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 Libav.
12 * Libav 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 * Libav 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 Libav; 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/opt.h"
36 #include "aac_ac3_parser.h"
37 #include "ac3_parser.h"
39 #include "ac3dec_data.h"
43 * table for ungrouping 3 values in 7 bits.
44 * used for exponents and bap=2 mantissas
46 static uint8_t ungroup_3_in_7_bits_tab[128][3];
48 /** tables for ungrouping mantissas */
49 static int b1_mantissas[32][3];
50 static int b2_mantissas[128][3];
51 static int b3_mantissas[8];
52 static int b4_mantissas[128][2];
53 static int b5_mantissas[16];
56 * Quantization table: levels for symmetric. bits for asymmetric.
57 * reference: Table 7.18 Mapping of bap to Quantizer
59 static const uint8_t quantization_tab[16] = {
61 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
64 /** dynamic range table. converts codes to scale factors. */
65 static float dynamic_range_tab[256];
67 /** Adjustments in dB gain */
68 static const float gain_levels[9] = {
72 LEVEL_MINUS_1POINT5DB,
74 LEVEL_MINUS_4POINT5DB,
81 * Table for default stereo downmixing coefficients
82 * reference: Section 7.8.2 Downmixing Into Two Channels
84 static const uint8_t ac3_default_coeffs[8][5][2] = {
85 { { 2, 7 }, { 7, 2 }, },
87 { { 2, 7 }, { 7, 2 }, },
88 { { 2, 7 }, { 5, 5 }, { 7, 2 }, },
89 { { 2, 7 }, { 7, 2 }, { 6, 6 }, },
90 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 8, 8 }, },
91 { { 2, 7 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
92 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
96 * Symmetrical Dequantization
97 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
101 symmetric_dequant(int code, int levels)
103 return ((code - (levels >> 1)) << 24) / levels;
107 * Initialize tables at runtime.
109 static av_cold void ac3_tables_init(void)
113 /* generate table for ungrouping 3 values in 7 bits
114 reference: Section 7.1.3 Exponent Decoding */
115 for (i = 0; i < 128; i++) {
116 ungroup_3_in_7_bits_tab[i][0] = i / 25;
117 ungroup_3_in_7_bits_tab[i][1] = (i % 25) / 5;
118 ungroup_3_in_7_bits_tab[i][2] = (i % 25) % 5;
121 /* generate grouped mantissa tables
122 reference: Section 7.3.5 Ungrouping of Mantissas */
123 for (i = 0; i < 32; i++) {
124 /* bap=1 mantissas */
125 b1_mantissas[i][0] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][0], 3);
126 b1_mantissas[i][1] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][1], 3);
127 b1_mantissas[i][2] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][2], 3);
129 for (i = 0; i < 128; i++) {
130 /* bap=2 mantissas */
131 b2_mantissas[i][0] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][0], 5);
132 b2_mantissas[i][1] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][1], 5);
133 b2_mantissas[i][2] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][2], 5);
135 /* bap=4 mantissas */
136 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
137 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
139 /* generate ungrouped mantissa tables
140 reference: Tables 7.21 and 7.23 */
141 for (i = 0; i < 7; i++) {
142 /* bap=3 mantissas */
143 b3_mantissas[i] = symmetric_dequant(i, 7);
145 for (i = 0; i < 15; i++) {
146 /* bap=5 mantissas */
147 b5_mantissas[i] = symmetric_dequant(i, 15);
150 /* generate dynamic range table
151 reference: Section 7.7.1 Dynamic Range Control */
152 for (i = 0; i < 256; i++) {
153 int v = (i >> 5) - ((i >> 7) << 3) - 5;
154 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
159 * AVCodec initialization
161 static av_cold int ac3_decode_init(AVCodecContext *avctx)
163 AC3DecodeContext *s = avctx->priv_data;
168 ff_ac3_common_init();
170 ff_mdct_init(&s->imdct_256, 8, 1, 1.0);
171 ff_mdct_init(&s->imdct_512, 9, 1, 1.0);
172 ff_kbd_window_init(s->window, 5.0, 256);
173 ff_dsputil_init(&s->dsp, avctx);
174 avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
175 ff_ac3dsp_init(&s->ac3dsp, avctx->flags & CODEC_FLAG_BITEXACT);
176 ff_fmt_convert_init(&s->fmt_conv, avctx);
177 av_lfg_init(&s->dith_state, 0);
179 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
181 /* allow downmixing to stereo or mono */
182 #if FF_API_REQUEST_CHANNELS
183 FF_DISABLE_DEPRECATION_WARNINGS
184 if (avctx->request_channels == 1)
185 avctx->request_channel_layout = AV_CH_LAYOUT_MONO;
186 else if (avctx->request_channels == 2)
187 avctx->request_channel_layout = AV_CH_LAYOUT_STEREO;
188 FF_ENABLE_DEPRECATION_WARNINGS
190 if (avctx->channels > 1 &&
191 avctx->request_channel_layout == AV_CH_LAYOUT_MONO)
193 else if (avctx->channels > 2 &&
194 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO)
198 for (i = 0; i < AC3_MAX_CHANNELS; i++) {
199 s->xcfptr[i] = s->transform_coeffs[i];
200 s->dlyptr[i] = s->delay[i];
207 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
208 * GetBitContext within AC3DecodeContext must point to
209 * the start of the synchronized AC-3 bitstream.
