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 }, },
110 * Symmetrical Dequantization
111 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
112 * Tables 7.19 to 7.23
115 symmetric_dequant(int code, int levels)
117 return ((code - (levels >> 1)) * (1 << 24)) / levels;
121 * Initialize tables at runtime.
123 static av_cold void ac3_tables_init(void)
127 /* generate table for ungrouping 3 values in 7 bits
128 reference: Section 7.1.3 Exponent Decoding */
129 for (i = 0; i < 128; i++) {
130 ungroup_3_in_7_bits_tab[i][0] = i / 25;
131 ungroup_3_in_7_bits_tab[i][1] = (i % 25) / 5;
132 ungroup_3_in_7_bits_tab[i][2] = (i % 25) % 5;
135 /* generate grouped mantissa tables
136 reference: Section 7.3.5 Ungrouping of Mantissas */
137 for (i = 0; i < 32; i++) {
138 /* bap=1 mantissas */
139 b1_mantissas[i][0] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][0], 3);
140 b1_mantissas[i][1] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][1], 3);
141 b1_mantissas[i][2] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][2], 3);
143 for (i = 0; i < 128; i++) {
144 /* bap=2 mantissas */
145 b2_mantissas[i][0] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][0], 5);
146 b2_mantissas[i][1] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][1], 5);
147 b2_mantissas[i][2] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][2], 5);
149 /* bap=4 mantissas */
150 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
151 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
153 /* generate ungrouped mantissa tables
154 reference: Tables 7.21 and 7.23 */
155 for (i = 0; i < 7; i++) {
156 /* bap=3 mantissas */
157 b3_mantissas[i] = symmetric_dequant(i, 7);
159 for (i = 0; i < 15; i++) {
160 /* bap=5 mantissas */
161 b5_mantissas[i] = symmetric_dequant(i, 15);
165 /* generate dynamic range table
166 reference: Section 7.7.1 Dynamic Range Control */
167 for (i = 0; i < 256; i++) {
168 int v = (i >> 5) - ((i >> 7) << 3) - 5;
169 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
172 /* generate compr dynamic range table
173 reference: Section 7.7.2 Heavy Compression */
174 for (i = 0; i < 256; i++) {
175 int v = (i >> 4) - ((i >> 7) << 4) - 4;
176 ff_ac3_heavy_dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0xF) | 0x10);
182 * AVCodec initialization
184 static av_cold int ac3_decode_init(AVCodecContext *avctx)
186 AC3DecodeContext *s = avctx->priv_data;
192 ff_mdct_init(&s->imdct_256, 8, 1, 1.0);
193 ff_mdct_init(&s->imdct_512, 9, 1, 1.0);
194 AC3_RENAME(ff_kbd_window_init)(s->window, 5.0, 256);
195 ff_bswapdsp_init(&s->bdsp);
198 s->fdsp = avpriv_alloc_fixed_dsp(avctx->flags & AV_CODEC_FLAG_BITEXACT);
200 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
201 ff_fmt_convert_init(&s->fmt_conv, avctx);
204 ff_ac3dsp_init(&s->ac3dsp, avctx->flags & AV_CODEC_FLAG_BITEXACT);
205 av_lfg_init(&s->dith_state, 0);
208 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
210 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
212 /* allow downmixing to stereo or mono */
213 if (avctx->channels > 1 &&
214 avctx->request_channel_layout == AV_CH_LAYOUT_MONO)
216 else if (avctx->channels > 2 &&
217 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO)
221 for (i = 0; i < AC3_MAX_CHANNELS; i++) {
222 s->xcfptr[i] = s->transform_coeffs[i];
223 s->dlyptr[i] = s->delay[i];
230 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
231 * GetBitContext within AC3DecodeContext must point to
232 * the start of the synchronized AC-3 bitstream.
234 static int ac3_parse_header(AC3DecodeContext *s)
236 GetBitContext *gbc = &s->gbc;
239 /* read the rest of the bsi. read twice for dual mono mode. */
240 i = !s->channel_mode;
242 s->dialog_normalization[(!s->channel_mode)-i] = -get_bits(gbc, 5);
243 if (s->dialog_normalization[(!s->channel_mode)-i] == 0) {
244 s->dialog_normalization[(!s->channel_mode)-i] = -31;
246 if (s->target_level != 0) {
247 s->level_gain[(!s->channel_mode)-i] = powf(2.0f,
248 (float)(s->target_level -
249 s->dialog_normalization[(!s->channel_mode)-i])/6.0f);
251 if (s->compression_exists[(!s->channel_mode)-i] = get_bits1(gbc)) {
252 s->heavy_dynamic_range[(!s->channel_mode)-i] =
253 AC3_HEAVY_RANGE(get_bits(gbc, 8));
256 skip_bits(gbc, 8); //skip language code
258 skip_bits(gbc, 7); //skip audio production information
261 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
263 /* skip the timecodes or parse the Alternate Bit Stream Syntax */
264 if (s->bitstream_id != 6) {
266 skip_bits(gbc, 14); //skip timecode1
268 skip_bits(gbc, 14); //skip timecode2
270 if (get_bits1(gbc)) {
271 s->preferred_downmix = get_bits(gbc, 2);
272 s->center_mix_level_ltrt = get_bits(gbc, 3);
273 s->surround_mix_level_ltrt = av_clip(get_bits(gbc, 3), 3, 7);
274 s->center_mix_level = get_bits(gbc, 3);
275 s->surround_mix_level = av_clip(get_bits(gbc, 3), 3, 7);
277 if (get_bits1(gbc)) {
278 s->dolby_surround_ex_mode = get_bits(gbc, 2);
279 s->dolby_headphone_mode = get_bits(gbc, 2);
280 skip_bits(gbc, 10); // skip adconvtyp (1), xbsi2 (8), encinfo (1)
284 /* skip additional bitstream info */
285 if (get_bits1(gbc)) {
286 i = get_bits(gbc, 6);
296 * Common function to parse AC-3 or E-AC-3 frame header
298 static int parse_frame_header(AC3DecodeContext *s)
303 err = ff_ac3_parse_header(&s->gbc, &hdr);
307 /* get decoding parameters from header info */
308 s->bit_alloc_params.sr_code = hdr.sr_code;
309 s->bitstream_id = hdr.bitstream_id;
310 s->bitstream_mode = hdr.bitstream_mode;
311 s->channel_mode = hdr.channel_mode;
312 s->lfe_on = hdr.lfe_on;
313 s->bit_alloc_params.sr_shift = hdr.sr_shift;
314 s->sample_rate = hdr.sample_rate;
315 s->bit_rate = hdr.bit_rate;
316 s->channels = hdr.channels;
317 s->fbw_channels = s->channels - s->lfe_on;
318 s->lfe_ch = s->fbw_channels + 1;
319 s->frame_size = hdr.frame_size;
320 s->superframe_size += hdr.frame_size;
321 s->preferred_downmix = AC3_DMIXMOD_NOTINDICATED;
322 s->center_mix_level = hdr.center_mix_level;
323 s->center_mix_level_ltrt = 4; // -3.0dB
324 s->surround_mix_level = hdr.surround_mix_level;
325 s->surround_mix_level_ltrt = 4; // -3.0dB
326 s->lfe_mix_level_exists = 0;
327 s->num_blocks = hdr.num_blocks;
328 s->frame_type = hdr.frame_type;
329 s->substreamid = hdr.substreamid;
330 s->dolby_surround_mode = hdr.dolby_surround_mode;
331 s->dolby_surround_ex_mode = AC3_DSUREXMOD_NOTINDICATED;
332 s->dolby_headphone_mode = AC3_DHEADPHONMOD_NOTINDICATED;
335 s->start_freq[s->lfe_ch] = 0;
336 s->end_freq[s->lfe_ch] = 7;
337 s->num_exp_groups[s->lfe_ch] = 2;
338 s->channel_in_cpl[s->lfe_ch] = 0;
341 if (s->bitstream_id <= 10) {
343 s->snr_offset_strategy = 2;
344 s->block_switch_syntax = 1;
345 s->dither_flag_syntax = 1;
346 s->bit_allocation_syntax = 1;
347 s->fast_gain_syntax = 0;
348 s->first_cpl_leak = 0;
351 memset(s->channel_uses_aht, 0, sizeof(s->channel_uses_aht));
352 return ac3_parse_header(s);
353 } else if (CONFIG_EAC3_DECODER) {
355 return ff_eac3_parse_header(s);
357 av_log(s->avctx, AV_LOG_ERROR, "E-AC-3 support not compiled in\n");
358 return AVERROR(ENOSYS);
363 * Set stereo downmixing coefficients based on frame header info.
