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
36 #include "ac3_parser.h"
37 #include "bitstream.h"
43 * Table of bin locations for rematrixing bands
44 * reference: Section 7.5.2 Rematrixing : Frequency Band Definitions
46 static const uint8_t rematrix_band_tab[5] = { 13, 25, 37, 61, 253 };
49 * table for exponent to scale_factor mapping
50 * scale_factors[i] = 2 ^ -i
52 static float scale_factors[25];
54 /** table for grouping exponents */
55 static uint8_t exp_ungroup_tab[128][3];
58 /** tables for ungrouping mantissas */
59 static float b1_mantissas[32][3];
60 static float b2_mantissas[128][3];
61 static float b3_mantissas[8];
62 static float b4_mantissas[128][2];
63 static float b5_mantissas[16];
66 * Quantization table: levels for symmetric. bits for asymmetric.
67 * reference: Table 7.18 Mapping of bap to Quantizer
69 static const uint8_t quantization_tab[16] = {
71 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
74 /** dynamic range table. converts codes to scale factors. */
75 static float dynamic_range_tab[256];
77 /** Adjustments in dB gain */
78 #define LEVEL_MINUS_3DB 0.7071067811865476
79 #define LEVEL_MINUS_4POINT5DB 0.5946035575013605
80 #define LEVEL_MINUS_6DB 0.5000000000000000
81 #define LEVEL_MINUS_9DB 0.3535533905932738
82 #define LEVEL_ZERO 0.0000000000000000
83 #define LEVEL_ONE 1.0000000000000000
85 static const float gain_levels[6] = {
89 LEVEL_MINUS_4POINT5DB,
95 * Table for center mix levels
96 * reference: Section 5.4.2.4 cmixlev
98 static const uint8_t center_levels[4] = { 2, 3, 4, 3 };
101 * Table for surround mix levels
102 * reference: Section 5.4.2.5 surmixlev
104 static const uint8_t surround_levels[4] = { 2, 4, 0, 4 };
107 * Table for default stereo downmixing coefficients
108 * reference: Section 7.8.2 Downmixing Into Two Channels
110 static const uint8_t ac3_default_coeffs[8][5][2] = {
111 { { 1, 0 }, { 0, 1 }, },
113 { { 1, 0 }, { 0, 1 }, },
114 { { 1, 0 }, { 3, 3 }, { 0, 1 }, },
115 { { 1, 0 }, { 0, 1 }, { 4, 4 }, },
116 { { 1, 0 }, { 3, 3 }, { 0, 1 }, { 5, 5 }, },
117 { { 1, 0 }, { 0, 1 }, { 4, 0 }, { 0, 4 }, },
118 { { 1, 0 }, { 3, 3 }, { 0, 1 }, { 4, 0 }, { 0, 4 }, },
121 /* override ac3.h to include coupling channel */
122 #undef AC3_MAX_CHANNELS
123 #define AC3_MAX_CHANNELS 7
126 #define AC3_OUTPUT_LFEON 8
129 int channel_mode; ///< channel mode (acmod)
130 int block_switch[AC3_MAX_CHANNELS]; ///< block switch flags
131 int dither_flag[AC3_MAX_CHANNELS]; ///< dither flags
132 int dither_all; ///< true if all channels are dithered
133 int cpl_in_use; ///< coupling in use
134 int channel_in_cpl[AC3_MAX_CHANNELS]; ///< channel in coupling
135 int phase_flags_in_use; ///< phase flags in use
136 int cpl_band_struct[18]; ///< coupling band structure
137 int rematrixing_strategy; ///< rematrixing strategy
138 int num_rematrixing_bands; ///< number of rematrixing bands
139 int rematrixing_flags[4]; ///< rematrixing flags
140 int exp_strategy[AC3_MAX_CHANNELS]; ///< exponent strategies
141 int snr_offset[AC3_MAX_CHANNELS]; ///< signal-to-noise ratio offsets
142 int fast_gain[AC3_MAX_CHANNELS]; ///< fast gain values (signal-to-mask ratio)
143 int dba_mode[AC3_MAX_CHANNELS]; ///< delta bit allocation mode
144 int dba_nsegs[AC3_MAX_CHANNELS]; ///< number of delta segments
145 uint8_t dba_offsets[AC3_MAX_CHANNELS][8]; ///< delta segment offsets
