Replace LDLATEFLAGS hackery by proper LDFLAGS tests.
[FFMpeg-mirror/ffmpeg-vdpau.git] / libavcodec / ac3dec.c
blobbd1310b1c45a115ac2ad23cb274bc7e9134b8657
1 /*
2 * AC-3 Audio Decoder
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
30 #include <stdio.h>
31 #include <stddef.h>
32 #include <math.h>
33 #include <string.h>
35 #include "libavutil/crc.h"
36 #include "libavutil/random.h"
37 #include "avcodec.h"
38 #include "ac3_parser.h"
39 #include "bitstream.h"
40 #include "dsputil.h"
41 #include "ac3dec.h"
42 #include "ac3dec_data.h"
44 /** Maximum possible frame size when the specification limit is ignored */
45 #define AC3_MAX_FRAME_SIZE 21695
47 /** table for grouping exponents */
48 static uint8_t exp_ungroup_tab[128][3];
51 /** tables for ungrouping mantissas */
52 static int b1_mantissas[32][3];
53 static int b2_mantissas[128][3];
54 static int b3_mantissas[8];
55 static int b4_mantissas[128][2];
56 static int b5_mantissas[16];
58 /**
59 * Quantization table: levels for symmetric. bits for asymmetric.
60 * reference: Table 7.18 Mapping of bap to Quantizer
62 static const uint8_t quantization_tab[16] = {
63 0, 3, 5, 7, 11, 15,
64 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
67 /** dynamic range table. converts codes to scale factors. */
68 static float dynamic_range_tab[256];
70 /** Adjustments in dB gain */
71 #define LEVEL_PLUS_3DB 1.4142135623730950
72 #define LEVEL_PLUS_1POINT5DB 1.1892071150027209
73 #define LEVEL_MINUS_1POINT5DB 0.8408964152537145
74 #define LEVEL_MINUS_3DB 0.7071067811865476
75 #define LEVEL_MINUS_4POINT5DB 0.5946035575013605
76 #define LEVEL_MINUS_6DB 0.5000000000000000
77 #define LEVEL_MINUS_9DB 0.3535533905932738
78 #define LEVEL_ZERO 0.0000000000000000
79 #define LEVEL_ONE 1.0000000000000000
81 static const float gain_levels[9] = {
82 LEVEL_PLUS_3DB,
83 LEVEL_PLUS_1POINT5DB,
84 LEVEL_ONE,
85 LEVEL_MINUS_1POINT5DB,
86 LEVEL_MINUS_3DB,
87 LEVEL_MINUS_4POINT5DB,
88 LEVEL_MINUS_6DB,
89 LEVEL_ZERO,
90 LEVEL_MINUS_9DB
93 /**
94 * Table for center mix levels
95 * reference: Section 5.4.2.4 cmixlev
97 static const uint8_t center_levels[4] = { 4, 5, 6, 5 };
99 /**
100 * Table for surround mix levels
101 * reference: Section 5.4.2.5 surmixlev
103 static const uint8_t surround_levels[4] = { 4, 6, 7, 6 };
106 * Table for default stereo downmixing coefficients
107 * reference: Section 7.8.2 Downmixing Into Two Channels
109 static const uint8_t ac3_default_coeffs[8][5][2] = {
110 { { 2, 7 }, { 7, 2 }, },
111 { { 4, 4 }, },
112 { { 2, 7 }, { 7, 2 }, },
113 { { 2, 7 }, { 5, 5 }, { 7, 2 }, },
114 { { 2, 7 }, { 7, 2 }, { 6, 6 }, },
115 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 8, 8 }, },
116 { { 2, 7 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
117 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
121 * Symmetrical Dequantization
122 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
123 * Tables 7.19 to 7.23
125 static inline int
126 symmetric_dequant(int code, int levels)
128 return ((code - (levels >> 1)) << 24) / levels;
132 * Initialize tables at runtime.