364 * reference: Section 7.8.2 Downmixing Into Two Channels
366 static int set_downmix_coeffs(AC3DecodeContext *s)
369 float cmix = gain_levels[s-> center_mix_level];
370 float smix = gain_levels[s->surround_mix_level];
372 float downmix_coeffs[2][AC3_MAX_CHANNELS];
374 if (!s->downmix_coeffs[0]) {
375 s->downmix_coeffs[0] = av_malloc_array(2 * AC3_MAX_CHANNELS,
376 sizeof(**s->downmix_coeffs));
377 if (!s->downmix_coeffs[0])
378 return AVERROR(ENOMEM);
379 s->downmix_coeffs[1] = s->downmix_coeffs[0] + AC3_MAX_CHANNELS;
382 for (i = 0; i < s->fbw_channels; i++) {
383 downmix_coeffs[0][i] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
384 downmix_coeffs[1][i] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
386 if (s->channel_mode > 1 && s->channel_mode & 1) {
387 downmix_coeffs[0][1] = downmix_coeffs[1][1] = cmix;
389 if (s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
390 int nf = s->channel_mode - 2;
391 downmix_coeffs[0][nf] = downmix_coeffs[1][nf] = smix * LEVEL_MINUS_3DB;
393 if (s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
394 int nf = s->channel_mode - 4;
395 downmix_coeffs[0][nf] = downmix_coeffs[1][nf+1] = smix;
400 for (i = 0; i < s->fbw_channels; i++) {
401 norm0 += downmix_coeffs[0][i];
402 norm1 += downmix_coeffs[1][i];
404 norm0 = 1.0f / norm0;
405 norm1 = 1.0f / norm1;
406 for (i = 0; i < s->fbw_channels; i++) {
407 downmix_coeffs[0][i] *= norm0;
408 downmix_coeffs[1][i] *= norm1;
411 if (s->output_mode == AC3_CHMODE_MONO) {
412 for (i = 0; i < s->fbw_channels; i++)
413 downmix_coeffs[0][i] = (downmix_coeffs[0][i] +
414 downmix_coeffs[1][i]) * LEVEL_MINUS_3DB;
416 for (i = 0; i < s->fbw_channels; i++) {
417 s->downmix_coeffs[0][i] = FIXR12(downmix_coeffs[0][i]);
418 s->downmix_coeffs[1][i] = FIXR12(downmix_coeffs[1][i]);
425 * Decode the grouped exponents according to exponent strategy.
426 * reference: Section 7.1.3 Exponent Decoding
428 static int decode_exponents(AC3DecodeContext *s,
429 GetBitContext *gbc, int exp_strategy, int ngrps,
430 uint8_t absexp, int8_t *dexps)
432 int i, j, grp, group_size;
437 group_size = exp_strategy + (exp_strategy == EXP_D45);
438 for (grp = 0, i = 0; grp < ngrps; grp++) {
439 expacc = get_bits(gbc, 7);
441 av_log(s->avctx, AV_LOG_ERROR, "expacc %d is out-of-range\n", expacc);
442 return AVERROR_INVALIDDATA;
444 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][0];
445 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][1];
446 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][2];
449 /* convert to absolute exps and expand groups */
451 for (i = 0, j = 0; i < ngrps * 3; i++) {
452 prevexp += dexp[i] - 2;
454 av_log(s->avctx, AV_LOG_ERROR, "exponent %d is out-of-range\n", prevexp);
457 switch (group_size) {
458 case 4: dexps[j++] = prevexp;
459 dexps[j++] = prevexp;
460 case 2: dexps[j++] = prevexp;
461 case 1: dexps[j++] = prevexp;
468 * Generate transform coefficients for each coupled channel in the coupling
469 * range using the coupling coefficients and coupling coordinates.
470 * reference: Section 7.4.3 Coupling Coordinate Format
472 static void calc_transform_coeffs_cpl(AC3DecodeContext *s)
476 bin = s->start_freq[CPL_CH];
477 for (band = 0; band < s->num_cpl_bands; band++) {
478 int band_start = bin;
479 int band_end = bin + s->cpl_band_sizes[band];
480 for (ch = 1; ch <= s->fbw_channels; ch++) {
481 if (s->channel_in_cpl[ch]) {
482 int cpl_coord = s->cpl_coords[ch][band] << 5;
483 for (bin = band_start; bin < band_end; bin++) {
484 s->fixed_coeffs[ch][bin] =
485 MULH(s->fixed_coeffs[CPL_CH][bin] * (1 << 4), cpl_coord);
487 if (ch == 2 && s->phase_flags[band]) {
488 for (bin = band_start; bin < band_end; bin++)
489 s->fixed_coeffs[2][bin] = -s->fixed_coeffs[2][bin];
498 * Grouped mantissas for 3-level 5-level and 11-level quantization
500 typedef struct mant_groups {
510 * Decode the transform coefficients for a particular channel
511 * reference: Section 7.3 Quantization and Decoding of Mantissas
513 static void ac3_decode_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
515 int start_freq = s->start_freq[ch_index];
516 int end_freq = s->end_freq[ch_index];
517 uint8_t *baps = s->bap[ch_index];
518 int8_t *exps = s->dexps[ch_index];
519 int32_t *coeffs = s->fixed_coeffs[ch_index];
520 int dither = (ch_index == CPL_CH) || s->dither_flag[ch_index];
521 GetBitContext *gbc = &s->gbc;
524 for (freq = start_freq; freq < end_freq; freq++) {
525 int bap = baps[freq];
529 /* random noise with approximate range of -0.707 to 0.707 */
531 mantissa = (((av_lfg_get(&s->dith_state)>>8)*181)>>8) - 5931008;
538 mantissa = m->b1_mant[m->b1];
540 int bits = get_bits(gbc, 5);
541 mantissa = b1_mantissas[bits][0];
542 m->b1_mant[1] = b1_mantissas[bits][1];
543 m->b1_mant[0] = b1_mantissas[bits][2];
550 mantissa = m->b2_mant[m->b2];
552 int bits = get_bits(gbc, 7);
553 mantissa = b2_mantissas[bits][0];
554 m->b2_mant[1] = b2_mantissas[bits][1];
555 m->b2_mant[0] = b2_mantissas[bits][2];
560 mantissa = b3_mantissas[get_bits(gbc, 3)];
565 mantissa = m->b4_mant;
567 int bits = get_bits(gbc, 7);
568 mantissa = b4_mantissas[bits][0];
569 m->b4_mant = b4_mantissas[bits][1];
574 mantissa = b5_mantissas[get_bits(gbc, 4)];
576 default: /* 6 to 15 */
577 /* Shift mantissa and sign-extend it. */
579 av_log(s->avctx, AV_LOG_ERROR, "bap %d is invalid in plain AC-3\n", bap);
582 mantissa = (unsigned)get_sbits(gbc, quantization_tab[bap]) << (24 - quantization_tab[bap]);
585 coeffs[freq] = mantissa >> exps[freq];
590 * Remove random dithering from coupling range coefficients with zero-bit
591 * mantissas for coupled channels which do not use dithering.
