3 * This code is developed as part of Google Summer of Code 2006 Program.
5 * Copyright (c) 2006 Kartikey Mahendra BHATT (bhattkm at gmail dot com).
6 * Copyright (c) 2007 Justin Ruggles
8 * Portions of this code are derived from liba52
9 * http://liba52.sourceforge.net
10 * Copyright (C) 2000-2003 Michel Lespinasse <walken@zoy.org>
11 * Copyright (C) 1999-2000 Aaron Holtzman <aholtzma@ess.engr.uvic.ca>
13 * This file is part of FFmpeg.
15 * FFmpeg is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public
17 * License as published by the Free Software Foundation; either
18 * version 2 of the License, or (at your option) any later version.
20 * FFmpeg is distributed in the hope that it will be useful,
21 * but WITHOUT ANY WARRANTY; without even the implied warranty of
22 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
23 * General Public License for more details.
25 * You should have received a copy of the GNU General Public
26 * License along with FFmpeg; if not, write to the Free Software
27 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
35 #include "libavutil/crc.h"
36 #include "libavutil/random.h"
38 #include "ac3_parser.h"
39 #include "bitstream.h"
42 /** Maximum possible frame size when the specification limit is ignored */
43 #define AC3_MAX_FRAME_SIZE 21695
46 * Table of bin locations for rematrixing bands
47 * reference: Section 7.5.2 Rematrixing : Frequency Band Definitions
49 static const uint8_t rematrix_band_tab[5] = { 13, 25, 37, 61, 253 };
51 /** table for grouping exponents */
52 static uint8_t exp_ungroup_tab[128][3];
55 /** tables for ungrouping mantissas */
56 static int b1_mantissas[32][3];
57 static int b2_mantissas[128][3];
58 static int b3_mantissas[8];
59 static int b4_mantissas[128][2];
60 static int b5_mantissas[16];
63 * Quantization table: levels for symmetric. bits for asymmetric.
64 * reference: Table 7.18 Mapping of bap to Quantizer
66 static const uint8_t quantization_tab[16] = {
68 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
71 /** dynamic range table. converts codes to scale factors. */
72 static float dynamic_range_tab[256];
74 /** Adjustments in dB gain */
75 #define LEVEL_MINUS_3DB 0.7071067811865476
76 #define LEVEL_MINUS_4POINT5DB 0.5946035575013605
77 #define LEVEL_MINUS_6DB 0.5000000000000000
78 #define LEVEL_MINUS_9DB 0.3535533905932738
79 #define LEVEL_ZERO 0.0000000000000000
80 #define LEVEL_ONE 1.0000000000000000
82 static const float gain_levels[6] = {
86 LEVEL_MINUS_4POINT5DB,
92 * Table for default stereo downmixing coefficients
93 * reference: Section 7.8.2 Downmixing Into Two Channels
95 static const uint8_t ac3_default_coeffs[8][5][2] = {
96 { { 1, 0 }, { 0, 1 }, },
98 { { 1, 0 }, { 0, 1 }, },
99 { { 1, 0 }, { 3, 3 }, { 0, 1 }, },
100 { { 1, 0 }, { 0, 1 }, { 4, 4 }, },
101 { { 1, 0 }, { 3, 3 }, { 0, 1 }, { 5, 5 }, },
102 { { 1, 0 }, { 0, 1 }, { 4, 0 }, { 0, 4 }, },
103 { { 1, 0 }, { 3, 3 }, { 0, 1 }, { 4, 0 }, { 0, 4 }, },
106 /* override ac3.h to include coupling channel */
107 #undef AC3_MAX_CHANNELS
108 #define AC3_MAX_CHANNELS 7
111 #define AC3_OUTPUT_LFEON 8
114 int channel_mode; ///< channel mode (acmod)
115 int block_switch[AC3_MAX_CHANNELS]; ///< block switch flags
116 int dither_flag[AC3_MAX_CHANNELS]; ///< dither flags
117 int dither_all; ///< true if all channels are dithered
118 int cpl_in_use; ///< coupling in use
119 int channel_in_cpl[AC3_MAX_CHANNELS]; ///< channel in coupling
120 int phase_flags_in_use; ///< phase flags in use
121 int phase_flags[18]; ///< phase flags
122 int cpl_band_struct[18]; ///< coupling band structure
123 int num_rematrixing_bands; ///< number of rematrixing bands
