2 * DCA compatible decoder
3 * Copyright (C) 2004 Gildas Bazin
4 * Copyright (C) 2004 Benjamin Zores
5 * Copyright (C) 2006 Benjamin Larsson
6 * Copyright (C) 2007 Konstantin Shishkov
8 * This file is part of FFmpeg.
10 * FFmpeg is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation; either
13 * version 2.1 of the License, or (at your option) any later version.
15 * FFmpeg is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with FFmpeg; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
29 #include "libavutil/channel_layout.h"
30 #include "libavutil/common.h"
31 #include "libavutil/float_dsp.h"
32 #include "libavutil/internal.h"
33 #include "libavutil/intreadwrite.h"
34 #include "libavutil/mathematics.h"
35 #include "libavutil/opt.h"
36 #include "libavutil/samplefmt.h"
44 #include "fmtconvert.h"
48 #include "synth_filter.h"
69 enum DCAXxchSpeakerMask {
70 DCA_XXCH_FRONT_CENTER = 0x0000001,
71 DCA_XXCH_FRONT_LEFT = 0x0000002,
72 DCA_XXCH_FRONT_RIGHT = 0x0000004,
73 DCA_XXCH_SIDE_REAR_LEFT = 0x0000008,
74 DCA_XXCH_SIDE_REAR_RIGHT = 0x0000010,
75 DCA_XXCH_LFE1 = 0x0000020,
76 DCA_XXCH_REAR_CENTER = 0x0000040,
77 DCA_XXCH_SURROUND_REAR_LEFT = 0x0000080,
78 DCA_XXCH_SURROUND_REAR_RIGHT = 0x0000100,
79 DCA_XXCH_SIDE_SURROUND_LEFT = 0x0000200,
80 DCA_XXCH_SIDE_SURROUND_RIGHT = 0x0000400,
81 DCA_XXCH_FRONT_CENTER_LEFT = 0x0000800,
82 DCA_XXCH_FRONT_CENTER_RIGHT = 0x0001000,
83 DCA_XXCH_FRONT_HIGH_LEFT = 0x0002000,
84 DCA_XXCH_FRONT_HIGH_CENTER = 0x0004000,
85 DCA_XXCH_FRONT_HIGH_RIGHT = 0x0008000,
86 DCA_XXCH_LFE2 = 0x0010000,
87 DCA_XXCH_SIDE_FRONT_LEFT = 0x0020000,
88 DCA_XXCH_SIDE_FRONT_RIGHT = 0x0040000,
89 DCA_XXCH_OVERHEAD = 0x0080000,
90 DCA_XXCH_SIDE_HIGH_LEFT = 0x0100000,
91 DCA_XXCH_SIDE_HIGH_RIGHT = 0x0200000,
92 DCA_XXCH_REAR_HIGH_CENTER = 0x0400000,
93 DCA_XXCH_REAR_HIGH_LEFT = 0x0800000,
94 DCA_XXCH_REAR_HIGH_RIGHT = 0x1000000,
95 DCA_XXCH_REAR_LOW_CENTER = 0x2000000,
96 DCA_XXCH_REAR_LOW_LEFT = 0x4000000,
97 DCA_XXCH_REAR_LOW_RIGHT = 0x8000000,
100 #define DCA_DOLBY 101 /* FIXME */
102 #define DCA_CHANNEL_BITS 6
103 #define DCA_CHANNEL_MASK 0x3F
107 #define HEADER_SIZE 14
109 #define DCA_NSYNCAUX 0x9A1105A0
112 /** Bit allocation */
113 typedef struct BitAlloc {
114 int offset; ///< code values offset
115 int maxbits[8]; ///< max bits in VLC
116 int wrap; ///< wrap for get_vlc2()
117 VLC vlc[8]; ///< actual codes
120 static BitAlloc dca_bitalloc_index; ///< indexes for samples VLC select
121 static BitAlloc dca_tmode; ///< transition mode VLCs
122 static BitAlloc dca_scalefactor; ///< scalefactor VLCs
123 static BitAlloc dca_smpl_bitalloc[11]; ///< samples VLCs
125 static av_always_inline int get_bitalloc(GetBitContext *gb, BitAlloc *ba,
128 return get_vlc2(gb, ba->vlc[idx].table, ba->vlc[idx].bits, ba->wrap) +
132 static float dca_dmix_code(unsigned code);
134 static av_cold void dca_init_vlcs(void)
136 static int vlcs_initialized = 0;
138 static VLC_TYPE dca_table[23622][2];
140 if (vlcs_initialized)
143 dca_bitalloc_index.offset = 1;
144 dca_bitalloc_index.wrap = 2;
145 for (i = 0; i < 5; i++) {
146 dca_bitalloc_index.vlc[i].table = &dca_table[ff_dca_vlc_offs[i]];
147 dca_bitalloc_index.vlc[i].table_allocated = ff_dca_vlc_offs[i + 1] - ff_dca_vlc_offs[i];
148 init_vlc(&dca_bitalloc_index.vlc[i], bitalloc_12_vlc_bits[i], 12,
149 bitalloc_12_bits[i], 1, 1,
150 bitalloc_12_codes[i], 2, 2, INIT_VLC_USE_NEW_STATIC);
152 dca_scalefactor.offset = -64;
153 dca_scalefactor.wrap = 2;
154 for (i = 0; i < 5; i++) {
155 dca_scalefactor.vlc[i].table = &dca_table[ff_dca_vlc_offs[i + 5]];
156 dca_scalefactor.vlc[i].table_allocated = ff_dca_vlc_offs[i + 6] - ff_dca_vlc_offs[i + 5];
157 init_vlc(&dca_scalefactor.vlc[i], SCALES_VLC_BITS, 129,
158 scales_bits[i], 1, 1,
159 scales_codes[i], 2, 2, INIT_VLC_USE_NEW_STATIC);