211 static int ac3_parse_header(AC3DecodeContext *s)
213 GetBitContext *gbc = &s->gbc;
216 /* read the rest of the bsi. read twice for dual mono mode. */
217 i = !s->channel_mode;
219 skip_bits(gbc, 5); // skip dialog normalization
221 skip_bits(gbc, 8); //skip compression
223 skip_bits(gbc, 8); //skip language code
225 skip_bits(gbc, 7); //skip audio production information
228 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
230 /* default dolby matrix encoding modes */
231 s->dolby_surround_ex_mode = AC3_DSUREXMOD_NOTINDICATED;
232 s->dolby_headphone_mode = AC3_DHEADPHONMOD_NOTINDICATED;
234 /* skip the timecodes or parse the Alternate Bit Stream Syntax
235 TODO: read & use the xbsi1 downmix levels */
236 if (s->bitstream_id != 6) {
238 skip_bits(gbc, 14); //skip timecode1
240 skip_bits(gbc, 14); //skip timecode2
243 skip_bits(gbc, 14); //skip xbsi1
244 if (get_bits1(gbc)) {
245 s->dolby_surround_ex_mode = get_bits(gbc, 2);
246 s->dolby_headphone_mode = get_bits(gbc, 2);
247 skip_bits(gbc, 10); // skip adconvtyp (1), xbsi2 (8), encinfo (1)
251 /* skip additional bitstream info */
252 if (get_bits1(gbc)) {
253 i = get_bits(gbc, 6);
263 * Common function to parse AC-3 or E-AC-3 frame header
265 static int parse_frame_header(AC3DecodeContext *s)
270 err = avpriv_ac3_parse_header(&s->gbc, &hdr);
274 /* get decoding parameters from header info */
275 s->bit_alloc_params.sr_code = hdr.sr_code;
276 s->bitstream_id = hdr.bitstream_id;
277 s->bitstream_mode = hdr.bitstream_mode;
278 s->channel_mode = hdr.channel_mode;
279 s->lfe_on = hdr.lfe_on;
280 s->bit_alloc_params.sr_shift = hdr.sr_shift;
281 s->sample_rate = hdr.sample_rate;
282 s->bit_rate = hdr.bit_rate;
283 s->channels = hdr.channels;
284 s->fbw_channels = s->channels - s->lfe_on;
285 s->lfe_ch = s->fbw_channels + 1;
286 s->frame_size = hdr.frame_size;
287 s->center_mix_level = hdr.center_mix_level;
288 s->surround_mix_level = hdr.surround_mix_level;
289 s->num_blocks = hdr.num_blocks;
290 s->frame_type = hdr.frame_type;
291 s->substreamid = hdr.substreamid;
292 s->dolby_surround_mode = hdr.dolby_surround_mode;
295 s->start_freq[s->lfe_ch] = 0;
296 s->end_freq[s->lfe_ch] = 7;
297 s->num_exp_groups[s->lfe_ch] = 2;
298 s->channel_in_cpl[s->lfe_ch] = 0;
301 if (s->bitstream_id <= 10) {
303 s->snr_offset_strategy = 2;
304 s->block_switch_syntax = 1;
305 s->dither_flag_syntax = 1;
306 s->bit_allocation_syntax = 1;
307 s->fast_gain_syntax = 0;
308 s->first_cpl_leak = 0;
311 memset(s->channel_uses_aht, 0, sizeof(s->channel_uses_aht));
312 return ac3_parse_header(s);
313 } else if (CONFIG_EAC3_DECODER) {
315 return ff_eac3_parse_header(s);
317 av_log(s->avctx, AV_LOG_ERROR, "E-AC-3 support not compiled in\n");
318 return AVERROR(ENOSYS);
323 * Set stereo downmixing coefficients based on frame header info.
324 * reference: Section 7.8.2 Downmixing Into Two Channels
326 static void set_downmix_coeffs(AC3DecodeContext *s)
329 float cmix = gain_levels[s-> center_mix_level];
330 float smix = gain_levels[s->surround_mix_level];
333 for (i = 0; i < s->fbw_channels; i++) {
334 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
335 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
337 if (s->channel_mode > 1 && s->channel_mode & 1) {
338 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
340 if (s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
341 int nf = s->channel_mode - 2;
342 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
344 if (s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
345 int nf = s->channel_mode - 4;
346 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
351 for (i = 0; i < s->fbw_channels; i++) {
352 norm0 += s->downmix_coeffs[i][0];
353 norm1 += s->downmix_coeffs[i][1];
355 norm0 = 1.0f / norm0;
356 norm1 = 1.0f / norm1;
357 for (i = 0; i < s->fbw_channels; i++) {
358 s->downmix_coeffs[i][0] *= norm0;
359 s->downmix_coeffs[i][1] *= norm1;
362 if (s->output_mode == AC3_CHMODE_MONO) {
363 for (i = 0; i < s->fbw_channels; i++)
364 s->downmix_coeffs[i][0] = (s->downmix_coeffs[i][0] +
365 s->downmix_coeffs[i][1]) * LEVEL_MINUS_3DB;
370 * Decode the grouped exponents according to exponent strategy.
371 * reference: Section 7.1.3 Exponent Decoding
373 static int decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
374 uint8_t absexp, int8_t *dexps)
376 int i, j, grp, group_size;
381 group_size = exp_strategy + (exp_strategy == EXP_D45);
382 for (grp = 0, i = 0; grp < ngrps; grp++) {
383 expacc = get_bits(gbc, 7);
384 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][0];
385 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][1];
386 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][2];
389 /* convert to absolute exps and expand groups */
391 for (i = 0, j = 0; i < ngrps * 3; i++) {
392 prevexp += dexp[i] - 2;
395 switch (group_size) {
396 case 4: dexps[j++] = prevexp;
397 dexps[j++] = prevexp;
398 case 2: dexps[j++] = prevexp;
399 case 1: dexps[j++] = prevexp;
406 * Generate transform coefficients for each coupled channel in the coupling
407 * range using the coupling coefficients and coupling coordinates.