146 uint8_t dba_lengths[AC3_MAX_CHANNELS][8]; ///< delta segment lengths
147 uint8_t dba_values[AC3_MAX_CHANNELS][8]; ///< delta values for each segment
149 int sample_rate; ///< sample frequency, in Hz
150 int bit_rate; ///< stream bit rate, in bits-per-second
151 int frame_size; ///< current frame size, in bytes
153 int channels; ///< number of total channels
154 int fbw_channels; ///< number of full-bandwidth channels
155 int lfe_on; ///< lfe channel in use
156 int lfe_ch; ///< index of LFE channel
157 int output_mode; ///< output channel configuration
158 int out_channels; ///< number of output channels
160 float downmix_coeffs[AC3_MAX_CHANNELS][2]; ///< stereo downmix coefficients
161 float dynamic_range[2]; ///< dynamic range
162 float cpl_coords[AC3_MAX_CHANNELS][18]; ///< coupling coordinates
163 int num_cpl_bands; ///< number of coupling bands
164 int num_cpl_subbands; ///< number of coupling sub bands
165 int start_freq[AC3_MAX_CHANNELS]; ///< start frequency bin
166 int end_freq[AC3_MAX_CHANNELS]; ///< end frequency bin
167 AC3BitAllocParameters bit_alloc_params; ///< bit allocation parameters
169 int8_t dexps[AC3_MAX_CHANNELS][256]; ///< decoded exponents
170 uint8_t bap[AC3_MAX_CHANNELS][256]; ///< bit allocation pointers
171 int16_t psd[AC3_MAX_CHANNELS][256]; ///< scaled exponents
172 int16_t band_psd[AC3_MAX_CHANNELS][50]; ///< interpolated exponents
173 int16_t mask[AC3_MAX_CHANNELS][50]; ///< masking curve values
175 DECLARE_ALIGNED_16(float, transform_coeffs[AC3_MAX_CHANNELS][256]); ///< transform coefficients
178 MDCTContext imdct_512; ///< for 512 sample IMDCT
179 MDCTContext imdct_256; ///< for 256 sample IMDCT
180 DSPContext dsp; ///< for optimization
181 float add_bias; ///< offset for float_to_int16 conversion
182 float mul_bias; ///< scaling for float_to_int16 conversion
184 DECLARE_ALIGNED_16(float, output[AC3_MAX_CHANNELS-1][256]); ///< output after imdct transform and windowing
185 DECLARE_ALIGNED_16(short, int_output[AC3_MAX_CHANNELS-1][256]); ///< final 16-bit integer output
186 DECLARE_ALIGNED_16(float, delay[AC3_MAX_CHANNELS-1][256]); ///< delay - added to the next block
187 DECLARE_ALIGNED_16(float, tmp_imdct[256]); ///< temporary storage for imdct transform
188 DECLARE_ALIGNED_16(float, tmp_output[512]); ///< temporary storage for output before windowing
189 DECLARE_ALIGNED_16(float, window[256]); ///< window coefficients
192 GetBitContext gbc; ///< bitstream reader
193 AVRandomState dith_state; ///< for dither generation
194 AVCodecContext *avctx; ///< parent context
198 * Generate a Kaiser-Bessel Derived Window.
200 static void ac3_window_init(float *window)
203 double sum = 0.0, bessel, tmp;
204 double local_window[256];
205 double alpha2 = (5.0 * M_PI / 256.0) * (5.0 * M_PI / 256.0);
207 for (i = 0; i < 256; i++) {
208 tmp = i * (256 - i) * alpha2;
210 for (j = 100; j > 0; j--) /* default to 100 iterations */
211 bessel = bessel * tmp / (j * j) + 1;
213 local_window[i] = sum;
217 for (i = 0; i < 256; i++)
218 window[i] = sqrt(local_window[i] / sum);
222 * Symmetrical Dequantization
223 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
224 * Tables 7.19 to 7.23
227 symmetric_dequant(int code, int levels)
229 return (code - (levels >> 1)) * (2.0f / levels);
233 * Initialize tables at runtime.