124 int rematrixing_flags[4]; ///< rematrixing flags
125 int exp_strategy[AC3_MAX_CHANNELS]; ///< exponent strategies
126 int snr_offset[AC3_MAX_CHANNELS]; ///< signal-to-noise ratio offsets
127 int fast_gain[AC3_MAX_CHANNELS]; ///< fast gain values (signal-to-mask ratio)
128 int dba_mode[AC3_MAX_CHANNELS]; ///< delta bit allocation mode
129 int dba_nsegs[AC3_MAX_CHANNELS]; ///< number of delta segments
130 uint8_t dba_offsets[AC3_MAX_CHANNELS][8]; ///< delta segment offsets
131 uint8_t dba_lengths[AC3_MAX_CHANNELS][8]; ///< delta segment lengths
132 uint8_t dba_values[AC3_MAX_CHANNELS][8]; ///< delta values for each segment
134 int sample_rate; ///< sample frequency, in Hz
135 int bit_rate; ///< stream bit rate, in bits-per-second
136 int frame_size; ///< current frame size, in bytes
138 int channels; ///< number of total channels
139 int fbw_channels; ///< number of full-bandwidth channels
140 int lfe_on; ///< lfe channel in use
141 int lfe_ch; ///< index of LFE channel
142 int output_mode; ///< output channel configuration
143 int out_channels; ///< number of output channels
145 int center_mix_level; ///< Center mix level index
146 int surround_mix_level; ///< Surround mix level index
147 float downmix_coeffs[AC3_MAX_CHANNELS][2]; ///< stereo downmix coefficients
148 float downmix_coeff_adjust[2]; ///< adjustment needed for each output channel when downmixing
149 float dynamic_range[2]; ///< dynamic range
150 int cpl_coords[AC3_MAX_CHANNELS][18]; ///< coupling coordinates
151 int num_cpl_bands; ///< number of coupling bands
152 int num_cpl_subbands; ///< number of coupling sub bands
153 int start_freq[AC3_MAX_CHANNELS]; ///< start frequency bin
154 int end_freq[AC3_MAX_CHANNELS]; ///< end frequency bin
155 AC3BitAllocParameters bit_alloc_params; ///< bit allocation parameters
157 int8_t dexps[AC3_MAX_CHANNELS][256]; ///< decoded exponents
158 uint8_t bap[AC3_MAX_CHANNELS][256]; ///< bit allocation pointers
159 int16_t psd[AC3_MAX_CHANNELS][256]; ///< scaled exponents
160 int16_t band_psd[AC3_MAX_CHANNELS][50]; ///< interpolated exponents
161 int16_t mask[AC3_MAX_CHANNELS][50]; ///< masking curve values
163 int fixed_coeffs[AC3_MAX_CHANNELS][256]; ///> fixed-point transform coefficients
164 DECLARE_ALIGNED_16(float, transform_coeffs[AC3_MAX_CHANNELS][256]); ///< transform coefficients
165 int downmixed; ///< indicates if coeffs are currently downmixed
168 MDCTContext imdct_512; ///< for 512 sample IMDCT
169 MDCTContext imdct_256; ///< for 256 sample IMDCT
170 DSPContext dsp; ///< for optimization
171 float add_bias; ///< offset for float_to_int16 conversion
172 float mul_bias; ///< scaling for float_to_int16 conversion
174 DECLARE_ALIGNED_16(float, output[AC3_MAX_CHANNELS][256]); ///< output after imdct transform and windowing
175 DECLARE_ALIGNED_16(short, int_output[AC3_MAX_CHANNELS-1][256]); ///< final 16-bit integer output
176 DECLARE_ALIGNED_16(float, delay[AC3_MAX_CHANNELS][256]); ///< delay - added to the next block
177 DECLARE_ALIGNED_16(float, tmp_imdct[256]); ///< temporary storage for imdct transform
178 DECLARE_ALIGNED_16(float, tmp_output[512]); ///< temporary storage for output before windowing
179 DECLARE_ALIGNED_16(float, window[256]); ///< window coefficients
182 GetBitContext gbc; ///< bitstream reader
183 AVRandomState dith_state; ///< for dither generation
184 AVCodecContext *avctx; ///< parent context
185 uint8_t *input_buffer; ///< temp buffer to prevent overread
189 * Symmetrical Dequantization
190 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
191 * Tables 7.19 to 7.23
194 symmetric_dequant(int code, int levels)
196 return ((code - (levels >> 1)) << 24) / levels;
200 * Initialize tables at runtime.