408 * reference: Section 7.4.3 Coupling Coordinate Format
410 static void calc_transform_coeffs_cpl(AC3DecodeContext *s)
414 bin = s->start_freq[CPL_CH];
415 for (band = 0; band < s->num_cpl_bands; band++) {
416 int band_start = bin;
417 int band_end = bin + s->cpl_band_sizes[band];
418 for (ch = 1; ch <= s->fbw_channels; ch++) {
419 if (s->channel_in_cpl[ch]) {
420 int cpl_coord = s->cpl_coords[ch][band] << 5;
421 for (bin = band_start; bin < band_end; bin++) {
422 s->fixed_coeffs[ch][bin] =
423 MULH(s->fixed_coeffs[CPL_CH][bin] << 4, cpl_coord);
425 if (ch == 2 && s->phase_flags[band]) {
426 for (bin = band_start; bin < band_end; bin++)
427 s->fixed_coeffs[2][bin] = -s->fixed_coeffs[2][bin];
436 * Grouped mantissas for 3-level 5-level and 11-level quantization
448 * Decode the transform coefficients for a particular channel
449 * reference: Section 7.3 Quantization and Decoding of Mantissas
451 static void ac3_decode_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
453 int start_freq = s->start_freq[ch_index];
454 int end_freq = s->end_freq[ch_index];
455 uint8_t *baps = s->bap[ch_index];
456 int8_t *exps = s->dexps[ch_index];
457 int32_t *coeffs = s->fixed_coeffs[ch_index];
458 int dither = (ch_index == CPL_CH) || s->dither_flag[ch_index];
459 GetBitContext *gbc = &s->gbc;
462 for (freq = start_freq; freq < end_freq; freq++) {
463 int bap = baps[freq];
467 /* random noise with approximate range of -0.707 to 0.707 */
469 mantissa = (av_lfg_get(&s->dith_state) / 362) - 5932275;
476 mantissa = m->b1_mant[m->b1];
478 int bits = get_bits(gbc, 5);
479 mantissa = b1_mantissas[bits][0];
480 m->b1_mant[1] = b1_mantissas[bits][1];
481 m->b1_mant[0] = b1_mantissas[bits][2];
488 mantissa = m->b2_mant[m->b2];
490 int bits = get_bits(gbc, 7);
491 mantissa = b2_mantissas[bits][0];
492 m->b2_mant[1] = b2_mantissas[bits][1];
493 m->b2_mant[0] = b2_mantissas[bits][2];
498 mantissa = b3_mantissas[get_bits(gbc, 3)];
503 mantissa = m->b4_mant;
505 int bits = get_bits(gbc, 7);
506 mantissa = b4_mantissas[bits][0];
507 m->b4_mant = b4_mantissas[bits][1];
512 mantissa = b5_mantissas[get_bits(gbc, 4)];
514 default: /* 6 to 15 */
515 /* Shift mantissa and sign-extend it. */
516 mantissa = get_sbits(gbc, quantization_tab[bap]);
517 mantissa <<= 24 - quantization_tab[bap];
520 coeffs[freq] = mantissa >> exps[freq];
525 * Remove random dithering from coupling range coefficients with zero-bit
526 * mantissas for coupled channels which do not use dithering.
527 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
529 static void remove_dithering(AC3DecodeContext *s) {
532 for (ch = 1; ch <= s->fbw_channels; ch++) {
533 if (!s->dither_flag[ch] && s->channel_in_cpl[ch]) {
534 for (i = s->start_freq[CPL_CH]; i < s->end_freq[CPL_CH]; i++) {
535 if (!s->bap[CPL_CH][i])
536 s->fixed_coeffs[ch][i] = 0;
542 static void decode_transform_coeffs_ch(AC3DecodeContext *s, int blk, int ch,
545 if (!s->channel_uses_aht[ch]) {
546 ac3_decode_transform_coeffs_ch(s, ch, m);
548 /* if AHT is used, mantissas for all blocks are encoded in the first
549 block of the frame. */
551 if (!blk && CONFIG_EAC3_DECODER)
552 ff_eac3_decode_transform_coeffs_aht_ch(s, ch);
553 for (bin = s->start_freq[ch]; bin < s->end_freq[ch]; bin++) {
554 s->fixed_coeffs[ch][bin] = s->pre_mantissa[ch][bin][blk] >> s->dexps[ch][bin];
560 * Decode the transform coefficients.
562 static void decode_transform_coeffs(AC3DecodeContext *s, int blk)
568 m.b1 = m.b2 = m.b4 = 0;
570 for (ch = 1; ch <= s->channels; ch++) {
571 /* transform coefficients for full-bandwidth channel */
572 decode_transform_coeffs_ch(s, blk, ch, &m);
573 /* transform coefficients for coupling channel come right after the
574 coefficients for the first coupled channel*/
575 if (s->channel_in_cpl[ch]) {
577 decode_transform_coeffs_ch(s, blk, CPL_CH, &m);
578 calc_transform_coeffs_cpl(s);
581 end = s->end_freq[CPL_CH];
583 end = s->end_freq[ch];
586 s->fixed_coeffs[ch][end] = 0;
590 /* zero the dithered coefficients for appropriate channels */
595 * Stereo rematrixing.
596 * reference: Section 7.5.4 Rematrixing : Decoding Technique
598 static void do_rematrixing(AC3DecodeContext *s)
603 end = FFMIN(s->end_freq[1], s->end_freq[2]);
605 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++) {
606 if (s->rematrixing_flags[bnd]) {
607 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd + 1]);
608 for (i = ff_ac3_rematrix_band_tab[bnd]; i < bndend; i++) {
609 int tmp0 = s->fixed_coeffs[1][i];
610 s->fixed_coeffs[1][i] += s->fixed_coeffs[2][i];
611 s->fixed_coeffs[2][i] = tmp0 - s->fixed_coeffs[2][i];
618 * Inverse MDCT Transform.
619 * Convert frequency domain coefficients to time-domain audio samples.
620 * reference: Section 7.9.4 Transformation Equations
622 static inline void do_imdct(AC3DecodeContext *s, int channels)
626 for (ch = 1; ch <= channels; ch++) {
627 if (s->block_switch[ch]) {
629 float *x = s->tmp_output + 128;
630 for (i = 0; i < 128; i++)
631 x[i] = s->transform_coeffs[ch][2 * i];
632 s->imdct_256.imdct_half(&s->imdct_256, s->tmp_output, x);
633 s->fdsp.vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1],
634 s->tmp_output, s->window, 128);
635 for (i = 0; i < 128; i++)
636 x[i] = s->transform_coeffs[ch][2 * i + 1];
637 s->imdct_256.imdct_half(&s->imdct_256, s->delay[ch - 1], x);
639 s->imdct_512.imdct_half(&s->imdct_512, s->tmp_output, s->transform_coeffs[ch]);
640 s->fdsp.vector_fmul_window(s->outptr[ch - 1], s->delay[ch - 1],
641 s->tmp_output, s->window, 128);
642 memcpy(s->delay[ch - 1], s->tmp_output + 128, 128 * sizeof(float));
648 * Upmix delay samples from stereo to original channel layout.