592 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
594 static void remove_dithering(AC3DecodeContext *s) {
597 for (ch = 1; ch <= s->fbw_channels; ch++) {
598 if (!s->dither_flag[ch] && s->channel_in_cpl[ch]) {
599 for (i = s->start_freq[CPL_CH]; i < s->end_freq[CPL_CH]; i++) {
600 if (!s->bap[CPL_CH][i])
601 s->fixed_coeffs[ch][i] = 0;
607 static inline void decode_transform_coeffs_ch(AC3DecodeContext *s, int blk,
608 int ch, mant_groups *m)
610 if (!s->channel_uses_aht[ch]) {
611 ac3_decode_transform_coeffs_ch(s, ch, m);
613 /* if AHT is used, mantissas for all blocks are encoded in the first
614 block of the frame. */
616 if (CONFIG_EAC3_DECODER && !blk)
617 ff_eac3_decode_transform_coeffs_aht_ch(s, ch);
618 for (bin = s->start_freq[ch]; bin < s->end_freq[ch]; bin++) {
619 s->fixed_coeffs[ch][bin] = s->pre_mantissa[ch][bin][blk] >> s->dexps[ch][bin];
625 * Decode the transform coefficients.
627 static inline void decode_transform_coeffs(AC3DecodeContext *s, int blk)
633 m.b1 = m.b2 = m.b4 = 0;
635 for (ch = 1; ch <= s->channels; ch++) {
636 /* transform coefficients for full-bandwidth channel */
637 decode_transform_coeffs_ch(s, blk, ch, &m);
638 /* transform coefficients for coupling channel come right after the
639 coefficients for the first coupled channel*/
640 if (s->channel_in_cpl[ch]) {
642 decode_transform_coeffs_ch(s, blk, CPL_CH, &m);
643 calc_transform_coeffs_cpl(s);
646 end = s->end_freq[CPL_CH];
648 end = s->end_freq[ch];
651 s->fixed_coeffs[ch][end] = 0;
655 /* zero the dithered coefficients for appropriate channels */
660 * Stereo rematrixing.
661 * reference: Section 7.5.4 Rematrixing : Decoding Technique
663 static void do_rematrixing(AC3DecodeContext *s)
668 end = FFMIN(s->end_freq[1], s->end_freq[2]);
670 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++) {
671 if (s->rematrixing_flags[bnd]) {
672 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd + 1]);
673 for (i = ff_ac3_rematrix_band_tab[bnd]; i < bndend; i++) {
674 int tmp0 = s->fixed_coeffs[1][i];
675 s->fixed_coeffs[1][i] += s->fixed_coeffs[2][i];
676 s->fixed_coeffs[2][i] = tmp0 - s->fixed_coeffs[2][i];
683 * Inverse MDCT Transform.
684 * Convert frequency domain coefficients to time-domain audio samples.
685 * reference: Section 7.9.4 Transformation Equations
687 static inline void do_imdct(AC3DecodeContext *s, int channels, int offset)
691 for (ch = 1; ch <= channels; ch++) {
692 if (s->block_switch[ch]) {
694 FFTSample *x = s->tmp_output + 128;
695 for (i = 0; i < 128; i++)
696 x[i] = s->transform_coeffs[ch][2 * i];
697 s->imdct_256.imdct_half(&s->imdct_256, s->tmp_output, x);
699 s->fdsp->vector_fmul_window_scaled(s->outptr[ch - 1], s->delay[ch - 1 + offset],
700 s->tmp_output, s->window, 128, 8);
702 s->fdsp->vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1 + offset],
703 s->tmp_output, s->window, 128);
705 for (i = 0; i < 128; i++)
706 x[i] = s->transform_coeffs[ch][2 * i + 1];
707 s->imdct_256.imdct_half(&s->imdct_256, s->delay[ch - 1 + offset], x);
709 s->imdct_512.imdct_half(&s->imdct_512, s->tmp_output, s->transform_coeffs[ch]);
711 s->fdsp->vector_fmul_window_scaled(s->outptr[ch - 1], s->delay[ch - 1 + offset],
712 s->tmp_output, s->window, 128, 8);
714 s->fdsp->vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1 + offset],
715 s->tmp_output, s->window, 128);
717 memcpy(s->delay[ch - 1 + offset], s->tmp_output + 128, 128 * sizeof(FFTSample));
723 * Upmix delay samples from stereo to original channel layout.
725 static void ac3_upmix_delay(AC3DecodeContext *s)
727 int channel_data_size = sizeof(s->delay[0]);
728 switch (s->channel_mode) {
729 case AC3_CHMODE_DUALMONO:
730 case AC3_CHMODE_STEREO:
731 /* upmix mono to stereo */
732 memcpy(s->delay[1], s->delay[0], channel_data_size);
734 case AC3_CHMODE_2F2R:
735 memset(s->delay[3], 0, channel_data_size);
736 case AC3_CHMODE_2F1R:
737 memset(s->delay[2], 0, channel_data_size);
739 case AC3_CHMODE_3F2R:
740 memset(s->delay[4], 0, channel_data_size);
741 case AC3_CHMODE_3F1R:
742 memset(s->delay[3], 0, channel_data_size);
744 memcpy(s->delay[2], s->delay[1], channel_data_size);
745 memset(s->delay[1], 0, channel_data_size);
751 * Decode band structure for coupling, spectral extension, or enhanced coupling.
752 * The band structure defines how many subbands are in each band. For each
753 * subband in the range, 1 means it is combined with the previous band, and 0
754 * means that it starts a new band.
756 * @param[in] gbc bit reader context
757 * @param[in] blk block number
758 * @param[in] eac3 flag to indicate E-AC-3
759 * @param[in] ecpl flag to indicate enhanced coupling
760 * @param[in] start_subband subband number for start of range
761 * @param[in] end_subband subband number for end of range
762 * @param[in] default_band_struct default band structure table
763 * @param[out] num_bands number of bands (optionally NULL)
764 * @param[out] band_sizes array containing the number of bins in each band (optionally NULL)
765 * @param[in,out] band_struct current band structure
767 static void decode_band_structure(GetBitContext *gbc, int blk, int eac3,
768 int ecpl, int start_subband, int end_subband,
769 const uint8_t *default_band_struct,
770 int *num_bands, uint8_t *band_sizes,
771 uint8_t *band_struct, int band_struct_size)
773 int subbnd, bnd, n_subbands, n_bands=0;
776 n_subbands = end_subband - start_subband;
779 memcpy(band_struct, default_band_struct, band_struct_size);
781 av_assert0(band_struct_size >= start_subband + n_subbands);
783 band_struct += start_subband + 1;
785 /* decode band structure from bitstream or use default */
786 if (!eac3 || get_bits1(gbc)) {
787 for (subbnd = 0; subbnd < n_subbands - 1; subbnd++) {
788 band_struct[subbnd] = get_bits1(gbc);
792 /* calculate number of bands and band sizes based on band structure.