134 static av_cold void ac3_tables_init(void)
136 int i;
138 /* generate grouped mantissa tables
139 reference: Section 7.3.5 Ungrouping of Mantissas */
140 for(i=0; i<32; i++) {
141 /* bap=1 mantissas */
142 b1_mantissas[i][0] = symmetric_dequant( i / 9 , 3);
143 b1_mantissas[i][1] = symmetric_dequant((i % 9) / 3, 3);
144 b1_mantissas[i][2] = symmetric_dequant((i % 9) % 3, 3);
146 for(i=0; i<128; i++) {
147 /* bap=2 mantissas */
148 b2_mantissas[i][0] = symmetric_dequant( i / 25 , 5);
149 b2_mantissas[i][1] = symmetric_dequant((i % 25) / 5, 5);
150 b2_mantissas[i][2] = symmetric_dequant((i % 25) % 5, 5);
152 /* bap=4 mantissas */
153 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
154 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
156 /* generate ungrouped mantissa tables
157 reference: Tables 7.21 and 7.23 */
158 for(i=0; i<7; i++) {
159 /* bap=3 mantissas */
160 b3_mantissas[i] = symmetric_dequant(i, 7);
162 for(i=0; i<15; i++) {
163 /* bap=5 mantissas */
164 b5_mantissas[i] = symmetric_dequant(i, 15);
167 /* generate dynamic range table
168 reference: Section 7.7.1 Dynamic Range Control */
169 for(i=0; i<256; i++) {
170 int v = (i >> 5) - ((i >> 7) << 3) - 5;
171 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
174 /* generate exponent tables
175 reference: Section 7.1.3 Exponent Decoding */
176 for(i=0; i<128; i++) {
177 exp_ungroup_tab[i][0] = i / 25;
178 exp_ungroup_tab[i][1] = (i % 25) / 5;
179 exp_ungroup_tab[i][2] = (i % 25) % 5;
185 * AVCodec initialization
187 static av_cold int ac3_decode_init(AVCodecContext *avctx)
189 AC3DecodeContext *s = avctx->priv_data;
190 s->avctx = avctx;
192 ac3_common_init();
193 ac3_tables_init();
194 ff_mdct_init(&s->imdct_256, 8, 1);
195 ff_mdct_init(&s->imdct_512, 9, 1);
196 ff_kbd_window_init(s->window, 5.0, 256);
197 dsputil_init(&s->dsp, avctx);
198 av_init_random(0, &s->dith_state);
200 /* set bias values for float to int16 conversion */
201 if(s->dsp.float_to_int16 == ff_float_to_int16_c) {
202 s->add_bias = 385.0f;
203 s->mul_bias = 1.0f;
204 } else {
205 s->add_bias = 0.0f;
206 s->mul_bias = 32767.0f;
209 /* allow downmixing to stereo or mono */
210 if (avctx->channels > 0 && avctx->request_channels > 0 &&
211 avctx->request_channels < avctx->channels &&
212 avctx->request_channels <= 2) {
213 avctx->channels = avctx->request_channels;
215 s->downmixed = 1;
217 /* allocate context input buffer */
218 if (avctx->error_resilience >= FF_ER_CAREFUL) {
219 s->input_buffer = av_mallocz(AC3_MAX_FRAME_SIZE + FF_INPUT_BUFFER_PADDING_SIZE);
220 if (!s->input_buffer)
221 return AVERROR_NOMEM;
224 return 0;
228 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
229 * GetBitContext within AC3DecodeContext must point to
230 * start of the synchronized ac3 bitstream.
232 static int ac3_parse_header(AC3DecodeContext *s)
234 GetBitContext *gbc = &s->gbc;
235 int i;
237 /* read the rest of the bsi. read twice for dual mono mode. */
238 i = !(s->channel_mode);
239 do {
240 skip_bits(gbc, 5); // skip dialog normalization
241 if (get_bits1(gbc))
242 skip_bits(gbc, 8); //skip compression
243 if (get_bits1(gbc))
244 skip_bits(gbc, 8); //skip language code
245 if (get_bits1(gbc))
246 skip_bits(gbc, 7); //skip audio production information
247 } while (i--);
249 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
251 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
252 TODO: read & use the xbsi1 downmix levels */
253 if (get_bits1(gbc))
254 skip_bits(gbc, 14); //skip timecode1 / xbsi1
255 if (get_bits1(gbc))
256 skip_bits(gbc, 14); //skip timecode2 / xbsi2
258 /* skip additional bitstream info */
259 if (get_bits1(gbc)) {
260 i = get_bits(gbc, 6);
261 do {
262 skip_bits(gbc, 8);
263 } while(i--);
266 return 0;
270 * Common function to parse AC3 or E-AC3 frame header
272 static int parse_frame_header(AC3DecodeContext *s)
274 AC3HeaderInfo hdr;
275 GetBitContext *gbc = &s->gbc;
276 int err;
278 err = ff_ac3_parse_header(gbc, &hdr);
279 if(err)
280 return err;
282 if(hdr.bitstream_id > 10)
283 return AC3_PARSE_ERROR_BSID;
285 /* get decoding parameters from header info */
286 s->bit_alloc_params.sr_code = hdr.sr_code;
287 s->channel_mode = hdr.channel_mode;
288 s->lfe_on = hdr.lfe_on;
289 s->bit_alloc_params.sr_shift = hdr.sr_shift;
290 s->sample_rate = hdr.sample_rate;
291 s->bit_rate = hdr.bit_rate;
292 s->channels = hdr.channels;
293 s->fbw_channels = s->channels - s->lfe_on;
294 s->lfe_ch = s->fbw_channels + 1;
295 s->frame_size = hdr.frame_size;
296 s->center_mix_level = hdr.center_mix_level;
297 s->surround_mix_level = hdr.surround_mix_level;
298 s->num_blocks = hdr.num_blocks;
299 s->frame_type = hdr.frame_type;
300 s->substreamid = hdr.substreamid;
302 if(s->lfe_on) {
303 s->start_freq[s->lfe_ch] = 0;
304 s->end_freq[s->lfe_ch] = 7;
305 s->num_exp_groups[s->lfe_ch] = 2;
306 s->channel_in_cpl[s->lfe_ch] = 0;
309 return ac3_parse_header(s);
313 * Set stereo downmixing coefficients based on frame header info.
