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"
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 center mix levels
93 * reference: Section 5.4.2.4 cmixlev
95 static const uint8_t center_levels[4] = { 2, 3, 4, 3 };
98 * Table for surround mix levels
99 * reference: Section 5.4.2.5 surmixlev
101 static const uint8_t surround_levels[4] = { 2, 4, 0, 4 };
104 * Table for default stereo downmixing coefficients
105 * reference: Section 7.8.2 Downmixing Into Two Channels
107 static const uint8_t ac3_default_coeffs[8][5][2] = {
108 { { 1, 0 }, { 0, 1 }, },
110 { { 1, 0 }, { 0, 1 }, },
111 { { 1, 0 }, { 3, 3 }, { 0, 1 }, },
112 { { 1, 0 }, { 0, 1 }, { 4, 4 }, },
113 { { 1, 0 }, { 3, 3 }, { 0, 1 }, { 5, 5 }, },
114 { { 1, 0 }, { 0, 1 }, { 4, 0 }, { 0, 4 }, },
115 { { 1, 0 }, { 3, 3 }, { 0, 1 }, { 4, 0 }, { 0, 4 }, },
118 /* override ac3.h to include coupling channel */
119 #undef AC3_MAX_CHANNELS
120 #define AC3_MAX_CHANNELS 7
123 #define AC3_OUTPUT_LFEON 8
126 int channel_mode; ///< channel mode (acmod)
127 int block_switch[AC3_MAX_CHANNELS]; ///< block switch flags
128 int dither_flag[AC3_MAX_CHANNELS]; ///< dither flags
129 int dither_all; ///< true if all channels are dithered
130 int cpl_in_use; ///< coupling in use
131 int channel_in_cpl[AC3_MAX_CHANNELS]; ///< channel in coupling
132 int phase_flags_in_use; ///< phase flags in use
133 int phase_flags[18]; ///< phase flags
134 int cpl_band_struct[18]; ///< coupling band structure
135 int num_rematrixing_bands; ///< number of rematrixing bands
136 int rematrixing_flags[4]; ///< rematrixing flags
137 int exp_strategy[AC3_MAX_CHANNELS]; ///< exponent strategies
138 int snr_offset[AC3_MAX_CHANNELS]; ///< signal-to-noise ratio offsets
139 int fast_gain[AC3_MAX_CHANNELS]; ///< fast gain values (signal-to-mask ratio)
140 int dba_mode[AC3_MAX_CHANNELS]; ///< delta bit allocation mode
141 int dba_nsegs[AC3_MAX_CHANNELS]; ///< number of delta segments
142 uint8_t dba_offsets[AC3_MAX_CHANNELS][8]; ///< delta segment offsets
143 uint8_t dba_lengths[AC3_MAX_CHANNELS][8]; ///< delta segment lengths
144 uint8_t dba_values[AC3_MAX_CHANNELS][8]; ///< delta values for each segment
146 int sample_rate; ///< sample frequency, in Hz
147 int bit_rate; ///< stream bit rate, in bits-per-second
148 int frame_size; ///< current frame size, in bytes
150 int channels; ///< number of total channels
151 int fbw_channels; ///< number of full-bandwidth channels
152 int lfe_on; ///< lfe channel in use
153 int lfe_ch; ///< index of LFE channel
154 int output_mode; ///< output channel configuration
155 int out_channels; ///< number of output channels
157 int center_mix_level; ///< Center mix level index
158 int surround_mix_level; ///< Surround mix level index
159 float downmix_coeffs[AC3_MAX_CHANNELS][2]; ///< stereo downmix coefficients
160 float downmix_coeff_adjust[2]; ///< adjustment needed for each output channel when downmixing
161 float dynamic_range[2]; ///< dynamic range
162 int 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 int fixed_coeffs[AC3_MAX_CHANNELS][256]; ///> fixed-point transform coefficients
176 DECLARE_ALIGNED_16(float, transform_coeffs[AC3_MAX_CHANNELS][256]); ///< transform coefficients
177 int downmixed; ///< indicates if coeffs are currently downmixed
180 MDCTContext imdct_512; ///< for 512 sample IMDCT
181 MDCTContext imdct_256; ///< for 256 sample IMDCT
182 DSPContext dsp; ///< for optimization
183 float add_bias; ///< offset for float_to_int16 conversion
184 float mul_bias; ///< scaling for float_to_int16 conversion
186 DECLARE_ALIGNED_16(float, output[AC3_MAX_CHANNELS][256]); ///< output after imdct transform and windowing
187 DECLARE_ALIGNED_16(short, int_output[AC3_MAX_CHANNELS-1][256]); ///< final 16-bit integer output
188 DECLARE_ALIGNED_16(float, delay[AC3_MAX_CHANNELS][256]); ///< delay - added to the next block
189 DECLARE_ALIGNED_16(float, tmp_imdct[256]); ///< temporary storage for imdct transform
190 DECLARE_ALIGNED_16(float, tmp_output[512]); ///< temporary storage for output before windowing
191 DECLARE_ALIGNED_16(float, window[256]); ///< window coefficients
194 GetBitContext gbc; ///< bitstream reader
195 AVRandomState dith_state; ///< for dither generation
196 AVCodecContext *avctx; ///< parent context
197 uint8_t input_buffer[AC3_MAX_FRAME_SIZE]; ///< temp buffer to prevent overread
201 * Symmetrical Dequantization
202 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
203 * Tables 7.19 to 7.23
206 symmetric_dequant(int code, int levels)
208 return ((code - (levels >> 1)) << 24) / levels;
212 * Initialize tables at runtime.