202 static av_cold void ac3_tables_init(void)
206 /* generate grouped mantissa tables
207 reference: Section 7.3.5 Ungrouping of Mantissas */
208 for(i=0; i<32; i++) {
209 /* bap=1 mantissas */
210 b1_mantissas[i][0] = symmetric_dequant( i / 9 , 3);
211 b1_mantissas[i][1] = symmetric_dequant((i % 9) / 3, 3);
212 b1_mantissas[i][2] = symmetric_dequant((i % 9) % 3, 3);
214 for(i=0; i<128; i++) {
215 /* bap=2 mantissas */
216 b2_mantissas[i][0] = symmetric_dequant( i / 25 , 5);
217 b2_mantissas[i][1] = symmetric_dequant((i % 25) / 5, 5);
218 b2_mantissas[i][2] = symmetric_dequant((i % 25) % 5, 5);
220 /* bap=4 mantissas */
221 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
222 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
224 /* generate ungrouped mantissa tables
225 reference: Tables 7.21 and 7.23 */
227 /* bap=3 mantissas */
228 b3_mantissas[i] = symmetric_dequant(i, 7);
230 for(i=0; i<15; i++) {
231 /* bap=5 mantissas */
232 b5_mantissas[i] = symmetric_dequant(i, 15);
235 /* generate dynamic range table
236 reference: Section 7.7.1 Dynamic Range Control */
237 for(i=0; i<256; i++) {
238 int v = (i >> 5) - ((i >> 7) << 3) - 5;
239 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
242 /* generate exponent tables
243 reference: Section 7.1.3 Exponent Decoding */
244 for(i=0; i<128; i++) {
245 exp_ungroup_tab[i][0] = i / 25;
246 exp_ungroup_tab[i][1] = (i % 25) / 5;
247 exp_ungroup_tab[i][2] = (i % 25) % 5;
253 * AVCodec initialization
255 static av_cold int ac3_decode_init(AVCodecContext *avctx)
257 AC3DecodeContext *s = avctx->priv_data;
262 ff_mdct_init(&s->imdct_256, 8, 1);
263 ff_mdct_init(&s->imdct_512, 9, 1);
264 ff_kbd_window_init(s->window, 5.0, 256);
265 dsputil_init(&s->dsp, avctx);
266 av_init_random(0, &s->dith_state);
268 /* set bias values for float to int16 conversion */
269 if(s->dsp.float_to_int16 == ff_float_to_int16_c) {
270 s->add_bias = 385.0f;
274 s->mul_bias = 32767.0f;
277 /* allow downmixing to stereo or mono */
278 if (avctx->channels > 0 && avctx->request_channels > 0 &&
279 avctx->request_channels < avctx->channels &&
280 avctx->request_channels <= 2) {
281 avctx->channels = avctx->request_channels;
285 /* allocate context input buffer */
286 if (avctx->error_resilience >= FF_ER_CAREFUL) {
287 s->input_buffer = av_mallocz(AC3_MAX_FRAME_SIZE + FF_INPUT_BUFFER_PADDING_SIZE);
288 if (!s->input_buffer)
289 return AVERROR_NOMEM;
296 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
297 * GetBitContext within AC3DecodeContext must point to
298 * start of the synchronized ac3 bitstream.
300 static int ac3_parse_header(AC3DecodeContext *s)
303 GetBitContext *gbc = &s->gbc;
306 err = ff_ac3_parse_header(gbc, &hdr);
310 if(hdr.bitstream_id > 10)
311 return AC3_PARSE_ERROR_BSID;
313 /* get decoding parameters from header info */
314 s->bit_alloc_params.sr_code = hdr.sr_code;
315 s->channel_mode = hdr.channel_mode;
316 s->lfe_on = hdr.lfe_on;
317 s->bit_alloc_params.sr_shift = hdr.sr_shift;
318 s->sample_rate = hdr.sample_rate;
319 s->bit_rate = hdr.bit_rate;
320 s->channels = hdr.channels;
321 s->fbw_channels = s->channels - s->lfe_on;
322 s->lfe_ch = s->fbw_channels + 1;
323 s->frame_size = hdr.frame_size;
324 s->center_mix_level = hdr.center_mix_level;
325 s->surround_mix_level = hdr.surround_mix_level;
327 /* read the rest of the bsi. read twice for dual mono mode. */
328 i = !(s->channel_mode);
330 skip_bits(gbc, 5); // skip dialog normalization
332 skip_bits(gbc, 8); //skip compression
334 skip_bits(gbc, 8); //skip language code
336 skip_bits(gbc, 7); //skip audio production information
339 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
341 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
342 TODO: read & use the xbsi1 downmix levels */
344 skip_bits(gbc, 14); //skip timecode1 / xbsi1
346 skip_bits(gbc, 14); //skip timecode2 / xbsi2
348 /* skip additional bitstream info */
349 if (get_bits1(gbc)) {
350 i = get_bits(gbc, 6);
360 * Set stereo downmixing coefficients based on frame header info.
