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
32 #include "libavutil/channel_layout.h"
33 #include "libavutil/crc.h"
34 #include "libavutil/downmix_info.h"
35 #include "libavutil/opt.h"
38 #include "aac_ac3_parser.h"
39 #include "ac3_parser_internal.h"
41 #include "ac3dec_data.h"
45 * table for ungrouping 3 values in 7 bits.
46 * used for exponents and bap=2 mantissas
48 static uint8_t ungroup_3_in_7_bits_tab[128][3];
50 /** tables for ungrouping mantissas */
51 static int b1_mantissas[32][3];
52 static int b2_mantissas[128][3];
53 static int b3_mantissas[8];
54 static int b4_mantissas[128][2];
55 static int b5_mantissas[16];
58 * Quantization table: levels for symmetric. bits for asymmetric.
59 * reference: Table 7.18 Mapping of bap to Quantizer
61 static const uint8_t quantization_tab[16] = {
63 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];
69 float ff_ac3_heavy_dynamic_range_tab[256];
72 /** Adjustments in dB gain */
73 static const float gain_levels[9] = {
77 LEVEL_MINUS_1POINT5DB,
79 LEVEL_MINUS_4POINT5DB,
85 /** Adjustments in dB gain (LFE, +10 to -21 dB) */
86 static const float gain_levels_lfe[32] = {
87 3.162275, 2.818382, 2.511886, 2.238719, 1.995261, 1.778278, 1.584893,
88 1.412536, 1.258924, 1.122018, 1.000000, 0.891251, 0.794328, 0.707946,
89 0.630957, 0.562341, 0.501187, 0.446683, 0.398107, 0.354813, 0.316227,
90 0.281838, 0.251188, 0.223872, 0.199526, 0.177828, 0.158489, 0.141253,
91 0.125892, 0.112201, 0.100000, 0.089125
95 * Table for default stereo downmixing coefficients
96 * reference: Section 7.8.2 Downmixing Into Two Channels
98 static const uint8_t ac3_default_coeffs[8][5][2] = {
99 { { 2, 7 }, { 7, 2 }, },
101 { { 2, 7 }, { 7, 2 }, },
102 { { 2, 7 }, { 5, 5 }, { 7, 2 }, },
103 { { 2, 7 }, { 7, 2 }, { 6, 6 }, },
104 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 8, 8 }, },
105 { { 2, 7 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
106 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
109 static const uint64_t custom_channel_map_locations[16][2] = {
110 { 1, AV_CH_FRONT_LEFT },
111 { 1, AV_CH_FRONT_CENTER },
112 { 1, AV_CH_FRONT_RIGHT },
113 { 1, AV_CH_SIDE_LEFT },
114 { 1, AV_CH_SIDE_RIGHT },
115 { 0, AV_CH_FRONT_LEFT_OF_CENTER | AV_CH_FRONT_RIGHT_OF_CENTER },
116 { 0, AV_CH_BACK_LEFT | AV_CH_BACK_RIGHT },
117 { 0, AV_CH_BACK_CENTER },
118 { 0, AV_CH_TOP_CENTER },
119 { 0, AV_CH_SURROUND_DIRECT_LEFT | AV_CH_SURROUND_DIRECT_RIGHT },
120 { 0, AV_CH_WIDE_LEFT | AV_CH_WIDE_RIGHT },
121 { 0, AV_CH_TOP_FRONT_LEFT | AV_CH_TOP_FRONT_RIGHT},
122 { 0, AV_CH_TOP_FRONT_CENTER },
123 { 0, AV_CH_TOP_BACK_LEFT | AV_CH_TOP_BACK_RIGHT },
124 { 0, AV_CH_LOW_FREQUENCY_2 },
125 { 1, AV_CH_LOW_FREQUENCY },
129 * Symmetrical Dequantization
130 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
131 * Tables 7.19 to 7.23
134 symmetric_dequant(int code, int levels)
136 return ((code - (levels >> 1)) * (1 << 24)) / levels;
140 * Initialize tables at runtime.
142 static av_cold void ac3_tables_init(void)
146 /* generate table for ungrouping 3 values in 7 bits
147 reference: Section 7.1.3 Exponent Decoding */
148 for (i = 0; i < 128; i++) {
149 ungroup_3_in_7_bits_tab[i][0] = i / 25;
150 ungroup_3_in_7_bits_tab[i][1] = (i % 25) / 5;
151 ungroup_3_in_7_bits_tab[i][2] = (i % 25) % 5;
154 /* generate grouped mantissa tables
155 reference: Section 7.3.5 Ungrouping of Mantissas */
156 for (i = 0; i < 32; i++) {
157 /* bap=1 mantissas */
158 b1_mantissas[i][0] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][0], 3);
159 b1_mantissas[i][1] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][1], 3);
160 b1_mantissas[i][2] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][2], 3);
162 for (i = 0; i < 128; i++) {
163 /* bap=2 mantissas */
164 b2_mantissas[i][0] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][0], 5);
165 b2_mantissas[i][1] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][1], 5);
166 b2_mantissas[i][2] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][2], 5);
168 /* bap=4 mantissas */
169 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
170 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
172 /* generate ungrouped mantissa tables
173 reference: Tables 7.21 and 7.23 */
174 for (i = 0; i < 7; i++) {
175 /* bap=3 mantissas */
176 b3_mantissas[i] = symmetric_dequant(i, 7);
178 for (i = 0; i < 15; i++) {
179 /* bap=5 mantissas */
180 b5_mantissas[i] = symmetric_dequant(i, 15);
184 /* generate dynamic range table
185 reference: Section 7.7.1 Dynamic Range Control */
186 for (i = 0; i < 256; i++) {
187 int v = (i >> 5) - ((i >> 7) << 3) - 5;
188 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
191 /* generate compr dynamic range table
192 reference: Section 7.7.2 Heavy Compression */
193 for (i = 0; i < 256; i++) {
194 int v = (i >> 4) - ((i >> 7) << 4) - 4;
195 ff_ac3_heavy_dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0xF) | 0x10);
201 * AVCodec initialization
203 static av_cold int ac3_decode_init(AVCodecContext *avctx)
205 AC3DecodeContext *s = avctx->priv_data;
211 ff_mdct_init(&s->imdct_256, 8, 1, 1.0);
212 ff_mdct_init(&s->imdct_512, 9, 1, 1.0);
213 AC3_RENAME(ff_kbd_window_init)(s->window, 5.0, 256);
214 ff_bswapdsp_init(&s->bdsp);
217 s->fdsp = avpriv_alloc_fixed_dsp(avctx->flags & AV_CODEC_FLAG_BITEXACT);
219 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
220 ff_fmt_convert_init(&s->fmt_conv, avctx);
223 ff_ac3dsp_init(&s->ac3dsp, avctx->flags & AV_CODEC_FLAG_BITEXACT);
224 av_lfg_init(&s->dith_state, 0);
227 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
229 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
231 /* allow downmixing to stereo or mono */
232 if (avctx->channels > 1 &&
233 avctx->request_channel_layout == AV_CH_LAYOUT_MONO)
235 else if (avctx->channels > 2 &&
236 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO)
240 for (i = 0; i < AC3_MAX_CHANNELS; i++) {
241 s->xcfptr[i] = s->transform_coeffs[i];
242 s->dlyptr[i] = s->delay[i];
249 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
250 * GetBitContext within AC3DecodeContext must point to
251 * the start of the synchronized AC-3 bitstream.
