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 int num_exp_groups[AC3_MAX_CHANNELS]; ///< Number of exponent groups
158 int8_t dexps[AC3_MAX_CHANNELS][256]; ///< decoded exponents
159 uint8_t bap[AC3_MAX_CHANNELS][256]; ///< bit allocation pointers
160 int16_t psd[AC3_MAX_CHANNELS][256]; ///< scaled exponents
161 int16_t band_psd[AC3_MAX_CHANNELS][50]; ///< interpolated exponents
162 int16_t mask[AC3_MAX_CHANNELS][50]; ///< masking curve values
164 int fixed_coeffs[AC3_MAX_CHANNELS][256]; ///> fixed-point transform coefficients
165 DECLARE_ALIGNED_16(float, transform_coeffs[AC3_MAX_CHANNELS][256]); ///< transform coefficients
166 int downmixed; ///< indicates if coeffs are currently downmixed
169 MDCTContext imdct_512; ///< for 512 sample IMDCT
170 MDCTContext imdct_256; ///< for 256 sample IMDCT
171 DSPContext dsp; ///< for optimization
172 float add_bias; ///< offset for float_to_int16 conversion
173 float mul_bias; ///< scaling for float_to_int16 conversion
175 DECLARE_ALIGNED_16(float, output[AC3_MAX_CHANNELS][256]); ///< output after imdct transform and windowing
176 DECLARE_ALIGNED_16(short, int_output[AC3_MAX_CHANNELS-1][256]); ///< final 16-bit integer output
177 DECLARE_ALIGNED_16(float, delay[AC3_MAX_CHANNELS][256]); ///< delay - added to the next block
178 DECLARE_ALIGNED_16(float, tmp_imdct[256]); ///< temporary storage for imdct transform
179 DECLARE_ALIGNED_16(float, tmp_output[512]); ///< temporary storage for output before windowing
180 DECLARE_ALIGNED_16(float, window[256]); ///< window coefficients
183 GetBitContext gbc; ///< bitstream reader
184 AVRandomState dith_state; ///< for dither generation
185 AVCodecContext *avctx; ///< parent context
186 uint8_t *input_buffer; ///< temp buffer to prevent overread
190 * Symmetrical Dequantization
191 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
192 * Tables 7.19 to 7.23
195 symmetric_dequant(int code, int levels)
197 return ((code - (levels >> 1)) << 24) / levels;
201 * Initialize tables at runtime.
203 static av_cold void ac3_tables_init(void)
207 /* generate grouped mantissa tables
208 reference: Section 7.3.5 Ungrouping of Mantissas */
209 for(i=0; i<32; i++) {
210 /* bap=1 mantissas */
211 b1_mantissas[i][0] = symmetric_dequant( i / 9 , 3);
212 b1_mantissas[i][1] = symmetric_dequant((i % 9) / 3, 3);
213 b1_mantissas[i][2] = symmetric_dequant((i % 9) % 3, 3);
215 for(i=0; i<128; i++) {
216 /* bap=2 mantissas */
217 b2_mantissas[i][0] = symmetric_dequant( i / 25 , 5);
218 b2_mantissas[i][1] = symmetric_dequant((i % 25) / 5, 5);
219 b2_mantissas[i][2] = symmetric_dequant((i % 25) % 5, 5);
221 /* bap=4 mantissas */
222 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
223 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
225 /* generate ungrouped mantissa tables
226 reference: Tables 7.21 and 7.23 */
228 /* bap=3 mantissas */
229 b3_mantissas[i] = symmetric_dequant(i, 7);
231 for(i=0; i<15; i++) {
232 /* bap=5 mantissas */
233 b5_mantissas[i] = symmetric_dequant(i, 15);
