1_mant[2] = b1_mantissas[gcode][2];
512 coeffs[i] = m->b1_mant[m->b1ptr++];
517 gcode = get_bits(gbc, 7);
518 m->b2_mant[0] = b2_mantissas[gcode][0];
519 m->b2_mant[1] = b2_mantissas[gcode][1];
520 m->b2_mant[2] = b2_mantissas[gcode][2];
523 coeffs[i] = m->b2_mant[m->b2ptr++];
527 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
532 gcode = get_bits(gbc, 7);
533 m->b4_mant[0] = b4_mantissas[gcode][0];
534 m->b4_mant[1] = b4_mantissas[gcode][1];
537 coeffs[i] = m->b4_mant[m->b4ptr++];
541 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
545 /* asymmetric dequantization */
546 int qlevel = quantization_tab[tbap];
547 coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
551 coeffs[i] >>= exps[i];
556 * Remove random dithering from coefficients with zero-bit mantissas
557 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
559 static void remove_dithering(AC3DecodeContext *s) {
565 for(ch=1; ch<=s->fbw_channels; ch++) {
566 if(!s->dither_flag[ch]) {
567 coeffs = s->fixed_coeffs[ch];
569 if(s->channel_in_cpl[ch])
570 end = s->start_freq[CPL_CH];
572 end = s->end_freq[ch];
573 for(i=0; i<end; i++) {
577 if(s->channel_in_cpl[ch]) {
578 bap = s->bap[CPL_CH];
579 for(; i<s->end_freq[CPL_CH]; i++) {
589 * Get the transform coefficients.
591 static void get_transform_coeffs(AC3DecodeContext *s)
597 m.b1ptr = m.b2ptr = m.b4ptr = 3;
599 for (ch = 1; ch <= s->channels; ch++) {
600 /* transform coefficients for full-bandwidth channel */
601 get_transform_coeffs_ch(s, ch, &m);
602 /* tranform coefficients for coupling channel come right after the
603 coefficients for the first coupled channel*/
604 if (s->channel_in_cpl[ch]) {
606 get_transform_coeffs_ch(s, CPL_CH, &m);
607 uncouple_channels(s);
610 end = s->end_freq[CPL_CH];
612 end = s->end_freq[ch];
615 s->fixed_coeffs[ch][end] = 0;
619 /* if any channel doesn't use dithering, zero appropriate coefficients */
625 * Stereo rematrixing.
626 * reference: Section 7.5.4 Rematrixing : Decoding Technique
628 static void do_rematrixing(AC3DecodeContext *s)
634 end = FFMIN(s->end_freq[1], s->end_freq[2]);
636 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
637 if(s->rematrixing_flags[bnd]) {
638 bndend = FFMIN(end, rematrix_band_tab[bnd+1]);
639 for(i=rematrix_band_tab[bnd]; i<bndend; i++) {
640 tmp0 = s->fixed_coeffs[1][i];
641 tmp1 = s->fixed_coeffs[2][i];
642 s->fixed_coeffs[1][i] = tmp0 + tmp1;
643 s->fixed_coeffs[2][i] = tmp0 - tmp1;
650 * Perform the 256-point IMDCT
652 static void do_imdct_256(AC3DecodeContext *s, int chindex)
655 DECLARE_ALIGNED_16(float, x[128]);
657 float *o_ptr = s->tmp_output;
660 /* de-interleave coefficients */
661 for(k=0; k<128; k++) {
662 x[k] = s->transform_coeffs[chindex][2*k+i];
665 /* run standard IMDCT */
666 s->imdct_256.fft.imdct_calc(&s->imdct_256, o_ptr, x, s->tmp_imdct);
668 /* reverse the post-rotation & reordering from standard IMDCT */
669 for(k=0; k<32; k++) {
670 z[i][32+k].re = -o_ptr[128+2*k];
671 z[i][32+k].im = -o_ptr[2*k];
672 z[i][31-k].re = o_ptr[2*k+1];
673 z[i][31-k].im = o_ptr[128+2*k+1];
677 /* apply AC-3 post-rotation & reordering */
678 for(k=0; k<64; k++) {
679 o_ptr[ 2*k ] = -z[0][ k].im;
680 o_ptr[ 2*k+1] = z[0][63-k].re;
681 o_ptr[128+2*k ] = -z[0][ k].re;
682 o_ptr[128+2*k+1] = z[0][63-k].im;
683 o_ptr[256+2*k ] = -z[1][ k].re;
684 o_ptr[256+2*k+1] = z[1][63-k].im;
685 o_ptr[384+2*k ] = z[1][ k].im;
686 o_ptr[384+2*k+1] = -z[1][63-k].re;
691 * Inverse MDCT Transform.