214 static av_cold void ac3_tables_init(void)
218 /* generate grouped mantissa tables
219 reference: Section 7.3.5 Ungrouping of Mantissas */
220 for(i=0; i<32; i++) {
221 /* bap=1 mantissas */
222 b1_mantissas[i][0] = symmetric_dequant( i / 9 , 3);
223 b1_mantissas[i][1] = symmetric_dequant((i % 9) / 3, 3);
224 b1_mantissas[i][2] = symmetric_dequant((i % 9) % 3, 3);
226 for(i=0; i<128; i++) {
227 /* bap=2 mantissas */
228 b2_mantissas[i][0] = symmetric_dequant( i / 25 , 5);
229 b2_mantissas[i][1] = symmetric_dequant((i % 25) / 5, 5);
230 b2_mantissas[i][2] = symmetric_dequant((i % 25) % 5, 5);
232 /* bap=4 mantissas */
233 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
234 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
236 /* generate ungrouped mantissa tables
237 reference: Tables 7.21 and 7.23 */
239 /* bap=3 mantissas */
240 b3_mantissas[i] = symmetric_dequant(i, 7);
242 for(i=0; i<15; i++) {
243 /* bap=5 mantissas */
244 b5_mantissas[i] = symmetric_dequant(i, 15);
247 /* generate dynamic range table
248 reference: Section 7.7.1 Dynamic Range Control */
249 for(i=0; i<256; i++) {
250 int v = (i >> 5) - ((i >> 7) << 3) - 5;
251 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
254 /* generate exponent tables
255 reference: Section 7.1.3 Exponent Decoding */
256 for(i=0; i<128; i++) {
257 exp_ungroup_tab[i][0] = i / 25;
258 exp_ungroup_tab[i][1] = (i % 25) / 5;
259 exp_ungroup_tab[i][2] = (i % 25) % 5;
265 * AVCodec initialization
267 static av_cold int ac3_decode_init(AVCodecContext *avctx)
269 AC3DecodeContext *s = avctx->priv_data;
274 ff_mdct_init(&s->imdct_256, 8, 1);
275 ff_mdct_init(&s->imdct_512, 9, 1);
276 ff_kbd_window_init(s->window, 5.0, 256);
277 dsputil_init(&s->dsp, avctx);
278 av_init_random(0, &s->dith_state);
280 /* set bias values for float to int16 conversion */
281 if(s->dsp.float_to_int16 == ff_float_to_int16_c) {
282 s->add_bias = 385.0f;
286 s->mul_bias = 32767.0f;
289 /* allow downmixing to stereo or mono */
290 if (avctx->channels > 0 && avctx->request_channels > 0 &&
291 avctx->request_channels < avctx->channels &&
292 avctx->request_channels <= 2) {
293 avctx->channels = avctx->request_channels;
301 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
302 * GetBitContext within AC3DecodeContext must point to
303 * start of the synchronized ac3 bitstream.