236 /* generate dynamic range table
237 reference: Section 7.7.1 Dynamic Range Control */
238 for(i=0; i<256; i++) {
239 int v = (i >> 5) - ((i >> 7) << 3) - 5;
240 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
243 /* generate exponent tables
244 reference: Section 7.1.3 Exponent Decoding */
245 for(i=0; i<128; i++) {
246 exp_ungroup_tab[i][0] = i / 25;
247 exp_ungroup_tab[i][1] = (i % 25) / 5;
248 exp_ungroup_tab[i][2] = (i % 25) % 5;
254 * AVCodec initialization
256 static av_cold int ac3_decode_init(AVCodecContext *avctx)
258 AC3DecodeContext *s = avctx->priv_data;
263 ff_mdct_init(&s->imdct_256, 8, 1);
264 ff_mdct_init(&s->imdct_512, 9, 1);
265 ff_kbd_window_init(s->window, 5.0, 256);
266 dsputil_init(&s->dsp, avctx);
267 av_init_random(0, &s->dith_state);
269 /* set bias values for float to int16 conversion */
270 if(s->dsp.float_to_int16 == ff_float_to_int16_c) {
271 s->add_bias = 385.0f;
275 s->mul_bias = 32767.0f;
278 /* allow downmixing to stereo or mono */
279 if (avctx->channels > 0 && avctx->request_channels > 0 &&
280 avctx->request_channels < avctx->channels &&
281 avctx->request_channels <= 2) {
282 avctx->channels = avctx->request_channels;
286 /* allocate context input buffer */
287 if (avctx->error_resilience >= FF_ER_CAREFUL) {
288 s->input_buffer = av_mallocz(AC3_MAX_FRAME_SIZE + FF_INPUT_BUFFER_PADDING_SIZE);
289 if (!s->input_buffer)
290 return AVERROR_NOMEM;
297 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
298 * GetBitContext within AC3DecodeContext must point to
299 * start of the synchronized ac3 bitstream.
301 static int ac3_parse_header(AC3DecodeContext *s)
304 GetBitContext *gbc = &s->gbc;
307 err = ff_ac3_parse_header(gbc, &hdr);
311 if(hdr.bitstream_id > 10)
312 return AC3_PARSE_ERROR_BSID;
314 /* get decoding parameters from header info */
315 s->bit_alloc_params.sr_code = hdr.sr_code;
316 s->channel_mode = hdr.channel_mode;
317 s->lfe_on = hdr.lfe_on;
318 s->bit_alloc_params.sr_shift = hdr.sr_shift;
319 s->sample_rate = hdr.sample_rate;
320 s->bit_rate = hdr.bit_rate;
321 s->channels = hdr.channels;
322 s->fbw_channels = s->channels - s->lfe_on;
323 s->lfe_ch = s->fbw_channels + 1;
324 s->frame_size = hdr.frame_size;
325 s->center_mix_level = hdr.center_mix_level;
326 s->surround_mix_level = hdr.surround_mix_level;
328 /* read the rest of the bsi. read twice for dual mono mode. */
329 i = !(s->channel_mode);
331 skip_bits(gbc, 5); // skip dialog normalization
333 skip_bits(gbc, 8); //skip compression
335 skip_bits(gbc, 8); //skip language code
337 skip_bits(gbc, 7); //skip audio production information
340 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
342 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
343 TODO: read & use the xbsi1 downmix levels */
345 skip_bits(gbc, 14); //skip timecode1 / xbsi1
347 skip_bits(gbc, 14); //skip timecode2 / xbsi2
349 /* skip additional bitstream info */
350 if (get_bits1(gbc)) {
351 i = get_bits(gbc, 6);
361 * Set stereo downmixing coefficients based on frame header info.