253 static int ac3_parse_header(AC3DecodeContext *s)
255 GetBitContext *gbc = &s->gbc;
258 /* read the rest of the bsi. read twice for dual mono mode. */
259 i = !s->channel_mode;
261 s->dialog_normalization[(!s->channel_mode)-i] = -get_bits(gbc, 5);
262 if (s->dialog_normalization[(!s->channel_mode)-i] == 0) {
263 s->dialog_normalization[(!s->channel_mode)-i] = -31;
265 if (s->target_level != 0) {
266 s->level_gain[(!s->channel_mode)-i] = powf(2.0f,
267 (float)(s->target_level -
268 s->dialog_normalization[(!s->channel_mode)-i])/6.0f);
270 if (s->compression_exists[(!s->channel_mode)-i] = get_bits1(gbc)) {
271 s->heavy_dynamic_range[(!s->channel_mode)-i] =
272 AC3_HEAVY_RANGE(get_bits(gbc, 8));
275 skip_bits(gbc, 8); //skip language code
277 skip_bits(gbc, 7); //skip audio production information
280 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
282 /* skip the timecodes or parse the Alternate Bit Stream Syntax */
283 if (s->bitstream_id != 6) {
285 skip_bits(gbc, 14); //skip timecode1
287 skip_bits(gbc, 14); //skip timecode2
289 if (get_bits1(gbc)) {
290 s->preferred_downmix = get_bits(gbc, 2);
291 s->center_mix_level_ltrt = get_bits(gbc, 3);
292 s->surround_mix_level_ltrt = av_clip(get_bits(gbc, 3), 3, 7);
293 s->center_mix_level = get_bits(gbc, 3);
294 s->surround_mix_level = av_clip(get_bits(gbc, 3), 3, 7);
296 if (get_bits1(gbc)) {
297 s->dolby_surround_ex_mode = get_bits(gbc, 2);
298 s->dolby_headphone_mode = get_bits(gbc, 2);
299 skip_bits(gbc, 10); // skip adconvtyp (1), xbsi2 (8), encinfo (1)
303 /* skip additional bitstream info */
304 if (get_bits1(gbc)) {
305 i = get_bits(gbc, 6);
315 * Common function to parse AC-3 or E-AC-3 frame header
317 static int parse_frame_header(AC3DecodeContext *s)
322 err = ff_ac3_parse_header(&s->gbc, &hdr);
326 /* get decoding parameters from header info */
327 s->bit_alloc_params.sr_code = hdr.sr_code;
328 s->bitstream_id = hdr.bitstream_id;
329 s->bitstream_mode = hdr.bitstream_mode;
330 s->channel_mode = hdr.channel_mode;
331 s->lfe_on = hdr.lfe_on;
332 s->bit_alloc_params.sr_shift = hdr.sr_shift;
333 s->sample_rate = hdr.sample_rate;
334 s->bit_rate = hdr.bit_rate;
335 s->channels = hdr.channels;
336 s->fbw_channels = s->channels - s->lfe_on;
337 s->lfe_ch = s->fbw_channels + 1;
338 s->frame_size = hdr.frame_size;
339 s->superframe_size += hdr.frame_size;
340 s->preferred_downmix = AC3_DMIXMOD_NOTINDICATED;
341 s->center_mix_level = hdr.center_mix_level;
342 s->center_mix_level_ltrt = 4; // -3.0dB
343 s->surround_mix_level = hdr.surround_mix_level;
344 s->surround_mix_level_ltrt = 4; // -3.0dB
345 s->lfe_mix_level_exists = 0;
346 s->num_blocks = hdr.num_blocks;
347 s->frame_type = hdr.frame_type;
348 s->substreamid = hdr.substreamid;
349 s->dolby_surround_mode = hdr.dolby_surround_mode;
350 s->dolby_surround_ex_mode = AC3_DSUREXMOD_NOTINDICATED;
351 s->dolby_headphone_mode = AC3_DHEADPHONMOD_NOTINDICATED;
354 s->start_freq[s->lfe_ch] = 0;
355 s->end_freq[s->lfe_ch] = 7;
356 s->num_exp_groups[s->lfe_ch] = 2;
357 s->channel_in_cpl[s->lfe_ch] = 0;
360 if (s->bitstream_id <= 10) {
362 s->snr_offset_strategy = 2;
363 s->block_switch_syntax = 1;
364 s->dither_flag_syntax = 1;
365 s->bit_allocation_syntax = 1;
366 s->fast_gain_syntax = 0;
367 s->first_cpl_leak = 0;
370 memset(s->channel_uses_aht, 0, sizeof(s->channel_uses_aht));
371 return ac3_parse_header(s);
372 } else if (CONFIG_EAC3_DECODER) {
374 return ff_eac3_parse_header(s);
376 av_log(s->avctx, AV_LOG_ERROR, "E-AC-3 support not compiled in\n");
377 return AVERROR(ENOSYS);
382 * Set stereo downmixing coefficients based on frame header info.
383 * reference: Section 7.8.2 Downmixing Into Two Channels
385 static int set_downmix_coeffs(AC3DecodeContext *s)
388 float cmix = gain_levels[s-> center_mix_level];
389 float smix = gain_levels[s->surround_mix_level];
391 float downmix_coeffs[2][AC3_MAX_CHANNELS];
393 if (!s->downmix_coeffs[0]) {
394 s->downmix_coeffs[0] = av_malloc_array(2 * AC3_MAX_CHANNELS,
395 sizeof(**s->downmix_coeffs));
396 if (!s->downmix_coeffs[0])
397 return AVERROR(ENOMEM);
398 s->downmix_coeffs[1] = s->downmix_coeffs[0] + AC3_MAX_CHANNELS;
401 for (i = 0; i < s->fbw_channels; i++) {
402 downmix_coeffs[0][i] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
403 downmix_coeffs[1][i] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
405 if (s->channel_mode > 1 && s->channel_mode & 1) {
406 downmix_coeffs[0][1] = downmix_coeffs[1][1] = cmix;
408 if (s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
409 int nf = s->channel_mode - 2;
410 downmix_coeffs[0][nf] = downmix_coeffs[1][nf] = smix * LEVEL_MINUS_3DB;
412 if (s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
413 int nf = s->channel_mode - 4;
414 downmix_coeffs[0][nf] = downmix_coeffs[1][nf+1] = smix;
419 for (i = 0; i < s->fbw_channels; i++) {
420 norm0 += downmix_coeffs[0][i];
421 norm1 += downmix_coeffs[1][i];
423 norm0 = 1.0f / norm0;
424 norm1 = 1.0f / norm1;
425 for (i = 0; i < s->fbw_channels; i++) {
426 downmix_coeffs[0][i] *= norm0;
427 downmix_coeffs[1][i] *= norm1;
430 if (s->output_mode == AC3_CHMODE_MONO) {
431 for (i = 0; i < s->fbw_channels; i++)
432 downmix_coeffs[0][i] = (downmix_coeffs[0][i] +
433 downmix_coeffs[1][i]) * LEVEL_MINUS_3DB;
435 for (i = 0; i < s->fbw_channels; i++) {
436 s->downmix_coeffs[0][i] = FIXR12(downmix_coeffs[0][i]);
437 s->downmix_coeffs[1][i] = FIXR12(downmix_coeffs[1][i]);
444 * Decode the grouped exponents according to exponent strategy.
445 * reference: Section 7.1.3 Exponent Decoding
447 static int decode_exponents(AC3DecodeContext *s,
448 GetBitContext *gbc, int exp_strategy, int ngrps,
449 uint8_t absexp, int8_t *dexps)
451 int i, j, grp, group_size;
456 group_size = exp_strategy + (exp_strategy == EXP_D45);
457 for (grp = 0, i = 0; grp < ngrps; grp++) {
458 expacc = get_bits(gbc, 7);
460 av_log(s->avctx, AV_LOG_ERROR, "expacc %d is out-of-range\n", expacc);
461 return AVERROR_INVALIDDATA;
463 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][0];
464 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][1];
465 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][2];
468 /* convert to absolute exps and expand groups */
470 for (i = 0, j = 0; i < ngrps * 3; i++) {
471 prevexp += dexp[i] - 2;
473 av_log(s->avctx, AV_LOG_ERROR, "exponent %d is out-of-range\n", prevexp);
476 switch (group_size) {
477 case 4: dexps[j++] = prevexp;
478 dexps[j++] = prevexp;
479 case 2: dexps[j++] = prevexp;
480 case 1: dexps[j++] = prevexp;
487 * Generate transform coefficients for each coupled channel in the coupling
488 * range using the coupling coefficients and coupling coordinates.
