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; ///< 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;
297 /* allocate context input buffer */
298 if (avctx->error_resilience >= FF_ER_CAREFUL) {
299 s->input_buffer = av_mallocz(AC3_MAX_FRAME_SIZE + FF_INPUT_BUFFER_PADDING_SIZE);
300 if (!s->input_buffer)
301 return AVERROR_NOMEM;
308 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
309 * GetBitContext within AC3DecodeContext must point to
310 * start of the synchronized ac3 bitstream.
312 static int ac3_parse_header(AC3DecodeContext *s)
315 GetBitContext *gbc = &s->gbc;
318 err = ff_ac3_parse_header(gbc->buffer, &hdr);
322 if(hdr.bitstream_id > 10)
323 return AC3_PARSE_ERROR_BSID;
325 /* get decoding parameters from header info */
326 s->bit_alloc_params.sr_code = hdr.sr_code;
327 s->channel_mode = hdr.channel_mode;
328 s->lfe_on = hdr.lfe_on;
329 s->bit_alloc_params.sr_shift = hdr.sr_shift;
330 s->sample_rate = hdr.sample_rate;
331 s->bit_rate = hdr.bit_rate;
332 s->channels = hdr.channels;
333 s->fbw_channels = s->channels - s->lfe_on;
334 s->lfe_ch = s->fbw_channels + 1;
335 s->frame_size = hdr.frame_size;
337 /* set default output to all source channels */
338 s->out_channels = s->channels;
339 s->output_mode = s->channel_mode;
341 s->output_mode |= AC3_OUTPUT_LFEON;
343 /* set default mix levels */
344 s->center_mix_level = 3; // -4.5dB
345 s->surround_mix_level = 4; // -6.0dB
347 /* skip over portion of header which has already been read */
348 skip_bits(gbc, 16); // skip the sync_word
349 skip_bits(gbc, 16); // skip crc1
350 skip_bits(gbc, 8); // skip fscod and frmsizecod
351 skip_bits(gbc, 11); // skip bsid, bsmod, and acmod
352 if(s->channel_mode == AC3_CHMODE_STEREO) {
353 skip_bits(gbc, 2); // skip dsurmod
355 if((s->channel_mode & 1) && s->channel_mode != AC3_CHMODE_MONO)
356 s->center_mix_level = center_levels[get_bits(gbc, 2)];
357 if(s->channel_mode & 4)
358 s->surround_mix_level = surround_levels[get_bits(gbc, 2)];
360 skip_bits1(gbc); // skip lfeon
362 /* read the rest of the bsi. read twice for dual mono mode. */
363 i = !(s->channel_mode);
365 skip_bits(gbc, 5); // skip dialog normalization
367 skip_bits(gbc, 8); //skip compression
369 skip_bits(gbc, 8); //skip language code
371 skip_bits(gbc, 7); //skip audio production information
374 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
376 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
377 TODO: read & use the xbsi1 downmix levels */
379 skip_bits(gbc, 14); //skip timecode1 / xbsi1
381 skip_bits(gbc, 14); //skip timecode2 / xbsi2
383 /* skip additional bitstream info */
384 if (get_bits1(gbc)) {
385 i = get_bits(gbc, 6);
395 * Set stereo downmixing coefficients based on frame header info.
396 * reference: Section 7.8.2 Downmixing Into Two Channels
398 static void set_downmix_coeffs(AC3DecodeContext *s)
401 float cmix = gain_levels[s->center_mix_level];
402 float smix = gain_levels[s->surround_mix_level];
404 for(i=0; i<s->fbw_channels; i++) {
405 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
406 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
408 if(s->channel_mode > 1 && s->channel_mode & 1) {
409 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
411 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
412 int nf = s->channel_mode - 2;
413 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
415 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
416 int nf = s->channel_mode - 4;
417 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
420 /* calculate adjustment needed for each channel to avoid clipping */
421 s->downmix_coeff_adjust[0] = s->downmix_coeff_adjust[1] = 0.0f;
422 for(i=0; i<s->fbw_channels; i++) {
423 s->downmix_coeff_adjust[0] += s->downmix_coeffs[i][0];
424 s->downmix_coeff_adjust[1] += s->downmix_coeffs[i][1];
426 s->downmix_coeff_adjust[0] = 1.0f / s->downmix_coeff_adjust[0];
427 s->downmix_coeff_adjust[1] = 1.0f / s->downmix_coeff_adjust[1];
431 * Decode the grouped exponents according to exponent strategy.