692 * Convert frequency domain coefficients to time-domain audio samples.
693 * reference: Section 7.9.4 Transformation Equations
695 static inline void do_imdct(AC3DecodeContext *s, int channels)
699 for (ch=1; ch<=channels; ch++) {
700 if (s->block_switch[ch]) {
703 s->imdct_512.fft.imdct_calc(&s->imdct_512, s->tmp_output,
704 s->transform_coeffs[ch], s->tmp_imdct);
706 /* For the first half of the block, apply the window, add the delay
707 from the previous block, and send to output */
708 s->dsp.vector_fmul_add_add(s->output[ch-1], s->tmp_output,
709 s->window, s->delay[ch-1], 0, 256, 1);
710 /* For the second half of the block, apply the window and store the
711 samples to delay, to be combined with the next block */
712 s->dsp.vector_fmul_reverse(s->delay[ch-1], s->tmp_output+256,
718 * Downmix the output to mono or stereo.
720 static void ac3_downmix(AC3DecodeContext *s,
721 float samples[AC3_MAX_CHANNELS][256], int ch_offset)
726 for(i=0; i<256; i++) {
728 for(j=0; j<s->fbw_channels; j++) {
729 v0 += samples[j+ch_offset][i] * s->downmix_coeffs[j][0];
730 v1 += samples[j+ch_offset][i] * s->downmix_coeffs[j][1];
732 v0 *= s->downmix_coeff_adjust[0];
733 v1 *= s->downmix_coeff_adjust[1];
734 if(s->output_mode == AC3_CHMODE_MONO) {
735 samples[ch_offset][i] = (v0 + v1) * LEVEL_MINUS_3DB;
736 } else if(s->output_mode == AC3_CHMODE_STEREO) {
737 samples[ ch_offset][i] = v0;
738 samples[1+ch_offset][i] = v1;
744 * Upmix delay samples from stereo to original channel layout.
746 static void ac3_upmix_delay(AC3DecodeContext *s)
748 int channel_data_size = sizeof(s->delay[0]);
749 switch(s->channel_mode) {
750 case AC3_CHMODE_DUALMONO:
751 case AC3_CHMODE_STEREO:
752 /* upmix mono to stereo */
753 memcpy(s->delay[1], s->delay[0], channel_data_size);
755 case AC3_CHMODE_2F2R:
756 memset(s->delay[3], 0, channel_data_size);
757 case AC3_CHMODE_2F1R:
758 memset(s->delay[2], 0, channel_data_size);
760 case AC3_CHMODE_3F2R:
761 memset(s->delay[4], 0, channel_data_size);
762 case AC3_CHMODE_3F1R:
763 memset(s->delay[3], 0, channel_data_size);
765 memcpy(s->delay[2], s->delay[1], channel_data_size);
766 memset(s->delay[1], 0, channel_data_size);
772 * Parse an audio block from AC-3 bitstream.