362 * reference: Section 7.8.2 Downmixing Into Two Channels
364 static void set_downmix_coeffs(AC3DecodeContext *s)
367 float cmix = gain_levels[s->center_mix_level];
368 float smix = gain_levels[s->surround_mix_level];
370 for(i=0; i<s->fbw_channels; i++) {
371 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
372 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
374 if(s->channel_mode > 1 && s->channel_mode & 1) {
375 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
377 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
378 int nf = s->channel_mode - 2;
379 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
381 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
382 int nf = s->channel_mode - 4;
383 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
386 /* calculate adjustment needed for each channel to avoid clipping */
387 s->downmix_coeff_adjust[0] = s->downmix_coeff_adjust[1] = 0.0f;
388 for(i=0; i<s->fbw_channels; i++) {
389 s->downmix_coeff_adjust[0] += s->downmix_coeffs[i][0];
390 s->downmix_coeff_adjust[1] += s->downmix_coeffs[i][1];
392 s->downmix_coeff_adjust[0] = 1.0f / s->downmix_coeff_adjust[0];
393 s->downmix_coeff_adjust[1] = 1.0f / s->downmix_coeff_adjust[1];
397 * Decode the grouped exponents according to exponent strategy.
398 * reference: Section 7.1.3 Exponent Decoding
400 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
401 uint8_t absexp, int8_t *dexps)
403 int i, j, grp, group_size;
408 group_size = exp_strategy + (exp_strategy == EXP_D45);
409 for(grp=0,i=0; grp<ngrps; grp++) {
410 expacc = get_bits(gbc, 7);
411 dexp[i++] = exp_ungroup_tab[expacc][0];
412 dexp[i++] = exp_ungroup_tab[expacc][1];
413 dexp[i++] = exp_ungroup_tab[expacc][2];
416 /* convert to absolute exps and expand groups */
418 for(i=0; i<ngrps*3; i++) {
419 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
420 for(j=0; j<group_size; j++) {
421 dexps[(i*group_size)+j] = prevexp;
427 * Generate transform coefficients for each coupled channel in the coupling
428 * range using the coupling coefficients and coupling coordinates.
429 * reference: Section 7.4.3 Coupling Coordinate Format
431 static void uncouple_channels(AC3DecodeContext *s)
433 int i, j, ch, bnd, subbnd;
436 i = s->start_freq[CPL_CH];
437 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
440 for(j=0; j<12; j++) {
441 for(ch=1; ch<=s->fbw_channels; ch++) {
442 if(s->channel_in_cpl[ch]) {
443 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
444 if (ch == 2 && s->phase_flags[bnd])
445 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
450 } while(s->cpl_band_struct[subbnd]);
455 * Grouped mantissas for 3-level 5-level and 11-level quantization
467 * Get the transform coefficients for a particular channel
468 * reference: Section 7.3 Quantization and Decoding of Mantissas
470 static void get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
472 GetBitContext *gbc = &s->gbc;
473 int i, gcode, tbap, start, end;
478 exps = s->dexps[ch_index];
479 bap = s->bap[ch_index];
480 coeffs = s->fixed_coeffs[ch_index];
481 start = s->start_freq[ch_index];
482 end = s->end_freq[ch_index];
484 for (i = start; i < end; i++) {
488 coeffs[i] = (av_random(&s->dith_state) & 0x7FFFFF) - 4194304;
493 gcode = get_bits(gbc, 5);
494 m->b1_mant[0] = b1_mantissas[gcode][0];
495 m->b1_mant[1] = b1_mantissas[gcode][1];
496 m->b1_mant[2] = b1_mantissas[gcode][2];
499 coeffs[i] = m->b1_mant[m->b1ptr++];
504 gcode = get_bits(gbc, 7);
505 m->b2_mant[0] = b2_mantissas[gcode][0];
506 m->b2_mant[1] = b2_mantissas[gcode][1];
507 m->b2_mant[2] = b2_mantissas[gcode][2];
510 coeffs[i] = m->b2_mant[m->b2ptr++];
514 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
519 gcode = get_bits(gbc, 7);
520 m->b4_mant[0] = b4_mantissas[gcode][0];
521 m->b4_mant[1] = b4_mantissas[gcode][1];
524 coeffs[i] = m->b4_mant[m->b4ptr++];
528 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
532 /* asymmetric dequantization */
533 int qlevel = quantization_tab[tbap];
534 coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
538 coeffs[i] >>= exps[i];
543 * Remove random dithering from coefficients with zero-bit mantissas
544 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
546 static void remove_dithering(AC3DecodeContext *s) {
552 for(ch=1; ch<=s->fbw_channels; ch++) {
553 if(!s->dither_flag[ch]) {
554 coeffs = s->fixed_coeffs[ch];
556 if(s->channel_in_cpl[ch])
557 end = s->start_freq[CPL_CH];
559 end = s->end_freq[ch];
560 for(i=0; i<end; i++) {
564 if(s->channel_in_cpl[ch]) {
565 bap = s->bap[CPL_CH];
566 for(; i<s->end_freq[CPL_CH]; i++) {
576 * Get the transform coefficients.