161 dca_tmode.offset = 0;
163 for (i = 0; i < 4; i++) {
164 dca_tmode.vlc[i].table = &dca_table[ff_dca_vlc_offs[i + 10]];
165 dca_tmode.vlc[i].table_allocated = ff_dca_vlc_offs[i + 11] - ff_dca_vlc_offs[i + 10];
166 init_vlc(&dca_tmode.vlc[i], tmode_vlc_bits[i], 4,
168 tmode_codes[i], 2, 2, INIT_VLC_USE_NEW_STATIC);
171 for (i = 0; i < 10; i++)
172 for (j = 0; j < 7; j++) {
173 if (!bitalloc_codes[i][j])
175 dca_smpl_bitalloc[i + 1].offset = bitalloc_offsets[i];
176 dca_smpl_bitalloc[i + 1].wrap = 1 + (j > 4);
177 dca_smpl_bitalloc[i + 1].vlc[j].table = &dca_table[ff_dca_vlc_offs[c]];
178 dca_smpl_bitalloc[i + 1].vlc[j].table_allocated = ff_dca_vlc_offs[c + 1] - ff_dca_vlc_offs[c];
180 init_vlc(&dca_smpl_bitalloc[i + 1].vlc[j], bitalloc_maxbits[i][j],
182 bitalloc_bits[i][j], 1, 1,
183 bitalloc_codes[i][j], 2, 2, INIT_VLC_USE_NEW_STATIC);
186 vlcs_initialized = 1;
189 static inline void get_array(GetBitContext *gb, int *dst, int len, int bits)
192 *dst++ = get_bits(gb, bits);
195 static inline int dca_xxch2index(DCAContext *s, int xxch_ch)
199 /* locate channel set containing the channel */
200 for (i = -1, base = 0, mask = (s->xxch_core_spkmask & ~DCA_XXCH_LFE1);
201 i <= s->xxch_chset && !(mask & xxch_ch); mask = s->xxch_spk_masks[++i])
202 base += av_popcount(mask);
204 return base + av_popcount(mask & (xxch_ch - 1));
207 static int dca_parse_audio_coding_header(DCAContext *s, int base_channel,
211 static const float adj_table[4] = { 1.0, 1.1250, 1.2500, 1.4375 };
212 static const int bitlen[11] = { 0, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3 };
213 static const int thr[11] = { 0, 1, 3, 3, 3, 3, 7, 7, 7, 7, 7 };
214 int hdr_pos = 0, hdr_size = 0;
216 int this_chans, acc_mask;
217 int embedded_downmix;
221 /* xxch has arbitrary sized audio coding headers */
223 hdr_pos = get_bits_count(&s->gb);
224 hdr_size = get_bits(&s->gb, 7) + 1;
227 nchans = get_bits(&s->gb, 3) + 1;
228 s->total_channels = nchans + base_channel;
229 s->prim_channels = s->total_channels;
231 /* obtain speaker layout mask & downmix coefficients for XXCH */
233 acc_mask = s->xxch_core_spkmask;
235 this_chans = get_bits(&s->gb, s->xxch_nbits_spk_mask - 6) << 6;
236 s->xxch_spk_masks[s->xxch_chset] = this_chans;
237 s->xxch_chset_nch[s->xxch_chset] = nchans;
239 for (i = 0; i <= s->xxch_chset; i++)
240 acc_mask |= s->xxch_spk_masks[i];
242 /* check for downmixing information */
243 if (get_bits1(&s->gb)) {
244 embedded_downmix = get_bits1(&s->gb);
245 coeff = get_bits(&s->gb, 6);
247 if (coeff<1 || coeff>61) {
248 av_log(s->avctx, AV_LOG_ERROR, "6bit coeff %d is out of range\n", coeff);
249 return AVERROR_INVALIDDATA;
252 scale_factor = -1.0f / dca_dmix_code((coeff<<2)-3);
254 s->xxch_dmix_sf[s->xxch_chset] = scale_factor;
256 for (i = base_channel; i < s->prim_channels; i++) {
257 mask[i] = get_bits(&s->gb, s->xxch_nbits_spk_mask);
260 for (j = base_channel; j < s->prim_channels; j++) {
261 memset(s->xxch_dmix_coeff[j], 0, sizeof(s->xxch_dmix_coeff[0]));
262 s->xxch_dmix_embedded |= (embedded_downmix << j);
263 for (i = 0; i < s->xxch_nbits_spk_mask; i++) {
264 if (mask[j] & (1 << i)) {
265 if ((1 << i) == DCA_XXCH_LFE1) {
266 av_log(s->avctx, AV_LOG_WARNING,
267 "DCA-XXCH: dmix to LFE1 not supported.\n");
271 coeff = get_bits(&s->gb, 7);
272 ichan = dca_xxch2index(s, 1 << i);
273 if ((coeff&63)<1 || (coeff&63)>61) {
274 av_log(s->avctx, AV_LOG_ERROR, "7bit coeff %d is out of range\n", coeff);
275 return AVERROR_INVALIDDATA;
277 s->xxch_dmix_coeff[j][ichan] = dca_dmix_code((coeff<<2)-3);
284 if (s->prim_channels > DCA_PRIM_CHANNELS_MAX)
285 s->prim_channels = DCA_PRIM_CHANNELS_MAX;
287 for (i = base_channel; i < s->prim_channels; i++) {
288 s->subband_activity[i] = get_bits(&s->gb, 5) + 2;
289 if (s->subband_activity[i] > DCA_SUBBANDS)
