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 channel_mode; ///< channel mode (acmod)
115 int block_switch[AC3_MAX_CHANNELS]; ///< block switch flags
116 int dither_flag[AC3_MAX_CHANNELS]; ///< dither flags
117 int dither_all; ///< true if all channels are dithered
118 int cpl_in_use; ///< coupling in use
119 int channel_in_cpl[AC3_MAX_CHANNELS]; ///< channel in coupling
120 int phase_flags_in_use; ///< phase flags in use
121 int phase_flags[18]; ///< phase flags
122 int cpl_band_struct[18]; ///< coupling band structure
123 int num_rematrixing_bands; ///< number of rematrixing bands
124 int rematrixing_flags[4]; ///< rematrixing flags
125 int exp_strategy[AC3_MAX_CHANNELS]; ///< exponent strategies
126 int snr_offset[AC3_MAX_CHANNELS]; ///< signal-to-noise ratio offsets
127 int fast_gain[AC3_MAX_CHANNELS]; ///< fast gain values (signal-to-mask ratio)
128 int dba_mode[AC3_MAX_CHANNELS]; ///< delta bit allocation mode
129 int dba_nsegs[AC3_MAX_CHANNELS]; ///< number of delta segments
130 uint8_t dba_offsets[AC3_MAX_CHANNELS][8]; ///< delta segment offsets
131 uint8_t dba_lengths[AC3_MAX_CHANNELS][8]; ///< delta segment lengths
132 uint8_t dba_values[AC3_MAX_CHANNELS][8]; ///< delta values for each segment
134 int sample_rate; ///< sample frequency, in Hz
135 int bit_rate; ///< stream bit rate, in bits-per-second
136 int frame_size; ///< current frame size, in bytes
138 int channels; ///< number of total channels
139 int fbw_channels; ///< number of full-bandwidth channels
140 int lfe_on; ///< lfe channel in use
141 int lfe_ch; ///< index of LFE channel
142 int output_mode; ///< output channel configuration
143 int out_channels; ///< number of output channels
145 int center_mix_level; ///< Center mix level index
146 int surround_mix_level; ///< Surround mix level index
147 float downmix_coeffs[AC3_MAX_CHANNELS][2]; ///< stereo downmix coefficients
148 float downmix_coeff_adjust[2]; ///< adjustment needed for each output channel when downmixing
149 float dynamic_range[2]; ///< dynamic range
150 int cpl_coords[AC3_MAX_CHANNELS][18]; ///< coupling coordinates
151 int num_cpl_bands; ///< number of coupling bands
152 int num_cpl_subbands; ///< number of coupling sub bands
153 int start_freq[AC3_MAX_CHANNELS]; ///< start frequency bin
154 int end_freq[AC3_MAX_CHANNELS]; ///< end frequency bin
155 AC3BitAllocParameters bit_alloc_params; ///< bit allocation parameters
157 int8_t dexps[AC3_MAX_CHANNELS][256]; ///< decoded exponents
158 uint8_t bap[AC3_MAX_CHANNELS][256]; ///< bit allocation pointers
159 int16_t psd[AC3_MAX_CHANNELS][256]; ///< scaled exponents
160 int16_t band_psd[AC3_MAX_CHANNELS][50]; ///< interpolated exponents
161 int16_t mask[AC3_MAX_CHANNELS][50]; ///< masking curve values
163 int fixed_coeffs[AC3_MAX_CHANNELS][256]; ///> fixed-point transform coefficients
164 DECLARE_ALIGNED_16(float, transform_coeffs[AC3_MAX_CHANNELS][256]); ///< transform coefficients
165 int downmixed; ///< indicates if coeffs are currently downmixed
168 MDCTContext imdct_512; ///< for 512 sample IMDCT
169 MDCTContext imdct_256; ///< for 256 sample IMDCT
170 DSPContext dsp; ///< for optimization
171 float add_bias; ///< offset for float_to_int16 conversion
172 float mul_bias; ///< scaling for float_to_int16 conversion
174 DECLARE_ALIGNED_16(float, output[AC3_MAX_CHANNELS][256]); ///< output after imdct transform and windowing
175 DECLARE_ALIGNED_16(short, int_output[AC3_MAX_CHANNELS-1][256]); ///< final 16-bit integer output
176 DECLARE_ALIGNED_16(float, delay[AC3_MAX_CHANNELS][256]); ///< delay - added to the next block
177 DECLARE_ALIGNED_16(float, tmp_imdct[256]); ///< temporary storage for imdct transform
178 DECLARE_ALIGNED_16(float, tmp_output[512]); ///< temporary storage for output before windowing
179 DECLARE_ALIGNED_16(float, window[256]); ///< window coefficients
182 GetBitContext gbc; ///< bitstream reader
183 AVRandomState dith_state; ///< for dither generation
184 AVCodecContext *avctx; ///< parent context
185 uint8_t *input_buffer; ///< temp buffer to prevent overread
189 * Symmetrical Dequantization
190 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
191 * Tables 7.19 to 7.23
194 symmetric_dequant(int code, int levels)
196 return ((code - (levels >> 1)) << 24) / levels;
200 * Initialize tables at runtime.