578 static void get_transform_coeffs(AC3DecodeContext *s)
584 m.b1ptr = m.b2ptr = m.b4ptr = 3;
586 for (ch = 1; ch <= s->channels; ch++) {
587 /* transform coefficients for full-bandwidth channel */
588 get_transform_coeffs_ch(s, ch, &m);
589 /* tranform coefficients for coupling channel come right after the
590 coefficients for the first coupled channel*/
591 if (s->channel_in_cpl[ch]) {
593 get_transform_coeffs_ch(s, CPL_CH, &m);
594 uncouple_channels(s);
597 end = s->end_freq[CPL_CH];
599 end = s->end_freq[ch];
602 s->fixed_coeffs[ch][end] = 0;
606 /* if any channel doesn't use dithering, zero appropriate coefficients */
612 * Stereo rematrixing.
613 * reference: Section 7.5.4 Rematrixing : Decoding Technique
615 static void do_rematrixing(AC3DecodeContext *s)
621 end = FFMIN(s->end_freq[1], s->end_freq[2]);
623 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
624 if(s->rematrixing_flags[bnd]) {
625 bndend = FFMIN(end, rematrix_band_tab[bnd+1]);
626 for(i=rematrix_band_tab[bnd]; i<bndend; i++) {
627 tmp0 = s->fixed_coeffs[1][i];
628 tmp1 = s->fixed_coeffs[2][i];
629 s->fixed_coeffs[1][i] = tmp0 + tmp1;
630 s->fixed_coeffs[2][i] = tmp0 - tmp1;
637 * Perform the 256-point IMDCT
639 static void do_imdct_256(AC3DecodeContext *s, int chindex)
642 DECLARE_ALIGNED_16(float, x[128]);
644 float *o_ptr = s->tmp_output;
647 /* de-interleave coefficients */
648 for(k=0; k<128; k++) {
649 x[k] = s->transform_coeffs[chindex][2*k+i];
652 /* run standard IMDCT */
653 s->imdct_256.fft.imdct_calc(&s->imdct_256, o_ptr, x, s->tmp_imdct);
655 /* reverse the post-rotation & reordering from standard IMDCT */
656 for(k=0; k<32; k++) {
657 z[i][32+k].re = -o_ptr[128+2*k];
658 z[i][32+k].im = -o_ptr[2*k];
659 z[i][31-k].re = o_ptr[2*k+1];
660 z[i][31-k].im = o_ptr[128+2*k+1];
664 /* apply AC-3 post-rotation & reordering */
665 for(k=0; k<64; k++) {
666 o_ptr[ 2*k ] = -z[0][ k].im;
667 o_ptr[ 2*k+1] = z[0][63-k].re;
668 o_ptr[128+2*k ] = -z[0][ k].re;
669 o_ptr[128+2*k+1] = z[0][63-k].im;
670 o_ptr[256+2*k ] = -z[1][ k].re;
671 o_ptr[256+2*k+1] = z[1][63-k].im;
672 o_ptr[384+2*k ] = z[1][ k].im;
673 o_ptr[384+2*k+1] = -z[1][63-k].re;
678 * Inverse MDCT Transform.
679 * Convert frequency domain coefficients to time-domain audio samples.