793 note that the first 4 subbands in enhanced coupling span only 6 bins
795 if (num_bands || band_sizes ) {
796 n_bands = n_subbands;
797 bnd_sz[0] = ecpl ? 6 : 12;
798 for (bnd = 0, subbnd = 1; subbnd < n_subbands; subbnd++) {
799 int subbnd_size = (ecpl && subbnd < 4) ? 6 : 12;
800 if (band_struct[subbnd - 1]) {
802 bnd_sz[bnd] += subbnd_size;
804 bnd_sz[++bnd] = subbnd_size;
809 /* set optional output params */
811 *num_bands = n_bands;
813 memcpy(band_sizes, bnd_sz, n_bands);
816 static inline int spx_strategy(AC3DecodeContext *s, int blk)
818 GetBitContext *bc = &s->gbc;
819 int fbw_channels = s->fbw_channels;
820 int dst_start_freq, dst_end_freq, src_start_freq,
821 start_subband, end_subband, ch;
823 /* determine which channels use spx */
824 if (s->channel_mode == AC3_CHMODE_MONO) {
825 s->channel_uses_spx[1] = 1;
827 for (ch = 1; ch <= fbw_channels; ch++)
828 s->channel_uses_spx[ch] = get_bits1(bc);
831 /* get the frequency bins of the spx copy region and the spx start
833 dst_start_freq = get_bits(bc, 2);
834 start_subband = get_bits(bc, 3) + 2;
835 if (start_subband > 7)
836 start_subband += start_subband - 7;
837 end_subband = get_bits(bc, 3) + 5;
839 s->spx_dst_end_freq = end_freq_inv_tab[end_subband-5];
842 end_subband += end_subband - 7;
843 dst_start_freq = dst_start_freq * 12 + 25;
844 src_start_freq = start_subband * 12 + 25;
845 dst_end_freq = end_subband * 12 + 25;
847 /* check validity of spx ranges */
848 if (start_subband >= end_subband) {
849 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
850 "range (%d >= %d)\n", start_subband, end_subband);
851 return AVERROR_INVALIDDATA;
853 if (dst_start_freq >= src_start_freq) {
854 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
855 "copy start bin (%d >= %d)\n", dst_start_freq, src_start_freq);
856 return AVERROR_INVALIDDATA;
859 s->spx_dst_start_freq = dst_start_freq;
860 s->spx_src_start_freq = src_start_freq;
862 s->spx_dst_end_freq = dst_end_freq;
864 decode_band_structure(bc, blk, s->eac3, 0,
865 start_subband, end_subband,
866 ff_eac3_default_spx_band_struct,
869 s->spx_band_struct, sizeof(s->spx_band_struct));
873 static inline void spx_coordinates(AC3DecodeContext *s)
875 GetBitContext *bc = &s->gbc;
876 int fbw_channels = s->fbw_channels;
879 for (ch = 1; ch <= fbw_channels; ch++) {
880 if (s->channel_uses_spx[ch]) {
881 if (s->first_spx_coords[ch] || get_bits1(bc)) {
883 int bin, master_spx_coord;
885 s->first_spx_coords[ch] = 0;
886 spx_blend = AC3_SPX_BLEND(get_bits(bc, 5));
887 master_spx_coord = get_bits(bc, 2) * 3;
889 bin = s->spx_src_start_freq;
890 for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
891 int bandsize = s->spx_band_sizes[bnd];
892 int spx_coord_exp, spx_coord_mant;
893 INTFLOAT nratio, sblend, nblend;
895 /* calculate blending factors */
896 int64_t accu = ((bin << 23) + (bandsize << 22))
897 * (int64_t)s->spx_dst_end_freq;
898 nratio = (int)(accu >> 32);
899 nratio -= spx_blend << 18;
904 } else if (nratio > 0x7fffff) {
905 nblend = 14529495; // sqrt(3) in FP.23
908 nblend = fixed_sqrt(nratio, 23);
909 accu = (int64_t)nblend * 1859775393;
910 nblend = (int)((accu + (1<<29)) >> 30);
911 sblend = fixed_sqrt(0x800000 - nratio, 23);
916 /* calculate blending factors */
917 nratio = ((float)((bin + (bandsize >> 1))) / s->spx_dst_end_freq) - spx_blend;
918 nratio = av_clipf(nratio, 0.0f, 1.0f);
919 nblend = sqrtf(3.0f * nratio); // noise is scaled by sqrt(3)
920 // to give unity variance
921 sblend = sqrtf(1.0f - nratio);
925 /* decode spx coordinates */
926 spx_coord_exp = get_bits(bc, 4);
927 spx_coord_mant = get_bits(bc, 2);
928 if (spx_coord_exp == 15) spx_coord_mant <<= 1;
929 else spx_coord_mant += 4;
930 spx_coord_mant <<= (25 - spx_coord_exp - master_spx_coord);
932 /* multiply noise and signal blending factors by spx coordinate */
934 accu = (int64_t)nblend * spx_coord_mant;
935 s->spx_noise_blend[ch][bnd] = (int)((accu + (1<<22)) >> 23);
936 accu = (int64_t)sblend * spx_coord_mant;
937 s->spx_signal_blend[ch][bnd] = (int)((accu + (1<<22)) >> 23);
939 spx_coord = spx_coord_mant * (1.0f / (1 << 23));
940 s->spx_noise_blend [ch][bnd] = nblend * spx_coord;
941 s->spx_signal_blend[ch][bnd] = sblend * spx_coord;
946 s->first_spx_coords[ch] = 1;
951 static inline int coupling_strategy(AC3DecodeContext *s, int blk,
952 uint8_t *bit_alloc_stages)
954 GetBitContext *bc = &s->gbc;
955 int fbw_channels = s->fbw_channels;
956 int channel_mode = s->channel_mode;
959 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
961 s->cpl_in_use[blk] = get_bits1(bc);
962 if (s->cpl_in_use[blk]) {
963 /* coupling in use */
964 int cpl_start_subband, cpl_end_subband;
966 if (channel_mode < AC3_CHMODE_STEREO) {
967 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
968 return AVERROR_INVALIDDATA;
971 /* check for enhanced coupling */
972 if (s->eac3 && get_bits1(bc)) {
973 /* TODO: parse enhanced coupling strategy info */
974 avpriv_request_sample(s->avctx, "Enhanced coupling");
975 return AVERROR_PATCHWELCOME;
978 /* determine which channels are coupled */
979 if (s->eac3 && s->channel_mode == AC3_CHMODE_STEREO) {
980 s->channel_in_cpl[1] = 1;
981 s->channel_in_cpl[2] = 1;
983 for (ch = 1; ch <= fbw_channels; ch++)
984 s->channel_in_cpl[ch] = get_bits1(bc);
987 /* phase flags in use */
988 if (channel_mode == AC3_CHMODE_STEREO)
989 s->phase_flags_in_use = get_bits1(bc);
991 /* coupling frequency range */
992 cpl_start_subband = get_bits(bc, 4);
993 cpl_end_subband = s->spx_in_use ? (s->spx_src_start_freq - 37) / 12 :
995 if (cpl_start_subband >= cpl_end_subband) {
996 av_log(s->avctx, AV_LOG_ERROR, "invalid coupling range (%d >= %d)\n",
997 cpl_start_subband, cpl_end_subband);
998 return AVERROR_INVALIDDATA;
1000 s->start_freq[CPL_CH] = cpl_start_subband * 12 + 37;
1001 s->end_freq[CPL_CH] = cpl_end_subband * 12 + 37;
1003 decode_band_structure(bc, blk, s->eac3, 0, cpl_start_subband,
1005 ff_eac3_default_cpl_band_struct,
1006 &s->num_cpl_bands, s->cpl_band_sizes,
1007 s->cpl_band_struct, sizeof(s->cpl_band_struct));
1009 /* coupling not in use */
1010 for (ch = 1; ch <= fbw_channels; ch++) {
1011 s->channel_in_cpl[ch] = 0;
1012 s->first_cpl_coords[ch] = 1;
1014 s->first_cpl_leak = s->eac3;
1015 s->phase_flags_in_use = 0;
1021 static inline int coupling_coordinates(AC3DecodeContext *s, int blk)
1023 GetBitContext *bc = &s->gbc;
1024 int fbw_channels = s->fbw_channels;
1026 int cpl_coords_exist = 0;
1028 for (ch = 1; ch <= fbw_channels; ch++) {
1029 if (s->channel_in_cpl[ch]) {
1030 if ((s->eac3 && s->first_cpl_coords[ch]) || get_bits1(bc)) {
1031 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
1032 s->first_cpl_coords[ch] = 0;
1033 cpl_coords_exist = 1;
1034 master_cpl_coord = 3 * get_bits(bc, 2);
1035 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
1036 cpl_coord_exp = get_bits(bc, 4);
1037 cpl_coord_mant = get_bits(bc, 4);
1038 if (cpl_coord_exp == 15)
1039 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
1041 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
1042 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
1045 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must "
1046 "be present in block 0\n");
1047 return AVERROR_INVALIDDATA;
1050 /* channel not in coupling */
1051 s->first_cpl_coords[ch] = 1;
1055 if (s->channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
1056 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
1057 s->phase_flags[bnd] = s->phase_flags_in_use ? get_bits1(bc) : 0;
1065 * Decode a single audio block from the AC-3 bitstream.