235 static void ac3_tables_init(void)
239 /* generate grouped mantissa tables
240 reference: Section 7.3.5 Ungrouping of Mantissas */
241 for(i=0; i<32; i++) {
242 /* bap=1 mantissas */
243 b1_mantissas[i][0] = symmetric_dequant( i / 9 , 3);
244 b1_mantissas[i][1] = symmetric_dequant((i % 9) / 3, 3);
245 b1_mantissas[i][2] = symmetric_dequant((i % 9) % 3, 3);
247 for(i=0; i<128; i++) {
248 /* bap=2 mantissas */
249 b2_mantissas[i][0] = symmetric_dequant( i / 25 , 5);
250 b2_mantissas[i][1] = symmetric_dequant((i % 25) / 5, 5);
251 b2_mantissas[i][2] = symmetric_dequant((i % 25) % 5, 5);
253 /* bap=4 mantissas */
254 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
255 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
257 /* generate ungrouped mantissa tables
258 reference: Tables 7.21 and 7.23 */
260 /* bap=3 mantissas */
261 b3_mantissas[i] = symmetric_dequant(i, 7);
263 for(i=0; i<15; i++) {
264 /* bap=5 mantissas */
265 b5_mantissas[i] = symmetric_dequant(i, 15);
268 /* generate dynamic range table
269 reference: Section 7.7.1 Dynamic Range Control */
270 for(i=0; i<256; i++) {
271 int v = (i >> 5) - ((i >> 7) << 3) - 5;
272 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
275 /* generate scale factors for exponents and asymmetrical dequantization
276 reference: Section 7.3.2 Expansion of Mantissas for Asymmetric Quantization */
277 for (i = 0; i < 25; i++)
278 scale_factors[i] = pow(2.0, -i);
280 /* generate exponent tables
281 reference: Section 7.1.3 Exponent Decoding */
282 for(i=0; i<128; i++) {
283 exp_ungroup_tab[i][0] = i / 25;
284 exp_ungroup_tab[i][1] = (i % 25) / 5;
285 exp_ungroup_tab[i][2] = (i % 25) % 5;
291 * AVCodec initialization
293 static int ac3_decode_init(AVCodecContext *avctx)
295 AC3DecodeContext *s = avctx->priv_data;
300 ff_mdct_init(&s->imdct_256, 8, 1);
301 ff_mdct_init(&s->imdct_512, 9, 1);
302 ac3_window_init(s->window);
303 dsputil_init(&s->dsp, avctx);
304 av_init_random(0, &s->dith_state);
306 /* set bias values for float to int16 conversion */
307 if(s->dsp.float_to_int16 == ff_float_to_int16_c) {
308 s->add_bias = 385.0f;
312 s->mul_bias = 32767.0f;
319 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
320 * GetBitContext within AC3DecodeContext must point to
321 * start of the synchronized ac3 bitstream.
323 static int ac3_parse_header(AC3DecodeContext *s)
326 GetBitContext *gbc = &s->gbc;
327 float center_mix_level, surround_mix_level;
330 err = ff_ac3_parse_header(gbc->buffer, &hdr);
334 /* get decoding parameters from header info */
335 s->bit_alloc_params.sr_code = hdr.sr_code;
336 s->channel_mode = hdr.channel_mode;
337 center_mix_level = gain_levels[center_levels[hdr.center_mix_level]];
338 surround_mix_level = gain_levels[surround_levels[hdr.surround_mix_level]];
339 s->lfe_on = hdr.lfe_on;
340 s->bit_alloc_params.sr_shift = hdr.sr_shift;
341 s->sample_rate = hdr.sample_rate;
342 s->bit_rate = hdr.bit_rate;
343 s->channels = hdr.channels;
344 s->fbw_channels = s->channels - s->lfe_on;
345 s->lfe_ch = s->fbw_channels + 1;
346 s->frame_size = hdr.frame_size;
348 /* set default output to all source channels */
349 s->out_channels = s->channels;
350 s->output_mode = s->channel_mode;
352 s->output_mode |= AC3_OUTPUT_LFEON;
354 /* skip over portion of header which has already been read */
355 skip_bits(gbc, 16); // skip the sync_word
356 skip_bits(gbc, 16); // skip crc1
357 skip_bits(gbc, 8); // skip fscod and frmsizecod
358 skip_bits(gbc, 11); // skip bsid, bsmod, and acmod
359 if(s->channel_mode == AC3_CHMODE_STEREO) {
360 skip_bits(gbc, 2); // skip dsurmod
362 if((s->channel_mode & 1) && s->channel_mode != AC3_CHMODE_MONO)
363 skip_bits(gbc, 2); // skip cmixlev
364 if(s->channel_mode & 4)
365 skip_bits(gbc, 2); // skip surmixlev
367 skip_bits1(gbc); // skip lfeon
369 /* read the rest of the bsi. read twice for dual mono mode. */
370 i = !(s->channel_mode);
372 skip_bits(gbc, 5); // skip dialog normalization
374 skip_bits(gbc, 8); //skip compression
376 skip_bits(gbc, 8); //skip language code
378 skip_bits(gbc, 7); //skip audio production information
381 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
383 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
384 TODO: read & use the xbsi1 downmix levels */
386 skip_bits(gbc, 14); //skip timecode1 / xbsi1
388 skip_bits(gbc, 14); //skip timecode2 / xbsi2
390 /* skip additional bitstream info */
391 if (get_bits1(gbc)) {
392 i = get_bits(gbc, 6);
398 /* set stereo downmixing coefficients
399 reference: Section 7.8.2 Downmixing Into Two Channels */
400 for(i=0; i<s->fbw_channels; i++) {
401 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
402 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
404 if(s->channel_mode > 1 && s->channel_mode & 1) {
405 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = center_mix_level;
407 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
408 int nf = s->channel_mode - 2;
409 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = surround_mix_level * LEVEL_MINUS_3DB;
411 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
412 int nf = s->channel_mode - 4;
413 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = surround_mix_level;
420 * Decode the grouped exponents according to exponent strategy.