305 static int ac3_parse_header(AC3DecodeContext *s)
308 GetBitContext *gbc = &s->gbc;
311 err = ff_ac3_parse_header(gbc->buffer, &hdr);
315 if(hdr.bitstream_id > 10)
316 return AC3_PARSE_ERROR_BSID;
318 /* get decoding parameters from header info */
319 s->bit_alloc_params.sr_code = hdr.sr_code;
320 s->channel_mode = hdr.channel_mode;
321 s->lfe_on = hdr.lfe_on;
322 s->bit_alloc_params.sr_shift = hdr.sr_shift;
323 s->sample_rate = hdr.sample_rate;
324 s->bit_rate = hdr.bit_rate;
325 s->channels = hdr.channels;
326 s->fbw_channels = s->channels - s->lfe_on;
327 s->lfe_ch = s->fbw_channels + 1;
328 s->frame_size = hdr.frame_size;
330 /* set default output to all source channels */
331 s->out_channels = s->channels;
332 s->output_mode = s->channel_mode;
334 s->output_mode |= AC3_OUTPUT_LFEON;
336 /* set default mix levels */
337 s->center_mix_level = 3; // -4.5dB
338 s->surround_mix_level = 4; // -6.0dB
340 /* skip over portion of header which has already been read */
341 skip_bits(gbc, 16); // skip the sync_word
342 skip_bits(gbc, 16); // skip crc1
343 skip_bits(gbc, 8); // skip fscod and frmsizecod
344 skip_bits(gbc, 11); // skip bsid, bsmod, and acmod
345 if(s->channel_mode == AC3_CHMODE_STEREO) {
346 skip_bits(gbc, 2); // skip dsurmod
348 if((s->channel_mode & 1) && s->channel_mode != AC3_CHMODE_MONO)
349 s->center_mix_level = center_levels[get_bits(gbc, 2)];
350 if(s->channel_mode & 4)
351 s->surround_mix_level = surround_levels[get_bits(gbc, 2)];
353 skip_bits1(gbc); // skip lfeon
355 /* read the rest of the bsi. read twice for dual mono mode. */
356 i = !(s->channel_mode);
358 skip_bits(gbc, 5); // skip dialog normalization
360 skip_bits(gbc, 8); //skip compression
362 skip_bits(gbc, 8); //skip language code
364 skip_bits(gbc, 7); //skip audio production information
367 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
369 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
370 TODO: read & use the xbsi1 downmix levels */
372 skip_bits(gbc, 14); //skip timecode1 / xbsi1
374 skip_bits(gbc, 14); //skip timecode2 / xbsi2
376 /* skip additional bitstream info */
377 if (get_bits1(gbc)) {
378 i = get_bits(gbc, 6);
388 * Set stereo downmixing coefficients based on frame header info.
389 * reference: Section 7.8.2 Downmixing Into Two Channels
391 static void set_downmix_coeffs(AC3DecodeContext *s)
394 float cmix = gain_levels[s->center_mix_level];
395 float smix = gain_levels[s->surround_mix_level];
397 for(i=0; i<s->fbw_channels; i++) {
398 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
399 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
401 if(s->channel_mode > 1 && s->channel_mode & 1) {
402 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
404 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
405 int nf = s->channel_mode - 2;
406 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
408 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
409 int nf = s->channel_mode - 4;
410 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
413 /* calculate adjustment needed for each channel to avoid clipping */
414 s->downmix_coeff_adjust[0] = s->downmix_coeff_adjust[1] = 0.0f;
415 for(i=0; i<s->fbw_channels; i++) {
416 s->downmix_coeff_adjust[0] += s->downmix_coeffs[i][0];
417 s->downmix_coeff_adjust[1] += s->downmix_coeffs[i][1];
419 s->downmix_coeff_adjust[0] = 1.0f / s->downmix_coeff_adjust[0];
420 s->downmix_coeff_adjust[1] = 1.0f / s->downmix_coeff_adjust[1];
424 * Decode the grouped exponents according to exponent strategy.
425 * reference: Section 7.1.3 Exponent Decoding
427 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
428 uint8_t absexp, int8_t *dexps)
430 int i, j, grp, group_size;
435 group_size = exp_strategy + (exp_strategy == EXP_D45);
436 for(grp=0,i=0; grp<ngrps; grp++) {
437 expacc = get_bits(gbc, 7);
438 dexp[i++] = exp_ungroup_tab[expacc][0];
439 dexp[i++] = exp_ungroup_tab[expacc][1];
440 dexp[i++] = exp_ungroup_tab[expacc][2];
443 /* convert to absolute exps and expand groups */
445 for(i=0; i<ngrps*3; i++) {
446 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
447 for(j=0; j<group_size; j++) {
448 dexps[(i*group_size)+j] = prevexp;
454 * Generate transform coefficients for each coupled channel in the coupling
455 * range using the coupling coefficients and coupling coordinates.