432 * reference: Section 7.1.3 Exponent Decoding
434 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
435 uint8_t absexp, int8_t *dexps)
437 int i, j, grp, group_size;
442 group_size = exp_strategy + (exp_strategy == EXP_D45);
443 for(grp=0,i=0; grp<ngrps; grp++) {
444 expacc = get_bits(gbc, 7);
445 dexp[i++] = exp_ungroup_tab[expacc][0];
446 dexp[i++] = exp_ungroup_tab[expacc][1];
447 dexp[i++] = exp_ungroup_tab[expacc][2];
450 /* convert to absolute exps and expand groups */
452 for(i=0; i<ngrps*3; i++) {
453 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
454 for(j=0; j<group_size; j++) {
455 dexps[(i*group_size)+j] = prevexp;
461 * Generate transform coefficients for each coupled channel in the coupling
462 * range using the coupling coefficients and coupling coordinates.
463 * reference: Section 7.4.3 Coupling Coordinate Format
465 static void uncouple_channels(AC3DecodeContext *s)
467 int i, j, ch, bnd, subbnd;
470 i = s->start_freq[CPL_CH];
471 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
474 for(j=0; j<12; j++) {
475 for(ch=1; ch<=s->fbw_channels; ch++) {
476 if(s->channel_in_cpl[ch]) {
477 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
478 if (ch == 2 && s->phase_flags[bnd])
479 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
484 } while(s->cpl_band_struct[subbnd]);
489 * Grouped mantissas for 3-level 5-level and 11-level quantization
501 * Get the transform coefficients for a particular channel
502 * reference: Section 7.3 Quantization and Decoding of Mantissas
504 static int get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
506 GetBitContext *gbc = &s->gbc;
507 int i, gcode, tbap, start, end;
512 exps = s->dexps[ch_index];
513 bap = s->bap[ch_index];
514 coeffs = s->fixed_coeffs[ch_index];
515 start = s->start_freq[ch_index];
516 end = s->end_freq[ch_index];
518 for (i = start; i < end; i++) {
522 coeffs[i] = (av_random(&s->dith_state) & 0x7FFFFF) - 4194304;
527 gcode = get_bits(gbc, 5);
528 m->b1_mant[0] = b1_mantissas[gcode][0];
529 m->b1_mant[1] = b1_mantissas[gcode][1];
530 m->b1_mant[2] = b1_mantissas[gcode][2];
533 coeffs[i] = m->b1_mant[m->b1ptr++];
538 gcode = get_bits(gbc, 7);
539 m->b2_mant[0] = b2_mantissas[gcode][0];
540 m->b2_mant[1] = b2_mantissas[gcode][1];
541 m->b2_mant[2] = b2_mantissas[gcode][2];
544 coeffs[i] = m->b2_mant[m->b2ptr++];
548 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
553 gcode = get_bits(gbc, 7);
554 m->b4_mant[0] = b4_mantissas[gcode][0];
555 m->b4_mant[1] = b4_mantissas[gcode][1];
558 coeffs[i] = m->b4_mant[m->b4ptr++];
562 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
566 /* asymmetric dequantization */
567 int qlevel = quantization_tab[tbap];
568 coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
572 coeffs[i] >>= exps[i];
579 * Remove random dithering from coefficients with zero-bit mantissas
580 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
582 static void remove_dithering(AC3DecodeContext *s) {
588 for(ch=1; ch<=s->fbw_channels; ch++) {
589 if(!s->dither_flag[ch]) {
590 coeffs = s->fixed_coeffs[ch];
592 if(s->channel_in_cpl[ch])
593 end = s->start_freq[CPL_CH];
595 end = s->end_freq[ch];
596 for(i=0; i<end; i++) {
600 if(s->channel_in_cpl[ch]) {
601 bap = s->bap[CPL_CH];
602 for(; i<s->end_freq[CPL_CH]; i++) {
612 * Get the transform coefficients.
614 static int get_transform_coeffs(AC3DecodeContext *s)
620 m.b1ptr = m.b2ptr = m.b4ptr = 3;
622 for (ch = 1; ch <= s->channels; ch++) {
623 /* transform coefficients for full-bandwidth channel */
624 if (get_transform_coeffs_ch(s, ch, &m))
626 /* tranform coefficients for coupling channel come right after the
627 coefficients for the first coupled channel*/
628 if (s->channel_in_cpl[ch]) {
630 if (get_transform_coeffs_ch(s, CPL_CH, &m)) {
631 av_log(s->avctx, AV_LOG_ERROR, "error in decoupling channels\n");
634 uncouple_channels(s);
637 end = s->end_freq[CPL_CH];
639 end = s->end_freq[ch];
642 s->transform_coeffs[ch][end] = 0;
646 /* if any channel doesn't use dithering, zero appropriate coefficients */
654 * Stereo rematrixing.