421 * reference: Section 7.1.3 Exponent Decoding
423 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
424 uint8_t absexp, int8_t *dexps)
426 int i, j, grp, group_size;
431 group_size = exp_strategy + (exp_strategy == EXP_D45);
432 for(grp=0,i=0; grp<ngrps; grp++) {
433 expacc = get_bits(gbc, 7);
434 dexp[i++] = exp_ungroup_tab[expacc][0];
435 dexp[i++] = exp_ungroup_tab[expacc][1];
436 dexp[i++] = exp_ungroup_tab[expacc][2];
439 /* convert to absolute exps and expand groups */
441 for(i=0; i<ngrps*3; i++) {
442 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
443 for(j=0; j<group_size; j++) {
444 dexps[(i*group_size)+j] = prevexp;
450 * Generate transform coefficients for each coupled channel in the coupling
451 * range using the coupling coefficients and coupling coordinates.
452 * reference: Section 7.4.3 Coupling Coordinate Format
454 static void uncouple_channels(AC3DecodeContext *s)
456 int i, j, ch, bnd, subbnd;
459 i = s->start_freq[CPL_CH];
460 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
463 for(j=0; j<12; j++) {
464 for(ch=1; ch<=s->fbw_channels; ch++) {
465 if(s->channel_in_cpl[ch])
466 s->transform_coeffs[ch][i] = s->transform_coeffs[CPL_CH][i] * s->cpl_coords[ch][bnd] * 8.0f;
470 } while(s->cpl_band_struct[subbnd]);
475 * Grouped mantissas for 3-level 5-level and 11-level quantization
487 * Get the transform coefficients for a particular channel
488 * reference: Section 7.3 Quantization and Decoding of Mantissas
490 static int get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
492 GetBitContext *gbc = &s->gbc;
493 int i, gcode, tbap, start, end;
498 exps = s->dexps[ch_index];
499 bap = s->bap[ch_index];
500 coeffs = s->transform_coeffs[ch_index];
501 start = s->start_freq[ch_index];
502 end = s->end_freq[ch_index];
504 for (i = start; i < end; i++) {
508 coeffs[i] = ((av_random(&s->dith_state) & 0xFFFF) / 65535.0f) - 0.5f;
513 gcode = get_bits(gbc, 5);
514 m->b1_mant[0] = b1_mantissas[gcode][0];
515 m->b1_mant[1] = b1_mantissas[gcode][1];
516 m->b1_mant[2] = b1_mantissas[gcode][2];
519 coeffs[i] = m->b1_mant[m->b1ptr++];
524 gcode = get_bits(gbc, 7);
525 m->b2_mant[0] = b2_mantissas[gcode][0];
526 m->b2_mant[1] = b2_mantissas[gcode][1];
527 m->b2_mant[2] = b2_mantissas[gcode][2];
530 coeffs[i] = m->b2_mant[m->b2ptr++];
534 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
539 gcode = get_bits(gbc, 7);
540 m->b4_mant[0] = b4_mantissas[gcode][0];
541 m->b4_mant[1] = b4_mantissas[gcode][1];
544 coeffs[i] = m->b4_mant[m->b4ptr++];
548 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
552 /* asymmetric dequantization */
553 coeffs[i] = get_sbits(gbc, quantization_tab[tbap]) * scale_factors[quantization_tab[tbap]-1];
556 coeffs[i] *= scale_factors[exps[i]];
563 * Remove random dithering from coefficients with zero-bit mantissas
564 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
566 static void remove_dithering(AC3DecodeContext *s) {
572 for(ch=1; ch<=s->fbw_channels; ch++) {
573 if(!s->dither_flag[ch]) {
574 coeffs = s->transform_coeffs[ch];
576 if(s->channel_in_cpl[ch])
577 end = s->start_freq[CPL_CH];
579 end = s->end_freq[ch];
580 for(i=0; i<end; i++) {
584 if(s->channel_in_cpl[ch]) {
585 bap = s->bap[CPL_CH];
586 for(; i<s->end_freq[CPL_CH]; i++) {
596 * Get the transform coefficients.