489 * reference: Section 7.4.3 Coupling Coordinate Format
491 static void calc_transform_coeffs_cpl(AC3DecodeContext *s)
495 bin = s->start_freq[CPL_CH];
496 for (band = 0; band < s->num_cpl_bands; band++) {
497 int band_start = bin;
498 int band_end = bin + s->cpl_band_sizes[band];
499 for (ch = 1; ch <= s->fbw_channels; ch++) {
500 if (s->channel_in_cpl[ch]) {
501 int cpl_coord = s->cpl_coords[ch][band] << 5;
502 for (bin = band_start; bin < band_end; bin++) {
503 s->fixed_coeffs[ch][bin] =
504 MULH(s->fixed_coeffs[CPL_CH][bin] * (1 << 4), cpl_coord);
506 if (ch == 2 && s->phase_flags[band]) {
507 for (bin = band_start; bin < band_end; bin++)
508 s->fixed_coeffs[2][bin] = -s->fixed_coeffs[2][bin];
517 * Grouped mantissas for 3-level 5-level and 11-level quantization
519 typedef struct mant_groups {
529 * Decode the transform coefficients for a particular channel
530 * reference: Section 7.3 Quantization and Decoding of Mantissas
532 static void ac3_decode_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
534 int start_freq = s->start_freq[ch_index];
535 int end_freq = s->end_freq[ch_index];
536 uint8_t *baps = s->bap[ch_index];
537 int8_t *exps = s->dexps[ch_index];
538 int32_t *coeffs = s->fixed_coeffs[ch_index];
539 int dither = (ch_index == CPL_CH) || s->dither_flag[ch_index];
540 GetBitContext *gbc = &s->gbc;
543 for (freq = start_freq; freq < end_freq; freq++) {
544 int bap = baps[freq];
548 /* random noise with approximate range of -0.707 to 0.707 */
550 mantissa = (((av_lfg_get(&s->dith_state)>>8)*181)>>8) - 5931008;
557 mantissa = m->b1_mant[m->b1];
559 int bits = get_bits(gbc, 5);
560 mantissa = b1_mantissas[bits][0];
561 m->b1_mant[1] = b1_mantissas[bits][1];
562 m->b1_mant[0] = b1_mantissas[bits][2];
569 mantissa = m->b2_mant[m->b2];
571 int bits = get_bits(gbc, 7);
572 mantissa = b2_mantissas[bits][0];
573 m->b2_mant[1] = b2_mantissas[bits][1];
574 m->b2_mant[0] = b2_mantissas[bits][2];
579 mantissa = b3_mantissas[get_bits(gbc, 3)];
584 mantissa = m->b4_mant;
586 int bits = get_bits(gbc, 7);
587 mantissa = b4_mantissas[bits][0];
588 m->b4_mant = b4_mantissas[bits][1];
593 mantissa = b5_mantissas[get_bits(gbc, 4)];
595 default: /* 6 to 15 */
596 /* Shift mantissa and sign-extend it. */
598 av_log(s->avctx, AV_LOG_ERROR, "bap %d is invalid in plain AC-3\n", bap);
601 mantissa = (unsigned)get_sbits(gbc, quantization_tab[bap]) << (24 - quantization_tab[bap]);
604 coeffs[freq] = mantissa >> exps[freq];
609 * Remove random dithering from coupling range coefficients with zero-bit
610 * mantissas for coupled channels which do not use dithering.
611 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
613 static void remove_dithering(AC3DecodeContext *s) {
616 for (ch = 1; ch <= s->fbw_channels; ch++) {
617 if (!s->dither_flag[ch] && s->channel_in_cpl[ch]) {
618 for (i = s->start_freq[CPL_CH]; i < s->end_freq[CPL_CH]; i++) {
619 if (!s->bap[CPL_CH][i])
620 s->fixed_coeffs[ch][i] = 0;
626 static inline void decode_transform_coeffs_ch(AC3DecodeContext *s, int blk,
627 int ch, mant_groups *m)
629 if (!s->channel_uses_aht[ch]) {
630 ac3_decode_transform_coeffs_ch(s, ch, m);
632 /* if AHT is used, mantissas for all blocks are encoded in the first
633 block of the frame. */
635 if (CONFIG_EAC3_DECODER && !blk)
636 ff_eac3_decode_transform_coeffs_aht_ch(s, ch);
637 for (bin = s->start_freq[ch]; bin < s->end_freq[ch]; bin++) {
638 s->fixed_coeffs[ch][bin] = s->pre_mantissa[ch][bin][blk] >> s->dexps[ch][bin];
644 * Decode the transform coefficients.
646 static inline void decode_transform_coeffs(AC3DecodeContext *s, int blk)
652 m.b1 = m.b2 = m.b4 = 0;
654 for (ch = 1; ch <= s->channels; ch++) {
655 /* transform coefficients for full-bandwidth channel */
656 decode_transform_coeffs_ch(s, blk, ch, &m);
657 /* transform coefficients for coupling channel come right after the
658 coefficients for the first coupled channel*/
659 if (s->channel_in_cpl[ch]) {
661 decode_transform_coeffs_ch(s, blk, CPL_CH, &m);
662 calc_transform_coeffs_cpl(s);
665 end = s->end_freq[CPL_CH];
667 end = s->end_freq[ch];
670 s->fixed_coeffs[ch][end] = 0;
674 /* zero the dithered coefficients for appropriate channels */
679 * Stereo rematrixing.
680 * reference: Section 7.5.4 Rematrixing : Decoding Technique
682 static void do_rematrixing(AC3DecodeContext *s)
687 end = FFMIN(s->end_freq[1], s->end_freq[2]);
689 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++) {
690 if (s->rematrixing_flags[bnd]) {
691 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd + 1]);
692 for (i = ff_ac3_rematrix_band_tab[bnd]; i < bndend; i++) {
693 int tmp0 = s->fixed_coeffs[1][i];
694 s->fixed_coeffs[1][i] += s->fixed_coeffs[2][i];
695 s->fixed_coeffs[2][i] = tmp0 - s->fixed_coeffs[2][i];
702 * Inverse MDCT Transform.
703 * Convert frequency domain coefficients to time-domain audio samples.
704 * reference: Section 7.9.4 Transformation Equations
706 static inline void do_imdct(AC3DecodeContext *s, int channels, int offset)
710 for (ch = 1; ch <= channels; ch++) {
711 if (s->block_switch[ch]) {
713 FFTSample *x = s->tmp_output + 128;
714 for (i = 0; i < 128; i++)
715 x[i] = s->transform_coeffs[ch][2 * i];
716 s->imdct_256.imdct_half(&s->imdct_256, s->tmp_output, x);
718 s->fdsp->vector_fmul_window_scaled(s->outptr[ch - 1], s->delay[ch - 1 + offset],
719 s->tmp_output, s->window, 128, 8);
721 s->fdsp->vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1 + offset],
722 s->tmp_output, s->window, 128);
724 for (i = 0; i < 128; i++)
725 x[i] = s->transform_coeffs[ch][2 * i + 1];
726 s->imdct_256.imdct_half(&s->imdct_256, s->delay[ch - 1 + offset], x);
728 s->imdct_512.imdct_half(&s->imdct_512, s->tmp_output, s->transform_coeffs[ch]);
730 s->fdsp->vector_fmul_window_scaled(s->outptr[ch - 1], s->delay[ch - 1 + offset],
731 s->tmp_output, s->window, 128, 8);
733 s->fdsp->vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1 + offset],
734 s->tmp_output, s->window, 128);
736 memcpy(s->delay[ch - 1 + offset], s->tmp_output + 128, 128 * sizeof(FFTSample));
742 * Upmix delay samples from stereo to original channel layout.