456 * reference: Section 7.4.3 Coupling Coordinate Format
458 static void uncouple_channels(AC3DecodeContext *s)
460 int i, j, ch, bnd, subbnd;
463 i = s->start_freq[CPL_CH];
464 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
467 for(j=0; j<12; j++) {
468 for(ch=1; ch<=s->fbw_channels; ch++) {
469 if(s->channel_in_cpl[ch]) {
470 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
471 if (ch == 2 && s->phase_flags[bnd])
472 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
477 } while(s->cpl_band_struct[subbnd]);
482 * Grouped mantissas for 3-level 5-level and 11-level quantization
494 * Get the transform coefficients for a particular channel
495 * reference: Section 7.3 Quantization and Decoding of Mantissas
497 static int get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
499 GetBitContext *gbc = &s->gbc;
500 int i, gcode, tbap, start, end;
505 exps = s->dexps[ch_index];
506 bap = s->bap[ch_index];
507 coeffs = s->fixed_coeffs[ch_index];
508 start = s->start_freq[ch_index];
509 end = s->end_freq[ch_index];
511 for (i = start; i < end; i++) {
515 coeffs[i] = (av_random(&s->dith_state) & 0x7FFFFF) - 4194304;
520 gcode = get_bits(gbc, 5);
521 m->b1_mant[0] = b1_mantissas[gcode][0];
522 m->b1_mant[1] = b1_mantissas[gcode][1];
523 m->b1_mant[2] = b1_mantissas[gcode][2];
526 coeffs[i] = m->b1_mant[m->b1ptr++];
531 gcode = get_bits(gbc, 7);
532 m->b2_mant[0] = b2_mantissas[gcode][0];
533 m->b2_mant[1] = b2_mantissas[gcode][1];
534 m->b2_mant[2] = b2_mantissas[gcode][2];
537 coeffs[i] = m->b2_mant[m->b2ptr++];
541 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
546 gcode = get_bits(gbc, 7);
547 m->b4_mant[0] = b4_mantissas[gcode][0];
548 m->b4_mant[1] = b4_mantissas[gcode][1];
551 coeffs[i] = m->b4_mant[m->b4ptr++];
555 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
559 /* asymmetric dequantization */
560 int qlevel = quantization_tab[tbap];
561 coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
565 coeffs[i] >>= exps[i];
572 * Remove random dithering from coefficients with zero-bit mantissas
573 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
575 static void remove_dithering(AC3DecodeContext *s) {
581 for(ch=1; ch<=s->fbw_channels; ch++) {
582 if(!s->dither_flag[ch]) {
583 coeffs = s->fixed_coeffs[ch];
585 if(s->channel_in_cpl[ch])
586 end = s->start_freq[CPL_CH];
588 end = s->end_freq[ch];
589 for(i=0; i<end; i++) {
593 if(s->channel_in_cpl[ch]) {
594 bap = s->bap[CPL_CH];
595 for(; i<s->end_freq[CPL_CH]; i++) {
605 * Get the transform coefficients.
607 static int get_transform_coeffs(AC3DecodeContext *s)
613 m.b1ptr = m.b2ptr = m.b4ptr = 3;
615 for (ch = 1; ch <= s->channels; ch++) {
616 /* transform coefficients for full-bandwidth channel */
617 if (get_transform_coeffs_ch(s, ch, &m))
619 /* tranform coefficients for coupling channel come right after the
620 coefficients for the first coupled channel*/
621 if (s->channel_in_cpl[ch]) {
623 if (get_transform_coeffs_ch(s, CPL_CH, &m)) {
624 av_log(s->avctx, AV_LOG_ERROR, "error in decoupling channels\n");
627 uncouple_channels(s);
630 end = s->end_freq[CPL_CH];
632 end = s->end_freq[ch];
635 s->transform_coeffs[ch][end] = 0;
639 /* if any channel doesn't use dithering, zero appropriate coefficients */
647 * Stereo rematrixing.