361 * reference: Section 7.8.2 Downmixing Into Two Channels
363 static void set_downmix_coeffs(AC3DecodeContext *s)
366 float cmix = gain_levels[s->center_mix_level];
367 float smix = gain_levels[s->surround_mix_level];
369 for(i=0; i<s->fbw_channels; i++) {
370 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
371 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
373 if(s->channel_mode > 1 && s->channel_mode & 1) {
374 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
376 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
377 int nf = s->channel_mode - 2;
378 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
380 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
381 int nf = s->channel_mode - 4;
382 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
385 /* calculate adjustment needed for each channel to avoid clipping */
386 s->downmix_coeff_adjust[0] = s->downmix_coeff_adjust[1] = 0.0f;
387 for(i=0; i<s->fbw_channels; i++) {
388 s->downmix_coeff_adjust[0] += s->downmix_coeffs[i][0];
389 s->downmix_coeff_adjust[1] += s->downmix_coeffs[i][1];
391 s->downmix_coeff_adjust[0] = 1.0f / s->downmix_coeff_adjust[0];
392 s->downmix_coeff_adjust[1] = 1.0f / s->downmix_coeff_adjust[1];
396 * Decode the grouped exponents according to exponent strategy.
397 * reference: Section 7.1.3 Exponent Decoding
399 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
400 uint8_t absexp, int8_t *dexps)
402 int i, j, grp, group_size;
407 group_size = exp_strategy + (exp_strategy == EXP_D45);
408 for(grp=0,i=0; grp<ngrps; grp++) {
409 expacc = get_bits(gbc, 7);
410 dexp[i++] = exp_ungroup_tab[expacc][0];
411 dexp[i++] = exp_ungroup_tab[expacc][1];
412 dexp[i++] = exp_ungroup_tab[expacc][2];
415 /* convert to absolute exps and expand groups */
417 for(i=0; i<ngrps*3; i++) {
418 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
419 for(j=0; j<group_size; j++) {
420 dexps[(i*group_size)+j] = prevexp;
426 * Generate transform coefficients for each coupled channel in the coupling
427 * range using the coupling coefficients and coupling coordinates.
428 * reference: Section 7.4.3 Coupling Coordinate Format
430 static void uncouple_channels(AC3DecodeContext *s)
432 int i, j, ch, bnd, subbnd;
435 i = s->start_freq[CPL_CH];
436 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
439 for(j=0; j<12; j++) {
440 for(ch=1; ch<=s->fbw_channels; ch++) {
441 if(s->channel_in_cpl[ch]) {
442 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
443 if (ch == 2 && s->phase_flags[bnd])
444 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
449 } while(s->cpl_band_struct[subbnd]);
454 * Grouped mantissas for 3-level 5-level and 11-level quantization
466 * Get the transform coefficients for a particular channel
467 * reference: Section 7.3 Quantization and Decoding of Mantissas
469 static int get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
471 GetBitContext *gbc = &s->gbc;
472 int i, gcode, tbap, start, end;
477 exps = s->dexps[ch_index];
478 bap = s->bap[ch_index];
479 coeffs = s->fixed_coeffs[ch_index];
480 start = s->start_freq[ch_index];
481 end = s->end_freq[ch_index];
483 for (i = start; i < end; i++) {
487 coeffs[i] = (av_random(&s->dith_state) & 0x7FFFFF) - 4194304;
492 gcode = get_bits(gbc, 5);
493 m->b1_mant[0] = b1_mantissas[gcode][0];
494 m->b1_mant[1] = b1_mantissas[gcode][1];
495 m->b1_mant[2] = b1_mantissas[gcode][2];
498 coeffs[i] = m->b1_mant[m->b1ptr++];
503 gcode = get_bits(gbc, 7);
504 m->b2_mant[0] = b2_mantissas[gcode][0];
505 m->b2_mant[1] = b2_mantissas[gcode][1];
506 m->b2_mant[2] = b2_mantissas[gcode][2];
509 coeffs[i] = m->b2_mant[m->b2ptr++];
513 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
518 gcode = get_bits(gbc, 7);
519 m->b4_mant[0] = b4_mantissas[gcode][0];
520 m->b4_mant[1] = b4_mantissas[gcode][1];
523 coeffs[i] = m->b4_mant[m->b4ptr++];
527 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
531 /* asymmetric dequantization */
532 int qlevel = quantization_tab[tbap];
533 coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
537 coeffs[i] >>= exps[i];
544 * Remove random dithering from coefficients with zero-bit mantissas
545 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
547 static void remove_dithering(AC3DecodeContext *s) {
553 for(ch=1; ch<=s->fbw_channels; ch++) {
554 if(!s->dither_flag[ch]) {
555 coeffs = s->fixed_coeffs[ch];
557 if(s->channel_in_cpl[ch])
558 end = s->start_freq[CPL_CH];
560 end = s->end_freq[ch];
561 for(i=0; i<end; i++) {
565 if(s->channel_in_cpl[ch]) {
566 bap = s->bap[CPL_CH];
567 for(; i<s->end_freq[CPL_CH]; i++) {
577 * Get the transform coefficients.