1067 static int decode_audio_block(AC3DecodeContext *s, int blk, int offset)
1069 int fbw_channels = s->fbw_channels;
1070 int channel_mode = s->channel_mode;
1071 int i, bnd, seg, ch, ret;
1072 int different_transforms;
1075 GetBitContext *gbc = &s->gbc;
1076 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS] = { 0 };
1078 /* block switch flags */
1079 different_transforms = 0;
1080 if (s->block_switch_syntax) {
1081 for (ch = 1; ch <= fbw_channels; ch++) {
1082 s->block_switch[ch] = get_bits1(gbc);
1083 if (ch > 1 && s->block_switch[ch] != s->block_switch[1])
1084 different_transforms = 1;
1088 /* dithering flags */
1089 if (s->dither_flag_syntax) {
1090 for (ch = 1; ch <= fbw_channels; ch++) {
1091 s->dither_flag[ch] = get_bits1(gbc);
1096 i = !s->channel_mode;
1098 if (get_bits1(gbc)) {
1099 /* Allow asymmetric application of DRC when drc_scale > 1.
1100 Amplification of quiet sounds is enhanced */
1101 int range_bits = get_bits(gbc, 8);
1102 INTFLOAT range = AC3_RANGE(range_bits);
1103 if (range_bits <= 127 || s->drc_scale <= 1.0)
1104 s->dynamic_range[i] = AC3_DYNAMIC_RANGE(range);
1106 s->dynamic_range[i] = range;
1107 } else if (blk == 0) {
1108 s->dynamic_range[i] = AC3_DYNAMIC_RANGE1;
1112 /* spectral extension strategy */
1113 if (s->eac3 && (!blk || get_bits1(gbc))) {
1114 s->spx_in_use = get_bits1(gbc);
1115 if (s->spx_in_use) {
1116 if ((ret = spx_strategy(s, blk)) < 0)
1120 if (!s->eac3 || !s->spx_in_use) {
1122 for (ch = 1; ch <= fbw_channels; ch++) {
1123 s->channel_uses_spx[ch] = 0;
1124 s->first_spx_coords[ch] = 1;
1128 /* spectral extension coordinates */
1132 /* coupling strategy */
1133 if (s->eac3 ? s->cpl_strategy_exists[blk] : get_bits1(gbc)) {
1134 if ((ret = coupling_strategy(s, blk, bit_alloc_stages)) < 0)
1136 } else if (!s->eac3) {
1138 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must "
1139 "be present in block 0\n");
1140 return AVERROR_INVALIDDATA;
1142 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
1145 cpl_in_use = s->cpl_in_use[blk];
1147 /* coupling coordinates */
1149 if ((ret = coupling_coordinates(s, blk)) < 0)
1153 /* stereo rematrixing strategy and band structure */
1154 if (channel_mode == AC3_CHMODE_STEREO) {
1155 if ((s->eac3 && !blk) || get_bits1(gbc)) {
1156 s->num_rematrixing_bands = 4;
1157 if (cpl_in_use && s->start_freq[CPL_CH] <= 61) {
1158 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
1159 } else if (s->spx_in_use && s->spx_src_start_freq <= 61) {
1160 s->num_rematrixing_bands--;
1162 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++)
1163 s->rematrixing_flags[bnd] = get_bits1(gbc);
1165 av_log(s->avctx, AV_LOG_WARNING, "Warning: "
1166 "new rematrixing strategy not present in block 0\n");
1167 s->num_rematrixing_bands = 0;
1171 /* exponent strategies for each channel */
1172 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1174 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
1175 if (s->exp_strategy[blk][ch] != EXP_REUSE)
1176 bit_alloc_stages[ch] = 3;
1179 /* channel bandwidth */
1180 for (ch = 1; ch <= fbw_channels; ch++) {
1181 s->start_freq[ch] = 0;
1182 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1184 int prev = s->end_freq[ch];
1185 if (s->channel_in_cpl[ch])
1186 s->end_freq[ch] = s->start_freq[CPL_CH];
1187 else if (s->channel_uses_spx[ch])
1188 s->end_freq[ch] = s->spx_src_start_freq;
1190 int bandwidth_code = get_bits(gbc, 6);
1191 if (bandwidth_code > 60) {
1192 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60\n", bandwidth_code);
1193 return AVERROR_INVALIDDATA;
1195 s->end_freq[ch] = bandwidth_code * 3 + 73;
1197 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
1198 s->num_exp_groups[ch] = (s->end_freq[ch] + group_size-4) / group_size;
1199 if (blk > 0 && s->end_freq[ch] != prev)
1200 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
1203 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
1204 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
1205 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
1208 /* decode exponents for each channel */
1209 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1210 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1211 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
1212 if (decode_exponents(s, gbc, s->exp_strategy[blk][ch],
1213 s->num_exp_groups[ch], s->dexps[ch][0],
1214 &s->dexps[ch][s->start_freq[ch]+!!ch])) {
1215 return AVERROR_INVALIDDATA;
1217 if (ch != CPL_CH && ch != s->lfe_ch)
1218 skip_bits(gbc, 2); /* skip gainrng */
1222 /* bit allocation information */
1223 if (s->bit_allocation_syntax) {
1224 if (get_bits1(gbc)) {
1225 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1226 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1227 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
1228 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
1229 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
1230 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1231 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1233 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must "
1234 "be present in block 0\n");
1235 return AVERROR_INVALIDDATA;
1239 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
1240 if (!s->eac3 || !blk) {
1241 if (s->snr_offset_strategy && get_bits1(gbc)) {
1244 csnr = (get_bits(gbc, 6) - 15) << 4;
1245 for (i = ch = !cpl_in_use; ch <= s->channels; ch++) {
1247 if (ch == i || s->snr_offset_strategy == 2)
1248 snr = (csnr + get_bits(gbc, 4)) << 2;
1249 /* run at least last bit allocation stage if snr offset changes */
1250 if (blk && s->snr_offset[ch] != snr) {
1251 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 1);
1253 s->snr_offset[ch] = snr;
1255 /* fast gain (normal AC-3 only) */
1257 int prev = s->fast_gain[ch];
1258 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1259 /* run last 2 bit allocation stages if fast gain changes */
1260 if (blk && prev != s->fast_gain[ch])
1261 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1264 } else if (!s->eac3 && !blk) {
1265 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
1266 return AVERROR_INVALIDDATA;