648 * reference: Section 7.5.4 Rematrixing : Decoding Technique
650 static void do_rematrixing(AC3DecodeContext *s)
656 end = FFMIN(s->end_freq[1], s->end_freq[2]);
658 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
659 if(s->rematrixing_flags[bnd]) {
660 bndend = FFMIN(end, rematrix_band_tab[bnd+1]);
661 for(i=rematrix_band_tab[bnd]; i<bndend; i++) {
662 tmp0 = s->fixed_coeffs[1][i];
663 tmp1 = s->fixed_coeffs[2][i];
664 s->fixed_coeffs[1][i] = tmp0 + tmp1;
665 s->fixed_coeffs[2][i] = tmp0 - tmp1;
672 * Perform the 256-point IMDCT
674 static void do_imdct_256(AC3DecodeContext *s, int chindex)
677 DECLARE_ALIGNED_16(float, x[128]);
679 float *o_ptr = s->tmp_output;
682 /* de-interleave coefficients */
683 for(k=0; k<128; k++) {
684 x[k] = s->transform_coeffs[chindex][2*k+i];
687 /* run standard IMDCT */
688 s->imdct_256.fft.imdct_calc(&s->imdct_256, o_ptr, x, s->tmp_imdct);
690 /* reverse the post-rotation & reordering from standard IMDCT */
691 for(k=0; k<32; k++) {
692 z[i][32+k].re = -o_ptr[128+2*k];
693 z[i][32+k].im = -o_ptr[2*k];
694 z[i][31-k].re = o_ptr[2*k+1];
695 z[i][31-k].im = o_ptr[128+2*k+1];
699 /* apply AC-3 post-rotation & reordering */
700 for(k=0; k<64; k++) {
701 o_ptr[ 2*k ] = -z[0][ k].im;
702 o_ptr[ 2*k+1] = z[0][63-k].re;
703 o_ptr[128+2*k ] = -z[0][ k].re;
704 o_ptr[128+2*k+1] = z[0][63-k].im;
705 o_ptr[256+2*k ] = -z[1][ k].re;
706 o_ptr[256+2*k+1] = z[1][63-k].im;
707 o_ptr[384+2*k ] = z[1][ k].im;
708 o_ptr[384+2*k+1] = -z[1][63-k].re;
713 * Inverse MDCT Transform.
714 * Convert frequency domain coefficients to time-domain audio samples.
715 * reference: Section 7.9.4 Transformation Equations
717 static inline void do_imdct(AC3DecodeContext *s, int channels)
721 for (ch=1; ch<=channels; ch++) {
722 if (s->block_switch[ch]) {
725 s->imdct_512.fft.imdct_calc(&s->imdct_512, s->tmp_output,
726 s->transform_coeffs[ch], s->tmp_imdct);
728 /* For the first half of the block, apply the window, add the delay
729 from the previous block, and send to output */
730 s->dsp.vector_fmul_add_add(s->output[ch-1], s->tmp_output,
731 s->window, s->delay[ch-1], 0, 256, 1);
732 /* For the second half of the block, apply the window and store the
733 samples to delay, to be combined with the next block */
734 s->dsp.vector_fmul_reverse(s->delay[ch-1], s->tmp_output+256,
740 * Downmix the output to mono or stereo.
742 static void ac3_downmix(AC3DecodeContext *s,
743 float samples[AC3_MAX_CHANNELS][256], int ch_offset)
748 for(i=0; i<256; i++) {
750 for(j=0; j<s->fbw_channels; j++) {
751 v0 += samples[j+ch_offset][i] * s->downmix_coeffs[j][0];
752 v1 += samples[j+ch_offset][i] * s->downmix_coeffs[j][1];
754 v0 *= s->downmix_coeff_adjust[0];
755 v1 *= s->downmix_coeff_adjust[1];
756 if(s->output_mode == AC3_CHMODE_MONO) {
757 samples[ch_offset][i] = (v0 + v1) * LEVEL_MINUS_3DB;
758 } else if(s->output_mode == AC3_CHMODE_STEREO) {
759 samples[ ch_offset][i] = v0;
760 samples[1+ch_offset][i] = v1;
766 * Upmix delay samples from stereo to original channel layout.