680 * reference: Section 7.9.4 Transformation Equations
682 static inline void do_imdct(AC3DecodeContext *s, int channels)
686 for (ch=1; ch<=channels; ch++) {
687 if (s->block_switch[ch]) {
690 s->imdct_512.fft.imdct_calc(&s->imdct_512, s->tmp_output,
691 s->transform_coeffs[ch], s->tmp_imdct);
693 /* For the first half of the block, apply the window, add the delay
694 from the previous block, and send to output */
695 s->dsp.vector_fmul_add_add(s->output[ch-1], s->tmp_output,
696 s->window, s->delay[ch-1], 0, 256, 1);
697 /* For the second half of the block, apply the window and store the
698 samples to delay, to be combined with the next block */
699 s->dsp.vector_fmul_reverse(s->delay[ch-1], s->tmp_output+256,
705 * Downmix the output to mono or stereo.
707 static void ac3_downmix(AC3DecodeContext *s,
708 float samples[AC3_MAX_CHANNELS][256], int ch_offset)
713 for(i=0; i<256; i++) {
715 for(j=0; j<s->fbw_channels; j++) {
716 v0 += samples[j+ch_offset][i] * s->downmix_coeffs[j][0];
717 v1 += samples[j+ch_offset][i] * s->downmix_coeffs[j][1];
719 v0 *= s->downmix_coeff_adjust[0];
720 v1 *= s->downmix_coeff_adjust[1];
721 if(s->output_mode == AC3_CHMODE_MONO) {
722 samples[ch_offset][i] = (v0 + v1) * LEVEL_MINUS_3DB;
723 } else if(s->output_mode == AC3_CHMODE_STEREO) {
724 samples[ ch_offset][i] = v0;
725 samples[1+ch_offset][i] = v1;
731 * Upmix delay samples from stereo to original channel layout.
733 static void ac3_upmix_delay(AC3DecodeContext *s)
735 int channel_data_size = sizeof(s->delay[0]);
736 switch(s->channel_mode) {
737 case AC3_CHMODE_DUALMONO:
738 case AC3_CHMODE_STEREO:
739 /* upmix mono to stereo */
740 memcpy(s->delay[1], s->delay[0], channel_data_size);
742 case AC3_CHMODE_2F2R:
743 memset(s->delay[3], 0, channel_data_size);
744 case AC3_CHMODE_2F1R:
745 memset(s->delay[2], 0, channel_data_size);
747 case AC3_CHMODE_3F2R:
748 memset(s->delay[4], 0, channel_data_size);
749 case AC3_CHMODE_3F1R:
750 memset(s->delay[3], 0, channel_data_size);
752 memcpy(s->delay[2], s->delay[1], channel_data_size);
753 memset(s->delay[1], 0, channel_data_size);
759 * Parse an audio block from AC-3 bitstream.
761 static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
763 int fbw_channels = s->fbw_channels;
764 int channel_mode = s->channel_mode;
766 int different_transforms;
768 GetBitContext *gbc = &s->gbc;
769 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
771 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
773 /* block switch flags */
774 different_transforms = 0;
775 for (ch = 1; ch <= fbw_channels; ch++) {
776 s->block_switch[ch] = get_bits1(gbc);
777 if(ch > 1 && s->block_switch[ch] != s->block_switch[1])
778 different_transforms = 1;
781 /* dithering flags */
783 for (ch = 1; ch <= fbw_channels; ch++) {
784 s->dither_flag[ch] = get_bits1(gbc);
785 if(!s->dither_flag[ch])
790 i = !(s->channel_mode);
793 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
794 s->avctx->drc_scale)+1.0;
795 } else if(blk == 0) {
796 s->dynamic_range[i] = 1.0f;
800 /* coupling strategy */
801 if (get_bits1(gbc)) {
802 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
803 s->cpl_in_use = get_bits1(gbc);
805 /* coupling in use */
806 int cpl_begin_freq, cpl_end_freq;
808 if (channel_mode < AC3_CHMODE_STEREO) {
809 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
813 /* determine which channels are coupled */
814 for (ch = 1; ch <= fbw_channels; ch++)
815 s->channel_in_cpl[ch] = get_bits1(gbc);
817 /* phase flags in use */
818 if (channel_mode == AC3_CHMODE_STEREO)
819 s->phase_flags_in_use = get_bits1(gbc);
821 /* coupling frequency range and band structure */
822 cpl_begin_freq = get_bits(gbc, 4);
823 cpl_end_freq = get_bits(gbc, 4);
824 if (3 + cpl_end_freq - cpl_begin_freq < 0) {
825 av_log(s->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
828 s->num_cpl_bands = s->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
829 s->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
830 s->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
831 for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
832 if (get_bits1(gbc)) {