744 static void ac3_upmix_delay(AC3DecodeContext *s)
746 int channel_data_size = sizeof(s->delay[0]);
747 switch (s->channel_mode) {
748 case AC3_CHMODE_DUALMONO:
749 case AC3_CHMODE_STEREO:
750 /* upmix mono to stereo */
751 memcpy(s->delay[1], s->delay[0], channel_data_size);
753 case AC3_CHMODE_2F2R:
754 memset(s->delay[3], 0, channel_data_size);
755 case AC3_CHMODE_2F1R:
756 memset(s->delay[2], 0, channel_data_size);
758 case AC3_CHMODE_3F2R:
759 memset(s->delay[4], 0, channel_data_size);
760 case AC3_CHMODE_3F1R:
761 memset(s->delay[3], 0, channel_data_size);
763 memcpy(s->delay[2], s->delay[1], channel_data_size);
764 memset(s->delay[1], 0, channel_data_size);
770 * Decode band structure for coupling, spectral extension, or enhanced coupling.
771 * The band structure defines how many subbands are in each band. For each
772 * subband in the range, 1 means it is combined with the previous band, and 0
773 * means that it starts a new band.
775 * @param[in] gbc bit reader context
776 * @param[in] blk block number
777 * @param[in] eac3 flag to indicate E-AC-3
778 * @param[in] ecpl flag to indicate enhanced coupling
779 * @param[in] start_subband subband number for start of range
780 * @param[in] end_subband subband number for end of range
781 * @param[in] default_band_struct default band structure table
782 * @param[out] num_bands number of bands (optionally NULL)
783 * @param[out] band_sizes array containing the number of bins in each band (optionally NULL)
784 * @param[in,out] band_struct current band structure
786 static void decode_band_structure(GetBitContext *gbc, int blk, int eac3,
787 int ecpl, int start_subband, int end_subband,
788 const uint8_t *default_band_struct,
789 int *num_bands, uint8_t *band_sizes,
790 uint8_t *band_struct, int band_struct_size)
792 int subbnd, bnd, n_subbands, n_bands=0;
795 n_subbands = end_subband - start_subband;
798 memcpy(band_struct, default_band_struct, band_struct_size);
800 av_assert0(band_struct_size >= start_subband + n_subbands);
802 band_struct += start_subband + 1;
804 /* decode band structure from bitstream or use default */
805 if (!eac3 || get_bits1(gbc)) {
806 for (subbnd = 0; subbnd < n_subbands - 1; subbnd++) {
807 band_struct[subbnd] = get_bits1(gbc);
811 /* calculate number of bands and band sizes based on band structure.
812 note that the first 4 subbands in enhanced coupling span only 6 bins
814 if (num_bands || band_sizes ) {
815 n_bands = n_subbands;
816 bnd_sz[0] = ecpl ? 6 : 12;
817 for (bnd = 0, subbnd = 1; subbnd < n_subbands; subbnd++) {
818 int subbnd_size = (ecpl && subbnd < 4) ? 6 : 12;
819 if (band_struct[subbnd - 1]) {
821 bnd_sz[bnd] += subbnd_size;
823 bnd_sz[++bnd] = subbnd_size;
828 /* set optional output params */
830 *num_bands = n_bands;
832 memcpy(band_sizes, bnd_sz, n_bands);
835 static inline int spx_strategy(AC3DecodeContext *s, int blk)
837 GetBitContext *bc = &s->gbc;
838 int fbw_channels = s->fbw_channels;
839 int dst_start_freq, dst_end_freq, src_start_freq,
840 start_subband, end_subband, ch;
842 /* determine which channels use spx */
843 if (s->channel_mode == AC3_CHMODE_MONO) {
844 s->channel_uses_spx[1] = 1;
846 for (ch = 1; ch <= fbw_channels; ch++)
847 s->channel_uses_spx[ch] = get_bits1(bc);
850 /* get the frequency bins of the spx copy region and the spx start
852 dst_start_freq = get_bits(bc, 2);
853 start_subband = get_bits(bc, 3) + 2;
854 if (start_subband > 7)
855 start_subband += start_subband - 7;
856 end_subband = get_bits(bc, 3) + 5;
858 s->spx_dst_end_freq = end_freq_inv_tab[end_subband-5];
861 end_subband += end_subband - 7;
862 dst_start_freq = dst_start_freq * 12 + 25;
863 src_start_freq = start_subband * 12 + 25;
864 dst_end_freq = end_subband * 12 + 25;
866 /* check validity of spx ranges */
867 if (start_subband >= end_subband) {
868 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
869 "range (%d >= %d)\n", start_subband, end_subband);
870 return AVERROR_INVALIDDATA;
872 if (dst_start_freq >= src_start_freq) {
873 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
874 "copy start bin (%d >= %d)\n", dst_start_freq, src_start_freq);
875 return AVERROR_INVALIDDATA;
878 s->spx_dst_start_freq = dst_start_freq;
879 s->spx_src_start_freq = src_start_freq;
881 s->spx_dst_end_freq = dst_end_freq;
883 decode_band_structure(bc, blk, s->eac3, 0,
884 start_subband, end_subband,
885 ff_eac3_default_spx_band_struct,
888 s->spx_band_struct, sizeof(s->spx_band_struct));
892 static inline void spx_coordinates(AC3DecodeContext *s)
894 GetBitContext *bc = &s->gbc;
895 int fbw_channels = s->fbw_channels;
898 for (ch = 1; ch <= fbw_channels; ch++) {
899 if (s->channel_uses_spx[ch]) {
900 if (s->first_spx_coords[ch] || get_bits1(bc)) {
902 int bin, master_spx_coord;
904 s->first_spx_coords[ch] = 0;
905 spx_blend = AC3_SPX_BLEND(get_bits(bc, 5));
906 master_spx_coord = get_bits(bc, 2) * 3;
908 bin = s->spx_src_start_freq;
909 for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
910 int bandsize = s->spx_band_sizes[bnd];
911 int spx_coord_exp, spx_coord_mant;
912 INTFLOAT nratio, sblend, nblend;
914 /* calculate blending factors */
915 int64_t accu = ((bin << 23) + (bandsize << 22))
916 * (int64_t)s->spx_dst_end_freq;
917 nratio = (int)(accu >> 32);
918 nratio -= spx_blend << 18;
923 } else if (nratio > 0x7fffff) {
924 nblend = 14529495; // sqrt(3) in FP.23
927 nblend = fixed_sqrt(nratio, 23);
928 accu = (int64_t)nblend * 1859775393;
929 nblend = (int)((accu + (1<<29)) >> 30);
930 sblend = fixed_sqrt(0x800000 - nratio, 23);
935 /* calculate blending factors */
936 nratio = ((float)((bin + (bandsize >> 1))) / s->spx_dst_end_freq) - spx_blend;
937 nratio = av_clipf(nratio, 0.0f, 1.0f);
938 nblend = sqrtf(3.0f * nratio); // noise is scaled by sqrt(3)
939 // to give unity variance
940 sblend = sqrtf(1.0f - nratio);
944 /* decode spx coordinates */
945 spx_coord_exp = get_bits(bc, 4);
946 spx_coord_mant = get_bits(bc, 2);
947 if (spx_coord_exp == 15) spx_coord_mant <<= 1;
948 else spx_coord_mant += 4;
949 spx_coord_mant <<= (25 - spx_coord_exp - master_spx_coord);
951 /* multiply noise and signal blending factors by spx coordinate */
953 accu = (int64_t)nblend * spx_coord_mant;
954 s->spx_noise_blend[ch][bnd] = (int)((accu + (1<<22)) >> 23);
955 accu = (int64_t)sblend * spx_coord_mant;
956 s->spx_signal_blend[ch][bnd] = (int)((accu + (1<<22)) >> 23);
958 spx_coord = spx_coord_mant * (1.0f / (1 << 23));
959 s->spx_noise_blend [ch][bnd] = nblend * spx_coord;
960 s->spx_signal_blend[ch][bnd] = sblend * spx_coord;
965 s->first_spx_coords[ch] = 1;
970 static inline int coupling_strategy(AC3DecodeContext *s, int blk,
971 uint8_t *bit_alloc_stages)
973 GetBitContext *bc = &s->gbc;
974 int fbw_channels = s->fbw_channels;
975 int channel_mode = s->channel_mode;
978 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
980 s->cpl_in_use[blk] = get_bits1(bc);
981 if (s->cpl_in_use[blk]) {
982 /* coupling in use */
983 int cpl_start_subband, cpl_end_subband;
985 if (channel_mode < AC3_CHMODE_STEREO) {
986 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
987 return AVERROR_INVALIDDATA;
990 /* check for enhanced coupling */
991 if (s->eac3 && get_bits1(bc)) {
992 /* TODO: parse enhanced coupling strategy info */
993 avpriv_request_sample(s->avctx, "Enhanced coupling");
994 return AVERROR_PATCHWELCOME;
997 /* determine which channels are coupled */