202 static av_cold void ac3_tables_init(void)
206 /* generate grouped mantissa tables
207 reference: Section 7.3.5 Ungrouping of Mantissas */
208 for(i=0; i<32; i++) {
209 /* bap=1 mantissas */
210 b1_mantissas[i][0] = symmetric_dequant( i / 9 , 3);
211 b1_mantissas[i][1] = symmetric_dequant((i % 9) / 3, 3);
212 b1_mantissas[i][2] = symmetric_dequant((i % 9) % 3, 3);
214 for(i=0; i<128; i++) {
215 /* bap=2 mantissas */
216 b2_mantissas[i][0] = symmetric_dequant( i / 25 , 5);
217 b2_mantissas[i][1] = symmetric_dequant((i % 25) / 5, 5);
218 b2_mantissas[i][2] = symmetric_dequant((i % 25) % 5, 5);
220 /* bap=4 mantissas */
221 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
222 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
224 /* generate ungrouped mantissa tables
225 reference: Tables 7.21 and 7.23 */
227 /* bap=3 mantissas */
228 b3_mantissas[i] = symmetric_dequant(i, 7);
230 for(i=0; i<15; i++) {
231 /* bap=5 mantissas */
232 b5_mantissas[i] = symmetric_dequant(i, 15);
235 /* generate dynamic range table
236 reference: Section 7.7.1 Dynamic Range Control */
237 for(i=0; i<256; i++) {
238 int v = (i >> 5) - ((i >> 7) << 3) - 5;
239 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
242 /* generate exponent tables
243 reference: Section 7.1.3 Exponent Decoding */
244 for(i=0; i<128; i++) {
245 exp_ungroup_tab[i][0] = i / 25;
246 exp_ungroup_tab[i][1] = (i % 25) / 5;
247 exp_ungroup_tab[i][2] = (i % 25) % 5;
253 * AVCodec initialization
255 static av_cold int ac3_decode_init(AVCodecContext *avctx)
257 AC3DecodeContext *s = avctx->priv_data;
262 ff_mdct_init(&s->imdct_256, 8, 1);
263 ff_mdct_init(&s->imdct_512, 9, 1);
264 ff_kbd_window_init(s->window, 5.0, 256);
265 dsputil_init(&s->dsp, avctx);
266 av_init_random(0, &s->dith_state);
268 /* set bias values for float to int16 conversion */
269 if(s->dsp.float_to_int16 == ff_float_to_int16_c) {
270 s->add_bias = 385.0f;
274 s->mul_bias = 32767.0f;
277 /* allow downmixing to stereo or mono */
278 if (avctx->channels > 0 && avctx->request_channels > 0 &&
279 avctx->request_channels < avctx->channels &&
280 avctx->request_channels <= 2) {
281 avctx->channels = avctx->request_channels;
285 /* allocate context input buffer */
286 if (avctx->error_resilience >= FF_ER_CAREFUL) {
287 s->input_buffer = av_mallocz(AC3_MAX_FRAME_SIZE + FF_INPUT_BUFFER_PADDING_SIZE);
288 if (!s->input_buffer)
289 return AVERROR_NOMEM;
296 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
297 * GetBitContext within AC3DecodeContext must point to
298 * start of the synchronized ac3 bitstream.