833 s->cpl_band_struct[bnd] = 1;
837 s->cpl_band_struct[s->num_cpl_subbands-1] = 0;
839 /* coupling not in use */
840 for (ch = 1; ch <= fbw_channels; ch++)
841 s->channel_in_cpl[ch] = 0;
844 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must be present in block 0\n");
848 /* coupling coordinates */
850 int cpl_coords_exist = 0;
852 for (ch = 1; ch <= fbw_channels; ch++) {
853 if (s->channel_in_cpl[ch]) {
854 if (get_bits1(gbc)) {
855 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
856 cpl_coords_exist = 1;
857 master_cpl_coord = 3 * get_bits(gbc, 2);
858 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
859 cpl_coord_exp = get_bits(gbc, 4);
860 cpl_coord_mant = get_bits(gbc, 4);
861 if (cpl_coord_exp == 15)
862 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
864 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
865 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
868 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must be present in block 0\n");
874 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
875 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
876 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
881 /* stereo rematrixing strategy and band structure */
882 if (channel_mode == AC3_CHMODE_STEREO) {
883 if (get_bits1(gbc)) {
884 s->num_rematrixing_bands = 4;
885 if(s->cpl_in_use && s->start_freq[CPL_CH] <= 61)
886 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
887 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
888 s->rematrixing_flags[bnd] = get_bits1(gbc);
890 av_log(s->avctx, AV_LOG_ERROR, "new rematrixing strategy must be present in block 0\n");
895 /* exponent strategies for each channel */
896 s->exp_strategy[CPL_CH] = EXP_REUSE;
897 s->exp_strategy[s->lfe_ch] = EXP_REUSE;
898 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
900 s->exp_strategy[ch] = get_bits(gbc, 1);
902 s->exp_strategy[ch] = get_bits(gbc, 2);
903 if(s->exp_strategy[ch] != EXP_REUSE)
904 bit_alloc_stages[ch] = 3;
907 /* channel bandwidth */
908 for (ch = 1; ch <= fbw_channels; ch++) {
909 s->start_freq[ch] = 0;
910 if (s->exp_strategy[ch] != EXP_REUSE) {
912 int prev = s->end_freq[ch];
913 if (s->channel_in_cpl[ch])
914 s->end_freq[ch] = s->start_freq[CPL_CH];
916 int bandwidth_code = get_bits(gbc, 6);
917 if (bandwidth_code > 60) {
918 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
921 s->end_freq[ch] = bandwidth_code * 3 + 73;
923 group_size = 3 << (s->exp_strategy[ch] - 1);
924 s->num_exp_groups[ch] = (s->end_freq[ch]+group_size-4) / group_size;
925 if(blk > 0 && s->end_freq[ch] != prev)
926 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
929 s->start_freq[s->lfe_ch] = 0;
930 s->end_freq[s->lfe_ch] = 7;
931 s->num_exp_groups[s->lfe_ch] = 2;
932 if (s->cpl_in_use && s->exp_strategy[CPL_CH] != EXP_REUSE) {
933 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
934 (3 << (s->exp_strategy[CPL_CH] - 1));
937 /* decode exponents for each channel */
938 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
939 if (s->exp_strategy[ch] != EXP_REUSE) {
940 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
941 decode_exponents(gbc, s->exp_strategy[ch],
942 s->num_exp_groups[ch], s->dexps[ch][0],
943 &s->dexps[ch][s->start_freq[ch]+!!ch]);
944 if(ch != CPL_CH && ch != s->lfe_ch)
945 skip_bits(gbc, 2); /* skip gainrng */
949 /* bit allocation information */
950 if (get_bits1(gbc)) {
951 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
952 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
953 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
954 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
955 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
956 for(ch=!s->cpl_in_use; ch<=s->channels; ch++)
957 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
959 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must be present in block 0\n");
963 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
964 if (get_bits1(gbc)) {
966 csnr = (get_bits(gbc, 6) - 15) << 4;
967 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
968 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