1270 /* fast gain (E-AC-3 only) */
1271 if (s->fast_gain_syntax && get_bits1(gbc)) {
1272 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1273 int prev = s->fast_gain[ch];
1274 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1275 /* run last 2 bit allocation stages if fast gain changes */
1276 if (blk && prev != s->fast_gain[ch])
1277 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1279 } else if (s->eac3 && !blk) {
1280 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1281 s->fast_gain[ch] = ff_ac3_fast_gain_tab[4];
1284 /* E-AC-3 to AC-3 converter SNR offset */
1285 if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT && get_bits1(gbc)) {
1286 skip_bits(gbc, 10); // skip converter snr offset
1289 /* coupling leak information */
1291 if (s->first_cpl_leak || get_bits1(gbc)) {
1292 int fl = get_bits(gbc, 3);
1293 int sl = get_bits(gbc, 3);
1294 /* run last 2 bit allocation stages for coupling channel if
1295 coupling leak changes */
1296 if (blk && (fl != s->bit_alloc_params.cpl_fast_leak ||
1297 sl != s->bit_alloc_params.cpl_slow_leak)) {
1298 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
1300 s->bit_alloc_params.cpl_fast_leak = fl;
1301 s->bit_alloc_params.cpl_slow_leak = sl;
1302 } else if (!s->eac3 && !blk) {
1303 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must "
1304 "be present in block 0\n");
1305 return AVERROR_INVALIDDATA;
1307 s->first_cpl_leak = 0;
1310 /* delta bit allocation information */
1311 if (s->dba_syntax && get_bits1(gbc)) {
1312 /* delta bit allocation exists (strategy) */
1313 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1314 s->dba_mode[ch] = get_bits(gbc, 2);
1315 if (s->dba_mode[ch] == DBA_RESERVED) {
1316 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1317 return AVERROR_INVALIDDATA;
1319 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1321 /* channel delta offset, len and bit allocation */
1322 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1323 if (s->dba_mode[ch] == DBA_NEW) {
1324 s->dba_nsegs[ch] = get_bits(gbc, 3) + 1;
1325 for (seg = 0; seg < s->dba_nsegs[ch]; seg++) {
1326 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1327 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1328 s->dba_values[ch][seg] = get_bits(gbc, 3);
1330 /* run last 2 bit allocation stages if new dba values */
1331 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1334 } else if (blk == 0) {
1335 for (ch = 0; ch <= s->channels; ch++) {
1336 s->dba_mode[ch] = DBA_NONE;
1340 /* Bit allocation */
1341 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1342 if (bit_alloc_stages[ch] > 2) {
1343 /* Exponent mapping into PSD and PSD integration */
1344 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1345 s->start_freq[ch], s->end_freq[ch],
1346 s->psd[ch], s->band_psd[ch]);
1348 if (bit_alloc_stages[ch] > 1) {
1349 /* Compute excitation function, Compute masking curve, and
1350 Apply delta bit allocation */
1351 if (ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1352 s->start_freq[ch], s->end_freq[ch],
1353 s->fast_gain[ch], (ch == s->lfe_ch),
1354 s->dba_mode[ch], s->dba_nsegs[ch],
1355 s->dba_offsets[ch], s->dba_lengths[ch],
1356 s->dba_values[ch], s->mask[ch])) {
1357 av_log(s->avctx, AV_LOG_ERROR, "error in bit allocation\n");
1358 return AVERROR_INVALIDDATA;
1361 if (bit_alloc_stages[ch] > 0) {
1362 /* Compute bit allocation */
1363 const uint8_t *bap_tab = s->channel_uses_aht[ch] ?
1364 ff_eac3_hebap_tab : ff_ac3_bap_tab;
1365 s->ac3dsp.bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1366 s->start_freq[ch], s->end_freq[ch],
1368 s->bit_alloc_params.floor,
1369 bap_tab, s->bap[ch]);
1373 /* unused dummy data */
1374 if (s->skip_syntax && get_bits1(gbc)) {
1375 int skipl = get_bits(gbc, 9);
1376 skip_bits_long(gbc, 8 * skipl);
1379 /* unpack the transform coefficients
1380 this also uncouples channels if coupling is in use. */
1381 decode_transform_coeffs(s, blk);
1383 /* TODO: generate enhanced coupling coordinates and uncouple */
1385 /* recover coefficients if rematrixing is in use */
1386 if (s->channel_mode == AC3_CHMODE_STEREO)
1389 /* apply scaling to coefficients (headroom, dynrng) */
1390 for (ch = 1; ch <= s->channels; ch++) {
1391 int audio_channel = 0;
1393 if (s->channel_mode == AC3_CHMODE_DUALMONO && ch <= 2)
1394 audio_channel = 2-ch;
1395 if (s->heavy_compression && s->compression_exists[audio_channel])
1396 gain = s->heavy_dynamic_range[audio_channel];
1398 gain = s->dynamic_range[audio_channel];
1401 scale_coefs(s->transform_coeffs[ch], s->fixed_coeffs[ch], gain, 256);
1403 if (s->target_level != 0)
1404 gain = gain * s->level_gain[audio_channel];
1405 gain *= 1.0 / 4194304.0f;
1406 s->fmt_conv.int32_to_float_fmul_scalar(s->transform_coeffs[ch],
1407 s->fixed_coeffs[ch], gain, 256);
1411 /* apply spectral extension to high frequency bins */
1412 if (CONFIG_EAC3_DECODER && s->spx_in_use) {
1413 ff_eac3_apply_spectral_extension(s);
1416 /* downmix and MDCT. order depends on whether block switching is used for
1417 any channel in this block. this is because coefficients for the long
1418 and short transforms cannot be mixed. */
1419 downmix_output = s->channels != s->out_channels &&
1420 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1421 s->fbw_channels == s->out_channels);
1422 if (different_transforms) {
1423 /* the delay samples have already been downmixed, so we upmix the delay
1424 samples in order to reconstruct all channels before downmixing. */
1430 do_imdct(s, s->channels, offset);
1432 if (downmix_output) {
1434 ac3_downmix_c_fixed16(s->outptr, s->downmix_coeffs,
1435 s->out_channels, s->fbw_channels, 256);
1437 ff_ac3dsp_downmix(&s->ac3dsp, s->outptr, s->downmix_coeffs,
1438 s->out_channels, s->fbw_channels, 256);
1442 if (downmix_output) {
1443 AC3_RENAME(ff_ac3dsp_downmix)(&s->ac3dsp, s->xcfptr + 1, s->downmix_coeffs,
1444 s->out_channels, s->fbw_channels, 256);
1447 if (downmix_output && !s->downmixed) {
1449 AC3_RENAME(ff_ac3dsp_downmix)(&s->ac3dsp, s->dlyptr, s->downmix_coeffs,
1450 s->out_channels, s->fbw_channels, 128);
1453 do_imdct(s, s->out_channels, offset);
1460 * Decode a single AC-3 frame.