768 static void ac3_upmix_delay(AC3DecodeContext *s)
770 int channel_data_size = sizeof(s->delay[0]);
771 switch(s->channel_mode) {
772 case AC3_CHMODE_DUALMONO:
773 case AC3_CHMODE_STEREO:
774 /* upmix mono to stereo */
775 memcpy(s->delay[1], s->delay[0], channel_data_size);
777 case AC3_CHMODE_2F2R:
778 memset(s->delay[3], 0, channel_data_size);
779 case AC3_CHMODE_2F1R:
780 memset(s->delay[2], 0, channel_data_size);
782 case AC3_CHMODE_3F2R:
783 memset(s->delay[4], 0, channel_data_size);
784 case AC3_CHMODE_3F1R:
785 memset(s->delay[3], 0, channel_data_size);
787 memcpy(s->delay[2], s->delay[1], channel_data_size);
788 memset(s->delay[1], 0, channel_data_size);
794 * Parse an audio block from AC-3 bitstream.
796 static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
798 int fbw_channels = s->fbw_channels;
799 int channel_mode = s->channel_mode;
801 int different_transforms;
803 GetBitContext *gbc = &s->gbc;
804 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
806 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
808 /* block switch flags */
809 different_transforms = 0;
810 for (ch = 1; ch <= fbw_channels; ch++) {
811 s->block_switch[ch] = get_bits1(gbc);
812 if(ch > 1 && s->block_switch[ch] != s->block_switch[1])
813 different_transforms = 1;
816 /* dithering flags */
818 for (ch = 1; ch <= fbw_channels; ch++) {
819 s->dither_flag[ch] = get_bits1(gbc);
820 if(!s->dither_flag[ch])
825 i = !(s->channel_mode);
828 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
829 s->avctx->drc_scale)+1.0;
830 } else if(blk == 0) {
831 s->dynamic_range[i] = 1.0f;
835 /* coupling strategy */
836 if (get_bits1(gbc)) {
837 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
838 s->cpl_in_use = get_bits1(gbc);
840 /* coupling in use */
841 int cpl_begin_freq, cpl_end_freq;
843 /* determine which channels are coupled */
844 for (ch = 1; ch <= fbw_channels; ch++)
845 s->channel_in_cpl[ch] = get_bits1(gbc);
847 /* phase flags in use */
848 if (channel_mode == AC3_CHMODE_STEREO)
849 s->phase_flags_in_use = get_bits1(gbc);
851 /* coupling frequency range and band structure */
852 cpl_begin_freq = get_bits(gbc, 4);
853 cpl_end_freq = get_bits(gbc, 4);
854 if (3 + cpl_end_freq - cpl_begin_freq < 0) {
855 av_log(s->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
858 s->num_cpl_bands = s->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
859 s->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
860 s->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
861 for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
862 if (get_bits1(gbc)) {
863 s->cpl_band_struct[bnd] = 1;
867 s->cpl_band_struct[s->num_cpl_subbands-1] = 0;
869 /* coupling not in use */
870 for (ch = 1; ch <= fbw_channels; ch++)
871 s->channel_in_cpl[ch] = 0;
875 /* coupling coordinates */
877 int cpl_coords_exist = 0;
879 for (ch = 1; ch <= fbw_channels; ch++) {
880 if (s->channel_in_cpl[ch]) {
881 if (get_bits1(gbc)) {
882 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
883 cpl_coords_exist = 1;
884 master_cpl_coord = 3 * get_bits(gbc, 2);
885 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
886 cpl_coord_exp = get_bits(gbc, 4);
887 cpl_coord_mant = get_bits(gbc, 4);
888 if (cpl_coord_exp == 15)
889 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
891 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
892 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
898 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
899 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
900 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
905 /* stereo rematrixing strategy and band structure */
906 if (channel_mode == AC3_CHMODE_STEREO) {
907 if (get_bits1(gbc)) {
908 s->num_rematrixing_bands = 4;
909 if(s->cpl_in_use && s->start_freq[CPL_CH] <= 61)
910 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
911 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
912 s->rematrixing_flags[bnd] = get_bits1(gbc);
916 /* exponent strategies for each channel */
917 s->exp_strategy[CPL_CH] = EXP_REUSE;
918 s->exp_strategy[s->lfe_ch] = EXP_REUSE;
919 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
921 s->exp_strategy[ch] = get_bits(gbc, 1);
923 s->exp_strategy[ch] = get_bits(gbc, 2);
924 if(s->exp_strategy[ch] != EXP_REUSE)
925 bit_alloc_stages[ch] = 3;
928 /* channel bandwidth */
929 for (ch = 1; ch <= fbw_channels; ch++) {