969 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
971 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
973 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
977 /* coupling leak information */
979 if (get_bits1(gbc)) {
980 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
981 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
982 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
984 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must be present in block 0\n");
989 /* delta bit allocation information */
990 if (get_bits1(gbc)) {
991 /* delta bit allocation exists (strategy) */
992 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
993 s->dba_mode[ch] = get_bits(gbc, 2);
994 if (s->dba_mode[ch] == DBA_RESERVED) {
995 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
998 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1000 /* channel delta offset, len and bit allocation */
1001 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
1002 if (s->dba_mode[ch] == DBA_NEW) {
1003 s->dba_nsegs[ch] = get_bits(gbc, 3);
1004 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
1005 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1006 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1007 s->dba_values[ch][seg] = get_bits(gbc, 3);
1009 /* run last 2 bit allocation stages if new dba values */
1010 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1013 } else if(blk == 0) {
1014 for(ch=0; ch<=s->channels; ch++) {
1015 s->dba_mode[ch] = DBA_NONE;
1019 /* Bit allocation */
1020 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
1021 if(bit_alloc_stages[ch] > 2) {
1022 /* Exponent mapping into PSD and PSD integration */
1023 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1024 s->start_freq[ch], s->end_freq[ch],
1025 s->psd[ch], s->band_psd[ch]);
1027 if(bit_alloc_stages[ch] > 1) {
1028 /* Compute excitation function, Compute masking curve, and
1029 Apply delta bit allocation */
1030 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1031 s->start_freq[ch], s->end_freq[ch],
1032 s->fast_gain[ch], (ch == s->lfe_ch),
1033 s->dba_mode[ch], s->dba_nsegs[ch],
1034 s->dba_offsets[ch], s->dba_lengths[ch],
1035 s->dba_values[ch], s->mask[ch]);
1037 if(bit_alloc_stages[ch] > 0) {
1038 /* Compute bit allocation */
1039 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1040 s->start_freq[ch], s->end_freq[ch],
1042 s->bit_alloc_params.floor,
1047 /* unused dummy data */
1048 if (get_bits1(gbc)) {
1049 int skipl = get_bits(gbc, 9);
1054 /* unpack the transform coefficients
1055 this also uncouples channels if coupling is in use. */
1056 get_transform_coeffs(s);
1058 /* recover coefficients if rematrixing is in use */
1059 if(s->channel_mode == AC3_CHMODE_STEREO)
1062 /* apply scaling to coefficients (headroom, dynrng) */
1063 for(ch=1; ch<=s->channels; ch++) {
1064 float gain = s->mul_bias / 4194304.0f;
1065 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1066 gain *= s->dynamic_range[ch-1];
1068 gain *= s->dynamic_range[0];
1070 for(i=0; i<256; i++) {
1071 s->transform_coeffs[ch][i] = s->fixed_coeffs[ch][i] * gain;
1075 /* downmix and MDCT. order depends on whether block switching is used for
1076 any channel in this block. this is because coefficients for the long
1077 and short transforms cannot be mixed. */
1078 downmix_output = s->channels != s->out_channels &&
1079 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1080 s->fbw_channels == s->out_channels);
1081 if(different_transforms) {
1082 /* the delay samples have already been downmixed, so we upmix the delay
1083 samples in order to reconstruct all channels before downmixing. */
1089 do_imdct(s, s->channels);
1091 if(downmix_output) {
1092 ac3_downmix(s, s->output, 0);
1095 if(downmix_output) {
1096 ac3_downmix(s, s->transform_coeffs, 1);
1101 ac3_downmix(s, s->delay, 0);
1104 do_imdct(s, s->out_channels);
1107 /* convert float to 16-bit integer */
1108 for(ch=0; ch<s->out_channels; ch++) {
1109 for(i=0; i<256; i++) {
1110 s->output[ch][i] += s->add_bias;
1112 s->dsp.float_to_int16(s->int_output[ch], s->output[ch], 256);
1119 * Decode a single AC-3 frame.