650 static void ac3_upmix_delay(AC3DecodeContext *s)
652 int channel_data_size = sizeof(s->delay[0]);
653 switch (s->channel_mode) {
654 case AC3_CHMODE_DUALMONO:
655 case AC3_CHMODE_STEREO:
656 /* upmix mono to stereo */
657 memcpy(s->delay[1], s->delay[0], channel_data_size);
659 case AC3_CHMODE_2F2R:
660 memset(s->delay[3], 0, channel_data_size);
661 case AC3_CHMODE_2F1R:
662 memset(s->delay[2], 0, channel_data_size);
664 case AC3_CHMODE_3F2R:
665 memset(s->delay[4], 0, channel_data_size);
666 case AC3_CHMODE_3F1R:
667 memset(s->delay[3], 0, channel_data_size);
669 memcpy(s->delay[2], s->delay[1], channel_data_size);
670 memset(s->delay[1], 0, channel_data_size);
676 * Decode band structure for coupling, spectral extension, or enhanced coupling.
677 * The band structure defines how many subbands are in each band. For each
678 * subband in the range, 1 means it is combined with the previous band, and 0
679 * means that it starts a new band.
681 * @param[in] gbc bit reader context
682 * @param[in] blk block number
683 * @param[in] eac3 flag to indicate E-AC-3
684 * @param[in] ecpl flag to indicate enhanced coupling
685 * @param[in] start_subband subband number for start of range
686 * @param[in] end_subband subband number for end of range
687 * @param[in] default_band_struct default band structure table
688 * @param[out] num_bands number of bands (optionally NULL)
689 * @param[out] band_sizes array containing the number of bins in each band (optionally NULL)
691 static void decode_band_structure(GetBitContext *gbc, int blk, int eac3,
692 int ecpl, int start_subband, int end_subband,
693 const uint8_t *default_band_struct,
694 int *num_bands, uint8_t *band_sizes)
696 int subbnd, bnd, n_subbands, n_bands=0;
698 uint8_t coded_band_struct[22];
699 const uint8_t *band_struct;
701 n_subbands = end_subband - start_subband;
703 /* decode band structure from bitstream or use default */
704 if (!eac3 || get_bits1(gbc)) {
705 for (subbnd = 0; subbnd < n_subbands - 1; subbnd++) {
706 coded_band_struct[subbnd] = get_bits1(gbc);
708 band_struct = coded_band_struct;
710 band_struct = &default_band_struct[start_subband+1];
712 /* no change in band structure */
716 /* calculate number of bands and band sizes based on band structure.
717 note that the first 4 subbands in enhanced coupling span only 6 bins
719 if (num_bands || band_sizes ) {
720 n_bands = n_subbands;
721 bnd_sz[0] = ecpl ? 6 : 12;
722 for (bnd = 0, subbnd = 1; subbnd < n_subbands; subbnd++) {
723 int subbnd_size = (ecpl && subbnd < 4) ? 6 : 12;
724 if (band_struct[subbnd - 1]) {
726 bnd_sz[bnd] += subbnd_size;
728 bnd_sz[++bnd] = subbnd_size;
733 /* set optional output params */
735 *num_bands = n_bands;
737 memcpy(band_sizes, bnd_sz, n_bands);
741 * Decode a single audio block from the AC-3 bitstream.
743 static int decode_audio_block(AC3DecodeContext *s, int blk)
745 int fbw_channels = s->fbw_channels;
746 int channel_mode = s->channel_mode;
748 int different_transforms;
751 GetBitContext *gbc = &s->gbc;
752 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS] = { 0 };
754 /* block switch flags */
755 different_transforms = 0;
756 if (s->block_switch_syntax) {
757 for (ch = 1; ch <= fbw_channels; ch++) {
758 s->block_switch[ch] = get_bits1(gbc);
759 if (ch > 1 && s->block_switch[ch] != s->block_switch[1])
760 different_transforms = 1;
764 /* dithering flags */
765 if (s->dither_flag_syntax) {
766 for (ch = 1; ch <= fbw_channels; ch++) {
767 s->dither_flag[ch] = get_bits1(gbc);
772 i = !s->channel_mode;
774 if (get_bits1(gbc)) {
775 s->dynamic_range[i] = powf(dynamic_range_tab[get_bits(gbc, 8)],
777 } else if (blk == 0) {
778 s->dynamic_range[i] = 1.0f;
782 /* spectral extension strategy */
783 if (s->eac3 && (!blk || get_bits1(gbc))) {
784 s->spx_in_use = get_bits1(gbc);
786 int dst_start_freq, dst_end_freq, src_start_freq,
787 start_subband, end_subband;
789 /* determine which channels use spx */
790 if (s->channel_mode == AC3_CHMODE_MONO) {
791 s->channel_uses_spx[1] = 1;
793 for (ch = 1; ch <= fbw_channels; ch++)
794 s->channel_uses_spx[ch] = get_bits1(gbc);
797 /* get the frequency bins of the spx copy region and the spx start
799 dst_start_freq = get_bits(gbc, 2);
800 start_subband = get_bits(gbc, 3) + 2;
801 if (start_subband > 7)
802 start_subband += start_subband - 7;
803 end_subband = get_bits(gbc, 3) + 5;
805 end_subband += end_subband - 7;
806 dst_start_freq = dst_start_freq * 12 + 25;
807 src_start_freq = start_subband * 12 + 25;
808 dst_end_freq = end_subband * 12 + 25;
810 /* check validity of spx ranges */
811 if (start_subband >= end_subband) {
812 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
813 "range (%d >= %d)\n", start_subband, end_subband);
814 return AVERROR_INVALIDDATA;
816 if (dst_start_freq >= src_start_freq) {
817 av_log(s->avctx, AV_LOG_ERROR, "invalid spectral extension "
818 "copy start bin (%d >= %d)\n", dst_start_freq, src_start_freq);
819 return AVERROR_INVALIDDATA;
822 s->spx_dst_start_freq = dst_start_freq;
823 s->spx_src_start_freq = src_start_freq;
824 s->spx_dst_end_freq = dst_end_freq;
826 decode_band_structure(gbc, blk, s->eac3, 0,
827 start_subband, end_subband,
828 ff_eac3_default_spx_band_struct,
832 for (ch = 1; ch <= fbw_channels; ch++) {
833 s->channel_uses_spx[ch] = 0;
834 s->first_spx_coords[ch] = 1;
839 /* spectral extension coordinates */
841 for (ch = 1; ch <= fbw_channels; ch++) {
842 if (s->channel_uses_spx[ch]) {
843 if (s->first_spx_coords[ch] || get_bits1(gbc)) {
845 int bin, master_spx_coord;
847 s->first_spx_coords[ch] = 0;
848 spx_blend = get_bits(gbc, 5) * (1.0f/32);
849 master_spx_coord = get_bits(gbc, 2) * 3;
851 bin = s->spx_src_start_freq;
852 for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
854 int spx_coord_exp, spx_coord_mant;