930 s->start_freq[ch] = 0;
931 if (s->exp_strategy[ch] != EXP_REUSE) {
932 int prev = s->end_freq[ch];
933 if (s->channel_in_cpl[ch])
934 s->end_freq[ch] = s->start_freq[CPL_CH];
936 int bandwidth_code = get_bits(gbc, 6);
937 if (bandwidth_code > 60) {
938 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
941 s->end_freq[ch] = bandwidth_code * 3 + 73;
943 if(blk > 0 && s->end_freq[ch] != prev)
944 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
947 s->start_freq[s->lfe_ch] = 0;
948 s->end_freq[s->lfe_ch] = 7;
950 /* decode exponents for each channel */
951 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
952 if (s->exp_strategy[ch] != EXP_REUSE) {
953 int group_size, num_groups;
954 group_size = 3 << (s->exp_strategy[ch] - 1);
956 num_groups = (s->end_freq[ch] - s->start_freq[ch]) / group_size;
957 else if(ch == s->lfe_ch)
960 num_groups = (s->end_freq[ch] + group_size - 4) / group_size;
961 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
962 decode_exponents(gbc, s->exp_strategy[ch], num_groups, s->dexps[ch][0],
963 &s->dexps[ch][s->start_freq[ch]+!!ch]);
964 if(ch != CPL_CH && ch != s->lfe_ch)
965 skip_bits(gbc, 2); /* skip gainrng */
969 /* bit allocation information */
970 if (get_bits1(gbc)) {
971 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
972 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
973 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
974 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
975 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
976 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
977 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
981 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
982 if (get_bits1(gbc)) {
984 csnr = (get_bits(gbc, 6) - 15) << 4;
985 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
986 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
987 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
989 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
992 /* coupling leak information */
993 if (s->cpl_in_use && get_bits1(gbc)) {
994 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
995 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
996 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
999 /* delta bit allocation information */
1000 if (get_bits1(gbc)) {
1001 /* delta bit allocation exists (strategy) */
1002 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
1003 s->dba_mode[ch] = get_bits(gbc, 2);
1004 if (s->dba_mode[ch] == DBA_RESERVED) {
1005 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1008 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1010 /* channel delta offset, len and bit allocation */
1011 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
1012 if (s->dba_mode[ch] == DBA_NEW) {
1013 s->dba_nsegs[ch] = get_bits(gbc, 3);
1014 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
1015 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1016 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1017 s->dba_values[ch][seg] = get_bits(gbc, 3);
1021 } else if(blk == 0) {
1022 for(ch=0; ch<=s->channels; ch++) {
1023 s->dba_mode[ch] = DBA_NONE;
1027 /* Bit allocation */
1028 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
1029 if(bit_alloc_stages[ch] > 2) {
1030 /* Exponent mapping into PSD and PSD integration */
1031 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1032 s->start_freq[ch], s->end_freq[ch],
1033 s->psd[ch], s->band_psd[ch]);
1035 if(bit_alloc_stages[ch] > 1) {
1036 /* Compute excitation function, Compute masking curve, and
1037 Apply delta bit allocation */
1038 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1039 s->start_freq[ch], s->end_freq[ch],
1040 s->fast_gain[ch], (ch == s->lfe_ch),
1041 s->dba_mode[ch], s->dba_nsegs[ch],
1042 s->dba_offsets[ch], s->dba_lengths[ch],
1043 s->dba_values[ch], s->mask[ch]);
1045 if(bit_alloc_stages[ch] > 0) {
1046 /* Compute bit allocation */
1047 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1048 s->start_freq[ch], s->end_freq[ch],
1050 s->bit_alloc_params.floor,
1055 /* unused dummy data */
1056 if (get_bits1(gbc)) {
1057 int skipl = get_bits(gbc, 9);
1062 /* unpack the transform coefficients
1063 this also uncouples channels if coupling is in use. */
1064 if (get_transform_coeffs(s)) {
1065 av_log(s->avctx, AV_LOG_ERROR, "Error in routine get_transform_coeffs\n");