314 * reference: Section 7.8.2 Downmixing Into Two Channels
316 static void set_downmix_coeffs(AC3DecodeContext *s)
318 int i;
319 float cmix = gain_levels[center_levels[s->center_mix_level]];
320 float smix = gain_levels[surround_levels[s->surround_mix_level]];
322 for(i=0; i<s->fbw_channels; i++) {
323 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
324 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
326 if(s->channel_mode > 1 && s->channel_mode & 1) {
327 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
329 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
330 int nf = s->channel_mode - 2;
331 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
333 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
334 int nf = s->channel_mode - 4;
335 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
338 /* calculate adjustment needed for each channel to avoid clipping */
339 s->downmix_coeff_adjust[0] = s->downmix_coeff_adjust[1] = 0.0f;
340 for(i=0; i<s->fbw_channels; i++) {
341 s->downmix_coeff_adjust[0] += s->downmix_coeffs[i][0];
342 s->downmix_coeff_adjust[1] += s->downmix_coeffs[i][1];
344 s->downmix_coeff_adjust[0] = 1.0f / s->downmix_coeff_adjust[0];
345 s->downmix_coeff_adjust[1] = 1.0f / s->downmix_coeff_adjust[1];
349 * Decode the grouped exponents according to exponent strategy.
350 * reference: Section 7.1.3 Exponent Decoding
352 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
353 uint8_t absexp, int8_t *dexps)
355 int i, j, grp, group_size;
356 int dexp[256];
357 int expacc, prevexp;
359 /* unpack groups */
360 group_size = exp_strategy + (exp_strategy == EXP_D45);
361 for(grp=0,i=0; grp<ngrps; grp++) {
362 expacc = get_bits(gbc, 7);
363 dexp[i++] = exp_ungroup_tab[expacc][0];
364 dexp[i++] = exp_ungroup_tab[expacc][1];
365 dexp[i++] = exp_ungroup_tab[expacc][2];
368 /* convert to absolute exps and expand groups */
369 prevexp = absexp;
370 for(i=0; i<ngrps*3; i++) {
371 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
372 for(j=0; j<group_size; j++) {
373 dexps[(i*group_size)+j] = prevexp;
379 * Generate transform coefficients for each coupled channel in the coupling
380 * range using the coupling coefficients and coupling coordinates.
381 * reference: Section 7.4.3 Coupling Coordinate Format
383 static void uncouple_channels(AC3DecodeContext *s)
385 int i, j, ch, bnd, subbnd;
387 subbnd = -1;
388 i = s->start_freq[CPL_CH];
389 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
390 do {
391 subbnd++;
392 for(j=0; j<12; j++) {
393 for(ch=1; ch<=s->fbw_channels; ch++) {
394 if(s->channel_in_cpl[ch]) {
395 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
396 if (ch == 2 && s->phase_flags[bnd])
397 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
400 i++;
402 } while(s->cpl_band_struct[subbnd]);
407 * Grouped mantissas for 3-level 5-level and 11-level quantization
409 typedef struct {
410 int b1_mant[3];
411 int b2_mant[3];
412 int b4_mant[2];
413 int b1ptr;
414 int b2ptr;
415 int b4ptr;
416 } mant_groups;
419 * Get the transform coefficients for a particular channel
420 * reference: Section 7.3 Quantization and Decoding of Mantissas
422 static void get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
424 GetBitContext *gbc = &s->gbc;
425 int i, gcode, tbap, start, end;
426 uint8_t *exps;
427 uint8_t *bap;
428 int *coeffs;
430 exps = s->dexps[ch_index];
431 bap = s->bap[ch_index];
432 coeffs = s->fixed_coeffs[ch_index];
433 start = s->start_freq[ch_index];
434 end = s->end_freq[ch_index];
436 for (i = start; i < end; i++) {
437 tbap = bap[i];
438 switch (tbap) {
439 case 0:
440 coeffs[i] = (av_random(&s->dith_state) & 0x7FFFFF) - 4194304;
441 break;
443 case 1:
444 if(m->b1ptr > 2) {
445 gcode = get_bits(gbc, 5);
446 m->b1_mant[0] = b1_mantissas[gcode][0];
447 m->b1_mant[1] = b1_mantissas[gcode][1];
448 m->b1_mant[2] = b1_mantissas[gcode][2];
449 m->b1ptr = 0;
451 coeffs[i] = m->b1_mant[m->b1ptr++];
452 break;
454 case 2:
455 if(m->b2ptr > 2) {
456 gcode = get_bits(gbc, 7);