855 float nratio, sblend, nblend, spx_coord;
857 /* calculate blending factors */
858 bandsize = s->spx_band_sizes[bnd];
859 nratio = ((float)((bin + (bandsize >> 1))) / s->spx_dst_end_freq) - spx_blend;
860 nratio = av_clipf(nratio, 0.0f, 1.0f);
861 nblend = sqrtf(3.0f * nratio); // noise is scaled by sqrt(3)
862 // to give unity variance
863 sblend = sqrtf(1.0f - nratio);
866 /* decode spx coordinates */
867 spx_coord_exp = get_bits(gbc, 4);
868 spx_coord_mant = get_bits(gbc, 2);
869 if (spx_coord_exp == 15) spx_coord_mant <<= 1;
870 else spx_coord_mant += 4;
871 spx_coord_mant <<= (25 - spx_coord_exp - master_spx_coord);
872 spx_coord = spx_coord_mant * (1.0f / (1 << 23));
874 /* multiply noise and signal blending factors by spx coordinate */
875 s->spx_noise_blend [ch][bnd] = nblend * spx_coord;
876 s->spx_signal_blend[ch][bnd] = sblend * spx_coord;
880 s->first_spx_coords[ch] = 1;
885 /* coupling strategy */
886 if (s->eac3 ? s->cpl_strategy_exists[blk] : get_bits1(gbc)) {
887 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
889 s->cpl_in_use[blk] = get_bits1(gbc);
890 if (s->cpl_in_use[blk]) {
891 /* coupling in use */
892 int cpl_start_subband, cpl_end_subband;
894 if (channel_mode < AC3_CHMODE_STEREO) {
895 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
896 return AVERROR_INVALIDDATA;
899 /* check for enhanced coupling */
900 if (s->eac3 && get_bits1(gbc)) {
901 /* TODO: parse enhanced coupling strategy info */
902 avpriv_request_sample(s->avctx, "Enhanced coupling");
903 return AVERROR_PATCHWELCOME;
906 /* determine which channels are coupled */
907 if (s->eac3 && s->channel_mode == AC3_CHMODE_STEREO) {
908 s->channel_in_cpl[1] = 1;
909 s->channel_in_cpl[2] = 1;
911 for (ch = 1; ch <= fbw_channels; ch++)
912 s->channel_in_cpl[ch] = get_bits1(gbc);
915 /* phase flags in use */
916 if (channel_mode == AC3_CHMODE_STEREO)
917 s->phase_flags_in_use = get_bits1(gbc);
919 /* coupling frequency range */
920 cpl_start_subband = get_bits(gbc, 4);
921 cpl_end_subband = s->spx_in_use ? (s->spx_src_start_freq - 37) / 12 :
922 get_bits(gbc, 4) + 3;
923 if (cpl_start_subband >= cpl_end_subband) {
924 av_log(s->avctx, AV_LOG_ERROR, "invalid coupling range (%d >= %d)\n",
925 cpl_start_subband, cpl_end_subband);
926 return AVERROR_INVALIDDATA;
928 s->start_freq[CPL_CH] = cpl_start_subband * 12 + 37;
929 s->end_freq[CPL_CH] = cpl_end_subband * 12 + 37;
931 decode_band_structure(gbc, blk, s->eac3, 0, cpl_start_subband,
933 ff_eac3_default_cpl_band_struct,
934 &s->num_cpl_bands, s->cpl_band_sizes);
936 /* coupling not in use */
937 for (ch = 1; ch <= fbw_channels; ch++) {
938 s->channel_in_cpl[ch] = 0;
939 s->first_cpl_coords[ch] = 1;
941 s->first_cpl_leak = s->eac3;
942 s->phase_flags_in_use = 0;
944 } else if (!s->eac3) {
946 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must "
947 "be present in block 0\n");
948 return AVERROR_INVALIDDATA;
950 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
953 cpl_in_use = s->cpl_in_use[blk];
955 /* coupling coordinates */
957 int cpl_coords_exist = 0;
959 for (ch = 1; ch <= fbw_channels; ch++) {
960 if (s->channel_in_cpl[ch]) {
961 if ((s->eac3 && s->first_cpl_coords[ch]) || get_bits1(gbc)) {
962 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
963 s->first_cpl_coords[ch] = 0;
964 cpl_coords_exist = 1;
965 master_cpl_coord = 3 * get_bits(gbc, 2);
966 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
967 cpl_coord_exp = get_bits(gbc, 4);
968 cpl_coord_mant = get_bits(gbc, 4);
969 if (cpl_coord_exp == 15)
970 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
972 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
973 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
976 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must "
977 "be present in block 0\n");
978 return AVERROR_INVALIDDATA;
981 /* channel not in coupling */
982 s->first_cpl_coords[ch] = 1;
986 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
987 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
988 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
993 /* stereo rematrixing strategy and band structure */
994 if (channel_mode == AC3_CHMODE_STEREO) {
995 if ((s->eac3 && !blk) || get_bits1(gbc)) {
996 s->num_rematrixing_bands = 4;
997 if (cpl_in_use && s->start_freq[CPL_CH] <= 61) {
998 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
999 } else if (s->spx_in_use && s->spx_src_start_freq <= 61) {
1000 s->num_rematrixing_bands--;
1002 for (bnd = 0; bnd < s->num_rematrixing_bands; bnd++)
1003 s->rematrixing_flags[bnd] = get_bits1(gbc);
1005 av_log(s->avctx, AV_LOG_WARNING, "Warning: "
1006 "new rematrixing strategy not present in block 0\n");
1007 s->num_rematrixing_bands = 0;
1011 /* exponent strategies for each channel */
1012 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1014 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
1015 if (s->exp_strategy[blk][ch] != EXP_REUSE)
1016 bit_alloc_stages[ch] = 3;
1019 /* channel bandwidth */
1020 for (ch = 1; ch <= fbw_channels; ch++) {
1021 s->start_freq[ch] = 0;
1022 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1024 int prev = s->end_freq[ch];
1025 if (s->channel_in_cpl[ch])
1026 s->end_freq[ch] = s->start_freq[CPL_CH];
1027 else if (s->channel_uses_spx[ch])
1028 s->end_freq[ch] = s->spx_src_start_freq;
1030 int bandwidth_code = get_bits(gbc, 6);
1031 if (bandwidth_code > 60) {
1032 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60\n", bandwidth_code);
1033 return AVERROR_INVALIDDATA;
1035 s->end_freq[ch] = bandwidth_code * 3 + 73;
1037 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
1038 s->num_exp_groups[ch] = (s->end_freq[ch] + group_size-4) / group_size;
1039 if (blk > 0 && s->end_freq[ch] != prev)
1040 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
