Modify decode_band_structure() so that the actual band structure is only
[FFMpeg-mirror/lagarith.git] / libavcodec / ac3dec.c
blob7f7efe56ba1f599483fc62a1c5bfa5cd5a5d30c5
1 /*
2 * AC-3 Audio Decoder
3 * This code was developed as part of Google Summer of Code 2006.
4 * E-AC-3 support was added as part of Google Summer of Code 2007.
6 * Copyright (c) 2006 Kartikey Mahendra BHATT (bhattkm at gmail dot com)
7 * Copyright (c) 2007-2008 Bartlomiej Wolowiec <bartek.wolowiec@gmail.com>
8 * Copyright (c) 2007 Justin Ruggles <justin.ruggles@gmail.com>
10 * This file is part of FFmpeg.
12 * FFmpeg is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public
14 * License as published by the Free Software Foundation; either
15 * version 2.1 of the License, or (at your option) any later version.
17 * FFmpeg is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with FFmpeg; if not, write to the Free Software
24 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
27 #include <stdio.h>
28 #include <stddef.h>
29 #include <math.h>
30 #include <string.h>
32 #include "libavutil/crc.h"
33 #include "internal.h"
34 #include "aac_ac3_parser.h"
35 #include "ac3_parser.h"
36 #include "ac3dec.h"
37 #include "ac3dec_data.h"
39 /** Large enough for maximum possible frame size when the specification limit is ignored */
40 #define AC3_FRAME_BUFFER_SIZE 32768
42 /**
43 * table for ungrouping 3 values in 7 bits.
44 * used for exponents and bap=2 mantissas
46 static uint8_t ungroup_3_in_7_bits_tab[128][3];
49 /** tables for ungrouping mantissas */
50 static int b1_mantissas[32][3];
51 static int b2_mantissas[128][3];
52 static int b3_mantissas[8];
53 static int b4_mantissas[128][2];
54 static int b5_mantissas[16];
56 /**
57 * Quantization table: levels for symmetric. bits for asymmetric.
58 * reference: Table 7.18 Mapping of bap to Quantizer
60 static const uint8_t quantization_tab[16] = {
61 0, 3, 5, 7, 11, 15,
62 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
65 /** dynamic range table. converts codes to scale factors. */
66 static float dynamic_range_tab[256];
68 /** Adjustments in dB gain */
69 #define LEVEL_PLUS_3DB 1.4142135623730950
70 #define LEVEL_PLUS_1POINT5DB 1.1892071150027209
71 #define LEVEL_MINUS_1POINT5DB 0.8408964152537145
72 #define LEVEL_MINUS_3DB 0.7071067811865476
73 #define LEVEL_MINUS_4POINT5DB 0.5946035575013605
74 #define LEVEL_MINUS_6DB 0.5000000000000000
75 #define LEVEL_MINUS_9DB 0.3535533905932738
76 #define LEVEL_ZERO 0.0000000000000000
77 #define LEVEL_ONE 1.0000000000000000
79 static const float gain_levels[9] = {
80 LEVEL_PLUS_3DB,
81 LEVEL_PLUS_1POINT5DB,
82 LEVEL_ONE,
83 LEVEL_MINUS_1POINT5DB,
84 LEVEL_MINUS_3DB,
85 LEVEL_MINUS_4POINT5DB,
86 LEVEL_MINUS_6DB,
87 LEVEL_ZERO,
88 LEVEL_MINUS_9DB
91 /**
92 * Table for center mix levels
93 * reference: Section 5.4.2.4 cmixlev
95 static const uint8_t center_levels[4] = { 4, 5, 6, 5 };
97 /**
98 * Table for surround mix levels
99 * reference: Section 5.4.2.5 surmixlev
101 static const uint8_t surround_levels[4] = { 4, 6, 7, 6 };
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 { { 2, 7 }, { 7, 2 }, },
109 { { 4, 4 }, },
110 { { 2, 7 }, { 7, 2 }, },
111 { { 2, 7 }, { 5, 5 }, { 7, 2 }, },
112 { { 2, 7 }, { 7, 2 }, { 6, 6 }, },
113 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 8, 8 }, },
114 { { 2, 7 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
115 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
119 * Symmetrical Dequantization
120 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
121 * Tables 7.19 to 7.23
123 static inline int
124 symmetric_dequant(int code, int levels)
126 return ((code - (levels >> 1)) << 24) / levels;
130 * Initialize tables at runtime.