774 static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
776 int fbw_channels = s->fbw_channels;
777 int channel_mode = s->channel_mode;
779 int different_transforms;
781 GetBitContext *gbc = &s->gbc;
782 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
784 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
786 /* block switch flags */
787 different_transforms = 0;
788 for (ch = 1; ch <= fbw_channels; ch++) {
789 s->block_switch[ch] = get_bits1(gbc);
790 if(ch > 1 && s->block_switch[ch] != s->block_switch[1])
791 different_transforms = 1;
794 /* dithering flags */
796 for (ch = 1; ch <= fbw_channels; ch++) {
797 s->dither_flag[ch] = get_bits1(gbc);
798 if(!s->dither_flag[ch])
803 i = !(s->channel_mode);
806 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
807 s->avctx->drc_scale)+1.0;
808 } else if(blk == 0) {
809 s->dynamic_range[i] = 1.0f;
813 /* coupling strategy */
814 if (get_bits1(gbc)) {
815 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
816 s->cpl_in_use = get_bits1(gbc);
818 /* coupling in use */
819 int cpl_begin_freq, cpl_end_freq;
821 if (channel_mode < AC3_CHMODE_STEREO) {
822 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
826 /* determine which channels are coupled */
827 for (ch = 1; ch <= fbw_channels; ch++)
828 s->channel_in_cpl[ch] = get_bits1(gbc);
830 /* phase flags in use */
831 if (channel_mode == AC3_CHMODE_STEREO)
832 s->phase_flags_in_use = get_bits1(gbc);
834 /* coupling frequency range and band structure */
835 cpl_begin_freq = get_bits(gbc, 4);
836 cpl_end_freq = get_bits(gbc, 4);
837 if (3 + cpl_end_freq - cpl_begin_freq < 0) {
838 av_log(s->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
841 s->num_cpl_bands = s->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
842 s->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
843 s->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
844 for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
845 if (get_bits1(gbc)) {
846 s->cpl_band_struct[bnd] = 1;
850 s->cpl_band_struct[s->num_cpl_subbands-1] = 0;
852 /* coupling not in use */
853 for (ch = 1; ch <= fbw_channels; ch++)
854 s->channel_in_cpl[ch] = 0;
857 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must be present in block 0\n");
861 /* coupling coordinates */
863 int cpl_coords_exist = 0;
865 for (ch = 1; ch <= fbw_channels; ch++) {
866 if (s->channel_in_cpl[ch]) {
867 if (get_bits1(gbc)) {
868 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
869 cpl_coords_exist = 1;
870 master_cpl_coord = 3 * get_bits(gbc, 2);
871 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
872 cpl_coord_exp = get_bits(gbc, 4);
873 cpl_coord_mant = get_bits(gbc, 4);
874 if (cpl_coord_exp == 15)
875 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
877 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
878 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
881 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must be present in block 0\n");
887 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
888 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
889 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
894 /* stereo rematrixing strategy and band structure */
895 if (channel_mode == AC3_CHMODE_STEREO) {
896 if (get_bits1(gbc)) {
897 s->num_rematrixing_bands = 4;
898 if(s->cpl_in_use && s->start_freq[CPL_CH] <= 61)
899 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
900 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
901 s->rematrixing_flags[bnd] = get_bits1(gbc);
903 av_log(s->avctx, AV_LOG_ERROR, "new rematrixing strategy must be present in block 0\n");
908 /* exponent strategies for each channel */
909 s->exp_strategy[CPL_CH] = EXP_REUSE;
910 s->exp_strategy[s->lfe_ch] = EXP_REUSE;
911 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
912 s->exp_strategy[ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
913 if(s->exp_strategy[ch] != EXP_REUSE)
914 bit_alloc_stages[ch] = 3;
917 /* channel bandwidth */
918 for (ch = 1; ch <= fbw_channels; ch++) {
919 s->start_freq[ch] = 0;
920 if (s->exp_strategy[ch] != EXP_REUSE) {
922 int prev = s->end_freq[ch];
923 if (s->channel_in_cpl[ch])
924 s->end_freq[ch] = s->start_freq[CPL_CH];
926 int bandwidth_code = get_bits(gbc, 6);
927 if (bandwidth_code > 60) {
928 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
931 s->end_freq[ch] = bandwidth_code * 3 + 73;
933 group_size = 3 << (s->exp_strategy[ch] - 1);
934 s->num_exp_groups[ch] = (s->end_freq[ch]+group_size-4) / group_size;
935 if(blk > 0 && s->end_freq[ch] != prev)
936 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
939 if (s->cpl_in_use && s->exp_strategy[CPL_CH] != EXP_REUSE) {
940 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
941 (3 << (s->exp_strategy[CPL_CH] - 1));
944 /* decode exponents for each channel */
945 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
946 if (s->exp_strategy[ch] != EXP_REUSE) {
947 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
948 decode_exponents(gbc, s->exp_strategy[ch],
949 s->num_exp_groups[ch], s->dexps[ch][0],