1043 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
1044 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
1045 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
1048 /* decode exponents for each channel */
1049 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1050 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
1051 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
1052 if (decode_exponents(gbc, s->exp_strategy[blk][ch],
1053 s->num_exp_groups[ch], s->dexps[ch][0],
1054 &s->dexps[ch][s->start_freq[ch]+!!ch])) {
1055 av_log(s->avctx, AV_LOG_ERROR, "exponent out-of-range\n");
1056 return AVERROR_INVALIDDATA;
1058 if (ch != CPL_CH && ch != s->lfe_ch)
1059 skip_bits(gbc, 2); /* skip gainrng */
1063 /* bit allocation information */
1064 if (s->bit_allocation_syntax) {
1065 if (get_bits1(gbc)) {
1066 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1067 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1068 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
1069 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
1070 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
1071 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1072 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1074 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must "
1075 "be present in block 0\n");
1076 return AVERROR_INVALIDDATA;
1080 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
1081 if (!s->eac3 || !blk) {
1082 if (s->snr_offset_strategy && get_bits1(gbc)) {
1085 csnr = (get_bits(gbc, 6) - 15) << 4;
1086 for (i = ch = !cpl_in_use; ch <= s->channels; ch++) {
1088 if (ch == i || s->snr_offset_strategy == 2)
1089 snr = (csnr + get_bits(gbc, 4)) << 2;
1090 /* run at least last bit allocation stage if snr offset changes */
1091 if (blk && s->snr_offset[ch] != snr) {
1092 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 1);
1094 s->snr_offset[ch] = snr;
1096 /* fast gain (normal AC-3 only) */
1098 int prev = s->fast_gain[ch];
1099 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1100 /* run last 2 bit allocation stages if fast gain changes */
1101 if (blk && prev != s->fast_gain[ch])
1102 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1105 } else if (!s->eac3 && !blk) {
1106 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
1107 return AVERROR_INVALIDDATA;
1111 /* fast gain (E-AC-3 only) */
1112 if (s->fast_gain_syntax && get_bits1(gbc)) {
1113 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1114 int prev = s->fast_gain[ch];
1115 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1116 /* run last 2 bit allocation stages if fast gain changes */
1117 if (blk && prev != s->fast_gain[ch])
1118 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1120 } else if (s->eac3 && !blk) {
1121 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1122 s->fast_gain[ch] = ff_ac3_fast_gain_tab[4];
1125 /* E-AC-3 to AC-3 converter SNR offset */
1126 if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT && get_bits1(gbc)) {
1127 skip_bits(gbc, 10); // skip converter snr offset
1130 /* coupling leak information */
1132 if (s->first_cpl_leak || get_bits1(gbc)) {
1133 int fl = get_bits(gbc, 3);
1134 int sl = get_bits(gbc, 3);
1135 /* run last 2 bit allocation stages for coupling channel if
1136 coupling leak changes */
1137 if (blk && (fl != s->bit_alloc_params.cpl_fast_leak ||
1138 sl != s->bit_alloc_params.cpl_slow_leak)) {
1139 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
1141 s->bit_alloc_params.cpl_fast_leak = fl;
1142 s->bit_alloc_params.cpl_slow_leak = sl;
1143 } else if (!s->eac3 && !blk) {
1144 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must "
1145 "be present in block 0\n");
1146 return AVERROR_INVALIDDATA;
1148 s->first_cpl_leak = 0;
1151 /* delta bit allocation information */
1152 if (s->dba_syntax && get_bits1(gbc)) {
1153 /* delta bit allocation exists (strategy) */
1154 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1155 s->dba_mode[ch] = get_bits(gbc, 2);
1156 if (s->dba_mode[ch] == DBA_RESERVED) {
1157 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1158 return AVERROR_INVALIDDATA;
1160 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1162 /* channel delta offset, len and bit allocation */
1163 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1164 if (s->dba_mode[ch] == DBA_NEW) {
1165 s->dba_nsegs[ch] = get_bits(gbc, 3) + 1;
1166 for (seg = 0; seg < s->dba_nsegs[ch]; seg++) {
1167 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1168 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1169 s->dba_values[ch][seg] = get_bits(gbc, 3);
1171 /* run last 2 bit allocation stages if new dba values */
1172 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1175 } else if (blk == 0) {
1176 for (ch = 0; ch <= s->channels; ch++) {
1177 s->dba_mode[ch] = DBA_NONE;
1181 /* Bit allocation */
1182 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1183 if (bit_alloc_stages[ch] > 2) {
1184 /* Exponent mapping into PSD and PSD integration */
1185 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1186 s->start_freq[ch], s->end_freq[ch],
1187 s->psd[ch], s->band_psd[ch]);
1189 if (bit_alloc_stages[ch] > 1) {
1190 /* Compute excitation function, Compute masking curve, and
1191 Apply delta bit allocation */
1192 if (ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1193 s->start_freq[ch], s->end_freq[ch],
1194 s->fast_gain[ch], (ch == s->lfe_ch),
1195 s->dba_mode[ch], s->dba_nsegs[ch],
1196 s->dba_offsets[ch], s->dba_lengths[ch],
1197 s->dba_values[ch], s->mask[ch])) {
1198 av_log(s->avctx, AV_LOG_ERROR, "error in bit allocation\n");
1199 return AVERROR_INVALIDDATA;
1202 if (bit_alloc_stages[ch] > 0) {
1203 /* Compute bit allocation */
1204 const uint8_t *bap_tab = s->channel_uses_aht[ch] ?