655 * reference: Section 7.5.4 Rematrixing : Decoding Technique
657 static void do_rematrixing(AC3DecodeContext *s)
663 end = FFMIN(s->end_freq[1], s->end_freq[2]);
665 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
666 if(s->rematrixing_flags[bnd]) {
667 bndend = FFMIN(end, rematrix_band_tab[bnd+1]);
668 for(i=rematrix_band_tab[bnd]; i<bndend; i++) {
669 tmp0 = s->fixed_coeffs[1][i];
670 tmp1 = s->fixed_coeffs[2][i];
671 s->fixed_coeffs[1][i] = tmp0 + tmp1;
672 s->fixed_coeffs[2][i] = tmp0 - tmp1;
679 * Perform the 256-point IMDCT
681 static void do_imdct_256(AC3DecodeContext *s, int chindex)
684 DECLARE_ALIGNED_16(float, x[128]);
686 float *o_ptr = s->tmp_output;
689 /* de-interleave coefficients */
690 for(k=0; k<128; k++) {
691 x[k] = s->transform_coeffs[chindex][2*k+i];
694 /* run standard IMDCT */
695 s->imdct_256.fft.imdct_calc(&s->imdct_256, o_ptr, x, s->tmp_imdct);
697 /* reverse the post-rotation & reordering from standard IMDCT */
698 for(k=0; k<32; k++) {
699 z[i][32+k].re = -o_ptr[128+2*k];
700 z[i][32+k].im = -o_ptr[2*k];
701 z[i][31-k].re = o_ptr[2*k+1];
702 z[i][31-k].im = o_ptr[128+2*k+1];
706 /* apply AC-3 post-rotation & reordering */
707 for(k=0; k<64; k++) {
708 o_ptr[ 2*k ] = -z[0][ k].im;
709 o_ptr[ 2*k+1] = z[0][63-k].re;
710 o_ptr[128+2*k ] = -z[0][ k].re;
711 o_ptr[128+2*k+1] = z[0][63-k].im;
712 o_ptr[256+2*k ] = -z[1][ k].re;
713 o_ptr[256+2*k+1] = z[1][63-k].im;
714 o_ptr[384+2*k ] = z[1][ k].im;
715 o_ptr[384+2*k+1] = -z[1][63-k].re;
720 * Inverse MDCT Transform.
721 * Convert frequency domain coefficients to time-domain audio samples.
722 * reference: Section 7.9.4 Transformation Equations
724 static inline void do_imdct(AC3DecodeContext *s, int channels)
728 for (ch=1; ch<=channels; ch++) {
729 if (s->block_switch[ch]) {
732 s->imdct_512.fft.imdct_calc(&s->imdct_512, s->tmp_output,
733 s->transform_coeffs[ch], s->tmp_imdct);
735 /* For the first half of the block, apply the window, add the delay
736 from the previous block, and send to output */
737 s->dsp.vector_fmul_add_add(s->output[ch-1], s->tmp_output,
738 s->window, s->delay[ch-1], 0, 256, 1);
739 /* For the second half of the block, apply the window and store the
740 samples to delay, to be combined with the next block */
741 s->dsp.vector_fmul_reverse(s->delay[ch-1], s->tmp_output+256,
747 * Downmix the output to mono or stereo.
749 static void ac3_downmix(AC3DecodeContext *s,
750 float samples[AC3_MAX_CHANNELS][256], int ch_offset)
755 for(i=0; i<256; i++) {
757 for(j=0; j<s->fbw_channels; j++) {
758 v0 += samples[j+ch_offset][i] * s->downmix_coeffs[j][0];
759 v1 += samples[j+ch_offset][i] * s->downmix_coeffs[j][1];
761 v0 *= s->downmix_coeff_adjust[0];
762 v1 *= s->downmix_coeff_adjust[1];
763 if(s->output_mode == AC3_CHMODE_MONO) {
764 samples[ch_offset][i] = (v0 + v1) * LEVEL_MINUS_3DB;
765 } else if(s->output_mode == AC3_CHMODE_STEREO) {
766 samples[ ch_offset][i] = v0;
767 samples[1+ch_offset][i] = v1;
773 * Upmix delay samples from stereo to original channel layout.