132 static av_cold void ac3_tables_init(void)
134 int i;
136 /* generate table for ungrouping 3 values in 7 bits
137 reference: Section 7.1.3 Exponent Decoding */
138 for(i=0; i<128; i++) {
139 ungroup_3_in_7_bits_tab[i][0] = i / 25;
140 ungroup_3_in_7_bits_tab[i][1] = (i % 25) / 5;
141 ungroup_3_in_7_bits_tab[i][2] = (i % 25) % 5;
144 /* generate grouped mantissa tables
145 reference: Section 7.3.5 Ungrouping of Mantissas */
146 for(i=0; i<32; i++) {
147 /* bap=1 mantissas */
148 b1_mantissas[i][0] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][0], 3);
149 b1_mantissas[i][1] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][1], 3);
150 b1_mantissas[i][2] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][2], 3);
152 for(i=0; i<128; i++) {
153 /* bap=2 mantissas */
154 b2_mantissas[i][0] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][0], 5);
155 b2_mantissas[i][1] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][1], 5);
156 b2_mantissas[i][2] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][2], 5);
158 /* bap=4 mantissas */
159 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
160 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
162 /* generate ungrouped mantissa tables
163 reference: Tables 7.21 and 7.23 */
164 for(i=0; i<7; i++) {
165 /* bap=3 mantissas */
166 b3_mantissas[i] = symmetric_dequant(i, 7);
168 for(i=0; i<15; i++) {
169 /* bap=5 mantissas */
170 b5_mantissas[i] = symmetric_dequant(i, 15);
173 /* generate dynamic range table
174 reference: Section 7.7.1 Dynamic Range Control */
175 for(i=0; i<256; i++) {
176 int v = (i >> 5) - ((i >> 7) << 3) - 5;
177 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
183 * AVCodec initialization
185 static av_cold int ac3_decode_init(AVCodecContext *avctx)
187 AC3DecodeContext *s = avctx->priv_data;
188 s->avctx = avctx;
190 ac3_common_init();
191 ac3_tables_init();
192 ff_mdct_init(&s->imdct_256, 8, 1, 1.0);
193 ff_mdct_init(&s->imdct_512, 9, 1, 1.0);
194 ff_kbd_window_init(s->window, 5.0, 256);
195 dsputil_init(&s->dsp, avctx);
196 av_lfg_init(&s->dith_state, 0);
198 /* set bias values for float to int16 conversion */
199 if(s->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
200 s->add_bias = 385.0f;
201 s->mul_bias = 1.0f;
202 } else {
203 s->add_bias = 0.0f;
204 s->mul_bias = 32767.0f;
207 /* allow downmixing to stereo or mono */
208 if (avctx->channels > 0 && avctx->request_channels > 0 &&
209 avctx->request_channels < avctx->channels &&
210 avctx->request_channels <= 2) {
211 avctx->channels = avctx->request_channels;
213 s->downmixed = 1;
215 /* allocate context input buffer */
216 if (avctx->error_recognition >= FF_ER_CAREFUL) {
217 s->input_buffer = av_mallocz(AC3_FRAME_BUFFER_SIZE + FF_INPUT_BUFFER_PADDING_SIZE);
218 if (!s->input_buffer)
219 return AVERROR_NOMEM;
222 avctx->sample_fmt = SAMPLE_FMT_S16;
223 return 0;
227 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
228 * GetBitContext within AC3DecodeContext must point to
229 * the start of the synchronized AC-3 bitstream.
231 static int ac3_parse_header(AC3DecodeContext *s)
233 GetBitContext *gbc = &s->gbc;
234 int i;
236 /* read the rest of the bsi. read twice for dual mono mode. */
237 i = !(s->channel_mode);
238 do {
239 skip_bits(gbc, 5); // skip dialog normalization
240 if (get_bits1(gbc))
241 skip_bits(gbc, 8); //skip compression
242 if (get_bits1(gbc))
243 skip_bits(gbc, 8); //skip language code
244 if (get_bits1(gbc))
245 skip_bits(gbc, 7); //skip audio production information
246 } while (i--);
248 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
250 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
251 TODO: read & use the xbsi1 downmix levels */
252 if (get_bits1(gbc))
253 skip_bits(gbc, 14); //skip timecode1 / xbsi1
254 if (get_bits1(gbc))
255 skip_bits(gbc, 14); //skip timecode2 / xbsi2
257 /* skip additional bitstream info */
258 if (get_bits1(gbc)) {
259 i = get_bits(gbc, 6);
260 do {
261 skip_bits(gbc, 8);
262 } while(i--);
265 return 0;
269 * Common function to parse AC-3 or E-AC-3 frame header
271 static int parse_frame_header(AC3DecodeContext *s)
273 AC3HeaderInfo hdr;
274 int err;
276 err = ff_ac3_parse_header(&s->gbc, &hdr);
277 if(err)
278 return err;
280 /* get decoding parameters from header info */
281 s->bit_alloc_params.sr_code = hdr.sr_code;
282 s->channel_mode = hdr.channel_mode;
283 s->channel_layout = hdr.channel_layout;
284 s->lfe_on = hdr.lfe_on;
285 s->bit_alloc_params.sr_shift = hdr.sr_shift;
286 s->sample_rate = hdr.sample_rate;
287 s->bit_rate = hdr.bit_rate;
288 s->channels = hdr.channels;
289 s->fbw_channels = s->channels - s->lfe_on;
290 s->lfe_ch = s->fbw_channels + 1;
291 s->frame_size = hdr.frame_size;
292 s->center_mix_level = hdr.center_mix_level;
293 s->surround_mix_level = hdr.surround_mix_level;
294 s->num_blocks = hdr.num_blocks;
295 s->frame_type = hdr.frame_type;
296 s->substreamid = hdr.substreamid;
298 if(s->lfe_on) {
299 s->start_freq[s->lfe_ch] = 0;
300 s->end_freq[s->lfe_ch] = 7;
301 s->num_exp_groups[s->lfe_ch] = 2;
302 s->channel_in_cpl[s->lfe_ch] = 0;
305 if (hdr.bitstream_id <= 10) {
306 s->eac3 = 0;
307 s->snr_offset_strategy = 2;
308 s->block_switch_syntax = 1;
309 s->dither_flag_syntax = 1;
310 s->bit_allocation_syntax = 1;
311 s->fast_gain_syntax = 0;
312 s->first_cpl_leak = 0;
313 s->dba_syntax = 1;
314 s->skip_syntax = 1;
315 memset(s->channel_uses_aht, 0, sizeof(s->channel_uses_aht));
316 return ac3_parse_header(s);
317 } else if (CONFIG_EAC3_DECODER) {
318 s->eac3 = 1;
319 return ff_eac3_parse_header(s);
320 } else {
321 av_log(s->avctx, AV_LOG_ERROR, "E-AC-3 support not compiled in\n");
322 return -1;
327 * Set stereo downmixing coefficients based on frame header info.