1205 ff_eac3_hebap_tab : ff_ac3_bap_tab;
1206 s->ac3dsp.bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1207 s->start_freq[ch], s->end_freq[ch],
1209 s->bit_alloc_params.floor,
1210 bap_tab, s->bap[ch]);
1214 /* unused dummy data */
1215 if (s->skip_syntax && get_bits1(gbc)) {
1216 int skipl = get_bits(gbc, 9);
1221 /* unpack the transform coefficients
1222 this also uncouples channels if coupling is in use. */
1223 decode_transform_coeffs(s, blk);
1225 /* TODO: generate enhanced coupling coordinates and uncouple */
1227 /* recover coefficients if rematrixing is in use */
1228 if (s->channel_mode == AC3_CHMODE_STEREO)
1231 /* apply scaling to coefficients (headroom, dynrng) */
1232 for (ch = 1; ch <= s->channels; ch++) {
1233 float gain = 1.0 / 4194304.0f;
1234 if (s->channel_mode == AC3_CHMODE_DUALMONO) {
1235 gain *= s->dynamic_range[2 - ch];
1237 gain *= s->dynamic_range[0];
1239 s->fmt_conv.int32_to_float_fmul_scalar(s->transform_coeffs[ch],
1240 s->fixed_coeffs[ch], gain, 256);
1243 /* apply spectral extension to high frequency bins */
1244 if (s->spx_in_use && CONFIG_EAC3_DECODER) {
1245 ff_eac3_apply_spectral_extension(s);
1248 /* downmix and MDCT. order depends on whether block switching is used for
1249 any channel in this block. this is because coefficients for the long
1250 and short transforms cannot be mixed. */
1251 downmix_output = s->channels != s->out_channels &&
1252 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1253 s->fbw_channels == s->out_channels);
1254 if (different_transforms) {
1255 /* the delay samples have already been downmixed, so we upmix the delay
1256 samples in order to reconstruct all channels before downmixing. */
1262 do_imdct(s, s->channels);
1264 if (downmix_output) {
1265 s->ac3dsp.downmix(s->outptr, s->downmix_coeffs,
1266 s->out_channels, s->fbw_channels, 256);
1269 if (downmix_output) {
1270 s->ac3dsp.downmix(s->xcfptr + 1, s->downmix_coeffs,
1271 s->out_channels, s->fbw_channels, 256);
1274 if (downmix_output && !s->downmixed) {
1276 s->ac3dsp.downmix(s->dlyptr, s->downmix_coeffs, s->out_channels,
1277 s->fbw_channels, 128);
1280 do_imdct(s, s->out_channels);
1287 * Decode a single AC-3 frame.
1289 static int ac3_decode_frame(AVCodecContext * avctx, void *data,
1290 int *got_frame_ptr, AVPacket *avpkt)
1292 AVFrame *frame = data;
1293 const uint8_t *buf = avpkt->data;
1294 int buf_size = avpkt->size;
1295 AC3DecodeContext *s = avctx->priv_data;
1296 int blk, ch, err, ret;
1297 const uint8_t *channel_map;
1298 const float *output[AC3_MAX_CHANNELS];
1299 enum AVMatrixEncoding matrix_encoding;
1301 /* copy input buffer to decoder context to avoid reading past the end
1302 of the buffer, which can be caused by a damaged input stream. */
1303 if (buf_size >= 2 && AV_RB16(buf) == 0x770B) {
1304 // seems to be byte-swapped AC-3
1305 int cnt = FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE) >> 1;
1306 s->dsp.bswap16_buf((uint16_t *)s->input_buffer, (const uint16_t *)buf, cnt);
1308 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE));
1309 buf = s->input_buffer;
1310 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1311 init_get_bits(&s->gbc, buf, buf_size * 8);
1313 /* parse the syncinfo */
1314 err = parse_frame_header(s);
1318 case AAC_AC3_PARSE_ERROR_SYNC:
1319 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1320 return AVERROR_INVALIDDATA;
1321 case AAC_AC3_PARSE_ERROR_BSID:
1322 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1324 case AAC_AC3_PARSE_ERROR_SAMPLE_RATE:
1325 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1327 case AAC_AC3_PARSE_ERROR_FRAME_SIZE:
1328 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1330 case AAC_AC3_PARSE_ERROR_FRAME_TYPE:
1331 /* skip frame if CRC is ok. otherwise use error concealment. */
1332 /* TODO: add support for substreams and dependent frames */
1333 if (s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1334 av_log(avctx, AV_LOG_WARNING, "unsupported frame type : "
1335 "skipping frame\n");
1339 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1342 case AAC_AC3_PARSE_ERROR_CRC:
1343 case AAC_AC3_PARSE_ERROR_CHANNEL_CFG:
1345 default: // Normal AVERROR do not try to recover.