598 static int get_transform_coeffs(AC3DecodeContext *s)
604 m.b1ptr = m.b2ptr = m.b4ptr = 3;
606 for (ch = 1; ch <= s->channels; ch++) {
607 /* transform coefficients for full-bandwidth channel */
608 if (get_transform_coeffs_ch(s, ch, &m))
610 /* tranform coefficients for coupling channel come right after the
611 coefficients for the first coupled channel*/
612 if (s->channel_in_cpl[ch]) {
614 if (get_transform_coeffs_ch(s, CPL_CH, &m)) {
615 av_log(s->avctx, AV_LOG_ERROR, "error in decoupling channels\n");
618 uncouple_channels(s);
621 end = s->end_freq[CPL_CH];
623 end = s->end_freq[ch];
626 s->transform_coeffs[ch][end] = 0;
630 /* if any channel doesn't use dithering, zero appropriate coefficients */
638 * Stereo rematrixing.
639 * reference: Section 7.5.4 Rematrixing : Decoding Technique
641 static void do_rematrixing(AC3DecodeContext *s)
647 end = FFMIN(s->end_freq[1], s->end_freq[2]);
649 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
650 if(s->rematrixing_flags[bnd]) {
651 bndend = FFMIN(end, rematrix_band_tab[bnd+1]);
652 for(i=rematrix_band_tab[bnd]; i<bndend; i++) {
653 tmp0 = s->transform_coeffs[1][i];
654 tmp1 = s->transform_coeffs[2][i];
655 s->transform_coeffs[1][i] = tmp0 + tmp1;
656 s->transform_coeffs[2][i] = tmp0 - tmp1;
663 * Perform the 256-point IMDCT
665 static void do_imdct_256(AC3DecodeContext *s, int chindex)
668 DECLARE_ALIGNED_16(float, x[128]);
670 float *o_ptr = s->tmp_output;
673 /* de-interleave coefficients */
674 for(k=0; k<128; k++) {
675 x[k] = s->transform_coeffs[chindex][2*k+i];
678 /* run standard IMDCT */
679 s->imdct_256.fft.imdct_calc(&s->imdct_256, o_ptr, x, s->tmp_imdct);
681 /* reverse the post-rotation & reordering from standard IMDCT */
682 for(k=0; k<32; k++) {
683 z[i][32+k].re = -o_ptr[128+2*k];
684 z[i][32+k].im = -o_ptr[2*k];
685 z[i][31-k].re = o_ptr[2*k+1];
686 z[i][31-k].im = o_ptr[128+2*k+1];
690 /* apply AC-3 post-rotation & reordering */
691 for(k=0; k<64; k++) {
692 o_ptr[ 2*k ] = -z[0][ k].im;
693 o_ptr[ 2*k+1] = z[0][63-k].re;
694 o_ptr[128+2*k ] = -z[0][ k].re;
695 o_ptr[128+2*k+1] = z[0][63-k].im;
696 o_ptr[256+2*k ] = -z[1][ k].re;
697 o_ptr[256+2*k+1] = z[1][63-k].im;
698 o_ptr[384+2*k ] = z[1][ k].im;
699 o_ptr[384+2*k+1] = -z[1][63-k].re;
704 * Inverse MDCT Transform.
705 * Convert frequency domain coefficients to time-domain audio samples.