300 static int ac3_parse_header(AC3DecodeContext *s)
303 GetBitContext *gbc = &s->gbc;
306 err = ff_ac3_parse_header(gbc, &hdr);
310 if(hdr.bitstream_id > 10)
311 return AC3_PARSE_ERROR_BSID;
313 /* get decoding parameters from header info */
314 s->bit_alloc_params.sr_code = hdr.sr_code;
315 s->channel_mode = hdr.channel_mode;
316 s->lfe_on = hdr.lfe_on;
317 s->bit_alloc_params.sr_shift = hdr.sr_shift;
318 s->sample_rate = hdr.sample_rate;
319 s->bit_rate = hdr.bit_rate;
320 s->channels = hdr.channels;
321 s->fbw_channels = s->channels - s->lfe_on;
322 s->lfe_ch = s->fbw_channels + 1;
323 s->frame_size = hdr.frame_size;
324 s->center_mix_level = hdr.center_mix_level;
325 s->surround_mix_level = hdr.surround_mix_level;
327 /* set default output to all source channels */
328 s->out_channels = s->channels;
329 s->output_mode = s->channel_mode;
331 s->output_mode |= AC3_OUTPUT_LFEON;
333 /* read the rest of the bsi. read twice for dual mono mode. */
334 i = !(s->channel_mode);
336 skip_bits(gbc, 5); // skip dialog normalization
338 skip_bits(gbc, 8); //skip compression
340 skip_bits(gbc, 8); //skip language code
342 skip_bits(gbc, 7); //skip audio production information
345 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
347 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
348 TODO: read & use the xbsi1 downmix levels */
350 skip_bits(gbc, 14); //skip timecode1 / xbsi1
352 skip_bits(gbc, 14); //skip timecode2 / xbsi2
354 /* skip additional bitstream info */
355 if (get_bits1(gbc)) {
356 i = get_bits(gbc, 6);
366 * Set stereo downmixing coefficients based on frame header info.
367 * reference: Section 7.8.2 Downmixing Into Two Channels
369 static void set_downmix_coeffs(AC3DecodeContext *s)
372 float cmix = gain_levels[s->center_mix_level];
373 float smix = gain_levels[s->surround_mix_level];
375 for(i=0; i<s->fbw_channels; i++) {
376 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
377 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
379 if(s->channel_mode > 1 && s->channel_mode & 1) {
380 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
382 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
383 int nf = s->channel_mode - 2;
384 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
386 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
387 int nf = s->channel_mode - 4;
388 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
391 /* calculate adjustment needed for each channel to avoid clipping */
392 s->downmix_coeff_adjust[0] = s->downmix_coeff_adjust[1] = 0.0f;
393 for(i=0; i<s->fbw_channels; i++) {
394 s->downmix_coeff_adjust[0] += s->downmix_coeffs[i][0];
395 s->downmix_coeff_adjust[1] += s->downmix_coeffs[i][1];
397 s->downmix_coeff_adjust[0] = 1.0f / s->downmix_coeff_adjust[0];
398 s->downmix_coeff_adjust[1] = 1.0f / s->downmix_coeff_adjust[1];
402 * Decode the grouped exponents according to exponent strategy.
403 * reference: Section 7.1.3 Exponent Decoding
405 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
406 uint8_t absexp, int8_t *dexps)
408 int i, j, grp, group_size;
413 group_size = exp_strategy + (exp_strategy == EXP_D45);
414 for(grp=0,i=0; grp<ngrps; grp++) {
415 expacc = get_bits(gbc, 7);
416 dexp[i++] = exp_ungroup_tab[expacc][0];
417 dexp[i++] = exp_ungroup_tab[expacc][1];
418 dexp[i++] = exp_ungroup_tab[expacc][2];
421 /* convert to absolute exps and expand groups */
423 for(i=0; i<ngrps*3; i++) {
424 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
425 for(j=0; j<group_size; j++) {
426 dexps[(i*group_size)+j] = prevexp;
432 * Generate transform coefficients for each coupled channel in the coupling
433 * range using the coupling coefficients and coupling coordinates.