579 static int get_transform_coeffs(AC3DecodeContext *s)
585 m.b1ptr = m.b2ptr = m.b4ptr = 3;
587 for (ch = 1; ch <= s->channels; ch++) {
588 /* transform coefficients for full-bandwidth channel */
589 if (get_transform_coeffs_ch(s, ch, &m))
591 /* tranform coefficients for coupling channel come right after the
592 coefficients for the first coupled channel*/
593 if (s->channel_in_cpl[ch]) {
595 if (get_transform_coeffs_ch(s, CPL_CH, &m)) {
596 av_log(s->avctx, AV_LOG_ERROR, "error in decoupling channels\n");
599 uncouple_channels(s);
602 end = s->end_freq[CPL_CH];
604 end = s->end_freq[ch];
607 s->fixed_coeffs[ch][end] = 0;
611 /* if any channel doesn't use dithering, zero appropriate coefficients */
619 * Stereo rematrixing.
620 * reference: Section 7.5.4 Rematrixing : Decoding Technique
622 static void do_rematrixing(AC3DecodeContext *s)
628 end = FFMIN(s->end_freq[1], s->end_freq[2]);
630 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
631 if(s->rematrixing_flags[bnd]) {
632 bndend = FFMIN(end, rematrix_band_tab[bnd+1]);
633 for(i=rematrix_band_tab[bnd]; i<bndend; i++) {
634 tmp0 = s->fixed_coeffs[1][i];
635 tmp1 = s->fixed_coeffs[2][i];
636 s->fixed_coeffs[1][i] = tmp0 + tmp1;
637 s->fixed_coeffs[2][i] = tmp0 - tmp1;
644 * Perform the 256-point IMDCT
646 static void do_imdct_256(AC3DecodeContext *s, int chindex)
649 DECLARE_ALIGNED_16(float, x[128]);
651 float *o_ptr = s->tmp_output;
654 /* de-interleave coefficients */
655 for(k=0; k<128; k++) {
656 x[k] = s->transform_coeffs[chindex][2*k+i];
659 /* run standard IMDCT */
660 s->imdct_256.fft.imdct_calc(&s->imdct_256, o_ptr, x, s->tmp_imdct);
662 /* reverse the post-rotation & reordering from standard IMDCT */
663 for(k=0; k<32; k++) {
664 z[i][32+k].re = -o_ptr[128+2*k];
665 z[i][32+k].im = -o_ptr[2*k];
666 z[i][31-k].re = o_ptr[2*k+1];
667 z[i][31-k].im = o_ptr[128+2*k+1];
671 /* apply AC-3 post-rotation & reordering */
672 for(k=0; k<64; k++) {
673 o_ptr[ 2*k ] = -z[0][ k].im;
674 o_ptr[ 2*k+1] = z[0][63-k].re;
675 o_ptr[128+2*k ] = -z[0][ k].re;
676 o_ptr[128+2*k+1] = z[0][63-k].im;
677 o_ptr[256+2*k ] = -z[1][ k].re;
678 o_ptr[256+2*k+1] = z[1][63-k].im;
679 o_ptr[384+2*k ] = z[1][ k].im;
680 o_ptr[384+2*k+1] = -z[1][63-k].re;
685 * Inverse MDCT Transform.
686 * Convert frequency domain coefficients to time-domain audio samples.
687 * reference: Section 7.9.4 Transformation Equations
689 static inline void do_imdct(AC3DecodeContext *s, int channels)
693 for (ch=1; ch<=channels; ch++) {
694 if (s->block_switch[ch]) {
697 s->imdct_512.fft.imdct_calc(&s->imdct_512, s->tmp_output,
698 s->transform_coeffs[ch], s->tmp_imdct);
700 /* For the first half of the block, apply the window, add the delay
701 from the previous block, and send to output */
702 s->dsp.vector_fmul_add_add(s->output[ch-1], s->tmp_output,
703 s->window, s->delay[ch-1], 0, 256, 1);
704 /* For the second half of the block, apply the window and store the
705 samples to delay, to be combined with the next block */
706 s->dsp.vector_fmul_reverse(s->delay[ch-1], s->tmp_output+256,
712 * Downmix the output to mono or stereo.
714 static void ac3_downmix(AC3DecodeContext *s,
715 float samples[AC3_MAX_CHANNELS][256], int ch_offset)
720 for(i=0; i<256; i++) {
722 for(j=0; j<s->fbw_channels; j++) {
723 v0 += samples[j+ch_offset][i] * s->downmix_coeffs[j][0];
724 v1 += samples[j+ch_offset][i] * s->downmix_coeffs[j][1];
726 v0 *= s->downmix_coeff_adjust[0];
727 v1 *= s->downmix_coeff_adjust[1];
728 if(s->output_mode == AC3_CHMODE_MONO) {
729 samples[ch_offset][i] = (v0 + v1) * LEVEL_MINUS_3DB;
730 } else if(s->output_mode == AC3_CHMODE_STEREO) {
731 samples[ ch_offset][i] = v0;
732 samples[1+ch_offset][i] = v1;
738 * Upmix delay samples from stereo to original channel layout.