706 * reference: Section 7.9.4 Transformation Equations
708 static inline void do_imdct(AC3DecodeContext *s)
713 /* Don't perform the IMDCT on the LFE channel unless it's used in the output */
714 channels = s->fbw_channels;
715 if(s->output_mode & AC3_OUTPUT_LFEON)
718 for (ch=1; ch<=channels; ch++) {
719 if (s->block_switch[ch]) {
722 s->imdct_512.fft.imdct_calc(&s->imdct_512, s->tmp_output,
723 s->transform_coeffs[ch], s->tmp_imdct);
725 /* For the first half of the block, apply the window, add the delay
726 from the previous block, and send to output */
727 s->dsp.vector_fmul_add_add(s->output[ch-1], s->tmp_output,
728 s->window, s->delay[ch-1], 0, 256, 1);
729 /* For the second half of the block, apply the window and store the
730 samples to delay, to be combined with the next block */
731 s->dsp.vector_fmul_reverse(s->delay[ch-1], s->tmp_output+256,
737 * Downmix the output to mono or stereo.
739 static void ac3_downmix(AC3DecodeContext *s)
742 float v0, v1, s0, s1;
744 for(i=0; i<256; i++) {
745 v0 = v1 = s0 = s1 = 0.0f;
746 for(j=0; j<s->fbw_channels; j++) {
747 v0 += s->output[j][i] * s->downmix_coeffs[j][0];
748 v1 += s->output[j][i] * s->downmix_coeffs[j][1];
749 s0 += s->downmix_coeffs[j][0];
750 s1 += s->downmix_coeffs[j][1];
754 if(s->output_mode == AC3_CHMODE_MONO) {
755 s->output[0][i] = (v0 + v1) * LEVEL_MINUS_3DB;
756 } else if(s->output_mode == AC3_CHMODE_STEREO) {
757 s->output[0][i] = v0;
758 s->output[1][i] = v1;
764 * Parse an audio block from AC-3 bitstream.
766 static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
768 int fbw_channels = s->fbw_channels;
769 int channel_mode = s->channel_mode;
771 GetBitContext *gbc = &s->gbc;
772 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
774 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
776 /* block switch flags */
777 for (ch = 1; ch <= fbw_channels; ch++)
778 s->block_switch[ch] = get_bits1(gbc);
780 /* dithering flags */
782 for (ch = 1; ch <= fbw_channels; ch++) {
783 s->dither_flag[ch] = get_bits1(gbc);
784 if(!s->dither_flag[ch])
789 i = !(s->channel_mode);
792 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
793 s->avctx->drc_scale)+1.0;
794 } else if(blk == 0) {
795 s->dynamic_range[i] = 1.0f;
799 /* coupling strategy */
800 if (get_bits1(gbc)) {
801 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
802 s->cpl_in_use = get_bits1(gbc);
804 /* coupling in use */
805 int cpl_begin_freq, cpl_end_freq;
807 /* determine which channels are coupled */
808 for (ch = 1; ch <= fbw_channels; ch++)
809 s->channel_in_cpl[ch] = get_bits1(gbc);
811 /* phase flags in use */
812 if (channel_mode == AC3_CHMODE_STEREO)
813 s->phase_flags_in_use = get_bits1(gbc);
815 /* coupling frequency range and band structure */
816 cpl_begin_freq = get_bits(gbc, 4);
817 cpl_end_freq = get_bits(gbc, 4);
818 if (3 + cpl_end_freq - cpl_begin_freq < 0) {
819 av_log(s->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
822 s->num_cpl_bands = s->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
823 s->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
824 s->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
825 for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
826 if (get_bits1(gbc)) {
827 s->cpl_band_struct[bnd] = 1;
832 /* coupling not in use */
833 for (ch = 1; ch <= fbw_channels; ch++)
834 s->channel_in_cpl[ch] = 0;
838 /* coupling coordinates */
840 int cpl_coords_exist = 0;
842 for (ch = 1; ch <= fbw_channels; ch++) {
843 if (s->channel_in_cpl[ch]) {
844 if (get_bits1(gbc)) {
845 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
846 cpl_coords_exist = 1;
847 master_cpl_coord = 3 * get_bits(gbc, 2);
848 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
849 cpl_coord_exp = get_bits(gbc, 4);
850 cpl_coord_mant = get_bits(gbc, 4);
851 if (cpl_coord_exp == 15)
852 s->cpl_coords[ch][bnd] = cpl_coord_mant / 16.0f;
854 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16.0f) / 32.0f;
855 s->cpl_coords[ch][bnd] *= scale_factors[cpl_coord_exp + master_cpl_coord];
861 if (channel_mode == AC3_CHMODE_STEREO && s->phase_flags_in_use && cpl_coords_exist) {
862 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
864 s->cpl_coords[2][bnd] = -s->cpl_coords[2][bnd];