457 m->b2_mant[0] = b2_mantissas[gcode][0];
458 m->b2_mant[1] = b2_mantissas[gcode][1];
459 m->b2_mant[2] = b2_mantissas[gcode][2];
460 m->b2ptr = 0;
462 coeffs[i] = m->b2_mant[m->b2ptr++];
463 break;
465 case 3:
466 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
467 break;
469 case 4:
470 if(m->b4ptr > 1) {
471 gcode = get_bits(gbc, 7);
472 m->b4_mant[0] = b4_mantissas[gcode][0];
473 m->b4_mant[1] = b4_mantissas[gcode][1];
474 m->b4ptr = 0;
476 coeffs[i] = m->b4_mant[m->b4ptr++];
477 break;
479 case 5:
480 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
481 break;
483 default: {
484 /* asymmetric dequantization */
485 int qlevel = quantization_tab[tbap];
486 coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
487 break;
490 coeffs[i] >>= exps[i];
495 * Remove random dithering from coefficients with zero-bit mantissas
496 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
498 static void remove_dithering(AC3DecodeContext *s) {
499 int ch, i;
500 int end=0;
501 int *coeffs;
502 uint8_t *bap;
504 for(ch=1; ch<=s->fbw_channels; ch++) {
505 if(!s->dither_flag[ch]) {
506 coeffs = s->fixed_coeffs[ch];
507 bap = s->bap[ch];
508 if(s->channel_in_cpl[ch])
509 end = s->start_freq[CPL_CH];
510 else
511 end = s->end_freq[ch];
512 for(i=0; i<end; i++) {
513 if(!bap[i])
514 coeffs[i] = 0;
516 if(s->channel_in_cpl[ch]) {
517 bap = s->bap[CPL_CH];
518 for(; i<s->end_freq[CPL_CH]; i++) {
519 if(!bap[i])
520 coeffs[i] = 0;
528 * Get the transform coefficients.
530 static void get_transform_coeffs(AC3DecodeContext *s)
532 int ch, end;
533 int got_cplchan = 0;
534 mant_groups m;
536 m.b1ptr = m.b2ptr = m.b4ptr = 3;
538 for (ch = 1; ch <= s->channels; ch++) {
539 /* transform coefficients for full-bandwidth channel */
540 get_transform_coeffs_ch(s, ch, &m);
541 /* tranform coefficients for coupling channel come right after the
542 coefficients for the first coupled channel*/
543 if (s->channel_in_cpl[ch]) {
544 if (!got_cplchan) {
545 get_transform_coeffs_ch(s, CPL_CH, &m);
546 uncouple_channels(s);
547 got_cplchan = 1;
549 end = s->end_freq[CPL_CH];
550 } else {
551 end = s->end_freq[ch];
554 s->fixed_coeffs[ch][end] = 0;
555 while(++end < 256);
558 /* if any channel doesn't use dithering, zero appropriate coefficients */
559 if(!s->dither_all)
560 remove_dithering(s);
564 * Stereo rematrixing.
565 * reference: Section 7.5.4 Rematrixing : Decoding Technique
567 static void do_rematrixing(AC3DecodeContext *s)
569 int bnd, i;
570 int end, bndend;
571 int tmp0, tmp1;
573 end = FFMIN(s->end_freq[1], s->end_freq[2]);
575 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
576 if(s->rematrixing_flags[bnd]) {
577 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd+1]);
578 for(i=ff_ac3_rematrix_band_tab[bnd]; i<bndend; i++) {
579 tmp0 = s->fixed_coeffs[1][i];
580 tmp1 = s->fixed_coeffs[2][i];
581 s->fixed_coeffs[1][i] = tmp0 + tmp1;
582 s->fixed_coeffs[2][i] = tmp0 - tmp1;
589 * Perform the 256-point IMDCT
591 static void do_imdct_256(AC3DecodeContext *s, int chindex)
593 int i, k;
594 DECLARE_ALIGNED_16(float, x[128]);
595 FFTComplex z[2][64];
596 float *o_ptr = s->tmp_output;
598 for(i=0; i<2; i++) {
599 /* de-interleave coefficients */
600 for(k=0; k<128; k++) {
601 x[k] = s->transform_coeffs[chindex][2*k+i];
604 /* run standard IMDCT */
605 s->imdct_256.fft.imdct_calc(&s->imdct_256, o_ptr, x, s->tmp_imdct);
607 /* reverse the post-rotation & reordering from standard IMDCT */
608 for(k=0; k<32; k++) {
609 z[i][32+k].re = -o_ptr[128+2*k];
610 z[i][32+k].im = -o_ptr[2*k];
611 z[i][31-k].re = o_ptr[2*k+1];
612 z[i][31-k].im = o_ptr[128+2*k+1];
616 /* apply AC-3 post-rotation & reordering */
617 for(k=0; k<64; k++) {
618 o_ptr[ 2*k ] = -z[0][ k].im;
619 o_ptr[ 2*k+1] = z[0][63-k].re;
620 o_ptr[128+2*k ] = -z[0][ k].re;
621 o_ptr[128+2*k+1] = z[0][63-k].im;
622 o_ptr[256+2*k ] = -z[1][ k].re;
623 o_ptr[256+2*k+1] = z[1][63-k].im;
624 o_ptr[384+2*k ] = z[1][ k].im;
625 o_ptr[384+2*k+1] = -z[1][63-k].re;
630 * Inverse MDCT Transform.