740 static void ac3_upmix_delay(AC3DecodeContext *s)
742 int channel_data_size = sizeof(s->delay[0]);
743 switch(s->channel_mode) {
744 case AC3_CHMODE_DUALMONO:
745 case AC3_CHMODE_STEREO:
746 /* upmix mono to stereo */
747 memcpy(s->delay[1], s->delay[0], channel_data_size);
749 case AC3_CHMODE_2F2R:
750 memset(s->delay[3], 0, channel_data_size);
751 case AC3_CHMODE_2F1R:
752 memset(s->delay[2], 0, channel_data_size);
754 case AC3_CHMODE_3F2R:
755 memset(s->delay[4], 0, channel_data_size);
756 case AC3_CHMODE_3F1R:
757 memset(s->delay[3], 0, channel_data_size);
759 memcpy(s->delay[2], s->delay[1], channel_data_size);
760 memset(s->delay[1], 0, channel_data_size);
766 * Parse an audio block from AC-3 bitstream.
768 static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
770 int fbw_channels = s->fbw_channels;
771 int channel_mode = s->channel_mode;
773 int different_transforms;
775 GetBitContext *gbc = &s->gbc;
776 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
778 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
780 /* block switch flags */
781 different_transforms = 0;
782 for (ch = 1; ch <= fbw_channels; ch++) {
783 s->block_switch[ch] = get_bits1(gbc);
784 if(ch > 1 && s->block_switch[ch] != s->block_switch[1])
785 different_transforms = 1;
788 /* dithering flags */
790 for (ch = 1; ch <= fbw_channels; ch++) {
791 s->dither_flag[ch] = get_bits1(gbc);
792 if(!s->dither_flag[ch])
797 i = !(s->channel_mode);
800 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
801 s->avctx->drc_scale)+1.0;
802 } else if(blk == 0) {
803 s->dynamic_range[i] = 1.0f;
807 /* coupling strategy */
808 if (get_bits1(gbc)) {
809 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
810 s->cpl_in_use = get_bits1(gbc);
812 /* coupling in use */
813 int cpl_begin_freq, cpl_end_freq;
815 if (channel_mode < AC3_CHMODE_STEREO) {
816 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
820 /* determine which channels are coupled */
821 for (ch = 1; ch <= fbw_channels; ch++)
822 s->channel_in_cpl[ch] = get_bits1(gbc);
824 /* phase flags in use */
825 if (channel_mode == AC3_CHMODE_STEREO)
826 s->phase_flags_in_use = get_bits1(gbc);
828 /* coupling frequency range and band structure */
829 cpl_begin_freq = get_bits(gbc, 4);
830 cpl_end_freq = get_bits(gbc, 4);
831 if (3 + cpl_end_freq - cpl_begin_freq < 0) {
832 av_log(s->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
835 s->num_cpl_bands = s->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
836 s->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
837 s->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
838 for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
839 if (get_bits1(gbc)) {
840 s->cpl_band_struct[bnd] = 1;
844 s->cpl_band_struct[s->num_cpl_subbands-1] = 0;
846 /* coupling not in use */
847 for (ch = 1; ch <= fbw_channels; ch++)
848 s->channel_in_cpl[ch] = 0;
851 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must be present in block 0\n");
855 /* coupling coordinates */
857 int cpl_coords_exist = 0;
859 for (ch = 1; ch <= fbw_channels; ch++) {
860 if (s->channel_in_cpl[ch]) {
861 if (get_bits1(gbc)) {
862 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
863 cpl_coords_exist = 1;
864 master_cpl_coord = 3 * get_bits(gbc, 2);
865 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
866 cpl_coord_exp = get_bits(gbc, 4);
867 cpl_coord_mant = get_bits(gbc, 4);
868 if (cpl_coord_exp == 15)
869 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
871 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
872 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
875 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must be present in block 0\n");
881 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
882 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
883 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
888 /* stereo rematrixing strategy and band structure */
889 if (channel_mode == AC3_CHMODE_STEREO) {
890 if (get_bits1(gbc)) {
891 s->num_rematrixing_bands = 4;
892 if(s->cpl_in_use && s->start_freq[CPL_CH] <= 61)
893 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
894 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
895 s->rematrixing_flags[bnd] = get_bits1(gbc);
897 av_log(s->avctx, AV_LOG_ERROR, "new rematrixing strategy must be present in block 0\n");
902 /* exponent strategies for each channel */
903 s->exp_strategy[CPL_CH] = EXP_REUSE;
904 s->exp_strategy[s->lfe_ch] = EXP_REUSE;
905 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
907 s->exp_strategy[ch] = get_bits(gbc, 1);
909 s->exp_strategy[ch] = get_bits(gbc, 2);