290 s->subband_activity[i] = DCA_SUBBANDS;
292 for (i = base_channel; i < s->prim_channels; i++) {
293 s->vq_start_subband[i] = get_bits(&s->gb, 5) + 1;
294 if (s->vq_start_subband[i] > DCA_SUBBANDS)
295 s->vq_start_subband[i] = DCA_SUBBANDS;
297 get_array(&s->gb, s->joint_intensity + base_channel, s->prim_channels - base_channel, 3);
298 get_array(&s->gb, s->transient_huffman + base_channel, s->prim_channels - base_channel, 2);
299 get_array(&s->gb, s->scalefactor_huffman + base_channel, s->prim_channels - base_channel, 3);
300 get_array(&s->gb, s->bitalloc_huffman + base_channel, s->prim_channels - base_channel, 3);
302 /* Get codebooks quantization indexes */
304 memset(s->quant_index_huffman, 0, sizeof(s->quant_index_huffman));
305 for (j = 1; j < 11; j++)
306 for (i = base_channel; i < s->prim_channels; i++)
307 s->quant_index_huffman[i][j] = get_bits(&s->gb, bitlen[j]);
309 /* Get scale factor adjustment */
310 for (j = 0; j < 11; j++)
311 for (i = base_channel; i < s->prim_channels; i++)
312 s->scalefactor_adj[i][j] = 1;
314 for (j = 1; j < 11; j++)
315 for (i = base_channel; i < s->prim_channels; i++)
316 if (s->quant_index_huffman[i][j] < thr[j])
317 s->scalefactor_adj[i][j] = adj_table[get_bits(&s->gb, 2)];
320 if (s->crc_present) {
321 /* Audio header CRC check */
322 get_bits(&s->gb, 16);
325 /* Skip to the end of the header, also ignore CRC if present */
326 i = get_bits_count(&s->gb);
327 if (hdr_pos + 8 * hdr_size > i)
328 skip_bits_long(&s->gb, hdr_pos + 8 * hdr_size - i);
331 s->current_subframe = 0;
332 s->current_subsubframe = 0;
337 static int dca_parse_frame_header(DCAContext *s)
339 init_get_bits(&s->gb, s->dca_buffer, s->dca_buffer_size * 8);
342 skip_bits_long(&s->gb, 32);
345 s->frame_type = get_bits(&s->gb, 1);
346 s->samples_deficit = get_bits(&s->gb, 5) + 1;
347 s->crc_present = get_bits(&s->gb, 1);
348 s->sample_blocks = get_bits(&s->gb, 7) + 1;
349 s->frame_size = get_bits(&s->gb, 14) + 1;
350 if (s->frame_size < 95)
351 return AVERROR_INVALIDDATA;
352 s->amode = get_bits(&s->gb, 6);
353 s->sample_rate = avpriv_dca_sample_rates[get_bits(&s->gb, 4)];
355 return AVERROR_INVALIDDATA;
356 s->bit_rate_index = get_bits(&s->gb, 5);
357 s->bit_rate = ff_dca_bit_rates[s->bit_rate_index];
359 return AVERROR_INVALIDDATA;
361 skip_bits1(&s->gb); // always 0 (reserved, cf. ETSI TS 102 114 V1.4.1)
362 s->dynrange = get_bits(&s->gb, 1);
363 s->timestamp = get_bits(&s->gb, 1);
364 s->aux_data = get_bits(&s->gb, 1);
365 s->hdcd = get_bits(&s->gb, 1);
366 s->ext_descr = get_bits(&s->gb, 3);
367 s->ext_coding = get_bits(&s->gb, 1);
368 s->aspf = get_bits(&s->gb, 1);
369 s->lfe = get_bits(&s->gb, 2);
370 s->predictor_history = get_bits(&s->gb, 1);
374 av_log(s->avctx, AV_LOG_ERROR, "Invalid LFE value: %d\n", s->lfe);
375 return AVERROR_INVALIDDATA;
378 /* TODO: check CRC */
380 s->header_crc = get_bits(&s->gb, 16);
382 s->multirate_inter = get_bits(&s->gb, 1);
383 s->version = get_bits(&s->gb, 4);
384 s->copy_history = get_bits(&s->gb, 2);
385 s->source_pcm_res = get_bits(&s->gb, 3);
386 s->front_sum = get_bits(&s->gb, 1);
387 s->surround_sum = get_bits(&s->gb, 1);
388 s->dialog_norm = get_bits(&s->gb, 4);
390 /* FIXME: channels mixing levels */
391 s->output = s->amode;
393 s->output |= DCA_LFE;
395 /* Primary audio coding header */
396 s->subframes = get_bits(&s->gb, 4) + 1;
398 return dca_parse_audio_coding_header(s, 0, 0);
401 static inline int get_scale(GetBitContext *gb, int level, int value, int log2range)
404 /* huffman encoded */
405 value += get_bitalloc(gb, &dca_scalefactor, level);
406 value = av_clip(value, 0, (1 << log2range) - 1);
407 } else if (level < 8) {
408 if (level + 1 > log2range) {
409 skip_bits(gb, level + 1 - log2range);
410 value = get_bits(gb, log2range);
412 value = get_bits(gb, level + 1);