631 * Convert frequency domain coefficients to time-domain audio samples.
632 * reference: Section 7.9.4 Transformation Equations
634 static inline void do_imdct(AC3DecodeContext *s, int channels)
636 int ch;
638 for (ch=1; ch<=channels; ch++) {
639 if (s->block_switch[ch]) {
640 do_imdct_256(s, ch);
641 } else {
642 s->imdct_512.fft.imdct_calc(&s->imdct_512, s->tmp_output,
643 s->transform_coeffs[ch], s->tmp_imdct);
645 /* For the first half of the block, apply the window, add the delay
646 from the previous block, and send to output */
647 s->dsp.vector_fmul_add_add(s->output[ch-1], s->tmp_output,
648 s->window, s->delay[ch-1], 0, 256, 1);
649 /* For the second half of the block, apply the window and store the
650 samples to delay, to be combined with the next block */
651 s->dsp.vector_fmul_reverse(s->delay[ch-1], s->tmp_output+256,
652 s->window, 256);
657 * Downmix the output to mono or stereo.
659 static void ac3_downmix(AC3DecodeContext *s,
660 float samples[AC3_MAX_CHANNELS][256], int ch_offset)
662 int i, j;
663 float v0, v1;
665 for(i=0; i<256; i++) {
666 v0 = v1 = 0.0f;
667 for(j=0; j<s->fbw_channels; j++) {
668 v0 += samples[j+ch_offset][i] * s->downmix_coeffs[j][0];
669 v1 += samples[j+ch_offset][i] * s->downmix_coeffs[j][1];
671 v0 *= s->downmix_coeff_adjust[0];
672 v1 *= s->downmix_coeff_adjust[1];
673 if(s->output_mode == AC3_CHMODE_MONO) {
674 samples[ch_offset][i] = (v0 + v1) * LEVEL_MINUS_3DB;
675 } else if(s->output_mode == AC3_CHMODE_STEREO) {
676 samples[ ch_offset][i] = v0;
677 samples[1+ch_offset][i] = v1;
683 * Upmix delay samples from stereo to original channel layout.
685 static void ac3_upmix_delay(AC3DecodeContext *s)
687 int channel_data_size = sizeof(s->delay[0]);
688 switch(s->channel_mode) {
689 case AC3_CHMODE_DUALMONO:
690 case AC3_CHMODE_STEREO:
691 /* upmix mono to stereo */
692 memcpy(s->delay[1], s->delay[0], channel_data_size);
693 break;
694 case AC3_CHMODE_2F2R:
695 memset(s->delay[3], 0, channel_data_size);
696 case AC3_CHMODE_2F1R:
697 memset(s->delay[2], 0, channel_data_size);
698 break;
699 case AC3_CHMODE_3F2R:
700 memset(s->delay[4], 0, channel_data_size);
701 case AC3_CHMODE_3F1R:
702 memset(s->delay[3], 0, channel_data_size);
703 case AC3_CHMODE_3F:
704 memcpy(s->delay[2], s->delay[1], channel_data_size);
705 memset(s->delay[1], 0, channel_data_size);
706 break;
711 * Parse an audio block from AC-3 bitstream.