869 /* stereo rematrixing strategy and band structure */
870 if (channel_mode == AC3_CHMODE_STEREO) {
871 s->rematrixing_strategy = get_bits1(gbc);
872 if (s->rematrixing_strategy) {
873 s->num_rematrixing_bands = 4;
874 if(s->cpl_in_use && s->start_freq[CPL_CH] <= 61)
875 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
876 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
877 s->rematrixing_flags[bnd] = get_bits1(gbc);
881 /* exponent strategies for each channel */
882 s->exp_strategy[CPL_CH] = EXP_REUSE;
883 s->exp_strategy[s->lfe_ch] = EXP_REUSE;
884 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
886 s->exp_strategy[ch] = get_bits(gbc, 1);
888 s->exp_strategy[ch] = get_bits(gbc, 2);
889 if(s->exp_strategy[ch] != EXP_REUSE)
890 bit_alloc_stages[ch] = 3;
893 /* channel bandwidth */
894 for (ch = 1; ch <= fbw_channels; ch++) {
895 s->start_freq[ch] = 0;
896 if (s->exp_strategy[ch] != EXP_REUSE) {
897 int prev = s->end_freq[ch];
898 if (s->channel_in_cpl[ch])
899 s->end_freq[ch] = s->start_freq[CPL_CH];
901 int bandwidth_code = get_bits(gbc, 6);
902 if (bandwidth_code > 60) {
903 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
906 s->end_freq[ch] = bandwidth_code * 3 + 73;
908 if(blk > 0 && s->end_freq[ch] != prev)
909 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
912 s->start_freq[s->lfe_ch] = 0;
913 s->end_freq[s->lfe_ch] = 7;
915 /* decode exponents for each channel */
916 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
917 if (s->exp_strategy[ch] != EXP_REUSE) {
918 int group_size, num_groups;
919 group_size = 3 << (s->exp_strategy[ch] - 1);
921 num_groups = (s->end_freq[ch] - s->start_freq[ch]) / group_size;
922 else if(ch == s->lfe_ch)
925 num_groups = (s->end_freq[ch] + group_size - 4) / group_size;
926 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
927 decode_exponents(gbc, s->exp_strategy[ch], num_groups, s->dexps[ch][0],
928 &s->dexps[ch][s->start_freq[ch]+!!ch]);
929 if(ch != CPL_CH && ch != s->lfe_ch)
930 skip_bits(gbc, 2); /* skip gainrng */
934 /* bit allocation information */
935 if (get_bits1(gbc)) {
936 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
937 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
938 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
939 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
940 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
941 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
942 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
946 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
947 if (get_bits1(gbc)) {
949 csnr = (get_bits(gbc, 6) - 15) << 4;
950 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
951 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
952 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
954 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
957 /* coupling leak information */
958 if (s->cpl_in_use && get_bits1(gbc)) {
959 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
960 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
961 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
964 /* delta bit allocation information */
965 if (get_bits1(gbc)) {
966 /* delta bit allocation exists (strategy) */
967 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
968 s->dba_mode[ch] = get_bits(gbc, 2);
969 if (s->dba_mode[ch] == DBA_RESERVED) {
970 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
973 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
975 /* channel delta offset, len and bit allocation */
976 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
977 if (s->dba_mode[ch] == DBA_NEW) {
978 s->dba_nsegs[ch] = get_bits(gbc, 3);
979 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
980 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
981 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
982 s->dba_values[ch][seg] = get_bits(gbc, 3);
986 } else if(blk == 0) {
987 for(ch=0; ch<=s->channels; ch++) {
988 s->dba_mode[ch] = DBA_NONE;
993 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
994 if(bit_alloc_stages[ch] > 2) {
995 /* Exponent mapping into PSD and PSD integration */
996 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
997 s->start_freq[ch], s->end_freq[ch],