910 if(s->exp_strategy[ch] != EXP_REUSE)
911 bit_alloc_stages[ch] = 3;
914 /* channel bandwidth */
915 for (ch = 1; ch <= fbw_channels; ch++) {
916 s->start_freq[ch] = 0;
917 if (s->exp_strategy[ch] != EXP_REUSE) {
918 int prev = s->end_freq[ch];
919 if (s->channel_in_cpl[ch])
920 s->end_freq[ch] = s->start_freq[CPL_CH];
922 int bandwidth_code = get_bits(gbc, 6);
923 if (bandwidth_code > 60) {
924 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
927 s->end_freq[ch] = bandwidth_code * 3 + 73;
929 if(blk > 0 && s->end_freq[ch] != prev)
930 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
933 s->start_freq[s->lfe_ch] = 0;
934 s->end_freq[s->lfe_ch] = 7;
936 /* decode exponents for each channel */
937 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
938 if (s->exp_strategy[ch] != EXP_REUSE) {
939 int group_size, num_groups;
940 group_size = 3 << (s->exp_strategy[ch] - 1);
942 num_groups = (s->end_freq[ch] - s->start_freq[ch]) / group_size;
943 else if(ch == s->lfe_ch)
946 num_groups = (s->end_freq[ch] + group_size - 4) / group_size;
947 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
948 decode_exponents(gbc, s->exp_strategy[ch], num_groups, s->dexps[ch][0],
949 &s->dexps[ch][s->start_freq[ch]+!!ch]);
950 if(ch != CPL_CH && ch != s->lfe_ch)
951 skip_bits(gbc, 2); /* skip gainrng */
955 /* bit allocation information */
956 if (get_bits1(gbc)) {
957 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
958 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
959 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
960 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
961 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
962 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
963 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
966 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must be present in block 0\n");
970 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
971 if (get_bits1(gbc)) {
973 csnr = (get_bits(gbc, 6) - 15) << 4;
974 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
975 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
976 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
978 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
980 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
984 /* coupling leak information */
986 if (get_bits1(gbc)) {
987 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
988 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
989 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
991 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must be present in block 0\n");
996 /* delta bit allocation information */
997 if (get_bits1(gbc)) {
998 /* delta bit allocation exists (strategy) */
999 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
1000 s->dba_mode[ch] = get_bits(gbc, 2);
1001 if (s->dba_mode[ch] == DBA_RESERVED) {
1002 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1005 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1007 /* channel delta offset, len and bit allocation */
1008 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
1009 if (s->dba_mode[ch] == DBA_NEW) {
1010 s->dba_nsegs[ch] = get_bits(gbc, 3);
1011 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
1012 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1013 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1014 s->dba_values[ch][seg] = get_bits(gbc, 3);
1018 } else if(blk == 0) {
1019 for(ch=0; ch<=s->channels; ch++) {
1020 s->dba_mode[ch] = DBA_NONE;
1024 /* Bit allocation */
1025 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
1026 if(bit_alloc_stages[ch] > 2) {
1027 /* Exponent mapping into PSD and PSD integration */
1028 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1029 s->start_freq[ch], s->end_freq[ch],
1030 s->psd[ch], s->band_psd[ch]);
1032 if(bit_alloc_stages[ch] > 1) {
1033 /* Compute excitation function, Compute masking curve, and
1034 Apply delta bit allocation */
1035 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1036 s->start_freq[ch], s->end_freq[ch],
1037 s->fast_gain[ch], (ch == s->lfe_ch),
1038 s->dba_mode[ch], s->dba_nsegs[ch],
1039 s->dba_offsets[ch], s->dba_lengths[ch],
1040 s->dba_values[ch], s->mask[ch]);
1042 if(bit_alloc_stages[ch] > 0) {
1043 /* Compute bit allocation */
1044 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1045 s->start_freq[ch], s->end_freq[ch],
1047 s->bit_alloc_params.floor,
1052 /* unused dummy data */
1053 if (get_bits1(gbc)) {
1054 int skipl = get_bits(gbc, 9);
1059 /* unpack the transform coefficients
1060 this also uncouples channels if coupling is in use. */
1061 if (get_transform_coeffs(s)) {