998 if (s->eac3 && s->channel_mode == AC3_CHMODE_STEREO) {
999 s->channel_in_cpl[1] = 1;
1000 s->channel_in_cpl[2] = 1;
1002 for (ch = 1; ch <= fbw_channels; ch++)
1003 s->channel_in_cpl[ch] = get_bits1(bc);
1006 /* phase flags in use */
1007 if (channel_mode == AC3_CHMODE_STEREO)
1008 s->phase_flags_in_use = get_bits1(bc);
1010 /* coupling frequency range */
1011 cpl_start_subband = get_bits(bc, 4);
1012 cpl_end_subband = s->spx_in_use ? (s->spx_src_start_freq - 37) / 12 :
1013 get_bits(bc, 4) + 3;
1014 if (cpl_start_subband >= cpl_end_subband) {
1015 av_log(s->avctx, AV_LOG_ERROR, "invalid coupling range (%d >= %d)\n",
1016 cpl_start_subband, cpl_end_subband);
1017 return AVERROR_INVALIDDATA;
1019 s->start_freq[CPL_CH] = cpl_start_subband * 12 + 37;
1020 s->end_freq[CPL_CH] = cpl_end_subband * 12 + 37;
1022 decode_band_structure(bc, blk, s->eac3, 0, cpl_start_subband,
1024 ff_eac3_default_cpl_band_struct,
1025 &s->num_cpl_bands, s->cpl_band_sizes,
1026 s->cpl_band_struct, sizeof(s->cpl_band_struct));
1028 /* coupling not in use */
1029 for (ch = 1; ch <= fbw_channels; ch++) {
1030 s->channel_in_cpl[ch] = 0;
1031 s->first_cpl_coords[ch] = 1;
1033 s->first_cpl_leak = s->eac3;
1034 s->phase_flags_in_use = 0;
1040 static inline int coupling_coordinates(AC3DecodeContext *s, int blk)
1042 GetBitContext *bc = &s->gbc;
1043 int fbw_channels = s->fbw_channels;
1045 int cpl_coords_exist = 0;
1047 for (ch = 1; ch <= fbw_channels; ch++) {
1048 if (s->channel_in_cpl[ch]) {
1049 if ((s->eac3 && s->first_cpl_coords[ch]) || get_bits1(bc)) {
1050 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
1051 s->first_cpl_coords[ch] = 0;
1052 cpl_coords_exist = 1;
1053 master_cpl_coord = 3 * get_bits(bc, 2);
1054 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
1055 cpl_coord_exp = get_bits(bc, 4);
1056 cpl_coord_mant = get_bits(bc, 4);
1057 if (cpl_coord_exp == 15)
1058 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
1060 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
1061 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
1064 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must "
1065 "be present in block 0\n");
1066 return AVERROR_INVALIDDATA;
1069 /* channel not in coupling */
1070 s->first_cpl_coords[ch] = 1;
1074 if (s->channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
1075 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
1076 s->phase_flags[bnd] = s->phase_flags_in_use ? get_bits1(bc) : 0;
1084 * Decode a single audio block from the AC-3 bitstream.
1086 static int decode_audio_block(AC3DecodeContext *s, int blk, int offset)
1088 int fbw_channels = s->fbw_channels;
1089 int channel_mode = s->channel_mode;
1090 int i, bnd, seg, ch, ret;
1091 int different_transforms;
1094 GetBitContext *gbc = &s->gbc;
1095 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS] = { 0 };
1097 /* block switch flags */
1098 different_transforms = 0;
1099 if (s->block_switch_syntax) {
1100 for (ch = 1; ch <= fbw_channels; ch++) {
1101 s->block_switch[ch] = get_bits1(gbc);
1102 if (ch > 1 && s->block_switch[ch] != s->block_switch[1])
1103 different_transforms = 1;
1107 /* dithering flags */
1108 if (s->dither_flag_syntax) {
1109 for (ch = 1; ch <= fbw_channels; ch++) {
1110 s->dither_flag[ch] = get_bits1(gbc);
1115 i = !s->channel_mode;
1117 if (get_bits1(gbc)) {
1118 /* Allow asymmetric application of DRC when drc_scale > 1.
1119 Amplification of quiet sounds is enhanced */
1120 int range_bits = get_bits(gbc, 8);
1121 INTFLOAT range = AC3_RANGE(range_bits);
1122 if (range_bits <= 127 || s->drc_scale <= 1.0)
1123 s->dynamic_range[i] = AC3_DYNAMIC_RANGE(range);
1125 s->dynamic_range[i] = range;
1126 } else if (blk == 0) {
1127 s->dynamic_range[i] = AC3_DYNAMIC_RANGE1;
1131 /* spectral extension strategy */
1132 if (s->eac3 && (!blk || get_bits1(gbc))) {
1133 s->spx_in_use = get_bits1(gbc);
1134 if (s->spx_in_use) {
1135 if ((ret = spx_strategy(s, blk)) < 0)
1139 if (!s->eac3 || !s->spx_in_use) {
1141 for (ch = 1; ch <= fbw_channels; ch++) {
1142 s->channel_uses_spx[ch] = 0;
1143 s->first_spx_coords[ch] = 1;
1147 /* spectral extension coordinates */
1151 /* coupling strategy */
1152 if (s->eac3 ? s->cpl_strategy_exists[blk] : get_bits1(gbc)) {
1153 if ((ret = coupling_strategy(s, blk, bit_alloc_stages)) < 0)
1155 } else if (!s->eac3) {
1157 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must "
1158 "be present in block 0\n");
1159 return AVERROR_INVALIDDATA;
1161 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
1164 cpl_in_use = s->cpl_in_use[blk];
1166 /* coupling coordinates */
1168 if ((ret = coupling_coordinates(s, blk)) < 0)
1172 /* stereo rematrixing strategy and band structure */
1173 if (channel_mode == AC3_CHMODE_STEREO) {
1174 if ((s->eac3 && !blk) || get_bits1(gbc)) {
1175 s->num_rematrixing_bands = 4;
1176 if (cpl_in_use && s->start_freq[CPL_CH] <= 61) {
1177 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
1178 } else if (s->spx_in_use && s->spx_src_start_freq <= 61) {
1179 s->num_rematrixing_bands--;
1181 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++)
1182 s->rematrixing_flags[bnd] = get_bits1(gbc);
1184 av_log(s->avctx, AV_LOG_WARNING, "Warning: "
1185 "new rematrixing strategy not present in block 0\n");
1186 s->num_rematrixing_bands = 0;
1190 /* exponent strategies for each channel */
1191 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1193 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
1194 if (s->exp_strategy[blk][ch] != EXP_REUSE)
1195 bit_alloc_stages[ch] = 3;
1198 /* channel bandwidth */
1199 for (ch = 1; ch <= fbw_channels; ch++) {
1200 s->start_freq[ch] = 0;
1201 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1203 int prev = s->end_freq[ch];
1204 if (s->channel_in_cpl[ch])
1205 s->end_freq[ch] = s->start_freq[CPL_CH];
1206 else if (s->channel_uses_spx[ch])
1207 s->end_freq[ch] = s->spx_src_start_freq;
1209 int bandwidth_code = get_bits(gbc, 6);
1210 if (bandwidth_code > 60) {
1211 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60\n", bandwidth_code);
1212 return AVERROR_INVALIDDATA;
1214 s->end_freq[ch] = bandwidth_code * 3 + 73;
1216 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
1217 s->num_exp_groups[ch] = (s->end_freq[ch] + group_size-4) / group_size;
1218 if (blk > 0 && s->end_freq[ch] != prev)
1219 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
1222 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
1223 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
1224 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
1227 /* decode exponents for each channel */
1228 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1229 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1230 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
1231 if (decode_exponents(s, gbc, s->exp_strategy[blk][ch],
1232 s->num_exp_groups[ch], s->dexps[ch][0],
1233 &s->dexps[ch][s->start_freq[ch]+!!ch])) {
1234 return AVERROR_INVALIDDATA;
1236 if (ch != CPL_CH && ch != s->lfe_ch)
1237 skip_bits(gbc, 2); /* skip gainrng */
1241 /* bit allocation information */
1242 if (s->bit_allocation_syntax) {
1243 if (get_bits1(gbc)) {
1244 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1245 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1246 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
1247 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
1248 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