775 static void ac3_upmix_delay(AC3DecodeContext *s)
777 int channel_data_size = sizeof(s->delay[0]);
778 switch(s->channel_mode) {
779 case AC3_CHMODE_DUALMONO:
780 case AC3_CHMODE_STEREO:
781 /* upmix mono to stereo */
782 memcpy(s->delay[1], s->delay[0], channel_data_size);
784 case AC3_CHMODE_2F2R:
785 memset(s->delay[3], 0, channel_data_size);
786 case AC3_CHMODE_2F1R:
787 memset(s->delay[2], 0, channel_data_size);
789 case AC3_CHMODE_3F2R:
790 memset(s->delay[4], 0, channel_data_size);
791 case AC3_CHMODE_3F1R:
792 memset(s->delay[3], 0, channel_data_size);
794 memcpy(s->delay[2], s->delay[1], channel_data_size);
795 memset(s->delay[1], 0, channel_data_size);
801 * Parse an audio block from AC-3 bitstream.
803 static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
805 int fbw_channels = s->fbw_channels;
806 int channel_mode = s->channel_mode;
808 int different_transforms;
810 GetBitContext *gbc = &s->gbc;
811 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
813 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
815 /* block switch flags */
816 different_transforms = 0;
817 for (ch = 1; ch <= fbw_channels; ch++) {
818 s->block_switch[ch] = get_bits1(gbc);
819 if(ch > 1 && s->block_switch[ch] != s->block_switch[1])
820 different_transforms = 1;
823 /* dithering flags */
825 for (ch = 1; ch <= fbw_channels; ch++) {
826 s->dither_flag[ch] = get_bits1(gbc);
827 if(!s->dither_flag[ch])
832 i = !(s->channel_mode);
835 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
836 s->avctx->drc_scale)+1.0;
837 } else if(blk == 0) {
838 s->dynamic_range[i] = 1.0f;
842 /* coupling strategy */
843 if (get_bits1(gbc)) {
844 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
845 s->cpl_in_use = get_bits1(gbc);
847 /* coupling in use */
848 int cpl_begin_freq, cpl_end_freq;
850 /* determine which channels are coupled */
851 for (ch = 1; ch <= fbw_channels; ch++)
852 s->channel_in_cpl[ch] = get_bits1(gbc);
854 /* phase flags in use */
855 if (channel_mode == AC3_CHMODE_STEREO)
856 s->phase_flags_in_use = get_bits1(gbc);
858 /* coupling frequency range and band structure */
859 cpl_begin_freq = get_bits(gbc, 4);
860 cpl_end_freq = get_bits(gbc, 4);
861 if (3 + cpl_end_freq - cpl_begin_freq < 0) {
862 av_log(s->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
865 s->num_cpl_bands = s->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
866 s->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
867 s->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
868 for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
869 if (get_bits1(gbc)) {
870 s->cpl_band_struct[bnd] = 1;
874 s->cpl_band_struct[s->num_cpl_subbands-1] = 0;
876 /* coupling not in use */
877 for (ch = 1; ch <= fbw_channels; ch++)
878 s->channel_in_cpl[ch] = 0;
882 /* coupling coordinates */
884 int cpl_coords_exist = 0;
886 for (ch = 1; ch <= fbw_channels; ch++) {
887 if (s->channel_in_cpl[ch]) {
888 if (get_bits1(gbc)) {
889 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
890 cpl_coords_exist = 1;
891 master_cpl_coord = 3 * get_bits(gbc, 2);
892 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
893 cpl_coord_exp = get_bits(gbc, 4);
894 cpl_coord_mant = get_bits(gbc, 4);
895 if (cpl_coord_exp == 15)
896 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
898 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
899 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
905 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
906 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
907 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
912 /* stereo rematrixing strategy and band structure */
913 if (channel_mode == AC3_CHMODE_STEREO) {
914 if (get_bits1(gbc)) {
915 s->num_rematrixing_bands = 4;
916 if(s->cpl_in_use && s->start_freq[CPL_CH] <= 61)
917 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
918 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
919 s->rematrixing_flags[bnd] = get_bits1(gbc);
923 /* exponent strategies for each channel */
924 s->exp_strategy[CPL_CH] = EXP_REUSE;
925 s->exp_strategy[s->lfe_ch] = EXP_REUSE;
926 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
928 s->exp_strategy[ch] = get_bits(gbc, 1);
930 s->exp_strategy[ch] = get_bits(gbc, 2);
931 if(s->exp_strategy[ch] != EXP_REUSE)
932 bit_alloc_stages[ch] = 3;
935 /* channel bandwidth */
936 for (ch = 1; ch <= fbw_channels; ch++) {
937 s->start_freq[ch] = 0;
938 if (s->exp_strategy[ch] != EXP_REUSE) {