1462 static int ac3_decode_frame(AVCodecContext * avctx, void *data,
1463 int *got_frame_ptr, AVPacket *avpkt)
1465 AVFrame *frame = data;
1466 const uint8_t *buf = avpkt->data;
1467 int buf_size, full_buf_size = avpkt->size;
1468 AC3DecodeContext *s = avctx->priv_data;
1469 int blk, ch, err, offset, ret;
1470 int got_independent_frame = 0;
1471 const uint8_t *channel_map;
1472 uint8_t extended_channel_map[EAC3_MAX_CHANNELS];
1473 const SHORTFLOAT *output[AC3_MAX_CHANNELS];
1474 enum AVMatrixEncoding matrix_encoding;
1475 AVDownmixInfo *downmix_info;
1477 s->superframe_size = 0;
1479 buf_size = full_buf_size;
1480 /* copy input buffer to decoder context to avoid reading past the end
1481 of the buffer, which can be caused by a damaged input stream. */
1482 if (buf_size >= 2 && AV_RB16(buf) == 0x770B) {
1483 // seems to be byte-swapped AC-3
1484 int cnt = FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE) >> 1;
1485 s->bdsp.bswap16_buf((uint16_t *) s->input_buffer,
1486 (const uint16_t *) buf, cnt);
1488 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE));
1490 /* if consistent noise generation is enabled, seed the linear feedback generator
1491 * with the contents of the AC-3 frame so that the noise is identical across
1492 * decodes given the same AC-3 frame data, for use with non-linear edititing software. */
1493 if (s->consistent_noise_generation)
1494 av_lfg_init_from_data(&s->dith_state, s->input_buffer, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE));
1496 buf = s->input_buffer;
1498 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1499 if ((ret = init_get_bits8(&s->gbc, buf, buf_size)) < 0)
1502 /* parse the syncinfo */
1503 err = parse_frame_header(s);
1507 case AAC_AC3_PARSE_ERROR_SYNC:
1508 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1509 return AVERROR_INVALIDDATA;
1510 case AAC_AC3_PARSE_ERROR_BSID:
1511 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1513 case AAC_AC3_PARSE_ERROR_SAMPLE_RATE:
1514 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1516 case AAC_AC3_PARSE_ERROR_FRAME_SIZE:
1517 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1519 case AAC_AC3_PARSE_ERROR_FRAME_TYPE:
1520 /* skip frame if CRC is ok. otherwise use error concealment. */
1521 /* TODO: add support for substreams */
1522 if (s->substreamid) {
1523 av_log(avctx, AV_LOG_DEBUG,
1524 "unsupported substream %d: skipping frame\n",
1529 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1532 case AAC_AC3_PARSE_ERROR_CRC:
1533 case AAC_AC3_PARSE_ERROR_CHANNEL_CFG:
1535 default: // Normal AVERROR do not try to recover.
1540 /* check that reported frame size fits in input buffer */
1541 if (s->frame_size > buf_size) {
1542 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1543 err = AAC_AC3_PARSE_ERROR_FRAME_SIZE;
1544 } else if (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL)) {
1545 /* check for crc mismatch */
1546 if (av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2],
1547 s->frame_size - 2)) {
1548 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1549 if (avctx->err_recognition & AV_EF_EXPLODE)
1550 return AVERROR_INVALIDDATA;
1551 err = AAC_AC3_PARSE_ERROR_CRC;
1556 if (s->frame_type == EAC3_FRAME_TYPE_DEPENDENT && !got_independent_frame) {
1557 av_log(avctx, AV_LOG_WARNING, "Ignoring dependent frame without independent frame.\n");
1559 return FFMIN(full_buf_size, s->frame_size);
1562 /* channel config */
1563 if (!err || (s->channels && s->out_channels != s->channels)) {
1564 s->out_channels = s->channels;
1565 s->output_mode = s->channel_mode;
1567 s->output_mode |= AC3_OUTPUT_LFEON;
1568 if (s->channels > 1 &&
1569 avctx->request_channel_layout == AV_CH_LAYOUT_MONO) {
1570 s->out_channels = 1;
1571 s->output_mode = AC3_CHMODE_MONO;
1572 } else if (s->channels > 2 &&
1573 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
1574 s->out_channels = 2;
1575 s->output_mode = AC3_CHMODE_STEREO;
1578 s->loro_center_mix_level = gain_levels[s-> center_mix_level];
1579 s->loro_surround_mix_level = gain_levels[s->surround_mix_level];
1580 s->ltrt_center_mix_level = LEVEL_MINUS_3DB;
1581 s->ltrt_surround_mix_level = LEVEL_MINUS_3DB;
1582 /* set downmixing coefficients if needed */
1583 if (s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1584 s->fbw_channels == s->out_channels)) {
1585 if ((ret = set_downmix_coeffs(s)) < 0) {
1586 av_log(avctx, AV_LOG_ERROR, "error setting downmix coeffs\n");
1590 } else if (!s->channels) {
1591 av_log(avctx, AV_LOG_ERROR, "unable to determine channel mode\n");
1592 return AVERROR_INVALIDDATA;
1594 avctx->channels = s->out_channels;
1595 avctx->channel_layout = avpriv_ac3_channel_layout_tab[s->output_mode & ~AC3_OUTPUT_LFEON];
1596 if (s->output_mode & AC3_OUTPUT_LFEON)
1597 avctx->channel_layout |= AV_CH_LOW_FREQUENCY;
1599 /* set audio service type based on bitstream mode for AC-3 */
1600 avctx->audio_service_type = s->bitstream_mode;
1601 if (s->bitstream_mode == 0x7 && s->channels > 1)
1602 avctx->audio_service_type = AV_AUDIO_SERVICE_TYPE_KARAOKE;
1604 /* decode the audio blocks */
1605 channel_map = ff_ac3_dec_channel_map[s->output_mode & ~AC3_OUTPUT_LFEON][s->lfe_on];
1606 offset = s->frame_type == EAC3_FRAME_TYPE_DEPENDENT ? AC3_MAX_CHANNELS : 0;
1607 for (ch = 0; ch < AC3_MAX_CHANNELS; ch++) {
1608 output[ch] = s->output[ch + offset];
1609 s->outptr[ch] = s->output[ch + offset];
1611 for (ch = 0; ch < s->channels; ch++) {
1612 if (ch < s->out_channels)
1613 s->outptr[channel_map[ch]] = s->output_buffer[ch + offset];
1615 for (blk = 0; blk < s->num_blocks; blk++) {
1616 if (!err && decode_audio_block(s, blk, offset)) {
1617 av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n");
1621 for (ch = 0; ch < s->out_channels; ch++)
1622 memcpy(s->output_buffer[ch + offset] + AC3_BLOCK_SIZE*blk, output[ch], AC3_BLOCK_SIZE*sizeof(SHORTFLOAT));
1623 for (ch = 0; ch < s->out_channels; ch++)
1624 output[ch] = s->outptr[channel_map[ch]];