1249 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1250 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1252 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must "
1253 "be present in block 0\n");
1254 return AVERROR_INVALIDDATA;
1258 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
1259 if (!s->eac3 || !blk) {
1260 if (s->snr_offset_strategy && get_bits1(gbc)) {
1263 csnr = (get_bits(gbc, 6) - 15) << 4;
1264 for (i = ch = !cpl_in_use; ch <= s->channels; ch++) {
1266 if (ch == i || s->snr_offset_strategy == 2)
1267 snr = (csnr + get_bits(gbc, 4)) << 2;
1268 /* run at least last bit allocation stage if snr offset changes */
1269 if (blk && s->snr_offset[ch] != snr) {
1270 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 1);
1272 s->snr_offset[ch] = snr;
1274 /* fast gain (normal AC-3 only) */
1276 int prev = s->fast_gain[ch];
1277 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1278 /* run last 2 bit allocation stages if fast gain changes */
1279 if (blk && prev != s->fast_gain[ch])
1280 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1283 } else if (!s->eac3 && !blk) {
1284 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
1285 return AVERROR_INVALIDDATA;
1289 /* fast gain (E-AC-3 only) */
1290 if (s->fast_gain_syntax && get_bits1(gbc)) {
1291 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1292 int prev = s->fast_gain[ch];
1293 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1294 /* run last 2 bit allocation stages if fast gain changes */
1295 if (blk && prev != s->fast_gain[ch])
1296 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1298 } else if (s->eac3 && !blk) {
1299 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1300 s->fast_gain[ch] = ff_ac3_fast_gain_tab[4];
1303 /* E-AC-3 to AC-3 converter SNR offset */
1304 if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT && get_bits1(gbc)) {
1305 skip_bits(gbc, 10); // skip converter snr offset
1308 /* coupling leak information */
1310 if (s->first_cpl_leak || get_bits1(gbc)) {
1311 int fl = get_bits(gbc, 3);
1312 int sl = get_bits(gbc, 3);
1313 /* run last 2 bit allocation stages for coupling channel if
1314 coupling leak changes */
1315 if (blk && (fl != s->bit_alloc_params.cpl_fast_leak ||
1316 sl != s->bit_alloc_params.cpl_slow_leak)) {
1317 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
1319 s->bit_alloc_params.cpl_fast_leak = fl;
1320 s->bit_alloc_params.cpl_slow_leak = sl;
1321 } else if (!s->eac3 && !blk) {
1322 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must "
1323 "be present in block 0\n");
1324 return AVERROR_INVALIDDATA;
1326 s->first_cpl_leak = 0;
1329 /* delta bit allocation information */
1330 if (s->dba_syntax && get_bits1(gbc)) {
1331 /* delta bit allocation exists (strategy) */
1332 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1333 s->dba_mode[ch] = get_bits(gbc, 2);
1334 if (s->dba_mode[ch] == DBA_RESERVED) {
1335 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1336 return AVERROR_INVALIDDATA;
1338 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1340 /* channel delta offset, len and bit allocation */
1341 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1342 if (s->dba_mode[ch] == DBA_NEW) {
1343 s->dba_nsegs[ch] = get_bits(gbc, 3) + 1;
1344 for (seg = 0; seg < s->dba_nsegs[ch]; seg++) {
1345 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1346 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1347 s->dba_values[ch][seg] = get_bits(gbc, 3);
1349 /* run last 2 bit allocation stages if new dba values */
1350 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1353 } else if (blk == 0) {
1354 for (ch = 0; ch <= s->channels; ch++) {
1355 s->dba_mode[ch] = DBA_NONE;
1359 /* Bit allocation */
1360 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1361 if (bit_alloc_stages[ch] > 2) {
1362 /* Exponent mapping into PSD and PSD integration */
1363 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1364 s->start_freq[ch], s->end_freq[ch],
1365 s->psd[ch], s->band_psd[ch]);
1367 if (bit_alloc_stages[ch] > 1) {
1368 /* Compute excitation function, Compute masking curve, and
1369 Apply delta bit allocation */
1370 if (ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1371 s->start_freq[ch], s->end_freq[ch],
1372 s->fast_gain[ch], (ch == s->lfe_ch),
1373 s->dba_mode[ch], s->dba_nsegs[ch],
1374 s->dba_offsets[ch], s->dba_lengths[ch],
1375 s->dba_values[ch], s->mask[ch])) {
1376 av_log(s->avctx, AV_LOG_ERROR, "error in bit allocation\n");
1377 return AVERROR_INVALIDDATA;
1380 if (bit_alloc_stages[ch] > 0) {
1381 /* Compute bit allocation */
1382 const uint8_t *bap_tab = s->channel_uses_aht[ch] ?
1383 ff_eac3_hebap_tab : ff_ac3_bap_tab;
1384 s->ac3dsp.bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1385 s->start_freq[ch], s->end_freq[ch],
1387 s->bit_alloc_params.floor,
1388 bap_tab, s->bap[ch]);
1392 /* unused dummy data */
1393 if (s->skip_syntax && get_bits1(gbc)) {
1394 int skipl = get_bits(gbc, 9);
1395 skip_bits_long(gbc, 8 * skipl);
1398 /* unpack the transform coefficients
1399 this also uncouples channels if coupling is in use. */
1400 decode_transform_coeffs(s, blk);
1402 /* TODO: generate enhanced coupling coordinates and uncouple */
1404 /* recover coefficients if rematrixing is in use */
1405 if (s->channel_mode == AC3_CHMODE_STEREO)
1408 /* apply scaling to coefficients (headroom, dynrng) */
1409 for (ch = 1; ch <= s->channels; ch++) {
1410 int audio_channel = 0;
1412 if (s->channel_mode == AC3_CHMODE_DUALMONO && ch <= 2)
1413 audio_channel = 2-ch;
1414 if (s->heavy_compression && s->compression_exists[audio_channel])
1415 gain = s->heavy_dynamic_range[audio_channel];
1417 gain = s->dynamic_range[audio_channel];
1420 scale_coefs(s->transform_coeffs[ch], s->fixed_coeffs[ch], gain, 256);
1422 if (s->target_level != 0)
1423 gain = gain * s->level_gain[audio_channel];
1424 gain *= 1.0 / 4194304.0f;
1425 s->fmt_conv.int32_to_float_fmul_scalar(s->transform_coeffs[ch],
1426 s->fixed_coeffs[ch], gain, 256);
1430 /* apply spectral extension to high frequency bins */
1431 if (CONFIG_EAC3_DECODER && s->spx_in_use) {
1432 ff_eac3_apply_spectral_extension(s);
1435 /* downmix and MDCT. order depends on whether block switching is used for
1436 any channel in this block. this is because coefficients for the long
1437 and short transforms cannot be mixed. */
1438 downmix_output = s->channels != s->out_channels &&
1439 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1440 s->fbw_channels == s->out_channels);
1441 if (different_transforms) {
1442 /* the delay samples have already been downmixed, so we upmix the delay
1443 samples in order to reconstruct all channels before downmixing. */
1449 do_imdct(s, s->channels, offset);
1451 if (downmix_output) {
1453 ac3_downmix_c_fixed16(s->outptr, s->downmix_coeffs,
1454 s->out_channels, s->fbw_channels, 256);
1456 ff_ac3dsp_downmix(&s->ac3dsp, s->outptr, s->downmix_coeffs,
1457 s->out_channels, s->fbw_channels, 256);
1461 if (downmix_output) {
1462 AC3_RENAME(ff_ac3dsp_downmix)(&s->ac3dsp, s->xcfptr + 1, s->downmix_coeffs,
1463 s->out_channels, s->fbw_channels, 256);
1466 if (downmix_output && !s->downmixed) {
1468 AC3_RENAME(ff_ac3dsp_downmix)(&s->ac3dsp, s->dlyptr, s->downmix_coeffs,
1469 s->out_channels, s->fbw_channels, 128);
1472 do_imdct(s, s->out_channels, offset);
1479 * Decode a single AC-3 frame.