418 static int dca_subframe_header(DCAContext *s, int base_channel, int block_index)
420 /* Primary audio coding side information */
423 if (get_bits_left(&s->gb) < 0)
424 return AVERROR_INVALIDDATA;
427 s->subsubframes[s->current_subframe] = get_bits(&s->gb, 2) + 1;
428 s->partial_samples[s->current_subframe] = get_bits(&s->gb, 3);
431 for (j = base_channel; j < s->prim_channels; j++) {
432 for (k = 0; k < s->subband_activity[j]; k++)
433 s->prediction_mode[j][k] = get_bits(&s->gb, 1);
436 /* Get prediction codebook */
437 for (j = base_channel; j < s->prim_channels; j++) {
438 for (k = 0; k < s->subband_activity[j]; k++) {
439 if (s->prediction_mode[j][k] > 0) {
440 /* (Prediction coefficient VQ address) */
441 s->prediction_vq[j][k] = get_bits(&s->gb, 12);
446 /* Bit allocation index */
447 for (j = base_channel; j < s->prim_channels; j++) {
448 for (k = 0; k < s->vq_start_subband[j]; k++) {
449 if (s->bitalloc_huffman[j] == 6)
450 s->bitalloc[j][k] = get_bits(&s->gb, 5);
451 else if (s->bitalloc_huffman[j] == 5)
452 s->bitalloc[j][k] = get_bits(&s->gb, 4);
453 else if (s->bitalloc_huffman[j] == 7) {
454 av_log(s->avctx, AV_LOG_ERROR,
455 "Invalid bit allocation index\n");
456 return AVERROR_INVALIDDATA;
459 get_bitalloc(&s->gb, &dca_bitalloc_index, s->bitalloc_huffman[j]);
462 if (s->bitalloc[j][k] > 26) {
463 av_dlog(s->avctx, "bitalloc index [%i][%i] too big (%i)\n",
464 j, k, s->bitalloc[j][k]);
465 return AVERROR_INVALIDDATA;
470 /* Transition mode */
471 for (j = base_channel; j < s->prim_channels; j++) {
472 for (k = 0; k < s->subband_activity[j]; k++) {
473 s->transition_mode[j][k] = 0;
474 if (s->subsubframes[s->current_subframe] > 1 &&
475 k < s->vq_start_subband[j] && s->bitalloc[j][k] > 0) {
476 s->transition_mode[j][k] =
477 get_bitalloc(&s->gb, &dca_tmode, s->transient_huffman[j]);
482 if (get_bits_left(&s->gb) < 0)
483 return AVERROR_INVALIDDATA;
485 for (j = base_channel; j < s->prim_channels; j++) {
486 const uint32_t *scale_table;
487 int scale_sum, log_size;
489 memset(s->scale_factor[j], 0,
490 s->subband_activity[j] * sizeof(s->scale_factor[0][0][0]) * 2);
492 if (s->scalefactor_huffman[j] == 6) {
493 scale_table = ff_dca_scale_factor_quant7;
496 scale_table = ff_dca_scale_factor_quant6;
500 /* When huffman coded, only the difference is encoded */
503 for (k = 0; k < s->subband_activity[j]; k++) {
504 if (k >= s->vq_start_subband[j] || s->bitalloc[j][k] > 0) {
505 scale_sum = get_scale(&s->gb, s->scalefactor_huffman[j], scale_sum, log_size);
506 s->scale_factor[j][k][0] = scale_table[scale_sum];
509 if (k < s->vq_start_subband[j] && s->transition_mode[j][k]) {
510 /* Get second scale factor */
511 scale_sum = get_scale(&s->gb, s->scalefactor_huffman[j], scale_sum, log_size);
512 s->scale_factor[j][k][1] = scale_table[scale_sum];
517 /* Joint subband scale factor codebook select */
518 for (j = base_channel; j < s->prim_channels; j++) {
519 /* Transmitted only if joint subband coding enabled */
520 if (s->joint_intensity[j] > 0)
521 s->joint_huff[j] = get_bits(&s->gb, 3);
524 if (get_bits_left(&s->gb) < 0)
525 return AVERROR_INVALIDDATA;
527 /* Scale factors for joint subband coding */
528 for (j = base_channel; j < s->prim_channels; j++) {
531 /* Transmitted only if joint subband coding enabled */
532 if (s->joint_intensity[j] > 0) {
534 source_channel = s->joint_intensity[j] - 1;
536 /* When huffman coded, only the difference is encoded
537 * (is this valid as well for joint scales ???) */
539 for (k = s->subband_activity[j]; k < s->subband_activity[source_channel]; k++) {
540 scale = get_scale(&s->gb, s->joint_huff[j], 64 /* bias */, 7);
541 s->joint_scale_factor[j][k] = scale; /*joint_scale_table[scale]; */
544 if (!(s->debug_flag & 0x02)) {
545 av_log(s->avctx, AV_LOG_DEBUG,