950 &s->dexps[ch][s->start_freq[ch]+!!ch]);
951 if(ch != CPL_CH && ch != s->lfe_ch)
952 skip_bits(gbc, 2); /* skip gainrng */
956 /* bit allocation information */
957 if (get_bits1(gbc)) {
958 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
959 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
960 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
961 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
962 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
963 for(ch=!s->cpl_in_use; ch<=s->channels; ch++)
964 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
966 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must be present in block 0\n");
970 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
971 if (get_bits1(gbc)) {
973 csnr = (get_bits(gbc, 6) - 15) << 4;
974 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
975 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
976 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
978 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
980 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
984 /* coupling leak information */
986 if (get_bits1(gbc)) {
987 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
988 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
989 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
991 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must be present in block 0\n");
996 /* delta bit allocation information */
997 if (get_bits1(gbc)) {
998 /* delta bit allocation exists (strategy) */
999 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
1000 s->dba_mode[ch] = get_bits(gbc, 2);
1001 if (s->dba_mode[ch] == DBA_RESERVED) {
1002 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1005 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1007 /* channel delta offset, len and bit allocation */
1008 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
1009 if (s->dba_mode[ch] == DBA_NEW) {
1010 s->dba_nsegs[ch] = get_bits(gbc, 3);
1011 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
1012 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1013 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1014 s->dba_values[ch][seg] = get_bits(gbc, 3);
1016 /* run last 2 bit allocation stages if new dba values */
1017 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1020 } else if(blk == 0) {
1021 for(ch=0; ch<=s->channels; ch++) {
1022 s->dba_mode[ch] = DBA_NONE;
1026 /* Bit allocation */
1027 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
1028 if(bit_alloc_stages[ch] > 2) {
1029 /* Exponent mapping into PSD and PSD integration */
1030 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1031 s->start_freq[ch], s->end_freq[ch],
1032 s->psd[ch], s->band_psd[ch]);
1034 if(bit_alloc_stages[ch] > 1) {
1035 /* Compute excitation function, Compute masking curve, and
1036 Apply delta bit allocation */
1037 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1038 s->start_freq[ch], s->end_freq[ch],
1039 s->fast_gain[ch], (ch == s->lfe_ch),
1040 s->dba_mode[ch], s->dba_nsegs[ch],
1041 s->dba_offsets[ch], s->dba_lengths[ch],
1042 s->dba_values[ch], s->mask[ch]);
1044 if(bit_alloc_stages[ch] > 0) {
1045 /* Compute bit allocation */
1046 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1047 s->start_freq[ch], s->end_freq[ch],
1049 s->bit_alloc_params.floor,
1054 /* unused dummy data */
1055 if (get_bits1(gbc)) {
1056 int skipl = get_bits(gbc, 9);
1061 /* unpack the transform coefficients
1062 this also uncouples channels if coupling is in use. */
1063 get_transform_coeffs(s);
1065 /* recover coefficients if rematrixing is in use */
1066 if(s->channel_mode == AC3_CHMODE_STEREO)
1069 /* apply scaling to coefficients (headroom, dynrng) */
1070 for(ch=1; ch<=s->channels; ch++) {
1071 float gain = s->mul_bias / 4194304.0f;
1072 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1073 gain *= s->dynamic_range[ch-1];
1075 gain *= s->dynamic_range[0];
1077 for(i=0; i<256; i++) {
1078 s->transform_coeffs[ch][i] = s->fixed_coeffs[ch][i] * gain;
1082 /* downmix and MDCT. order depends on whether block switching is used for
1083 any channel in this block. this is because coefficients for the long
1084 and short transforms cannot be mixed. */
1085 downmix_output = s->channels != s->out_channels &&
1086 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1087 s->fbw_channels == s->out_channels);
1088 if(different_transforms) {
1089 /* the delay samples have already been downmixed, so we upmix the delay
1090 samples in order to reconstruct all channels before downmixing. */
1096 do_imdct(s, s->channels);
1098 if(downmix_output) {
1099 ac3_downmix(s, s->output, 0);
1102 if(downmix_output) {
1103 ac3_downmix(s, s->transform_coeffs, 1);
1108 ac3_downmix(s, s->delay, 0);
1111 do_imdct(s, s->out_channels);
1114 /* convert float to 16-bit integer */
1115 for(ch=0; ch<s->out_channels; ch++) {
1116 for(i=0; i<256; i++) {
1117 s->output[ch][i] += s->add_bias;
1119 s->dsp.float_to_int16(s->int_output[ch], s->output[ch], 256);
1126 * Decode a single AC-3 frame.