328 * reference: Section 7.8.2 Downmixing Into Two Channels
330 static void set_downmix_coeffs(AC3DecodeContext *s)
332 int i;
333 float cmix = gain_levels[center_levels[s->center_mix_level]];
334 float smix = gain_levels[surround_levels[s->surround_mix_level]];
335 float norm0, norm1;
337 for(i=0; i<s->fbw_channels; i++) {
338 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
339 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
341 if(s->channel_mode > 1 && s->channel_mode & 1) {
342 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
344 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
345 int nf = s->channel_mode - 2;
346 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
348 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
349 int nf = s->channel_mode - 4;
350 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
353 /* renormalize */
354 norm0 = norm1 = 0.0;
355 for(i=0; i<s->fbw_channels; i++) {
356 norm0 += s->downmix_coeffs[i][0];
357 norm1 += s->downmix_coeffs[i][1];
359 norm0 = 1.0f / norm0;
360 norm1 = 1.0f / norm1;
361 for(i=0; i<s->fbw_channels; i++) {
362 s->downmix_coeffs[i][0] *= norm0;
363 s->downmix_coeffs[i][1] *= norm1;
366 if(s->output_mode == AC3_CHMODE_MONO) {
367 for(i=0; i<s->fbw_channels; i++)
368 s->downmix_coeffs[i][0] = (s->downmix_coeffs[i][0] + s->downmix_coeffs[i][1]) * LEVEL_MINUS_3DB;
373 * Decode the grouped exponents according to exponent strategy.
374 * reference: Section 7.1.3 Exponent Decoding
376 static int decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
377 uint8_t absexp, int8_t *dexps)
379 int i, j, grp, group_size;
380 int dexp[256];
381 int expacc, prevexp;
383 /* unpack groups */
384 group_size = exp_strategy + (exp_strategy == EXP_D45);
385 for(grp=0,i=0; grp<ngrps; grp++) {
386 expacc = get_bits(gbc, 7);
387 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][0];
388 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][1];
389 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][2];
392 /* convert to absolute exps and expand groups */
393 prevexp = absexp;
394 for(i=0,j=0; i<ngrps*3; i++) {
395 prevexp += dexp[i] - 2;
396 if (prevexp > 24U)
397 return -1;
398 switch (group_size) {
399 case 4: dexps[j++] = prevexp;
400 dexps[j++] = prevexp;
401 case 2: dexps[j++] = prevexp;
402 case 1: dexps[j++] = prevexp;
405 return 0;
409 * Generate transform coefficients for each coupled channel in the coupling
410 * range using the coupling coefficients and coupling coordinates.
411 * reference: Section 7.4.3 Coupling Coordinate Format
413 static void calc_transform_coeffs_cpl(AC3DecodeContext *s)
415 int i, j, ch, bnd;
417 i = s->start_freq[CPL_CH];
418 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
419 for (j = 0; j < s->cpl_band_sizes[bnd]; j++,i++) {
420 for(ch=1; ch<=s->fbw_channels; ch++) {
421 if(s->channel_in_cpl[ch]) {
422 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
423 if (ch == 2 && s->phase_flags[bnd])
424 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
432 * Grouped mantissas for 3-level 5-level and 11-level quantization
434 typedef struct {
435 int b1_mant[2];
436 int b2_mant[2];
437 int b4_mant;
438 int b1;
439 int b2;
440 int b4;
441 } mant_groups;
444 * Decode the transform coefficients for a particular channel
445 * reference: Section 7.3 Quantization and Decoding of Mantissas
447 static void ac3_decode_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
449 int start_freq = s->start_freq[ch_index];
450 int end_freq = s->end_freq[ch_index];
451 uint8_t *baps = s->bap[ch_index];
452 int8_t *exps = s->dexps[ch_index];
453 int *coeffs = s->fixed_coeffs[ch_index];
454 GetBitContext *gbc = &s->gbc;
455 int freq;
457 for(freq = start_freq; freq < end_freq; freq++){
458 int bap = baps[freq];
459 int mantissa;
460 switch(bap){
461 case 0:
462 mantissa = (av_lfg_get(&s->dith_state) & 0x7FFFFF) - 0x400000;
463 break;
464 case 1:
465 if(m->b1){
466 m->b1--;
467 mantissa = m->b1_mant[m->b1];
469 else{
470 int bits = get_bits(gbc, 5);
471 mantissa = b1_mantissas[bits][0];
472 m->b1_mant[1] = b1_mantissas[bits][1];
473 m->b1_mant[0] = b1_mantissas[bits][2];
474 m->b1 = 2;
476 break;
477 case 2:
478 if(m->b2){
479 m->b2--;
480 mantissa = m->b2_mant[m->b2];
482 else{
483 int bits = get_bits(gbc, 7);
484 mantissa = b2_mantissas[bits][0];
485 m->b2_mant[1] = b2_mantissas[bits][1];
486 m->b2_mant[0] = b2_mantissas[bits][2];
487 m->b2 = 2;
489 break;
490 case 3:
491 mantissa = b3_mantissas[get_bits(gbc, 3)];
492 break;
493 case 4:
494 if(m->b4){
495 m->b4 = 0;
496 mantissa = m->b4_mant;
498 else{
499 int bits = get_bits(gbc, 7);
500 mantissa = b4_mantissas[bits][0];
501 m->b4_mant = b4_mantissas[bits][1];
502 m->b4 = 1;
504 break;
505 case 5:
506 mantissa = b5_mantissas[get_bits(gbc, 4)];
507 break;
508 default: /* 6 to 15 */
509 mantissa = get_bits(gbc, quantization_tab[bap]);
510 /* Shift mantissa and sign-extend it. */
511 mantissa = (mantissa << (32-quantization_tab[bap]))>>8;
512 break;
514 coeffs[freq] = mantissa >> exps[freq];
519 * Remove random dithering from coefficients with zero-bit mantissas
520 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
522 static void remove_dithering(AC3DecodeContext *s) {
523 int ch, i;
524 int end=0;
525 int *coeffs;
526 uint8_t *bap;
528 for(ch=1; ch<=s->fbw_channels; ch++) {
529 if(!s->dither_flag[ch]) {
530 coeffs = s->fixed_coeffs[ch];
531 bap = s->bap[ch];
532 if(s->channel_in_cpl[ch])
533 end = s->start_freq[CPL_CH];
534 else
535 end = s->end_freq[ch];
536 for(i=0; i<end; i++) {
537 if(!bap[i])
538 coeffs[i] = 0;
540 if(s->channel_in_cpl[ch]) {
541 bap = s->bap[CPL_CH];
542 for(; i<s->end_freq[CPL_CH]; i++) {
543 if(!bap[i])
544 coeffs[i] = 0;
551 static void decode_transform_coeffs_ch(AC3DecodeContext *s, int blk, int ch,
552 mant_groups *m)
554 if (!s->channel_uses_aht[ch]) {
555 ac3_decode_transform_coeffs_ch(s, ch, m);
556 } else {
557 /* if AHT is used, mantissas for all blocks are encoded in the first
558 block of the frame. */
559 int bin;
560 if (!blk && CONFIG_EAC3_DECODER)
561 ff_eac3_decode_transform_coeffs_aht_ch(s, ch);
562 for (bin = s->start_freq[ch]; bin < s->end_freq[ch]; bin++) {
563 s->fixed_coeffs[ch][bin] = s->pre_mantissa[ch][bin][blk] >> s->dexps[ch][bin];
569 * Decode the transform coefficients.