1350 /* check that reported frame size fits in input buffer */
1351 if (s->frame_size > buf_size) {
1352 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1353 err = AAC_AC3_PARSE_ERROR_FRAME_SIZE;
1354 } else if (avctx->err_recognition & AV_EF_CRCCHECK) {
1355 /* check for crc mismatch */
1356 if (av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2],
1357 s->frame_size - 2)) {
1358 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1359 if (avctx->err_recognition & AV_EF_EXPLODE)
1360 return AVERROR_INVALIDDATA;
1361 err = AAC_AC3_PARSE_ERROR_CRC;
1366 /* if frame is ok, set audio parameters */
1368 avctx->sample_rate = s->sample_rate;
1369 avctx->bit_rate = s->bit_rate;
1372 /* channel config */
1373 if (!err || (s->channels && s->out_channels != s->channels)) {
1374 s->out_channels = s->channels;
1375 s->output_mode = s->channel_mode;
1377 s->output_mode |= AC3_OUTPUT_LFEON;
1378 if (s->channels > 1 &&
1379 avctx->request_channel_layout == AV_CH_LAYOUT_MONO) {
1380 s->out_channels = 1;
1381 s->output_mode = AC3_CHMODE_MONO;
1382 } else if (s->channels > 2 &&
1383 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
1384 s->out_channels = 2;
1385 s->output_mode = AC3_CHMODE_STEREO;
1388 /* set downmixing coefficients if needed */
1389 if (s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1390 s->fbw_channels == s->out_channels)) {
1391 set_downmix_coeffs(s);
1393 } else if (!s->channels) {
1394 av_log(avctx, AV_LOG_ERROR, "unable to determine channel mode\n");
1395 return AVERROR_INVALIDDATA;
1397 avctx->channels = s->out_channels;
1398 avctx->channel_layout = avpriv_ac3_channel_layout_tab[s->output_mode & ~AC3_OUTPUT_LFEON];
1399 if (s->output_mode & AC3_OUTPUT_LFEON)
1400 avctx->channel_layout |= AV_CH_LOW_FREQUENCY;
1402 /* set audio service type based on bitstream mode for AC-3 */
1403 avctx->audio_service_type = s->bitstream_mode;
1404 if (s->bitstream_mode == 0x7 && s->channels > 1)
1405 avctx->audio_service_type = AV_AUDIO_SERVICE_TYPE_KARAOKE;
1407 /* get output buffer */
1408 frame->nb_samples = s->num_blocks * AC3_BLOCK_SIZE;
1409 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1410 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1414 /* decode the audio blocks */
1415 channel_map = ff_ac3_dec_channel_map[s->output_mode & ~AC3_OUTPUT_LFEON][s->lfe_on];
1416 for (ch = 0; ch < s->channels; ch++) {
1417 if (ch < s->out_channels)
1418 s->outptr[channel_map[ch]] = (float *)frame->data[ch];
1420 s->outptr[ch] = s->output[ch];
1421 output[ch] = s->output[ch];
1423 for (blk = 0; blk < s->num_blocks; blk++) {
1424 if (!err && decode_audio_block(s, blk)) {
1425 av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n");
1429 for (ch = 0; ch < s->out_channels; ch++)
1430 memcpy(s->outptr[channel_map[ch]], output[ch], sizeof(**output) * AC3_BLOCK_SIZE);
1431 for (ch = 0; ch < s->out_channels; ch++)
1432 output[ch] = s->outptr[channel_map[ch]];
1433 for (ch = 0; ch < s->out_channels; ch++)
1434 s->outptr[ch] += AC3_BLOCK_SIZE;
1437 /* keep last block for error concealment in next frame */
1438 for (ch = 0; ch < s->out_channels; ch++)
1439 memcpy(s->output[ch], output[ch], sizeof(**output) * AC3_BLOCK_SIZE);
1444 * Check whether the input layout is compatible, and make sure we're not
1445 * downmixing (else the matrix encoding is no longer applicable).
1447 matrix_encoding = AV_MATRIX_ENCODING_NONE;
1448 if (s->channel_mode == AC3_CHMODE_STEREO &&
1449 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1450 if (s->dolby_surround_mode == AC3_DSURMOD_ON)
1451 matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
1452 else if (s->dolby_headphone_mode == AC3_DHEADPHONMOD_ON)
1453 matrix_encoding = AV_MATRIX_ENCODING_DOLBYHEADPHONE;
1454 } else if (s->channel_mode >= AC3_CHMODE_2F2R &&
1455 s->channel_mode == (s->output_mode & ~AC3_OUTPUT_LFEON)) {
1456 switch (s->dolby_surround_ex_mode) {
1457 case AC3_DSUREXMOD_ON: // EX or PLIIx
1458 matrix_encoding = AV_MATRIX_ENCODING_DOLBYEX;
1460 case AC3_DSUREXMOD_PLIIZ:
1461 matrix_encoding = AV_MATRIX_ENCODING_DPLIIZ;
1463 default: // not indicated or off
1467 if ((ret = ff_side_data_update_matrix_encoding(frame, matrix_encoding)) < 0)
1472 return FFMIN(buf_size, s->frame_size);
1476 * Uninitialize the AC-3 decoder.
1478 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1480 AC3DecodeContext *s = avctx->priv_data;
1481 ff_mdct_end(&s->imdct_512);
1482 ff_mdct_end(&s->imdct_256);
1487 #define OFFSET(x) offsetof(AC3DecodeContext, x)
1488 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)
1489 static const AVOption options[] = {
1490 { "drc_scale", "percentage of dynamic range compression to apply", OFFSET(drc_scale), AV_OPT_TYPE_FLOAT, {.dbl = 1.0}, 0.0, 1.0, PAR },
1494 static const AVClass ac3_decoder_class = {
1495 .class_name = "AC3 decoder",
1496 .item_name = av_default_item_name,
1498 .version = LIBAVUTIL_VERSION_INT,
1501 AVCodec ff_ac3_decoder = {
1503 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52A (AC-3)"),
1504 .type = AVMEDIA_TYPE_AUDIO,
1505 .id = AV_CODEC_ID_AC3,
1506 .priv_data_size = sizeof (AC3DecodeContext),
1507 .init = ac3_decode_init,
1508 .close = ac3_decode_end,
1509 .decode = ac3_decode_frame,
1510 .capabilities = CODEC_CAP_DR1,
1511 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1512 AV_SAMPLE_FMT_NONE },
1513 .priv_class = &ac3_decoder_class,
1516 #if CONFIG_EAC3_DECODER
1517 static const AVClass eac3_decoder_class = {
1518 .class_name = "E-AC3 decoder",
1519 .item_name = av_default_item_name,
1521 .version = LIBAVUTIL_VERSION_INT,
1524 AVCodec ff_eac3_decoder = {
1526 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52B (AC-3, E-AC-3)"),
1527 .type = AVMEDIA_TYPE_AUDIO,
1528 .id = AV_CODEC_ID_EAC3,
1529 .priv_data_size = sizeof (AC3DecodeContext),
1530 .init = ac3_decode_init,
1531 .close = ac3_decode_end,
1532 .decode = ac3_decode_frame,
1533 .capabilities = CODEC_CAP_DR1,
1534 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1535 AV_SAMPLE_FMT_NONE },
1536 .priv_class = &eac3_decoder_class,