546 "Joint stereo coding not supported\n");
547 s->debug_flag |= 0x02;
552 /* Dynamic range coefficient */
553 if (!base_channel && s->dynrange)
554 s->dynrange_coef = get_bits(&s->gb, 8);
556 /* Side information CRC check word */
557 if (s->crc_present) {
558 get_bits(&s->gb, 16);
562 * Primary audio data arrays
565 /* VQ encoded high frequency subbands */
566 for (j = base_channel; j < s->prim_channels; j++)
567 for (k = s->vq_start_subband[j]; k < s->subband_activity[j]; k++)
568 /* 1 vector -> 32 samples */
569 s->high_freq_vq[j][k] = get_bits(&s->gb, 10);
571 /* Low frequency effect data */
572 if (!base_channel && s->lfe) {
575 int lfe_samples = 2 * s->lfe * (4 + block_index);
576 int lfe_end_sample = 2 * s->lfe * (4 + block_index + s->subsubframes[s->current_subframe]);
579 for (j = lfe_samples; j < lfe_end_sample; j++) {
580 /* Signed 8 bits int */
581 s->lfe_data[j] = get_sbits(&s->gb, 8);
584 /* Scale factor index */
585 quant7 = get_bits(&s->gb, 8);
587 avpriv_request_sample(s->avctx, "LFEScaleIndex larger than 127");
588 return AVERROR_INVALIDDATA;
590 s->lfe_scale_factor = ff_dca_scale_factor_quant7[quant7];
592 /* Quantization step size * scale factor */
593 lfe_scale = 0.035 * s->lfe_scale_factor;
595 for (j = lfe_samples; j < lfe_end_sample; j++)
596 s->lfe_data[j] *= lfe_scale;
602 static void qmf_32_subbands(DCAContext *s, int chans,
603 float samples_in[32][8], float *samples_out,
606 const float *prCoeff;
608 int sb_act = s->subband_activity[chans];
610 scale *= sqrt(1 / 8.0);
613 if (!s->multirate_inter) /* Non-perfect reconstruction */
614 prCoeff = ff_dca_fir_32bands_nonperfect;
615 else /* Perfect reconstruction */
616 prCoeff = ff_dca_fir_32bands_perfect;
618 s->dcadsp.qmf_32_subbands(samples_in, sb_act, &s->synth, &s->imdct,
619 s->subband_fir_hist[chans],
620 &s->hist_index[chans],
621 s->subband_fir_noidea[chans], prCoeff,
622 samples_out, s->raXin, scale);
625 static void lfe_interpolation_fir(DCAContext *s, int decimation_select,
626 int num_deci_sample, float *samples_in,
629 /* samples_in: An array holding decimated samples.
630 * Samples in current subframe starts from samples_in[0],
631 * while samples_in[-1], samples_in[-2], ..., stores samples
632 * from last subframe as history.
634 * samples_out: An array holding interpolated samples
638 const float *prCoeff;
641 /* Select decimation filter */
642 if (decimation_select == 1) {
644 prCoeff = ff_dca_lfe_fir_128;
647 prCoeff = ff_dca_lfe_fir_64;
650 for (deciindex = 0; deciindex < num_deci_sample; deciindex++) {
651 s->dcadsp.lfe_fir[idx](samples_out, samples_in, prCoeff);
653 samples_out += 2 * 32 * (1 + idx);
657 /* downmixing routines */
658 #define MIX_REAR1(samples, s1, rs, coef) \
659 samples[0][i] += samples[s1][i] * coef[rs][0]; \
660 samples[1][i] += samples[s1][i] * coef[rs][1];
662 #define MIX_REAR2(samples, s1, s2, rs, coef) \
663 samples[0][i] += samples[s1][i] * coef[rs][0] + samples[s2][i] * coef[rs + 1][0]; \
664 samples[1][i] += samples[s1][i] * coef[rs][1] + samples[s2][i] * coef[rs + 1][1];
666 #define MIX_FRONT3(samples, coef) \
670 samples[0][i] = t * coef[0][0] + u * coef[1][0] + v * coef[2][0]; \
671 samples[1][i] = t * coef[0][1] + u * coef[1][1] + v * coef[2][1];
673 #define DOWNMIX_TO_STEREO(op1, op2) \
674 for (i = 0; i < 256; i++) { \
679 static void dca_downmix(float **samples, int srcfmt, int lfe_present,
680 float coef[DCA_PRIM_CHANNELS_MAX + 1][2],
681 const int8_t *channel_mapping)
683 int c, l, r, sl, sr, s;
690 av_log(NULL, AV_LOG_ERROR, "Not implemented!\n");
694 case DCA_STEREO_TOTAL:
695 case DCA_STEREO_SUMDIFF:
698 c = channel_mapping[0];
699 l = channel_mapping[1];
700 r = channel_mapping[2];
701 DOWNMIX_TO_STEREO(MIX_FRONT3(samples, coef), );
704 s = channel_mapping[2];
705 DOWNMIX_TO_STEREO(MIX_REAR1(samples, s, 2, coef), );