1121 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
1122 const uint8_t *buf, int buf_size)
1124 AC3DecodeContext *s = avctx->priv_data;
1125 int16_t *out_samples = (int16_t *)data;
1126 int i, blk, ch, err;
1128 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1129 if (s->input_buffer) {
1130 /* copy input buffer to decoder context to avoid reading past the end
1131 of the buffer, which can be caused by a damaged input stream. */
1132 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_MAX_FRAME_SIZE));
1133 init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
1135 init_get_bits(&s->gbc, buf, buf_size * 8);
1138 /* parse the syncinfo */
1140 err = ac3_parse_header(s);
1142 /* check that reported frame size fits in input buffer */
1143 if(s->frame_size > buf_size) {
1144 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1145 err = AC3_PARSE_ERROR_FRAME_SIZE;
1148 /* check for crc mismatch */
1149 if(err != AC3_PARSE_ERROR_FRAME_SIZE && avctx->error_resilience >= FF_ER_CAREFUL) {
1150 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1151 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1152 err = AC3_PARSE_ERROR_CRC;
1156 /* parse the syncinfo */
1157 if(err && err != AC3_PARSE_ERROR_CRC) {
1159 case AC3_PARSE_ERROR_SYNC:
1160 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1162 case AC3_PARSE_ERROR_BSID:
1163 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1165 case AC3_PARSE_ERROR_SAMPLE_RATE:
1166 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1168 case AC3_PARSE_ERROR_FRAME_SIZE:
1169 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1171 case AC3_PARSE_ERROR_FRAME_TYPE:
1172 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1175 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1180 /* if frame is ok, set audio parameters */
1182 avctx->sample_rate = s->sample_rate;
1183 avctx->bit_rate = s->bit_rate;
1185 /* channel config */
1186 s->out_channels = s->channels;
1187 s->output_mode = s->channel_mode;
1189 s->output_mode |= AC3_OUTPUT_LFEON;
1190 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1191 avctx->request_channels < s->channels) {
1192 s->out_channels = avctx->request_channels;
1193 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1195 avctx->channels = s->out_channels;
1197 /* set downmixing coefficients if needed */
1198 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1199 s->fbw_channels == s->out_channels)) {
1200 set_downmix_coeffs(s);
1202 } else if (!s->out_channels) {
1203 s->out_channels = avctx->channels;
1204 if(s->out_channels < s->channels)
1205 s->output_mode = s->out_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1208 /* parse the audio blocks */
1209 for (blk = 0; blk < NB_BLOCKS; blk++) {
1210 if (!err && ac3_parse_audio_block(s, blk)) {
1211 av_log(avctx, AV_LOG_ERROR, "error parsing the audio block\n");
1213 for (i = 0; i < 256; i++)
1214 for (ch = 0; ch < s->out_channels; ch++)
1215 *(out_samples++) = s->int_output[ch][i];
1217 *data_size = NB_BLOCKS * 256 * avctx->channels * sizeof (int16_t);
1218 return s->frame_size;
1222 * Uninitialize the AC-3 decoder.
1224 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1226 AC3DecodeContext *s = avctx->priv_data;
1227 ff_mdct_end(&s->imdct_512);
1228 ff_mdct_end(&s->imdct_256);
1230 av_freep(&s->input_buffer);
1235 AVCodec ac3_decoder = {
1237 .type = CODEC_TYPE_AUDIO,
1239 .priv_data_size = sizeof (AC3DecodeContext),
1240 .init = ac3_decode_init,
1241 .close = ac3_decode_end,
1242 .decode = ac3_decode_frame,
1243 .long_name = "ATSC A/52 / AC-3",