434 * reference: Section 7.4.3 Coupling Coordinate Format
436 static void uncouple_channels(AC3DecodeContext *s)
438 int i, j, ch, bnd, subbnd;
441 i = s->start_freq[CPL_CH];
442 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
445 for(j=0; j<12; j++) {
446 for(ch=1; ch<=s->fbw_channels; ch++) {
447 if(s->channel_in_cpl[ch]) {
448 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
449 if (ch == 2 && s->phase_flags[bnd])
450 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
455 } while(s->cpl_band_struct[subbnd]);
460 * Grouped mantissas for 3-level 5-level and 11-level quantization
472 * Get the transform coefficients for a particular channel
473 * reference: Section 7.3 Quantization and Decoding of Mantissas
475 static int get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
477 GetBitContext *gbc = &s->gbc;
478 int i, gcode, tbap, start, end;
483 exps = s->dexps[ch_index];
484 bap = s->bap[ch_index];
485 coeffs = s->fixed_coeffs[ch_index];
486 start = s->start_freq[ch_index];
487 end = s->end_freq[ch_index];
489 for (i = start; i < end; i++) {
493 coeffs[i] = (av_random(&s->dith_state) & 0x7FFFFF) - 4194304;
498 gcode = get_bits(gbc, 5);
499 m->b1_mant[0] = b1_mantissas[gcode][0];
500 m->b1_mant[1] = b1_mantissas[gcode][1];
501 m->b1_mant[2] = b1_mantissas[gcode][2];
504 coeffs[i] = m->b1_mant[m->b1ptr++];
509 gcode = get_bits(gbc, 7);
510 m->b2_mant[0] = b2_mantissas[gcode][0];
511 m->b2_mant[1] = b2_mantissas[gcode][1];
512 m->b2_mant[2] = b2_mantissas[gcode][2];
515 coeffs[i] = m->b2_mant[m->b2ptr++];
519 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
524 gcode = get_bits(gbc, 7);
525 m->b4_mant[0] = b4_mantissas[gcode][0];
526 m->b4_mant[1] = b4_mantissas[gcode][1];
529 coeffs[i] = m->b4_mant[m->b4ptr++];
533 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
537 /* asymmetric dequantization */
538 int qlevel = quantization_tab[tbap];
539 coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
543 coeffs[i] >>= exps[i];
550 * Remove random dithering from coefficients with zero-bit mantissas
551 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
553 static void remove_dithering(AC3DecodeContext *s) {
559 for(ch=1; ch<=s->fbw_channels; ch++) {
560 if(!s->dither_flag[ch]) {
561 coeffs = s->fixed_coeffs[ch];
563 if(s->channel_in_cpl[ch])
564 end = s->start_freq[CPL_CH];
566 end = s->end_freq[ch];
567 for(i=0; i<end; i++) {
571 if(s->channel_in_cpl[ch]) {
572 bap = s->bap[CPL_CH];
573 for(; i<s->end_freq[CPL_CH]; i++) {
583 * Get the transform coefficients.
585 static int get_transform_coeffs(AC3DecodeContext *s)
591 m.b1ptr = m.b2ptr = m.b4ptr = 3;
593 for (ch = 1; ch <= s->channels; ch++) {
594 /* transform coefficients for full-bandwidth channel */
595 if (get_transform_coeffs_ch(s, ch, &m))
597 /* tranform coefficients for coupling channel come right after the
598 coefficients for the first coupled channel*/
599 if (s->channel_in_cpl[ch]) {
601 if (get_transform_coeffs_ch(s, CPL_CH, &m)) {
602 av_log(s->avctx, AV_LOG_ERROR, "error in decoupling channels\n");
605 uncouple_channels(s);
608 end = s->end_freq[CPL_CH];
610 end = s->end_freq[ch];
613 s->transform_coeffs[ch][end] = 0;
617 /* 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++) {
913 s->exp_strategy[ch] = get_bits(gbc, 1);
915 s->exp_strategy[ch] = get_bits(gbc, 2);
916 if(s->exp_strategy[ch] != EXP_REUSE)
917 bit_alloc_stages[ch] = 3;
920 /* channel bandwidth */
921 for (ch = 1; ch <= fbw_channels; ch++) {
922 s->start_freq[ch] = 0;
923 if (s->exp_strategy[ch] != EXP_REUSE) {
924 int prev = s->end_freq[ch];
925 if (s->channel_in_cpl[ch])
926 s->end_freq[ch] = s->start_freq[CPL_CH];
928 int bandwidth_code = get_bits(gbc, 6);
929 if (bandwidth_code > 60) {
930 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
933 s->end_freq[ch] = bandwidth_code * 3 + 73;
935 if(blk > 0 && s->end_freq[ch] != prev)
936 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
939 s->start_freq[s->lfe_ch] = 0;
940 s->end_freq[s->lfe_ch] = 7;
942 /* decode exponents for each channel */
943 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
944 if (s->exp_strategy[ch] != EXP_REUSE) {
945 int group_size, num_groups;
946 group_size = 3 << (s->exp_strategy[ch] - 1);
948 num_groups = (s->end_freq[ch] - s->start_freq[ch]) / group_size;
949 else if(ch == s->lfe_ch)
952 num_groups = (s->end_freq[ch] + group_size - 4) / group_size;
953 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
954 decode_exponents(gbc, s->exp_strategy[ch], num_groups, s->dexps[ch][0],
955 &s->dexps[ch][s->start_freq[ch]+!!ch]);
956 if(ch != CPL_CH && ch != s->lfe_ch)
957 skip_bits(gbc, 2); /* skip gainrng */
961 /* bit allocation information */
962 if (get_bits1(gbc)) {
963 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