708 c = channel_mapping[0];
709 l = channel_mapping[1];
710 r = channel_mapping[2];
711 s = channel_mapping[3];
712 DOWNMIX_TO_STEREO(MIX_FRONT3(samples, coef),
713 MIX_REAR1(samples, s, 3, coef));
716 sl = channel_mapping[2];
717 sr = channel_mapping[3];
718 DOWNMIX_TO_STEREO(MIX_REAR2(samples, sl, sr, 2, coef), );
721 c = channel_mapping[0];
722 l = channel_mapping[1];
723 r = channel_mapping[2];
724 sl = channel_mapping[3];
725 sr = channel_mapping[4];
726 DOWNMIX_TO_STEREO(MIX_FRONT3(samples, coef),
727 MIX_REAR2(samples, sl, sr, 3, coef));
731 int lf_buf = ff_dca_lfe_index[srcfmt];
732 int lf_idx = ff_dca_channels[srcfmt];
733 for (i = 0; i < 256; i++) {
734 samples[0][i] += samples[lf_buf][i] * coef[lf_idx][0];
735 samples[1][i] += samples[lf_buf][i] * coef[lf_idx][1];
740 #ifndef decode_blockcodes
741 /* Very compact version of the block code decoder that does not use table
742 * look-up but is slightly slower */
743 static int decode_blockcode(int code, int levels, int32_t *values)
746 int offset = (levels - 1) >> 1;
748 for (i = 0; i < 4; i++) {
749 int div = FASTDIV(code, levels);
750 values[i] = code - offset - div * levels;
757 static int decode_blockcodes(int code1, int code2, int levels, int32_t *values)
759 return decode_blockcode(code1, levels, values) |
760 decode_blockcode(code2, levels, values + 4);
764 static const uint8_t abits_sizes[7] = { 7, 10, 12, 13, 15, 17, 19 };
765 static const uint8_t abits_levels[7] = { 3, 5, 7, 9, 13, 17, 25 };
767 static int dca_subsubframe(DCAContext *s, int base_channel, int block_index)
770 int subsubframe = s->current_subsubframe;
772 const float *quant_step_table;
775 float (*subband_samples)[DCA_SUBBANDS][8] = s->subband_samples[block_index];
776 LOCAL_ALIGNED_16(int32_t, block, [8 * DCA_SUBBANDS]);
782 /* Select quantization step size table */
783 if (s->bit_rate_index == 0x1f)
784 quant_step_table = ff_dca_lossless_quant_d;
786 quant_step_table = ff_dca_lossy_quant_d;
788 for (k = base_channel; k < s->prim_channels; k++) {
789 float rscale[DCA_SUBBANDS];
791 if (get_bits_left(&s->gb) < 0)
792 return AVERROR_INVALIDDATA;
794 for (l = 0; l < s->vq_start_subband[k]; l++) {
797 /* Select the mid-tread linear quantizer */
798 int abits = s->bitalloc[k][l];
800 float quant_step_size = quant_step_table[abits];
803 * Determine quantization index code book and its type
806 /* Select quantization index code book */
807 int sel = s->quant_index_huffman[k][abits];
810 * Extract bits from the bit stream
814 memset(block + 8 * l, 0, 8 * sizeof(block[0]));
816 /* Deal with transients */
817 int sfi = s->transition_mode[k][l] && subsubframe >= s->transition_mode[k][l];
818 rscale[l] = quant_step_size * s->scale_factor[k][l][sfi] *
819 s->scalefactor_adj[k][sel];
821 if (abits >= 11 || !dca_smpl_bitalloc[abits].vlc[sel].table) {
824 int block_code1, block_code2, size, levels, err;
826 size = abits_sizes[abits - 1];
827 levels = abits_levels[abits - 1];
829 block_code1 = get_bits(&s->gb, size);
830 block_code2 = get_bits(&s->gb, size);
831 err = decode_blockcodes(block_code1, block_code2,
832 levels, block + 8 * l);
834 av_log(s->avctx, AV_LOG_ERROR,
835 "ERROR: block code look-up failed\n");
836 return AVERROR_INVALIDDATA;
840 for (m = 0; m < 8; m++)
841 block[8 * l + m] = get_sbits(&s->gb, abits - 3);
845 for (m = 0; m < 8; m++)
846 block[8 * l + m] = get_bitalloc(&s->gb,
847 &dca_smpl_bitalloc[abits], sel);
852 s->fmt_conv.int32_to_float_fmul_array8(&s->fmt_conv, subband_samples[k][0],
853 block, rscale, 8 * s->vq_start_subband[k]);
855 for (l = 0; l < s->vq_start_subband[k]; l++) {
858 * Inverse ADPCM if in prediction mode
860 if (s->prediction_mode[k][l]) {
862 if (s->predictor_history)
863 subband_samples[k][l][0] += (ff_dca_adpcm_vb[s->prediction_vq[k][l]][0] *