571 static void decode_transform_coeffs(AC3DecodeContext *s, int blk)
573 int ch, end;
574 int got_cplchan = 0;
575 mant_groups m;
577 m.b1 = m.b2 = m.b4 = 0;
579 for (ch = 1; ch <= s->channels; ch++) {
580 /* transform coefficients for full-bandwidth channel */
581 decode_transform_coeffs_ch(s, blk, ch, &m);
582 /* tranform coefficients for coupling channel come right after the
583 coefficients for the first coupled channel*/
584 if (s->channel_in_cpl[ch]) {
585 if (!got_cplchan) {
586 decode_transform_coeffs_ch(s, blk, CPL_CH, &m);
587 calc_transform_coeffs_cpl(s);
588 got_cplchan = 1;
590 end = s->end_freq[CPL_CH];
591 } else {
592 end = s->end_freq[ch];
595 s->fixed_coeffs[ch][end] = 0;
596 while(++end < 256);
599 /* zero the dithered coefficients for appropriate channels */
600 remove_dithering(s);
604 * Stereo rematrixing.
605 * reference: Section 7.5.4 Rematrixing : Decoding Technique
607 static void do_rematrixing(AC3DecodeContext *s)
609 int bnd, i;
610 int end, bndend;
611 int tmp0, tmp1;
613 end = FFMIN(s->end_freq[1], s->end_freq[2]);
615 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
616 if(s->rematrixing_flags[bnd]) {
617 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd+1]);
618 for(i=ff_ac3_rematrix_band_tab[bnd]; i<bndend; i++) {
619 tmp0 = s->fixed_coeffs[1][i];
620 tmp1 = s->fixed_coeffs[2][i];
621 s->fixed_coeffs[1][i] = tmp0 + tmp1;
622 s->fixed_coeffs[2][i] = tmp0 - tmp1;
629 * Inverse MDCT Transform.
630 * Convert frequency domain coefficients to time-domain audio samples.
631 * reference: Section 7.9.4 Transformation Equations
633 static inline void do_imdct(AC3DecodeContext *s, int channels)
635 int ch;
636 float add_bias = s->add_bias;
637 if(s->out_channels==1 && channels>1)
638 add_bias *= LEVEL_MINUS_3DB; // compensate for the gain in downmix
640 for (ch=1; ch<=channels; ch++) {
641 if (s->block_switch[ch]) {
642 int i;
643 float *x = s->tmp_output+128;
644 for(i=0; i<128; i++)
645 x[i] = s->transform_coeffs[ch][2*i];
646 ff_imdct_half(&s->imdct_256, s->tmp_output, x);
647 s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, add_bias, 128);
648 for(i=0; i<128; i++)
649 x[i] = s->transform_coeffs[ch][2*i+1];
650 ff_imdct_half(&s->imdct_256, s->delay[ch-1], x);
651 } else {
652 ff_imdct_half(&s->imdct_512, s->tmp_output, s->transform_coeffs[ch]);
653 s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, add_bias, 128);
654 memcpy(s->delay[ch-1], s->tmp_output+128, 128*sizeof(float));
660 * Downmix the output to mono or stereo.
662 void ff_ac3_downmix_c(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len)
664 int i, j;
665 float v0, v1;
666 if(out_ch == 2) {
667 for(i=0; i<len; i++) {
668 v0 = v1 = 0.0f;
669 for(j=0; j<in_ch; j++) {
670 v0 += samples[j][i] * matrix[j][0];
671 v1 += samples[j][i] * matrix[j][1];
673 samples[0][i] = v0;
674 samples[1][i] = v1;
676 } else if(out_ch == 1) {
677 for(i=0; i<len; i++) {
678 v0 = 0.0f;
679 for(j=0; j<in_ch; j++)
680 v0 += samples[j][i] * matrix[j][0];
681 samples[0][i] = v0;
687 * Upmix delay samples from stereo to original channel layout.