964 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
965 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
966 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
967 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
968 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
969 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
972 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must be present in block 0\n");
976 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
977 if (get_bits1(gbc)) {
979 csnr = (get_bits(gbc, 6) - 15) << 4;
980 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
981 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
982 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
984 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
986 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
990 /* coupling leak information */
992 if (get_bits1(gbc)) {
993 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
994 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
995 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
997 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must be present in block 0\n");
1002 /* delta bit allocation information */
1003 if (get_bits1(gbc)) {
1004 /* delta bit allocation exists (strategy) */
1005 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
1006 s->dba_mode[ch] = get_bits(gbc, 2);
1007 if (s->dba_mode[ch] == DBA_RESERVED) {
1008 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1011 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1013 /* channel delta offset, len and bit allocation */
1014 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
1015 if (s->dba_mode[ch] == DBA_NEW) {
1016 s->dba_nsegs[ch] = get_bits(gbc, 3);
1017 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
1018 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1019 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1020 s->dba_values[ch][seg] = get_bits(gbc, 3);
1024 } else if(blk == 0) {
1025 for(ch=0; ch<=s->channels; ch++) {
1026 s->dba_mode[ch] = DBA_NONE;
1030 /* Bit allocation */
1031 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
1032 if(bit_alloc_stages[ch] > 2) {
1033 /* Exponent mapping into PSD and PSD integration */
1034 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1035 s->start_freq[ch], s->end_freq[ch],
1036 s->psd[ch], s->band_psd[ch]);
1038 if(bit_alloc_stages[ch] > 1) {
1039 /* Compute excitation function, Compute masking curve, and
1040 Apply delta bit allocation */
1041 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1042 s->start_freq[ch], s->end_freq[ch],
1043 s->fast_gain[ch], (ch == s->lfe_ch),
1044 s->dba_mode[ch], s->dba_nsegs[ch],
1045 s->dba_offsets[ch], s->dba_lengths[ch],
1046 s->dba_values[ch], s->mask[ch]);
1048 if(bit_alloc_stages[ch] > 0) {
1049 /* Compute bit allocation */
1050 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1051 s->start_freq[ch], s->end_freq[ch],
1053 s->bit_alloc_params.floor,
1058 /* unused dummy data */
1059 if (get_bits1(gbc)) {
1060 int skipl = get_bits(gbc, 9);
1065 /* unpack the transform coefficients
1066 this also uncouples channels if coupling is in use. */
1067 if (get_transform_coeffs(s)) {
1068 av_log(s->avctx, AV_LOG_ERROR, "Error in routine get_transform_coeffs\n");
1072 /* recover coefficients if rematrixing is in use */
1073 if(s->channel_mode == AC3_CHMODE_STEREO)
1076 /* apply scaling to coefficients (headroom, dynrng) */
1077 for(ch=1; ch<=s->channels; ch++) {
1078 float gain = s->mul_bias / 4194304.0f;
1079 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1080 gain *= s->dynamic_range[ch-1];
1082 gain *= s->dynamic_range[0];
1084 for(i=0; i<256; i++) {
1085 s->transform_coeffs[ch][i] = s->fixed_coeffs[ch][i] * gain;
1089 /* downmix and MDCT. order depends on whether block switching is used for
1090 any channel in this block. this is because coefficients for the long
1091 and short transforms cannot be mixed. */
1092 downmix_output = s->channels != s->out_channels &&
1093 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1094 s->fbw_channels == s->out_channels);
1095 if(different_transforms) {
1096 /* the delay samples have already been downmixed, so we upmix the delay
1097 samples in order to reconstruct all channels before downmixing. */
1103 do_imdct(s, s->channels);
1105 if(downmix_output) {
1106 ac3_downmix(s, s->output, 0);
1109 if(downmix_output) {
1110 ac3_downmix(s, s->transform_coeffs, 1);
1115 ac3_downmix(s, s->delay, 0);
1118 do_imdct(s, s->out_channels);
1121 /* convert float to 16-bit integer */
1122 for(ch=0; ch<s->out_channels; ch++) {
1123 for(i=0; i<256; i++) {
1124 s->output[ch][i] += s->add_bias;
1126 s->dsp.float_to_int16(s->int_output[ch], s->output[ch], 256);
1133 * Decode a single AC-3 frame.