713 static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
715 int fbw_channels = s->fbw_channels;
716 int channel_mode = s->channel_mode;
717 int i, bnd, seg, ch;
718 int different_transforms;
719 int downmix_output;
720 int cpl_in_use;
721 GetBitContext *gbc = &s->gbc;
722 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
724 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
726 /* block switch flags */
727 different_transforms = 0;
728 for (ch = 1; ch <= fbw_channels; ch++) {
729 s->block_switch[ch] = get_bits1(gbc);
730 if(ch > 1 && s->block_switch[ch] != s->block_switch[1])
731 different_transforms = 1;
734 /* dithering flags */
735 s->dither_all = 1;
736 for (ch = 1; ch <= fbw_channels; ch++) {
737 s->dither_flag[ch] = get_bits1(gbc);
738 if(!s->dither_flag[ch])
739 s->dither_all = 0;
742 /* dynamic range */
743 i = !(s->channel_mode);
744 do {
745 if(get_bits1(gbc)) {
746 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
747 s->avctx->drc_scale)+1.0;
748 } else if(blk == 0) {
749 s->dynamic_range[i] = 1.0f;
751 } while(i--);
753 /* coupling strategy */
754 if (get_bits1(gbc)) {
755 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
756 cpl_in_use = get_bits1(gbc);
757 if (cpl_in_use) {
758 /* coupling in use */
759 int cpl_begin_freq, cpl_end_freq;
761 if (channel_mode < AC3_CHMODE_STEREO) {
762 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
763 return -1;
766 /* determine which channels are coupled */
767 for (ch = 1; ch <= fbw_channels; ch++)
768 s->channel_in_cpl[ch] = get_bits1(gbc);
770 /* phase flags in use */
771 if (channel_mode == AC3_CHMODE_STEREO)
772 s->phase_flags_in_use = get_bits1(gbc);
774 /* coupling frequency range and band structure */
775 cpl_begin_freq = get_bits(gbc, 4);
776 cpl_end_freq = get_bits(gbc, 4);
777 if (3 + cpl_end_freq - cpl_begin_freq < 0) {
778 av_log(s->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
779 return -1;
781 s->num_cpl_bands = s->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
782 s->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
783 s->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
784 for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
785 if (get_bits1(gbc)) {
786 s->cpl_band_struct[bnd] = 1;
787 s->num_cpl_bands--;
790 s->cpl_band_struct[s->num_cpl_subbands-1] = 0;
791 } else {
792 /* coupling not in use */
793 for (ch = 1; ch <= fbw_channels; ch++)
794 s->channel_in_cpl[ch] = 0;
796 } else if (!blk) {
797 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must be present in block 0\n");
798 return -1;
799 } else {
800 cpl_in_use = s->cpl_in_use[blk-1];
802 s->cpl_in_use[blk] = cpl_in_use;
804 /* coupling coordinates */
805 if (cpl_in_use) {
806 int cpl_coords_exist = 0;
808 for (ch = 1; ch <= fbw_channels; ch++) {
809 if (s->channel_in_cpl[ch]) {
810 if (get_bits1(gbc)) {
811 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
812 cpl_coords_exist = 1;
813 master_cpl_coord = 3 * get_bits(gbc, 2);
814 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
815 cpl_coord_exp = get_bits(gbc, 4);
816 cpl_coord_mant = get_bits(gbc, 4);
817 if (cpl_coord_exp == 15)
818 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
819 else
820 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
821 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
823 } else if (!blk) {
824 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must be present in block 0\n");
825 return -1;
829 /* phase flags */
830 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
831 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
832 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
837 /* stereo rematrixing strategy and band structure */
838 if (channel_mode == AC3_CHMODE_STEREO) {
839 if (get_bits1(gbc)) {
840 s->num_rematrixing_bands = 4;
841 if(cpl_in_use && s->start_freq[CPL_CH] <= 61)
842 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
843 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
844 s->rematrixing_flags[bnd] = get_bits1(gbc);
845 } else if (!blk) {
846 av_log(s->avctx, AV_LOG_ERROR, "new rematrixing strategy must be present in block 0\n");
847 return -1;
851 /* exponent strategies for each channel */
852 s->exp_strategy[blk][CPL_CH] = EXP_REUSE;
853 s->exp_strategy[blk][s->lfe_ch] = EXP_REUSE;
854 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
855 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
856 if(s->exp_strategy[blk][ch] != EXP_REUSE)
857 bit_alloc_stages[ch] = 3;
860 /* channel bandwidth */
861 for (ch = 1; ch <= fbw_channels; ch++) {
862 s->start_freq[ch] = 0;
863 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
864 int group_size;
865 int prev = s->end_freq[ch];
866 if (s->channel_in_cpl[ch])
867 s->end_freq[ch] = s->start_freq[CPL_CH];
868 else {
869 int bandwidth_code = get_bits(gbc, 6);
870 if (bandwidth_code > 60) {
871 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
872 return -1;
874 s->end_freq[ch] = bandwidth_code * 3 + 73;
876 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
877 s->num_exp_groups[ch] = (s->end_freq[ch]+group_size-4) / group_size;
878 if(blk > 0 && s->end_freq[ch] != prev)
879 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
882 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
883 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
884 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
887 /* decode exponents for each channel */
888 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
889 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
890 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