1625 for (ch = 0; ch < s->out_channels; ch++) {
1626 if (!ch || channel_map[ch])
1627 s->outptr[channel_map[ch]] += AC3_BLOCK_SIZE;
1631 /* keep last block for error concealment in next frame */
1632 for (ch = 0; ch < s->out_channels; ch++)
1633 memcpy(s->output[ch + offset], output[ch], AC3_BLOCK_SIZE*sizeof(SHORTFLOAT));
1635 /* check if there is dependent frame */
1636 if (buf_size > s->frame_size) {
1640 if ((ret = init_get_bits8(&s->gbc, buf + s->frame_size, buf_size - s->frame_size)) < 0)
1643 err = ff_ac3_parse_header(&s->gbc, &hdr);
1647 if (hdr.frame_type == EAC3_FRAME_TYPE_DEPENDENT) {
1648 if (hdr.num_blocks != s->num_blocks || s->sample_rate != hdr.sample_rate) {
1649 av_log(avctx, AV_LOG_WARNING, "Ignoring non-compatible dependent frame.\n");
1651 buf += s->frame_size;
1652 buf_size -= s->frame_size;
1653 s->prev_output_mode = s->output_mode;
1654 s->prev_bit_rate = s->bit_rate;
1655 got_independent_frame = 1;
1656 goto dependent_frame;
1661 frame->decode_error_flags = err ? FF_DECODE_ERROR_INVALID_BITSTREAM : 0;
1663 /* if frame is ok, set audio parameters */
1665 avctx->sample_rate = s->sample_rate;
1666 avctx->bit_rate = s->bit_rate + s->prev_bit_rate;
1669 for (ch = 0; ch < EAC3_MAX_CHANNELS; ch++)
1670 extended_channel_map[ch] = ch;
1672 if (s->frame_type == EAC3_FRAME_TYPE_DEPENDENT) {
1673 uint64_t ich_layout = avpriv_ac3_channel_layout_tab[s->prev_output_mode & ~AC3_OUTPUT_LFEON];
1674 int channel_map_size = ff_ac3_channels_tab[s->output_mode & ~AC3_OUTPUT_LFEON] + s->lfe_on;
1675 uint64_t channel_layout;
1678 if (s->prev_output_mode & AC3_OUTPUT_LFEON)
1679 ich_layout |= AV_CH_LOW_FREQUENCY;
1681 channel_layout = ich_layout;
1682 for (ch = 0; ch < 16; ch++) {
1683 if (s->channel_map & (1 << (EAC3_MAX_CHANNELS - ch - 1))) {
1684 channel_layout |= ff_eac3_custom_channel_map_locations[ch][1];
1687 if (av_get_channel_layout_nb_channels(channel_layout) > EAC3_MAX_CHANNELS) {
1688 av_log(avctx, AV_LOG_ERROR, "Too many channels (%d) coded\n",
1689 av_get_channel_layout_nb_channels(channel_layout));
1690 return AVERROR_INVALIDDATA;
1693 avctx->channel_layout = channel_layout;
1694 avctx->channels = av_get_channel_layout_nb_channels(channel_layout);
1696 for (ch = 0; ch < EAC3_MAX_CHANNELS; ch++) {
1697 if (s->channel_map & (1 << (EAC3_MAX_CHANNELS - ch - 1))) {
1698 if (ff_eac3_custom_channel_map_locations[ch][0]) {
1699 int index = av_get_channel_layout_channel_index(channel_layout,
1700 ff_eac3_custom_channel_map_locations[ch][1]);
1702 return AVERROR_INVALIDDATA;
1703 if (extend >= channel_map_size)
1704 return AVERROR_INVALIDDATA;
1706 extended_channel_map[index] = offset + channel_map[extend++];
1710 for (i = 0; i < 64; i++) {
1711 if ((1ULL << i) & ff_eac3_custom_channel_map_locations[ch][1]) {
1712 int index = av_get_channel_layout_channel_index(channel_layout,
1715 return AVERROR_INVALIDDATA;
1716 if (extend >= channel_map_size)
1717 return AVERROR_INVALIDDATA;
1719 extended_channel_map[index] = offset + channel_map[extend++];
1727 /* get output buffer */
1728 frame->nb_samples = s->num_blocks * AC3_BLOCK_SIZE;
1729 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1732 for (ch = 0; ch < avctx->channels; ch++) {
1733 int map = extended_channel_map[ch];
1734 av_assert0(ch>=AV_NUM_DATA_POINTERS || frame->extended_data[ch] == frame->data[ch]);
1735 memcpy((SHORTFLOAT *)frame->extended_data[ch],
1736 s->output_buffer[map],
1737 s->num_blocks * AC3_BLOCK_SIZE * sizeof(SHORTFLOAT));
1743 * Check whether the input layout is compatible, and make sure we're not
1744 * downmixing (else the matrix encoding is no longer applicable).
1746 matrix_encoding = AV_MATRIX_ENCODING_NONE;
1747 if (s->channel_mode == AC3_CHMODE_STEREO &&
1748 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1749 if (s->dolby_surround_mode == AC3_DSURMOD_ON)
1750 matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
1751 else if (s->dolby_headphone_mode == AC3_DHEADPHONMOD_ON)
1752 matrix_encoding = AV_MATRIX_ENCODING_DOLBYHEADPHONE;
1753 } else if (s->channel_mode >= AC3_CHMODE_2F2R &&
1754 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1755 switch (s->dolby_surround_ex_mode) {
1756 case AC3_DSUREXMOD_ON: // EX or PLIIx
1757 matrix_encoding = AV_MATRIX_ENCODING_DOLBYEX;
1759 case AC3_DSUREXMOD_PLIIZ:
1760 matrix_encoding = AV_MATRIX_ENCODING_DPLIIZ;
1762 default: // not indicated or off
1766 if ((ret = ff_side_data_update_matrix_encoding(frame, matrix_encoding)) < 0)
1770 if ((downmix_info = av_downmix_info_update_side_data(frame))) {
1771 switch (s->preferred_downmix) {
1772 case AC3_DMIXMOD_LTRT:
1773 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LTRT;
1775 case AC3_DMIXMOD_LORO:
1776 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_LORO;
1778 case AC3_DMIXMOD_DPLII:
1779 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_DPLII;
1782 downmix_info->preferred_downmix_type = AV_DOWNMIX_TYPE_UNKNOWN;
1785 downmix_info->center_mix_level = gain_levels[s-> center_mix_level];
1786 downmix_info->center_mix_level_ltrt = gain_levels[s-> center_mix_level_ltrt];
1787 downmix_info->surround_mix_level = gain_levels[s-> surround_mix_level];
1788 downmix_info->surround_mix_level_ltrt = gain_levels[s->surround_mix_level_ltrt];
1789 if (s->lfe_mix_level_exists)
1790 downmix_info->lfe_mix_level = gain_levels_lfe[s->lfe_mix_level];
1792 downmix_info->lfe_mix_level = 0.0; // -inf dB
1794 return AVERROR(ENOMEM);
1798 if (!s->superframe_size)
1799 return FFMIN(full_buf_size, s->frame_size);
1801 return FFMIN(full_buf_size, s->superframe_size);
1805 * Uninitialize the AC-3 decoder.
1807 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1809 AC3DecodeContext *s = avctx->priv_data;
1810 ff_mdct_end(&s->imdct_512);
1811 ff_mdct_end(&s->imdct_256);
1813 av_freep(&s->downmix_coeffs[0]);
1818 #define OFFSET(x) offsetof(AC3DecodeContext, x)
1819 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)