998 s->psd[ch], s->band_psd[ch]);
1000 if(bit_alloc_stages[ch] > 1) {
1001 /* Compute excitation function, Compute masking curve, and
1002 Apply delta bit allocation */
1003 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1004 s->start_freq[ch], s->end_freq[ch],
1005 s->fast_gain[ch], (ch == s->lfe_ch),
1006 s->dba_mode[ch], s->dba_nsegs[ch],
1007 s->dba_offsets[ch], s->dba_lengths[ch],
1008 s->dba_values[ch], s->mask[ch]);
1010 if(bit_alloc_stages[ch] > 0) {
1011 /* Compute bit allocation */
1012 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1013 s->start_freq[ch], s->end_freq[ch],
1015 s->bit_alloc_params.floor,
1020 /* unused dummy data */
1021 if (get_bits1(gbc)) {
1022 int skipl = get_bits(gbc, 9);
1027 /* unpack the transform coefficients
1028 this also uncouples channels if coupling is in use. */
1029 if (get_transform_coeffs(s)) {
1030 av_log(s->avctx, AV_LOG_ERROR, "Error in routine get_transform_coeffs\n");
1034 /* recover coefficients if rematrixing is in use */
1035 if(s->channel_mode == AC3_CHMODE_STEREO)
1038 /* apply scaling to coefficients (headroom, dynrng) */
1039 for(ch=1; ch<=s->channels; ch++) {
1040 float gain = 2.0f * s->mul_bias;
1041 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1042 gain *= s->dynamic_range[ch-1];
1044 gain *= s->dynamic_range[0];
1046 for(i=0; i<s->end_freq[ch]; i++) {
1047 s->transform_coeffs[ch][i] *= gain;
1053 /* downmix output if needed */
1054 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1055 s->fbw_channels == s->out_channels)) {
1059 /* convert float to 16-bit integer */
1060 for(ch=0; ch<s->out_channels; ch++) {
1061 for(i=0; i<256; i++) {
1062 s->output[ch][i] += s->add_bias;
1064 s->dsp.float_to_int16(s->int_output[ch], s->output[ch], 256);
1071 * Decode a single AC-3 frame.
1073 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size, uint8_t *buf, int buf_size)
1075 AC3DecodeContext *s = avctx->priv_data;
1076 int16_t *out_samples = (int16_t *)data;
1077 int i, blk, ch, err;
1079 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1080 init_get_bits(&s->gbc, buf, buf_size * 8);
1082 /* parse the syncinfo */
1083 err = ac3_parse_header(s);
1086 case AC3_PARSE_ERROR_SYNC:
1087 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1089 case AC3_PARSE_ERROR_BSID:
1090 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1092 case AC3_PARSE_ERROR_SAMPLE_RATE:
1093 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1095 case AC3_PARSE_ERROR_FRAME_SIZE:
1096 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1099 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1105 /* check that reported frame size fits in input buffer */
1106 if(s->frame_size > buf_size) {
1107 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1111 /* check for crc mismatch */
1112 if(avctx->error_resilience > 0) {
1113 if(av_crc(av_crc8005, 0, &buf[2], s->frame_size-2)) {
1114 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1117 /* TODO: error concealment */
1120 avctx->sample_rate = s->sample_rate;
1121 avctx->bit_rate = s->bit_rate;
1123 /* channel config */
1124 s->out_channels = s->channels;
1125 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1126 avctx->request_channels < s->channels) {
1127 s->out_channels = avctx->request_channels;
1128 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1130 avctx->channels = s->out_channels;
1132 /* parse the audio blocks */
1133 for (blk = 0; blk < NB_BLOCKS; blk++) {
1134 if (ac3_parse_audio_block(s, blk)) {
1135 av_log(avctx, AV_LOG_ERROR, "error parsing the audio block\n");
1137 return s->frame_size;
1139 for (i = 0; i < 256; i++)
1140 for (ch = 0; ch < s->out_channels; ch++)
1141 *(out_samples++) = s->int_output[ch][i];
1143 *data_size = NB_BLOCKS * 256 * avctx->channels * sizeof (int16_t);
1144 return s->frame_size;
1148 * Uninitialize the AC-3 decoder.
1150 static int ac3_decode_end(AVCodecContext *avctx)
1152 AC3DecodeContext *s = avctx->priv_data;
1153 ff_mdct_end(&s->imdct_512);
1154 ff_mdct_end(&s->imdct_256);
1159 AVCodec ac3_decoder = {
1161 .type = CODEC_TYPE_AUDIO,
1163 .priv_data_size = sizeof (AC3DecodeContext),
1164 .init = ac3_decode_init,
1165 .close = ac3_decode_end,
1166 .decode = ac3_decode_frame,