1481 static int ac3_decode_frame(AVCodecContext * avctx, void *data,
1482 int *got_frame_ptr, AVPacket *avpkt)
1484 AVFrame *frame = data;
1485 const uint8_t *buf = avpkt->data;
1486 int buf_size, full_buf_size = avpkt->size;
1487 AC3DecodeContext *s = avctx->priv_data;
1488 int blk, ch, err, offset, ret;
1489 int got_independent_frame = 0;
1490 const uint8_t *channel_map;
1491 uint8_t extended_channel_map[EAC3_MAX_CHANNELS];
1492 const SHORTFLOAT *output[AC3_MAX_CHANNELS];
1493 enum AVMatrixEncoding matrix_encoding;
1494 AVDownmixInfo *downmix_info;
1496 s->superframe_size = 0;
1498 buf_size = full_buf_size;
1499 /* copy input buffer to decoder context to avoid reading past the end
1500 of the buffer, which can be caused by a damaged input stream. */
1501 if (buf_size >= 2 && AV_RB16(buf) == 0x770B) {
1502 // seems to be byte-swapped AC-3
1503 int cnt = FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE) >> 1;
1504 s->bdsp.bswap16_buf((uint16_t *) s->input_buffer,
1505 (const uint16_t *) buf, cnt);
1507 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE));
1509 /* if consistent noise generation is enabled, seed the linear feedback generator
1510 * with the contents of the AC-3 frame so that the noise is identical across
1511 * decodes given the same AC-3 frame data, for use with non-linear edititing software. */
1512 if (s->consistent_noise_generation)
1513 av_lfg_init_from_data(&s->dith_state, s->input_buffer, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE));
1515 buf = s->input_buffer;
1517 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1518 if ((ret = init_get_bits8(&s->gbc, buf, buf_size)) < 0)
1521 /* parse the syncinfo */
1522 err = parse_frame_header(s);
1526 case AAC_AC3_PARSE_ERROR_SYNC:
1527 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1528 return AVERROR_INVALIDDATA;
1529 case AAC_AC3_PARSE_ERROR_BSID:
1530 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1532 case AAC_AC3_PARSE_ERROR_SAMPLE_RATE:
1533 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1535 case AAC_AC3_PARSE_ERROR_FRAME_SIZE:
1536 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1538 case AAC_AC3_PARSE_ERROR_FRAME_TYPE:
1539 /* skip frame if CRC is ok. otherwise use error concealment. */
1540 /* TODO: add support for substreams */
1541 if (s->substreamid) {
1542 av_log(avctx, AV_LOG_DEBUG,
1543 "unsupported substream %d: skipping frame\n",
1548 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1551 case AAC_AC3_PARSE_ERROR_CRC:
1552 case AAC_AC3_PARSE_ERROR_CHANNEL_CFG:
1554 default: // Normal AVERROR do not try to recover.
1559 /* check that reported frame size fits in input buffer */
1560 if (s->frame_size > buf_size) {
1561 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1562 err = AAC_AC3_PARSE_ERROR_FRAME_SIZE;
1563 } else if (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL)) {
1564 /* check for crc mismatch */
1565 if (av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2],
1566 s->frame_size - 2)) {
1567 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1568 if (avctx->err_recognition & AV_EF_EXPLODE)
1569 return AVERROR_INVALIDDATA;
1570 err = AAC_AC3_PARSE_ERROR_CRC;
1575 if (s->frame_type == EAC3_FRAME_TYPE_DEPENDENT && !got_independent_frame) {
1576 av_log(avctx, AV_LOG_WARNING, "Ignoring dependent frame without independent frame.\n");
1578 return FFMIN(full_buf_size, s->frame_size);
1581 /* channel config */
1582 if (!err || (s->channels && s->out_channels != s->channels)) {
1583 s->out_channels = s->channels;
1584 s->output_mode = s->channel_mode;
1586 s->output_mode |= AC3_OUTPUT_LFEON;
1587 if (s->channels > 1 &&
1588 avctx->request_channel_layout == AV_CH_LAYOUT_MONO) {
1589 s->out_channels = 1;
1590 s->output_mode = AC3_CHMODE_MONO;
1591 } else if (s->channels > 2 &&
1592 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
1593 s->out_channels = 2;
1594 s->output_mode = AC3_CHMODE_STEREO;
1597 s->loro_center_mix_level = gain_levels[s-> center_mix_level];
1598 s->loro_surround_mix_level = gain_levels[s->surround_mix_level];
1599 s->ltrt_center_mix_level = LEVEL_MINUS_3DB;
1600 s->ltrt_surround_mix_level = LEVEL_MINUS_3DB;
1601 /* set downmixing coefficients if needed */
1602 if (s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1603 s->fbw_channels == s->out_channels)) {
1604 if ((ret = set_downmix_coeffs(s)) < 0) {
1605 av_log(avctx, AV_LOG_ERROR, "error setting downmix coeffs\n");
1609 } else if (!s->channels) {
1610 av_log(avctx, AV_LOG_ERROR, "unable to determine channel mode\n");
1611 return AVERROR_INVALIDDATA;
1613 avctx->channels = s->out_channels;
1614 avctx->channel_layout = avpriv_ac3_channel_layout_tab[s->output_mode & ~AC3_OUTPUT_LFEON];
1615 if (s->output_mode & AC3_OUTPUT_LFEON)
1616 avctx->channel_layout |= AV_CH_LOW_FREQUENCY;
1618 /* set audio service type based on bitstream mode for AC-3 */
1619 avctx->audio_service_type = s->bitstream_mode;
1620 if (s->bitstream_mode == 0x7 && s->channels > 1)
1621 avctx->audio_service_type = AV_AUDIO_SERVICE_TYPE_KARAOKE;
1623 /* decode the audio blocks */
1624 channel_map = ff_ac3_dec_channel_map[s->output_mode & ~AC3_OUTPUT_LFEON][s->lfe_on];
1625 offset = s->frame_type == EAC3_FRAME_TYPE_DEPENDENT ? AC3_MAX_CHANNELS : 0;
1626 for (ch = 0; ch < AC3_MAX_CHANNELS; ch++) {
1627 output[ch] = s->output[ch + offset];
1628 s->outptr[ch] = s->output[ch + offset];
1630 for (ch = 0; ch < s->channels; ch++) {
1631 if (ch < s->out_channels)
1632 s->outptr[channel_map[ch]] = s->output_buffer[ch + offset];
1634 for (blk = 0; blk < s->num_blocks; blk++) {
1635 if (!err && decode_audio_block(s, blk, offset)) {
1636 av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n");
1640 for (ch = 0; ch < s->out_channels; ch++)
1641 memcpy(s->output_buffer[ch + offset] + AC3_BLOCK_SIZE*blk, output[ch], AC3_BLOCK_SIZE*sizeof(SHORTFLOAT));
1642 for (ch = 0; ch < s->out_channels; ch++)