689 static void ac3_upmix_delay(AC3DecodeContext *s)
691 int channel_data_size = sizeof(s->delay[0]);
692 switch(s->channel_mode) {
693 case AC3_CHMODE_DUALMONO:
694 case AC3_CHMODE_STEREO:
695 /* upmix mono to stereo */
696 memcpy(s->delay[1], s->delay[0], channel_data_size);
697 break;
698 case AC3_CHMODE_2F2R:
699 memset(s->delay[3], 0, channel_data_size);
700 case AC3_CHMODE_2F1R:
701 memset(s->delay[2], 0, channel_data_size);
702 break;
703 case AC3_CHMODE_3F2R:
704 memset(s->delay[4], 0, channel_data_size);
705 case AC3_CHMODE_3F1R:
706 memset(s->delay[3], 0, channel_data_size);
707 case AC3_CHMODE_3F:
708 memcpy(s->delay[2], s->delay[1], channel_data_size);
709 memset(s->delay[1], 0, channel_data_size);
710 break;
715 * Decode band structure for coupling, spectral extension, or enhanced coupling.
716 * The band structure defines how many subbands are in each band. For each
717 * subband in the range, 1 means it is combined with the previous band, and 0
718 * means that it starts a new band.
720 * @param[in] gbc bit reader context
721 * @param[in] blk block number
722 * @param[in] eac3 flag to indicate E-AC-3
723 * @param[in] ecpl flag to indicate enhanced coupling
724 * @param[in] start_subband subband number for start of range
725 * @param[in] end_subband subband number for end of range
726 * @param[in] default_band_struct default band structure table
727 * @param[out] num_bands number of bands (optionally NULL)
728 * @param[out] band_sizes array containing the number of bins in each band (optionally NULL)
730 static void decode_band_structure(GetBitContext *gbc, int blk, int eac3,
731 int ecpl, int start_subband, int end_subband,
732 const uint8_t *default_band_struct,
733 int *num_bands,
734 uint8_t *band_sizes)
736 int subbnd, bnd, n_subbands, n_bands=0;
737 uint8_t bnd_sz[22];
738 uint8_t coded_band_struct[22];
739 const uint8_t *band_struct;
741 n_subbands = end_subband - start_subband;
743 /* decode band structure from bitstream or use default */
744 if (!eac3 || get_bits1(gbc)) {
745 for (subbnd = 0; subbnd < n_subbands - 1; subbnd++) {
746 coded_band_struct[subbnd] = get_bits1(gbc);
748 band_struct = coded_band_struct;
749 } else if (!blk) {
750 band_struct = &default_band_struct[start_subband+1];
751 } else {
752 /* no change in band structure */
753 return;
756 /* calculate number of bands and band sizes based on band structure.
757 note that the first 4 subbands in enhanced coupling span only 6 bins
758 instead of 12. */
759 if (num_bands || band_sizes ) {
760 n_bands = n_subbands;
761 bnd_sz[0] = ecpl ? 6 : 12;
762 for (bnd = 0, subbnd = 1; subbnd < n_subbands; subbnd++) {
763 int subbnd_size = (ecpl && subbnd < 4) ? 6 : 12;
764 if (band_struct[subbnd-1]) {
765 n_bands--;
766 bnd_sz[bnd] += subbnd_size;
767 } else {
768 bnd_sz[++bnd] = subbnd_size;
773 /* set optional output params */
774 if (num_bands)
775 *num_bands = n_bands;
776 if (band_sizes)
777 memcpy(band_sizes, bnd_sz, n_bands);
781 * Decode a single audio block from the AC-3 bitstream.
783 static int decode_audio_block(AC3DecodeContext *s, int blk)
785 int fbw_channels = s->fbw_channels;
786 int channel_mode = s->channel_mode;
787 int i, bnd, seg, ch;
788 int different_transforms;
789 int downmix_output;
790 int cpl_in_use;
791 GetBitContext *gbc = &s->gbc;
792 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
794 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
796 /* block switch flags */
797 different_transforms = 0;
798 if (s->block_switch_syntax) {
799 for (ch = 1; ch <= fbw_channels; ch++) {
800 s->block_switch[ch] = get_bits1(gbc);
801 if(ch > 1 && s->block_switch[ch] != s->block_switch[1])
802 different_transforms = 1;
806 /* dithering flags */
807 if (s->dither_flag_syntax) {
808 for (ch = 1; ch <= fbw_channels; ch++) {
809 s->dither_flag[ch] = get_bits1(gbc);
813 /* dynamic range */
814 i = !(s->channel_mode);
815 do {
816 if(get_bits1(gbc)) {
817 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
818 s->avctx->drc_scale)+1.0;
819 } else if(blk == 0) {
820 s->dynamic_range[i] = 1.0f;
822 } while(i--);
824 /* spectral extension strategy */
825 if (s->eac3 && (!blk || get_bits1(gbc))) {
826 if (get_bits1(gbc)) {
827 av_log_missing_feature(s->avctx, "Spectral extension", 1);
828 return -1;
830 /* TODO: parse spectral extension strategy info */
833 /* TODO: spectral extension coordinates */
835 /* coupling strategy */
836 if (s->eac3 ? s->cpl_strategy_exists[blk] : get_bits1(gbc)) {
837 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
838 if (!s->eac3)
839 s->cpl_in_use[blk] = get_bits1(gbc);
840 if (s->cpl_in_use[blk]) {
841 /* coupling in use */
842 int cpl_start_subband, cpl_end_subband;
844 if (channel_mode < AC3_CHMODE_STEREO) {
845 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
846 return -1;
849 /* check for enhanced coupling */
850 if (s->eac3 && get_bits1(gbc)) {
851 /* TODO: parse enhanced coupling strategy info */
852 av_log_missing_feature(s->avctx, "Enhanced coupling", 1);
853 return -1;
856 /* determine which channels are coupled */
857 if (s->eac3 && s->channel_mode == AC3_CHMODE_STEREO) {
858 s->channel_in_cpl[1] = 1;
859 s->channel_in_cpl[2] = 1;
860 } else {
861 for (ch = 1; ch <= fbw_channels; ch++)
862 s->channel_in_cpl[ch] = get_bits1(gbc);
865 /* phase flags in use */
866 if (channel_mode == AC3_CHMODE_STEREO)
867 s->phase_flags_in_use = get_bits1(gbc);
869 /* coupling frequency range */
870 /* TODO: modify coupling end freq if spectral extension is used */
871 cpl_start_subband = get_bits(gbc, 4);
872 cpl_end_subband = get_bits(gbc, 4) + 3;
873 if (cpl_start_subband >= cpl_end_subband) {
874 av_log(s->avctx, AV_LOG_ERROR, "invalid coupling range (%d >= %d)\n",
875 cpl_start_subband, cpl_end_subband);