864 s->subband_samples_hist[k][l][3] +
865 ff_dca_adpcm_vb[s->prediction_vq[k][l]][1] *
866 s->subband_samples_hist[k][l][2] +
867 ff_dca_adpcm_vb[s->prediction_vq[k][l]][2] *
868 s->subband_samples_hist[k][l][1] +
869 ff_dca_adpcm_vb[s->prediction_vq[k][l]][3] *
870 s->subband_samples_hist[k][l][0]) *
872 for (m = 1; m < 8; m++) {
873 float sum = ff_dca_adpcm_vb[s->prediction_vq[k][l]][0] *
874 subband_samples[k][l][m - 1];
875 for (n = 2; n <= 4; n++)
877 sum += ff_dca_adpcm_vb[s->prediction_vq[k][l]][n - 1] *
878 subband_samples[k][l][m - n];
879 else if (s->predictor_history)
880 sum += ff_dca_adpcm_vb[s->prediction_vq[k][l]][n - 1] *
881 s->subband_samples_hist[k][l][m - n + 4];
882 subband_samples[k][l][m] += sum * (1.0f / 8192);
888 * Decode VQ encoded high frequencies
890 if (s->subband_activity[k] > s->vq_start_subband[k]) {
891 if (!(s->debug_flag & 0x01)) {
892 av_log(s->avctx, AV_LOG_DEBUG,
893 "Stream with high frequencies VQ coding\n");
894 s->debug_flag |= 0x01;
896 s->dcadsp.decode_hf(subband_samples[k], s->high_freq_vq[k],
897 ff_dca_high_freq_vq, subsubframe * 8,
898 s->scale_factor[k], s->vq_start_subband[k],
899 s->subband_activity[k]);
903 /* Check for DSYNC after subsubframe */
904 if (s->aspf || subsubframe == s->subsubframes[s->current_subframe] - 1) {
905 if (get_bits(&s->gb, 16) != 0xFFFF) {
906 av_log(s->avctx, AV_LOG_ERROR, "Didn't get subframe DSYNC\n");
907 return AVERROR_INVALIDDATA;
911 /* Backup predictor history for adpcm */
912 for (k = base_channel; k < s->prim_channels; k++)
913 for (l = 0; l < s->vq_start_subband[k]; l++)
914 AV_COPY128(s->subband_samples_hist[k][l], &subband_samples[k][l][4]);
919 static int dca_filter_channels(DCAContext *s, int block_index)
921 float (*subband_samples)[DCA_SUBBANDS][8] = s->subband_samples[block_index];
924 /* 32 subbands QMF */
925 for (k = 0; k < s->prim_channels; k++) {
926 if (s->channel_order_tab[k] >= 0)
927 qmf_32_subbands(s, k, subband_samples[k],
928 s->samples_chanptr[s->channel_order_tab[k]],
929 M_SQRT1_2 / 32768.0);
932 /* Generate LFE samples for this subsubframe FIXME!!! */
934 lfe_interpolation_fir(s, s->lfe, 2 * s->lfe,
935 s->lfe_data + 2 * s->lfe * (block_index + 4),
936 s->samples_chanptr[s->lfe_index]);
937 /* Outputs 20bits pcm samples */
940 /* Downmixing to Stereo */
941 if (s->prim_channels + !!s->lfe > 2 &&
942 s->avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
943 dca_downmix(s->samples_chanptr, s->amode, !!s->lfe, s->downmix_coef,
944 s->channel_order_tab);
950 static int dca_subframe_footer(DCAContext *s, int base_channel)
952 int in, out, aux_data_count, aux_data_end, reserved;
956 * Unpack optional information
959 /* presumably optional information only appears in the core? */
962 skip_bits_long(&s->gb, 32);
965 aux_data_count = get_bits(&s->gb, 6);
968 skip_bits_long(&s->gb, (-get_bits_count(&s->gb)) & 31);
970 aux_data_end = 8 * aux_data_count + get_bits_count(&s->gb);
972 if ((nsyncaux = get_bits_long(&s->gb, 32)) != DCA_NSYNCAUX) {
973 av_log(s->avctx, AV_LOG_ERROR, "nSYNCAUX mismatch %#"PRIx32"\n",
975 return AVERROR_INVALIDDATA;
978 if (get_bits1(&s->gb)) { // bAUXTimeStampFlag
979 avpriv_request_sample(s->avctx,
980 "Auxiliary Decode Time Stamp Flag");
982 skip_bits(&s->gb, (-get_bits_count(&s->gb)) & 4);
983 // 44 bits: nMSByte (8), nMarker (4), nLSByte (28), nMarker (4)
984 skip_bits_long(&s->gb, 44);
987 if ((s->core_downmix = get_bits1(&s->gb))) {
988 int am = get_bits(&s->gb, 3);
991 s->core_downmix_amode = DCA_MONO;
994 s->core_downmix_amode = DCA_STEREO;
997 s->core_downmix_amode = DCA_STEREO_TOTAL;
1000 s->core_downmix_amode = DCA_3F;
1003 s->core_downmix_amode = DCA_2F1R;
1006 s->core_downmix_amode = DCA_2F2R;
1009 s->core_downmix_amode = DCA_3F1R;
1012 av_log(s->avctx, AV_LOG_ERROR,