1128 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
1129 const uint8_t *buf, int buf_size)
1131 AC3DecodeContext *s = avctx->priv_data;
1132 int16_t *out_samples = (int16_t *)data;
1133 int i, blk, ch, err;
1135 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1136 if (s->input_buffer) {
1137 /* copy input buffer to decoder context to avoid reading past the end
1138 of the buffer, which can be caused by a damaged input stream. */
1139 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_MAX_FRAME_SIZE));
1140 init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
1142 init_get_bits(&s->gbc, buf, buf_size * 8);
1145 /* parse the syncinfo */
1147 err = ac3_parse_header(s);
1149 /* check that reported frame size fits in input buffer */
1150 if(s->frame_size > buf_size) {
1151 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1152 err = AC3_PARSE_ERROR_FRAME_SIZE;
1155 /* check for crc mismatch */
1156 if(err != AC3_PARSE_ERROR_FRAME_SIZE && avctx->error_resilience >= FF_ER_CAREFUL) {
1157 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1158 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1159 err = AC3_PARSE_ERROR_CRC;
1163 if(err && err != AC3_PARSE_ERROR_CRC) {
1165 case AC3_PARSE_ERROR_SYNC:
1166 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1168 case AC3_PARSE_ERROR_BSID:
1169 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1171 case AC3_PARSE_ERROR_SAMPLE_RATE:
1172 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1174 case AC3_PARSE_ERROR_FRAME_SIZE:
1175 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1177 case AC3_PARSE_ERROR_FRAME_TYPE:
1178 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1181 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1186 /* if frame is ok, set audio parameters */
1188 avctx->sample_rate = s->sample_rate;
1189 avctx->bit_rate = s->bit_rate;
1191 /* channel config */
1192 s->out_channels = s->channels;
1193 s->output_mode = s->channel_mode;
1195 s->output_mode |= AC3_OUTPUT_LFEON;
1196 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1197 avctx->request_channels < s->channels) {
1198 s->out_channels = avctx->request_channels;
1199 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1201 avctx->channels = s->out_channels;
1203 /* set downmixing coefficients if needed */
1204 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1205 s->fbw_channels == s->out_channels)) {
1206 set_downmix_coeffs(s);
1208 } else if (!s->out_channels) {
1209 s->out_channels = avctx->channels;
1210 if(s->out_channels < s->channels)
1211 s->output_mode = s->out_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1214 /* parse the audio blocks */
1215 for (blk = 0; blk < s->num_blocks; blk++) {
1216 if (!err && ac3_parse_audio_block(s, blk)) {
1217 av_log(avctx, AV_LOG_ERROR, "error parsing the audio block\n");
1220 /* interleave output samples */
1221 for (i = 0; i < 256; i++)
1222 for (ch = 0; ch < s->out_channels; ch++)
1223 *(out_samples++) = s->int_output[ch][i];
1225 *data_size = s->num_blocks * 256 * avctx->channels * sizeof (int16_t);
1226 return s->frame_size;
1230 * Uninitialize the AC-3 decoder.
1232 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1234 AC3DecodeContext *s = avctx->priv_data;
1235 ff_mdct_end(&s->imdct_512);
1236 ff_mdct_end(&s->imdct_256);
1238 av_freep(&s->input_buffer);
1243 AVCodec ac3_decoder = {
1245 .type = CODEC_TYPE_AUDIO,
1247 .priv_data_size = sizeof (AC3DecodeContext),
1248 .init = ac3_decode_init,
1249 .close = ac3_decode_end,
1250 .decode = ac3_decode_frame,
1251 .long_name = "ATSC A/52 / AC-3",