876 return -1;
878 s->start_freq[CPL_CH] = cpl_start_subband * 12 + 37;
879 s->end_freq[CPL_CH] = cpl_end_subband * 12 + 37;
881 decode_band_structure(gbc, blk, s->eac3, 0, cpl_start_subband,
882 cpl_end_subband,
883 ff_eac3_default_cpl_band_struct,
884 &s->num_cpl_bands,
885 s->cpl_band_sizes);
886 } else {
887 /* coupling not in use */
888 for (ch = 1; ch <= fbw_channels; ch++) {
889 s->channel_in_cpl[ch] = 0;
890 s->first_cpl_coords[ch] = 1;
892 s->first_cpl_leak = s->eac3;
893 s->phase_flags_in_use = 0;
895 } else if (!s->eac3) {
896 if(!blk) {
897 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must be present in block 0\n");
898 return -1;
899 } else {
900 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
903 cpl_in_use = s->cpl_in_use[blk];
905 /* coupling coordinates */
906 if (cpl_in_use) {
907 int cpl_coords_exist = 0;
909 for (ch = 1; ch <= fbw_channels; ch++) {
910 if (s->channel_in_cpl[ch]) {
911 if ((s->eac3 && s->first_cpl_coords[ch]) || get_bits1(gbc)) {
912 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
913 s->first_cpl_coords[ch] = 0;
914 cpl_coords_exist = 1;
915 master_cpl_coord = 3 * get_bits(gbc, 2);
916 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
917 cpl_coord_exp = get_bits(gbc, 4);
918 cpl_coord_mant = get_bits(gbc, 4);
919 if (cpl_coord_exp == 15)
920 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
921 else
922 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
923 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
925 } else if (!blk) {
926 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must be present in block 0\n");
927 return -1;
929 } else {
930 /* channel not in coupling */
931 s->first_cpl_coords[ch] = 1;
934 /* phase flags */
935 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
936 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
937 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
942 /* stereo rematrixing strategy and band structure */
943 if (channel_mode == AC3_CHMODE_STEREO) {
944 if ((s->eac3 && !blk) || get_bits1(gbc)) {
945 s->num_rematrixing_bands = 4;
946 if(cpl_in_use && s->start_freq[CPL_CH] <= 61)
947 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
948 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
949 s->rematrixing_flags[bnd] = get_bits1(gbc);
950 } else if (!blk) {
951 av_log(s->avctx, AV_LOG_ERROR, "new rematrixing strategy must be present in block 0\n");
952 return -1;
956 /* exponent strategies for each channel */
957 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
958 if (!s->eac3)
959 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
960 if(s->exp_strategy[blk][ch] != EXP_REUSE)
961 bit_alloc_stages[ch] = 3;
964 /* channel bandwidth */
965 for (ch = 1; ch <= fbw_channels; ch++) {
966 s->start_freq[ch] = 0;
967 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
968 int group_size;
969 int prev = s->end_freq[ch];
970 if (s->channel_in_cpl[ch])
971 s->end_freq[ch] = s->start_freq[CPL_CH];
972 else {
973 int bandwidth_code = get_bits(gbc, 6);
974 if (bandwidth_code > 60) {
975 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60\n", bandwidth_code);
976 return -1;
978 s->end_freq[ch] = bandwidth_code * 3 + 73;
980 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
981 s->num_exp_groups[ch] = (s->end_freq[ch]+group_size-4) / group_size;
982 if(blk > 0 && s->end_freq[ch] != prev)
983 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
986 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
987 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
988 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
991 /* decode exponents for each channel */
992 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
993 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
994 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
995 if (decode_exponents(gbc, s->exp_strategy[blk][ch],
996 s->num_exp_groups[ch], s->dexps[ch][0],
997 &s->dexps[ch][s->start_freq[ch]+!!ch])) {
998 av_log(s->avctx, AV_LOG_ERROR, "exponent out-of-range\n");
999 return -1;
1001 if(ch != CPL_CH && ch != s->lfe_ch)
1002 skip_bits(gbc, 2); /* skip gainrng */
1006 /* bit allocation information */
1007 if (s->bit_allocation_syntax) {
1008 if (get_bits1(gbc)) {
1009 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1010 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1011 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
1012 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
1013 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
1014 for(ch=!cpl_in_use; ch<=s->channels; ch++)
1015 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1016 } else if (!blk) {
1017 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must be present in block 0\n");
1018 return -1;
1022 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
1023 if(!s->eac3 || !blk){
1024 if(s->snr_offset_strategy && get_bits1(gbc)) {
1025 int snr = 0;
1026 int csnr;
1027 csnr = (get_bits(gbc, 6) - 15) << 4;
1028 for (i = ch = !cpl_in_use; ch <= s->channels; ch++) {
1029 /* snr offset */
1030 if (ch == i || s->snr_offset_strategy == 2)
1031 snr = (csnr + get_bits(gbc, 4)) << 2;
1032 /* run at least last bit allocation stage if snr offset changes */
1033 if(blk && s->snr_offset[ch] != snr) {
1034 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 1);
1036 s->snr_offset[ch] = snr;
1038 /* fast gain (normal AC-3 only) */
1039 if (!s->eac3) {
1040 int prev = s->fast_gain[ch];
1041 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1042 /* run last 2 bit allocation stages if fast gain changes */
1043 if(blk && prev != s->fast_gain[ch])
1044 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1047 } else if (!s->eac3 && !blk) {