891 decode_exponents(gbc, s->exp_strategy[blk][ch],
892 s->num_exp_groups[ch], s->dexps[ch][0],
893 &s->dexps[ch][s->start_freq[ch]+!!ch]);
894 if(ch != CPL_CH && ch != s->lfe_ch)
895 skip_bits(gbc, 2); /* skip gainrng */
899 /* bit allocation information */
900 if (get_bits1(gbc)) {
901 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
902 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
903 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
904 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
905 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
906 for(ch=!cpl_in_use; ch<=s->channels; ch++)
907 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
908 } else if (!blk) {
909 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must be present in block 0\n");
910 return -1;
913 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
914 if (get_bits1(gbc)) {
915 int csnr;
916 csnr = (get_bits(gbc, 6) - 15) << 4;
917 for (ch = !cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
918 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
919 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
921 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
922 } else if (!blk) {
923 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
924 return -1;
927 /* coupling leak information */
928 if (cpl_in_use) {
929 if (get_bits1(gbc)) {
930 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
931 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
932 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
933 } else if (!blk) {
934 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must be present in block 0\n");
935 return -1;
939 /* delta bit allocation information */
940 if (get_bits1(gbc)) {
941 /* delta bit allocation exists (strategy) */
942 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
943 s->dba_mode[ch] = get_bits(gbc, 2);
944 if (s->dba_mode[ch] == DBA_RESERVED) {
945 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
946 return -1;
948 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
950 /* channel delta offset, len and bit allocation */
951 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
952 if (s->dba_mode[ch] == DBA_NEW) {
953 s->dba_nsegs[ch] = get_bits(gbc, 3);
954 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
955 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
956 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
957 s->dba_values[ch][seg] = get_bits(gbc, 3);
959 /* run last 2 bit allocation stages if new dba values */
960 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
963 } else if(blk == 0) {
964 for(ch=0; ch<=s->channels; ch++) {
965 s->dba_mode[ch] = DBA_NONE;
969 /* Bit allocation */
970 for(ch=!cpl_in_use; ch<=s->channels; ch++) {
971 if(bit_alloc_stages[ch] > 2) {
972 /* Exponent mapping into PSD and PSD integration */
973 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
974 s->start_freq[ch], s->end_freq[ch],
975 s->psd[ch], s->band_psd[ch]);
977 if(bit_alloc_stages[ch] > 1) {
978 /* Compute excitation function, Compute masking curve, and
979 Apply delta bit allocation */
980 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
981 s->start_freq[ch], s->end_freq[ch],
982 s->fast_gain[ch], (ch == s->lfe_ch),
983 s->dba_mode[ch], s->dba_nsegs[ch],
984 s->dba_offsets[ch], s->dba_lengths[ch],
985 s->dba_values[ch], s->mask[ch]);
987 if(bit_alloc_stages[ch] > 0) {
988 /* Compute bit allocation */
989 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
990 s->start_freq[ch], s->end_freq[ch],
991 s->snr_offset[ch],
992 s->bit_alloc_params.floor,
993 ff_ac3_bap_tab, s->bap[ch]);
997 /* unused dummy data */
998 if (get_bits1(gbc)) {
999 int skipl = get_bits(gbc, 9);
1000 while(skipl--)
1001 skip_bits(gbc, 8);
1004 /* unpack the transform coefficients
1005 this also uncouples channels if coupling is in use. */
1006 get_transform_coeffs(s);
1008 /* recover coefficients if rematrixing is in use */
1009 if(s->channel_mode == AC3_CHMODE_STEREO)
1010 do_rematrixing(s);
1012 /* apply scaling to coefficients (headroom, dynrng) */
1013 for(ch=1; ch<=s->channels; ch++) {
1014 float gain = s->mul_bias / 4194304.0f;
1015 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1016 gain *= s->dynamic_range[ch-1];
1017 } else {
1018 gain *= s->dynamic_range[0];
1020 for(i=0; i<256; i++) {
1021 s->transform_coeffs[ch][i] = s->fixed_coeffs[ch][i] * gain;
1025 /* downmix and MDCT. order depends on whether block switching is used for
1026 any channel in this block. this is because coefficients for the long
1027 and short transforms cannot be mixed. */
1028 downmix_output = s->channels != s->out_channels &&
1029 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1030 s->fbw_channels == s->out_channels);
1031 if(different_transforms) {
1032 /* the delay samples have already been downmixed, so we upmix the delay
1033 samples in order to reconstruct all channels before downmixing. */
1034 if(s->downmixed) {
1035 s->downmixed = 0;
1036 ac3_upmix_delay(s);
1039 do_imdct(s, s->channels);
1041 if(downmix_output) {
1042 ac3_downmix(s, s->output, 0);
1044 } else {
1045 if(downmix_output) {
1046 ac3_downmix(s, s->transform_coeffs, 1);
1049 if(!s->downmixed) {
1050 s->downmixed = 1;
1051 ac3_downmix(s, s->delay, 0);
1054 do_imdct(s, s->out_channels);
1057 /* convert float to 16-bit integer */
1058 for(ch=0; ch<s->out_channels; ch++) {
1059 for(i=0; i<256; i++) {
1060 s->output[ch][i] += s->add_bias;
1062 s->dsp.float_to_int16(s->int_output[ch], s->output[ch], 256);
1065 return 0;
1069 * Decode a single AC-3 frame.