1643 output[ch] = s->outptr[channel_map[ch]];
1644 for (ch = 0; ch < s->out_channels; ch++) {
1645 if (!ch || channel_map[ch])
1646 s->outptr[channel_map[ch]] += AC3_BLOCK_SIZE;
1650 /* keep last block for error concealment in next frame */
1651 for (ch = 0; ch < s->out_channels; ch++)
1652 memcpy(s->output[ch + offset], output[ch], AC3_BLOCK_SIZE*sizeof(SHORTFLOAT));
1654 /* check if there is dependent frame */
1655 if (buf_size > s->frame_size) {
1659 if ((ret = init_get_bits8(&s->gbc, buf + s->frame_size, buf_size - s->frame_size)) < 0)
1662 err = ff_ac3_parse_header(&s->gbc, &hdr);
1666 if (hdr.frame_type == EAC3_FRAME_TYPE_DEPENDENT) {
1667 if (hdr.num_blocks != s->num_blocks || s->sample_rate != hdr.sample_rate) {
1668 av_log(avctx, AV_LOG_WARNING, "Ignoring non-compatible dependent frame.\n");
1670 buf += s->frame_size;
1671 buf_size -= s->frame_size;
1672 s->prev_output_mode = s->output_mode;
1673 s->prev_bit_rate = s->bit_rate;
1674 got_independent_frame = 1;
1675 goto dependent_frame;
1680 frame->decode_error_flags = err ? FF_DECODE_ERROR_INVALID_BITSTREAM : 0;
1682 /* if frame is ok, set audio parameters */
1684 avctx->sample_rate = s->sample_rate;
1685 avctx->bit_rate = s->bit_rate + s->prev_bit_rate;
1688 for (ch = 0; ch < EAC3_MAX_CHANNELS; ch++)
1689 extended_channel_map[ch] = ch;
1691 if (s->frame_type == EAC3_FRAME_TYPE_DEPENDENT) {
1692 uint64_t ich_layout = avpriv_ac3_channel_layout_tab[s->prev_output_mode & ~AC3_OUTPUT_LFEON];
1693 int channel_map_size = ff_ac3_channels_tab[s->output_mode & ~AC3_OUTPUT_LFEON] + s->lfe_on;
1694 uint64_t channel_layout;
1697 if (s->prev_output_mode & AC3_OUTPUT_LFEON)
1698 ich_layout |= AV_CH_LOW_FREQUENCY;
1700 channel_layout = ich_layout;
1701 for (ch = 0; ch < 16; ch++) {
1702 if (s->channel_map & (1 << (EAC3_MAX_CHANNELS - ch - 1))) {
1703 channel_layout |= custom_channel_map_locations[ch][1];
1706 if (av_get_channel_layout_nb_channels(channel_layout) > EAC3_MAX_CHANNELS) {
1707 av_log(avctx, AV_LOG_ERROR, "Too many channels (%d) coded\n",
1708 av_get_channel_layout_nb_channels(channel_layout));
1709 return AVERROR_INVALIDDATA;
1712 avctx->channel_layout = channel_layout;
1713 avctx->channels = av_get_channel_layout_nb_channels(channel_layout);
1715 for (ch = 0; ch < EAC3_MAX_CHANNELS; ch++) {
1716 if (s->channel_map & (1 << (EAC3_MAX_CHANNELS - ch - 1))) {
1717 if (custom_channel_map_locations[ch][0]) {
1718 int index = av_get_channel_layout_channel_index(channel_layout,
1719 custom_channel_map_locations[ch][1]);
1721 return AVERROR_INVALIDDATA;
1722 if (extend >= channel_map_size)
1723 return AVERROR_INVALIDDATA;
1725 extended_channel_map[index] = offset + channel_map[extend++];
1729 for (i = 0; i < 64; i++) {
1730 if ((1LL << i) & custom_channel_map_locations[ch][1]) {
1731 int index = av_get_channel_layout_channel_index(channel_layout,
1734 return AVERROR_INVALIDDATA;
1735 if (extend >= channel_map_size)
1736 return AVERROR_INVALIDDATA;
1738 extended_channel_map[index] = offset + channel_map[extend++];
1746 /* get output buffer */
1747 frame->nb_samples = s->num_blocks * AC3_BLOCK_SIZE;
1748 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1751 for (ch = 0; ch < avctx->channels; ch++) {
1752 int map = extended_channel_map[ch];
1753 av_assert0(ch>=AV_NUM_DATA_POINTERS || frame->extended_data[ch] == frame->data[ch]);
1754 memcpy((SHORTFLOAT *)frame->extended_data[ch],
1755 s->output_buffer[map],
1756 s->num_blocks * AC3_BLOCK_SIZE * sizeof(SHORTFLOAT));
1762 * Check whether the input layout is compatible, and make sure we're not
1763 * downmixing (else the matrix encoding is no longer applicable).
1765 matrix_encoding = AV_MATRIX_ENCODING_NONE;
1766 if (s->channel_mode == AC3_CHMODE_STEREO &&
1767 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1768 if (s->dolby_surround_mode == AC3_DSURMOD_ON)
1769 matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
1770 else if (s->dolby_headphone_mode == AC3_DHEADPHONMOD_ON)
1771 matrix_encoding = AV_MATRIX_ENCODING_DOLBYHEADPHONE;
1772 } else if (s->channel_mode >= AC3_CHMODE_2F2R &&
1773 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1774 switch (s->dolby_surround_ex_mode) {
1775 case AC3_DSUREXMOD_ON: // EX or PLIIx
1776 matrix_encoding = AV_MATRIX_ENCODING_DOLBYEX;
1778 case AC3_DSUREXMOD_PLIIZ:
1779 matrix_encoding = AV_MATRIX_ENCODING_DPLIIZ;
1781 default: // not indicated or off
1785 if ((ret = ff_side_data_update_matrix_encoding(frame, matrix_encoding)) < 0)
1789 if ((downmix_info = av_downmix_info_update_side_data(frame))) {
1790 switch (s->preferred_downmix) {
1791 case AC3_DMIXMOD_LTRT:
1792 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LTRT;
1794 case AC3_DMIXMOD_LORO:
1795 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LORO;
1797 case AC3_DMIXMOD_DPLII:
1798 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_DPLII;
1801 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_UNKNOWN;
1804 downmix_info->center_mix_level = gain_levels[s-> center_mix_level];
1805 downmix_info->center_mix_level_ltrt = gain_levels[s-> center_mix_level_ltrt];
1806 downmix_info->surround_mix_level = gain_levels[s-> surround_mix_level];
1807 downmix_info->surround_mix_level_ltrt = gain_levels[s->surround_mix_level_ltrt];
1808 if (s->lfe_mix_level_exists)
1809 downmix_info->lfe_mix_level = gain_levels_lfe[s->lfe_mix_level];
1811 downmix_info->lfe_mix_level = 0.0; // -inf dB
1813 return AVERROR(ENOMEM);
1817 if (!s->superframe_size)
1818 return FFMIN(full_buf_size, s->frame_size);
1820 return FFMIN(full_buf_size, s->superframe_size);
1824 * Uninitialize the AC-3 decoder.
1826 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1828 AC3DecodeContext *s = avctx->priv_data;
1829 ff_mdct_end(&s->imdct_512);
1830 ff_mdct_end(&s->imdct_256);
1832 av_freep(&s->downmix_coeffs[0]);
1837 #define OFFSET(x) offsetof(AC3DecodeContext, x)
1838 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)