1048 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
1049 return -1;
1053 /* fast gain (E-AC-3 only) */
1054 if (s->fast_gain_syntax && get_bits1(gbc)) {
1055 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1056 int prev = s->fast_gain[ch];
1057 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1058 /* run last 2 bit allocation stages if fast gain changes */
1059 if(blk && prev != s->fast_gain[ch])
1060 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1062 } else if (s->eac3 && !blk) {
1063 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1064 s->fast_gain[ch] = ff_ac3_fast_gain_tab[4];
1067 /* E-AC-3 to AC-3 converter SNR offset */
1068 if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT && get_bits1(gbc)) {
1069 skip_bits(gbc, 10); // skip converter snr offset
1072 /* coupling leak information */
1073 if (cpl_in_use) {
1074 if (s->first_cpl_leak || get_bits1(gbc)) {
1075 int fl = get_bits(gbc, 3);
1076 int sl = get_bits(gbc, 3);
1077 /* run last 2 bit allocation stages for coupling channel if
1078 coupling leak changes */
1079 if(blk && (fl != s->bit_alloc_params.cpl_fast_leak ||
1080 sl != s->bit_alloc_params.cpl_slow_leak)) {
1081 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
1083 s->bit_alloc_params.cpl_fast_leak = fl;
1084 s->bit_alloc_params.cpl_slow_leak = sl;
1085 } else if (!s->eac3 && !blk) {
1086 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must be present in block 0\n");
1087 return -1;
1089 s->first_cpl_leak = 0;
1092 /* delta bit allocation information */
1093 if (s->dba_syntax && get_bits1(gbc)) {
1094 /* delta bit allocation exists (strategy) */
1095 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1096 s->dba_mode[ch] = get_bits(gbc, 2);
1097 if (s->dba_mode[ch] == DBA_RESERVED) {
1098 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1099 return -1;
1101 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1103 /* channel delta offset, len and bit allocation */
1104 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1105 if (s->dba_mode[ch] == DBA_NEW) {
1106 s->dba_nsegs[ch] = get_bits(gbc, 3);
1107 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
1108 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1109 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1110 s->dba_values[ch][seg] = get_bits(gbc, 3);
1112 /* run last 2 bit allocation stages if new dba values */
1113 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1116 } else if(blk == 0) {
1117 for(ch=0; ch<=s->channels; ch++) {
1118 s->dba_mode[ch] = DBA_NONE;
1122 /* Bit allocation */
1123 for(ch=!cpl_in_use; ch<=s->channels; ch++) {
1124 if(bit_alloc_stages[ch] > 2) {
1125 /* Exponent mapping into PSD and PSD integration */
1126 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1127 s->start_freq[ch], s->end_freq[ch],
1128 s->psd[ch], s->band_psd[ch]);
1130 if(bit_alloc_stages[ch] > 1) {
1131 /* Compute excitation function, Compute masking curve, and
1132 Apply delta bit allocation */
1133 if (ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1134 s->start_freq[ch], s->end_freq[ch],
1135 s->fast_gain[ch], (ch == s->lfe_ch),
1136 s->dba_mode[ch], s->dba_nsegs[ch],
1137 s->dba_offsets[ch], s->dba_lengths[ch],
1138 s->dba_values[ch], s->mask[ch])) {
1139 av_log(s->avctx, AV_LOG_ERROR, "error in bit allocation\n");
1140 return -1;
1143 if(bit_alloc_stages[ch] > 0) {
1144 /* Compute bit allocation */
1145 const uint8_t *bap_tab = s->channel_uses_aht[ch] ?
1146 ff_eac3_hebap_tab : ff_ac3_bap_tab;
1147 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1148 s->start_freq[ch], s->end_freq[ch],
1149 s->snr_offset[ch],
1150 s->bit_alloc_params.floor,
1151 bap_tab, s->bap[ch]);
1155 /* unused dummy data */
1156 if (s->skip_syntax && get_bits1(gbc)) {
1157 int skipl = get_bits(gbc, 9);
1158 while(skipl--)
1159 skip_bits(gbc, 8);
1162 /* unpack the transform coefficients
1163 this also uncouples channels if coupling is in use. */
1164 decode_transform_coeffs(s, blk);
1166 /* TODO: generate enhanced coupling coordinates and uncouple */
1168 /* TODO: apply spectral extension */
1170 /* recover coefficients if rematrixing is in use */
1171 if(s->channel_mode == AC3_CHMODE_STEREO)
1172 do_rematrixing(s);
1174 /* apply scaling to coefficients (headroom, dynrng) */
1175 for(ch=1; ch<=s->channels; ch++) {
1176 float gain = s->mul_bias / 4194304.0f;
1177 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1178 gain *= s->dynamic_range[ch-1];
1179 } else {
1180 gain *= s->dynamic_range[0];
1182 s->dsp.int32_to_float_fmul_scalar(s->transform_coeffs[ch], s->fixed_coeffs[ch], gain, 256);
1185 /* downmix and MDCT. order depends on whether block switching is used for
1186 any channel in this block. this is because coefficients for the long
1187 and short transforms cannot be mixed. */
1188 downmix_output = s->channels != s->out_channels &&
1189 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1190 s->fbw_channels == s->out_channels);
1191 if(different_transforms) {
1192 /* the delay samples have already been downmixed, so we upmix the delay
1193 samples in order to reconstruct all channels before downmixing. */
1194 if(s->downmixed) {
1195 s->downmixed = 0;
1196 ac3_upmix_delay(s);
1199 do_imdct(s, s->channels);
1201 if(downmix_output) {
1202 s->dsp.ac3_downmix(s->output, s->downmix_coeffs, s->out_channels, s->fbw_channels, 256);
1204 } else {
1205 if(downmix_output) {
1206 s->dsp.ac3_downmix(s->transform_coeffs+1, s->downmix_coeffs, s->out_channels, s->fbw_channels, 256);
1209 if(downmix_output && !s->downmixed) {
1210 s->downmixed = 1;
1211 s->dsp.ac3_downmix(s->delay, s->downmix_coeffs, s->out_channels, s->fbw_channels, 128);
1214 do_imdct(s, s->out_channels);
1217 return 0;
1221 * Decode a single AC-3 frame.