1062 av_log(s->avctx, AV_LOG_ERROR, "Error in routine get_transform_coeffs\n");
1066 /* recover coefficients if rematrixing is in use */
1067 if(s->channel_mode == AC3_CHMODE_STEREO)
1070 /* apply scaling to coefficients (headroom, dynrng) */
1071 for(ch=1; ch<=s->channels; ch++) {
1072 float gain = s->mul_bias / 4194304.0f;
1073 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1074 gain *= s->dynamic_range[ch-1];
1076 gain *= s->dynamic_range[0];
1078 for(i=0; i<256; i++) {
1079 s->transform_coeffs[ch][i] = s->fixed_coeffs[ch][i] * gain;
1083 /* downmix and MDCT. order depends on whether block switching is used for
1084 any channel in this block. this is because coefficients for the long
1085 and short transforms cannot be mixed. */
1086 downmix_output = s->channels != s->out_channels &&
1087 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1088 s->fbw_channels == s->out_channels);
1089 if(different_transforms) {
1090 /* the delay samples have already been downmixed, so we upmix the delay
1091 samples in order to reconstruct all channels before downmixing. */
1097 do_imdct(s, s->channels);
1099 if(downmix_output) {
1100 ac3_downmix(s, s->output, 0);
1103 if(downmix_output) {
1104 ac3_downmix(s, s->transform_coeffs, 1);
1109 ac3_downmix(s, s->delay, 0);
1112 do_imdct(s, s->out_channels);
1115 /* convert float to 16-bit integer */
1116 for(ch=0; ch<s->out_channels; ch++) {
1117 for(i=0; i<256; i++) {
1118 s->output[ch][i] += s->add_bias;
1120 s->dsp.float_to_int16(s->int_output[ch], s->output[ch], 256);
1127 * Decode a single AC-3 frame.
1129 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
1130 const uint8_t *buf, int buf_size)
1132 AC3DecodeContext *s = avctx->priv_data;
1133 int16_t *out_samples = (int16_t *)data;
1134 int i, blk, ch, err;
1136 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1137 if (s->input_buffer) {
1138 /* copy input buffer to decoder context to avoid reading past the end
1139 of the buffer, which can be caused by a damaged input stream. */
1140 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_MAX_FRAME_SIZE));
1141 init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
1143 init_get_bits(&s->gbc, buf, buf_size * 8);
1146 /* parse the syncinfo */
1147 err = ac3_parse_header(s);
1150 case AC3_PARSE_ERROR_SYNC:
1151 av_log(avctx, AV_LOG_ERROR, "frame sync error : cannot use error concealment\n");
1153 case AC3_PARSE_ERROR_BSID:
1154 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1156 case AC3_PARSE_ERROR_SAMPLE_RATE:
1157 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1159 case AC3_PARSE_ERROR_FRAME_SIZE:
1160 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1162 case AC3_PARSE_ERROR_FRAME_TYPE:
1163 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1166 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1171 /* check that reported frame size fits in input buffer */
1172 if(s->frame_size > buf_size) {
1173 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1177 /* check for crc mismatch */
1178 if(!err && avctx->error_resilience >= FF_ER_CAREFUL) {
1179 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1180 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1185 /* if frame is ok, set audio parameters */
1187 avctx->sample_rate = s->sample_rate;
1188 avctx->bit_rate = s->bit_rate;
1190 /* channel config */
1191 s->out_channels = s->channels;
1192 s->output_mode = s->channel_mode;
1194 s->output_mode |= AC3_OUTPUT_LFEON;
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);
1207 } else if (!s->out_channels) {
1208 s->out_channels = avctx->channels;
1209 if(s->out_channels < s->channels)
1210 s->output_mode = s->out_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1213 /* parse the audio blocks */
1214 for (blk = 0; blk < NB_BLOCKS; blk++) {
1215 if (!err && ac3_parse_audio_block(s, blk)) {
1216 av_log(avctx, AV_LOG_ERROR, "error parsing the audio block\n");
1218 for (i = 0; i < 256; i++)
1219 for (ch = 0; ch < s->out_channels; ch++)
1220 *(out_samples++) = s->int_output[ch][i];
1222 *data_size = NB_BLOCKS * 256 * avctx->channels * sizeof (int16_t);
1223 return s->frame_size;
1227 * Uninitialize the AC-3 decoder.
1229 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1231 AC3DecodeContext *s = avctx->priv_data;
1232 ff_mdct_end(&s->imdct_512);
1233 ff_mdct_end(&s->imdct_256);
1235 av_freep(&s->input_buffer);
1240 AVCodec ac3_decoder = {
1242 .type = CODEC_TYPE_AUDIO,
1244 .priv_data_size = sizeof (AC3DecodeContext),
1245 .init = ac3_decode_init,
1246 .close = ac3_decode_end,
1247 .decode = ac3_decode_frame,
1248 .long_name = "ATSC A/52 / AC-3",