939 int prev = s->end_freq[ch];
940 if (s->channel_in_cpl[ch])
941 s->end_freq[ch] = s->start_freq[CPL_CH];
943 int bandwidth_code = get_bits(gbc, 6);
944 if (bandwidth_code > 60) {
945 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
948 s->end_freq[ch] = bandwidth_code * 3 + 73;
950 if(blk > 0 && s->end_freq[ch] != prev)
951 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
954 s->start_freq[s->lfe_ch] = 0;
955 s->end_freq[s->lfe_ch] = 7;
957 /* decode exponents for each channel */
958 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
959 if (s->exp_strategy[ch] != EXP_REUSE) {
960 int group_size, num_groups;
961 group_size = 3 << (s->exp_strategy[ch] - 1);
963 num_groups = (s->end_freq[ch] - s->start_freq[ch]) / group_size;
964 else if(ch == s->lfe_ch)
967 num_groups = (s->end_freq[ch] + group_size - 4) / group_size;
968 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
969 decode_exponents(gbc, s->exp_strategy[ch], num_groups, s->dexps[ch][0],
970 &s->dexps[ch][s->start_freq[ch]+!!ch]);
971 if(ch != CPL_CH && ch != s->lfe_ch)
972 skip_bits(gbc, 2); /* skip gainrng */
976 /* bit allocation information */
977 if (get_bits1(gbc)) {
978 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
979 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
980 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
981 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
982 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
983 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
984 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
988 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
989 if (get_bits1(gbc)) {
991 csnr = (get_bits(gbc, 6) - 15) << 4;
992 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
993 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
994 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
996 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
999 /* coupling leak information */
1000 if (s->cpl_in_use && get_bits1(gbc)) {
1001 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
1002 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
1003 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
1006 /* delta bit allocation information */
1007 if (get_bits1(gbc)) {
1008 /* delta bit allocation exists (strategy) */
1009 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
1010 s->dba_mode[ch] = get_bits(gbc, 2);
1011 if (s->dba_mode[ch] == DBA_RESERVED) {
1012 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1015 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1017 /* channel delta offset, len and bit allocation */
1018 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
1019 if (s->dba_mode[ch] == DBA_NEW) {
1020 s->dba_nsegs[ch] = get_bits(gbc, 3);
1021 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
1022 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1023 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1024 s->dba_values[ch][seg] = get_bits(gbc, 3);
1028 } else if(blk == 0) {
1029 for(ch=0; ch<=s->channels; ch++) {
1030 s->dba_mode[ch] = DBA_NONE;
1034 /* Bit allocation */
1035 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
1036 if(bit_alloc_stages[ch] > 2) {
1037 /* Exponent mapping into PSD and PSD integration */
1038 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1039 s->start_freq[ch], s->end_freq[ch],
1040 s->psd[ch], s->band_psd[ch]);
1042 if(bit_alloc_stages[ch] > 1) {
1043 /* Compute excitation function, Compute masking curve, and
1044 Apply delta bit allocation */
1045 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1046 s->start_freq[ch], s->end_freq[ch],
1047 s->fast_gain[ch], (ch == s->lfe_ch),
1048 s->dba_mode[ch], s->dba_nsegs[ch],
1049 s->dba_offsets[ch], s->dba_lengths[ch],
1050 s->dba_values[ch], s->mask[ch]);
1052 if(bit_alloc_stages[ch] > 0) {
1053 /* Compute bit allocation */
1054 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1055 s->start_freq[ch], s->end_freq[ch],
1057 s->bit_alloc_params.floor,
1062 /* unused dummy data */
1063 if (get_bits1(gbc)) {
1064 int skipl = get_bits(gbc, 9);
1069 /* unpack the transform coefficients
1070 this also uncouples channels if coupling is in use. */
1071 if (get_transform_coeffs(s)) {
1072 av_log(s->avctx, AV_LOG_ERROR, "Error in routine get_transform_coeffs\n");
1076 /* recover coefficients if rematrixing is in use */
1077 if(s->channel_mode == AC3_CHMODE_STEREO)