1071 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
1072 const uint8_t *buf, int buf_size)
1074 AC3DecodeContext *s = avctx->priv_data;
1075 int16_t *out_samples = (int16_t *)data;
1076 int i, blk, ch, err;
1078 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1079 if (s->input_buffer) {
1080 /* copy input buffer to decoder context to avoid reading past the end
1081 of the buffer, which can be caused by a damaged input stream. */
1082 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_MAX_FRAME_SIZE));
1083 init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
1084 } else {
1085 init_get_bits(&s->gbc, buf, buf_size * 8);
1088 /* parse the syncinfo */
1089 *data_size = 0;
1090 err = parse_frame_header(s);
1092 /* check that reported frame size fits in input buffer */
1093 if(s->frame_size > buf_size) {
1094 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1095 err = AC3_PARSE_ERROR_FRAME_SIZE;
1098 /* check for crc mismatch */
1099 if(err != AC3_PARSE_ERROR_FRAME_SIZE && avctx->error_resilience >= FF_ER_CAREFUL) {
1100 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1101 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1102 err = AC3_PARSE_ERROR_CRC;
1106 if(err && err != AC3_PARSE_ERROR_CRC) {
1107 switch(err) {
1108 case AC3_PARSE_ERROR_SYNC:
1109 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1110 return -1;
1111 case AC3_PARSE_ERROR_BSID:
1112 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1113 break;
1114 case AC3_PARSE_ERROR_SAMPLE_RATE:
1115 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1116 break;
1117 case AC3_PARSE_ERROR_FRAME_SIZE:
1118 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1119 break;
1120 case AC3_PARSE_ERROR_FRAME_TYPE:
1121 /* skip frame if CRC is ok. otherwise use error concealment. */
1122 /* TODO: add support for substreams and dependent frames */
1123 if(s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1124 av_log(avctx, AV_LOG_ERROR, "unsupported frame type : skipping frame\n");
1125 return s->frame_size;
1126 } else {
1127 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1129 break;
1130 default:
1131 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1132 break;
1136 /* if frame is ok, set audio parameters */
1137 if (!err) {
1138 avctx->sample_rate = s->sample_rate;
1139 avctx->bit_rate = s->bit_rate;
1141 /* channel config */
1142 s->out_channels = s->channels;
1143 s->output_mode = s->channel_mode;
1144 if(s->lfe_on)
1145 s->output_mode |= AC3_OUTPUT_LFEON;
1146 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1147 avctx->request_channels < s->channels) {
1148 s->out_channels = avctx->request_channels;
1149 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1151 avctx->channels = s->out_channels;
1153 /* set downmixing coefficients if needed */
1154 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1155 s->fbw_channels == s->out_channels)) {
1156 set_downmix_coeffs(s);
1158 } else if (!s->out_channels) {
1159 s->out_channels = avctx->channels;
1160 if(s->out_channels < s->channels)
1161 s->output_mode = s->out_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1164 /* parse the audio blocks */
1165 for (blk = 0; blk < s->num_blocks; blk++) {
1166 if (!err && ac3_parse_audio_block(s, blk)) {
1167 av_log(avctx, AV_LOG_ERROR, "error parsing the audio block\n");
1170 /* interleave output samples */
1171 for (i = 0; i < 256; i++)
1172 for (ch = 0; ch < s->out_channels; ch++)
1173 *(out_samples++) = s->int_output[ch][i];
1175 *data_size = s->num_blocks * 256 * avctx->channels * sizeof (int16_t);
1176 return s->frame_size;
1180 * Uninitialize the AC-3 decoder.
1182 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1184 AC3DecodeContext *s = avctx->priv_data;
1185 ff_mdct_end(&s->imdct_512);
1186 ff_mdct_end(&s->imdct_256);
1188 av_freep(&s->input_buffer);
1190 return 0;
1193 AVCodec ac3_decoder = {
1194 .name = "ac3",
1195 .type = CODEC_TYPE_AUDIO,
1196 .id = CODEC_ID_AC3,
1197 .priv_data_size = sizeof (AC3DecodeContext),
1198 .init = ac3_decode_init,
1199 .close = ac3_decode_end,
1200 .decode = ac3_decode_frame,
1201 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52 / AC-3"),