1223 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
1224 AVPacket *avpkt)
1226 const uint8_t *buf = avpkt->data;
1227 int buf_size = avpkt->size;
1228 AC3DecodeContext *s = avctx->priv_data;
1229 int16_t *out_samples = (int16_t *)data;
1230 int blk, ch, err;
1231 const uint8_t *channel_map;
1232 const float *output[AC3_MAX_CHANNELS];
1234 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1235 if (s->input_buffer) {
1236 /* copy input buffer to decoder context to avoid reading past the end
1237 of the buffer, which can be caused by a damaged input stream. */
1238 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE));
1239 init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
1240 } else {
1241 init_get_bits(&s->gbc, buf, buf_size * 8);
1244 /* parse the syncinfo */
1245 *data_size = 0;
1246 err = parse_frame_header(s);
1248 /* check that reported frame size fits in input buffer */
1249 if(s->frame_size > buf_size) {
1250 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1251 err = AAC_AC3_PARSE_ERROR_FRAME_SIZE;
1254 /* check for crc mismatch */
1255 if(err != AAC_AC3_PARSE_ERROR_FRAME_SIZE && avctx->error_recognition >= FF_ER_CAREFUL) {
1256 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1257 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1258 err = AAC_AC3_PARSE_ERROR_CRC;
1262 if(err && err != AAC_AC3_PARSE_ERROR_CRC) {
1263 switch(err) {
1264 case AAC_AC3_PARSE_ERROR_SYNC:
1265 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1266 return -1;
1267 case AAC_AC3_PARSE_ERROR_BSID:
1268 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1269 break;
1270 case AAC_AC3_PARSE_ERROR_SAMPLE_RATE:
1271 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1272 break;
1273 case AAC_AC3_PARSE_ERROR_FRAME_SIZE:
1274 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1275 break;
1276 case AAC_AC3_PARSE_ERROR_FRAME_TYPE:
1277 /* skip frame if CRC is ok. otherwise use error concealment. */
1278 /* TODO: add support for substreams and dependent frames */
1279 if(s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1280 av_log(avctx, AV_LOG_ERROR, "unsupported frame type : skipping frame\n");
1281 return s->frame_size;
1282 } else {
1283 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1285 break;
1286 default:
1287 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1288 break;
1292 /* if frame is ok, set audio parameters */
1293 if (!err) {
1294 avctx->sample_rate = s->sample_rate;
1295 avctx->bit_rate = s->bit_rate;
1297 /* channel config */
1298 s->out_channels = s->channels;
1299 s->output_mode = s->channel_mode;
1300 if(s->lfe_on)
1301 s->output_mode |= AC3_OUTPUT_LFEON;
1302 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1303 avctx->request_channels < s->channels) {
1304 s->out_channels = avctx->request_channels;
1305 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1306 s->channel_layout = ff_ac3_channel_layout_tab[s->output_mode];
1308 avctx->channels = s->out_channels;
1309 avctx->channel_layout = s->channel_layout;
1311 /* set downmixing coefficients if needed */
1312 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1313 s->fbw_channels == s->out_channels)) {
1314 set_downmix_coeffs(s);
1316 } else if (!s->out_channels) {
1317 s->out_channels = avctx->channels;
1318 if(s->out_channels < s->channels)
1319 s->output_mode = s->out_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1322 /* decode the audio blocks */
1323 channel_map = ff_ac3_dec_channel_map[s->output_mode & ~AC3_OUTPUT_LFEON][s->lfe_on];
1324 for (ch = 0; ch < s->out_channels; ch++)
1325 output[ch] = s->output[channel_map[ch]];
1326 for (blk = 0; blk < s->num_blocks; blk++) {
1327 if (!err && decode_audio_block(s, blk)) {
1328 av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n");
1329 err = 1;
1331 s->dsp.float_to_int16_interleave(out_samples, output, 256, s->out_channels);
1332 out_samples += 256 * s->out_channels;
1334 *data_size = s->num_blocks * 256 * avctx->channels * sizeof (int16_t);
1335 return s->frame_size;
1339 * Uninitialize the AC-3 decoder.
1341 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1343 AC3DecodeContext *s = avctx->priv_data;
1344 ff_mdct_end(&s->imdct_512);
1345 ff_mdct_end(&s->imdct_256);
1347 av_freep(&s->input_buffer);
1349 return 0;
1352 AVCodec ac3_decoder = {
1353 .name = "ac3",
1354 .type = CODEC_TYPE_AUDIO,
1355 .id = CODEC_ID_AC3,
1356 .priv_data_size = sizeof (AC3DecodeContext),
1357 .init = ac3_decode_init,
1358 .close = ac3_decode_end,
1359 .decode = ac3_decode_frame,
1360 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52A (AC-3)"),
1363 #if CONFIG_EAC3_DECODER
1364 AVCodec eac3_decoder = {
1365 .name = "eac3",
1366 .type = CODEC_TYPE_AUDIO,
1367 .id = CODEC_ID_EAC3,
1368 .priv_data_size = sizeof (AC3DecodeContext),
1369 .init = ac3_decode_init,
1370 .close = ac3_decode_end,
1371 .decode = ac3_decode_frame,
1372 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52B (AC-3, E-AC-3)"),
1374 #endif