1135 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
1136 const uint8_t *buf, int buf_size)
1138 AC3DecodeContext *s = avctx->priv_data;
1139 int16_t *out_samples = (int16_t *)data;
1140 int i, blk, ch, err;
1142 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1143 if (s->input_buffer) {
1144 /* copy input buffer to decoder context to avoid reading past the end
1145 of the buffer, which can be caused by a damaged input stream. */
1146 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_MAX_FRAME_SIZE));
1147 init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
1149 init_get_bits(&s->gbc, buf, buf_size * 8);
1152 /* parse the syncinfo */
1153 err = ac3_parse_header(s);
1156 case AC3_PARSE_ERROR_SYNC:
1157 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1159 case AC3_PARSE_ERROR_BSID:
1160 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1162 case AC3_PARSE_ERROR_SAMPLE_RATE:
1163 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1165 case AC3_PARSE_ERROR_FRAME_SIZE:
1166 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1168 case AC3_PARSE_ERROR_FRAME_TYPE:
1169 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1172 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1178 /* check that reported frame size fits in input buffer */
1179 if(s->frame_size > buf_size) {
1180 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1184 /* check for crc mismatch */
1185 if(avctx->error_resilience >= FF_ER_CAREFUL) {
1186 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1187 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1190 /* TODO: error concealment */
1193 avctx->sample_rate = s->sample_rate;
1194 avctx->bit_rate = s->bit_rate;
1196 /* channel config */
1197 s->out_channels = s->channels;
1198 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1199 avctx->request_channels < s->channels) {
1200 s->out_channels = avctx->request_channels;
1201 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1203 avctx->channels = s->out_channels;
1205 /* set downmixing coefficients if needed */
1206 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1207 s->fbw_channels == s->out_channels)) {
1208 set_downmix_coeffs(s);
1211 /* parse the audio blocks */
1212 for (blk = 0; blk < NB_BLOCKS; blk++) {
1213 if (ac3_parse_audio_block(s, blk)) {
1214 av_log(avctx, AV_LOG_ERROR, "error parsing the audio block\n");
1216 return s->frame_size;
1218 for (i = 0; i < 256; i++)
1219 for (ch = 0; ch < s->out_channels; ch++)
1220 *(out_samples++) = s->int_output[ch][i];
1222 *data_size = NB_BLOCKS * 256 * avctx->channels * sizeof (int16_t);
1223 return s->frame_size;
1227 * Uninitialize the AC-3 decoder.
1229 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1231 AC3DecodeContext *s = avctx->priv_data;
1232 ff_mdct_end(&s->imdct_512);
1233 ff_mdct_end(&s->imdct_256);
1235 av_freep(&s->input_buffer);
1240 AVCodec ac3_decoder = {
1242 .type = CODEC_TYPE_AUDIO,
1244 .priv_data_size = sizeof (AC3DecodeContext),
1245 .init = ac3_decode_init,
1246 .close = ac3_decode_end,
1247 .decode = ac3_decode_frame,
1248 .long_name = "ATSC A/52 / AC-3",