1080 /* apply scaling to coefficients (headroom, dynrng) */
1081 for(ch=1; ch<=s->channels; ch++) {
1082 float gain = s->mul_bias / 4194304.0f;
1083 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1084 gain *= s->dynamic_range[ch-1];
1086 gain *= s->dynamic_range[0];
1088 for(i=0; i<256; i++) {
1089 s->transform_coeffs[ch][i] = s->fixed_coeffs[ch][i] * gain;
1093 /* downmix and MDCT. order depends on whether block switching is used for
1094 any channel in this block. this is because coefficients for the long
1095 and short transforms cannot be mixed. */
1096 downmix_output = s->channels != s->out_channels &&
1097 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1098 s->fbw_channels == s->out_channels);
1099 if(different_transforms) {
1100 /* the delay samples have already been downmixed, so we upmix the delay
1101 samples in order to reconstruct all channels before downmixing. */
1107 do_imdct(s, s->channels);
1109 if(downmix_output) {
1110 ac3_downmix(s, s->output, 0);
1113 if(downmix_output) {
1114 ac3_downmix(s, s->transform_coeffs, 1);
1119 ac3_downmix(s, s->delay, 0);
1122 do_imdct(s, s->out_channels);
1125 /* convert float to 16-bit integer */
1126 for(ch=0; ch<s->out_channels; ch++) {
1127 for(i=0; i<256; i++) {
1128 s->output[ch][i] += s->add_bias;
1130 s->dsp.float_to_int16(s->int_output[ch], s->output[ch], 256);
1137 * Decode a single AC-3 frame.
1139 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
1140 const uint8_t *buf, int buf_size)
1142 AC3DecodeContext *s = avctx->priv_data;
1143 int16_t *out_samples = (int16_t *)data;
1144 int i, blk, ch, err;
1146 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1147 if (s->input_buffer) {
1148 /* copy input buffer to decoder context to avoid reading past the end
1149 of the buffer, which can be caused by a damaged input stream. */
1150 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_MAX_FRAME_SIZE));
1151 init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
1153 init_get_bits(&s->gbc, buf, buf_size * 8);
1156 /* parse the syncinfo */
1157 err = ac3_parse_header(s);
1160 case AC3_PARSE_ERROR_SYNC:
1161 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1163 case AC3_PARSE_ERROR_BSID:
1164 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1166 case AC3_PARSE_ERROR_SAMPLE_RATE:
1167 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1169 case AC3_PARSE_ERROR_FRAME_SIZE:
1170 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1172 case AC3_PARSE_ERROR_FRAME_TYPE:
1173 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1176 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1182 /* check that reported frame size fits in input buffer */
1183 if(s->frame_size > buf_size) {
1184 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1188 /* check for crc mismatch */
1189 if(avctx->error_resilience >= FF_ER_CAREFUL) {
1190 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1191 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1194 /* TODO: error concealment */
1197 avctx->sample_rate = s->sample_rate;
1198 avctx->bit_rate = s->bit_rate;
1200 /* channel config */
1201 s->out_channels = s->channels;
1202 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1203 avctx->request_channels < s->channels) {
1204 s->out_channels = avctx->request_channels;
1205 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1207 avctx->channels = s->out_channels;
1209 /* set downmixing coefficients if needed */
1210 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1211 s->fbw_channels == s->out_channels)) {
1212 set_downmix_coeffs(s);
1215 /* parse the audio blocks */
1216 for (blk = 0; blk < NB_BLOCKS; blk++) {
1217 if (ac3_parse_audio_block(s, blk)) {
1218 av_log(avctx, AV_LOG_ERROR, "error parsing the audio block\n");
1220 return s->frame_size;
1222 for (i = 0; i < 256; i++)
1223 for (ch = 0; ch < s->out_channels; ch++)
1224 *(out_samples++) = s->int_output[ch][i];
1226 *data_size = NB_BLOCKS * 256 * avctx->channels * sizeof (int16_t);
1227 return s->frame_size;
1231 * Uninitialize the AC-3 decoder.
1233 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1235 AC3DecodeContext *s = avctx->priv_data;
1236 ff_mdct_end(&s->imdct_512);
1237 ff_mdct_end(&s->imdct_256);
1239 av_freep(&s->input_buffer);
1244 AVCodec ac3_decoder = {
1246 .type = CODEC_TYPE_AUDIO,
1248 .priv_data_size = sizeof (AC3DecodeContext),
1249 .init = ac3_decode_init,
1250 .close = ac3_decode_end,
1251 .decode = ac3_decode_frame,