1069 /* recover coefficients if rematrixing is in use */
1070 if(s->channel_mode == AC3_CHMODE_STEREO)
1073 /* apply scaling to coefficients (headroom, dynrng) */
1074 for(ch=1; ch<=s->channels; ch++) {
1075 float gain = s->mul_bias / 4194304.0f;
1076 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1077 gain *= s->dynamic_range[ch-1];
1079 gain *= s->dynamic_range[0];
1081 for(i=0; i<256; i++) {
1082 s->transform_coeffs[ch][i] = s->fixed_coeffs[ch][i] * gain;
1086 /* downmix and MDCT. order depends on whether block switching is used for
1087 any channel in this block. this is because coefficients for the long
1088 and short transforms cannot be mixed. */
1089 downmix_output = s->channels != s->out_channels &&
1090 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1091 s->fbw_channels == s->out_channels);
1092 if(different_transforms) {
1093 /* the delay samples have already been downmixed, so we upmix the delay
1094 samples in order to reconstruct all channels before downmixing. */
1100 do_imdct(s, s->channels);
1102 if(downmix_output) {
1103 ac3_downmix(s, s->output, 0);
1106 if(downmix_output) {
1107 ac3_downmix(s, s->transform_coeffs, 1);
1112 ac3_downmix(s, s->delay, 0);
1115 do_imdct(s, s->out_channels);
1118 /* convert float to 16-bit integer */
1119 for(ch=0; ch<s->out_channels; ch++) {
1120 for(i=0; i<256; i++) {
1121 s->output[ch][i] += s->add_bias;
1123 s->dsp.float_to_int16(s->int_output[ch], s->output[ch], 256);
1130 * Decode a single AC-3 frame.
1132 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
1133 const uint8_t *buf, int buf_size)
1135 AC3DecodeContext *s = avctx->priv_data;
1136 int16_t *out_samples = (int16_t *)data;
1137 int i, blk, ch, err;
1139 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1140 if(avctx->error_resilience >= FF_ER_CAREFUL) {
1141 /* copy input buffer to decoder context to avoid reading past the end
1142 of the buffer, which can be caused by a damaged input stream. */
1143 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_MAX_FRAME_SIZE));
1144 init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
1146 init_get_bits(&s->gbc, buf, buf_size * 8);
1149 /* parse the syncinfo */
1150 err = ac3_parse_header(s);
1153 case AC3_PARSE_ERROR_SYNC:
1154 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1156 case AC3_PARSE_ERROR_BSID:
1157 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1159 case AC3_PARSE_ERROR_SAMPLE_RATE:
1160 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1162 case AC3_PARSE_ERROR_FRAME_SIZE:
1163 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1165 case AC3_PARSE_ERROR_STREAM_TYPE:
1166 av_log(avctx, AV_LOG_ERROR, "invalid stream type\n");
1169 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1175 /* check that reported frame size fits in input buffer */
1176 if(s->frame_size > buf_size) {
1177 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1181 /* check for crc mismatch */
1182 if(avctx->error_resilience >= FF_ER_CAREFUL) {
1183 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1184 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1187 /* TODO: error concealment */
1190 avctx->sample_rate = s->sample_rate;
1191 avctx->bit_rate = s->bit_rate;
1193 /* channel config */
1194 s->out_channels = s->channels;
1195 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1196 avctx->request_channels < s->channels) {
1197 s->out_channels = avctx->request_channels;
1198 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1200 avctx->channels = s->out_channels;
1202 /* set downmixing coefficients if needed */
1203 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1204 s->fbw_channels == s->out_channels)) {
1205 set_downmix_coeffs(s);
1208 /* parse the audio blocks */
1209 for (blk = 0; blk < NB_BLOCKS; blk++) {
1210 if (ac3_parse_audio_block(s, blk)) {
1211 av_log(avctx, AV_LOG_ERROR, "error parsing the audio block\n");
1213 return s->frame_size;
1215 for (i = 0; i < 256; i++)
1216 for (ch = 0; ch < s->out_channels; ch++)
1217 *(out_samples++) = s->int_output[ch][i];
1219 *data_size = NB_BLOCKS * 256 * avctx->channels * sizeof (int16_t);
1220 return s->frame_size;
1224 * Uninitialize the AC-3 decoder.
1226 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1228 AC3DecodeContext *s = avctx->priv_data;
1229 ff_mdct_end(&s->imdct_512);
1230 ff_mdct_end(&s->imdct_256);
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,