1013 "Invalid mode %d for embedded downmix coefficients\n",
1015 return AVERROR_INVALIDDATA;
1017 for (out = 0; out < ff_dca_channels[s->core_downmix_amode]; out++) {
1018 for (in = 0; in < s->prim_channels + !!s->lfe; in++) {
1019 uint16_t tmp = get_bits(&s->gb, 9);
1020 if ((tmp & 0xFF) > 241) {
1021 av_log(s->avctx, AV_LOG_ERROR,
1022 "Invalid downmix coefficient code %"PRIu16"\n",
1024 return AVERROR_INVALIDDATA;
1026 s->core_downmix_codes[in][out] = tmp;
1031 align_get_bits(&s->gb); // byte align
1032 skip_bits(&s->gb, 16); // nAUXCRC16
1034 // additional data (reserved, cf. ETSI TS 102 114 V1.4.1)
1035 if ((reserved = (aux_data_end - get_bits_count(&s->gb))) < 0) {
1036 av_log(s->avctx, AV_LOG_ERROR,
1037 "Overread auxiliary data by %d bits\n", -reserved);
1038 return AVERROR_INVALIDDATA;
1039 } else if (reserved) {
1040 avpriv_request_sample(s->avctx,
1041 "Core auxiliary data reserved content");
1042 skip_bits_long(&s->gb, reserved);
1046 if (s->crc_present && s->dynrange)
1047 get_bits(&s->gb, 16);
1054 * Decode a dca frame block
1056 * @param s pointer to the DCAContext
1059 static int dca_decode_block(DCAContext *s, int base_channel, int block_index)
1064 if (s->current_subframe >= s->subframes) {
1065 av_log(s->avctx, AV_LOG_DEBUG, "check failed: %i>%i",
1066 s->current_subframe, s->subframes);
1067 return AVERROR_INVALIDDATA;
1070 if (!s->current_subsubframe) {
1071 /* Read subframe header */
1072 if ((ret = dca_subframe_header(s, base_channel, block_index)))
1076 /* Read subsubframe */
1077 if ((ret = dca_subsubframe(s, base_channel, block_index)))
1081 s->current_subsubframe++;
1082 if (s->current_subsubframe >= s->subsubframes[s->current_subframe]) {
1083 s->current_subsubframe = 0;
1084 s->current_subframe++;
1086 if (s->current_subframe >= s->subframes) {
1087 /* Read subframe footer */
1088 if ((ret = dca_subframe_footer(s, base_channel)))
1095 int ff_dca_xbr_parse_frame(DCAContext *s)
1097 int scale_table_high[DCA_CHSET_CHANS_MAX][DCA_SUBBANDS][2];
1098 int active_bands[DCA_CHSETS_MAX][DCA_CHSET_CHANS_MAX];
1099 int abits_high[DCA_CHSET_CHANS_MAX][DCA_SUBBANDS];
1100 int anctemp[DCA_CHSET_CHANS_MAX];
1101 int chset_fsize[DCA_CHSETS_MAX];
1102 int n_xbr_ch[DCA_CHSETS_MAX];
1103 int hdr_size, num_chsets, xbr_tmode, hdr_pos;
1104 int i, j, k, l, chset, chan_base;
1106 av_log(s->avctx, AV_LOG_DEBUG, "DTS-XBR: decoding XBR extension\n");
1108 /* get bit position of sync header */
1109 hdr_pos = get_bits_count(&s->gb) - 32;
1111 hdr_size = get_bits(&s->gb, 6) + 1;
1112 num_chsets = get_bits(&s->gb, 2) + 1;
1114 for(i = 0; i < num_chsets; i++)
1115 chset_fsize[i] = get_bits(&s->gb, 14) + 1;
1117 xbr_tmode = get_bits1(&s->gb);
1119 for(i = 0; i < num_chsets; i++) {
1120 n_xbr_ch[i] = get_bits(&s->gb, 3) + 1;
1121 k = get_bits(&s->gb, 2) + 5;
1122 for(j = 0; j < n_xbr_ch[i]; j++)
1123 active_bands[i][j] = get_bits(&s->gb, k) + 1;
1126 /* skip to the end of the header */
1127 i = get_bits_count(&s->gb);
1128 if(hdr_pos + hdr_size * 8 > i)
1129 skip_bits_long(&s->gb, hdr_pos + hdr_size * 8 - i);
1131 /* loop over the channel data sets */
1132 /* only decode as many channels as we've decoded base data for */
1133 for(chset = 0, chan_base = 0;
1134 chset < num_chsets && chan_base + n_xbr_ch[chset] <= s->prim_channels;
1135 chan_base += n_xbr_ch[chset++]) {
1136 int start_posn = get_bits_count(&s->gb);
1137 int subsubframe = 0;
1140 /* loop over subframes */
1141 for (k = 0; k < (s->sample_blocks / 8); k++) {
1142 /* parse header if we're on first subsubframe of a block */
1143 if(subsubframe == 0) {
1144 /* Parse subframe header */
1145 for(i = 0; i < n_xbr_ch[chset]; i++) {
1146 anctemp[i] = get_bits(&s->gb, 2) + 2;
1149 for(i = 0; i < n_xbr_ch[chset]; i++) {
1150 get_array(&s->gb, abits_high[i], active_bands[chset][i], anctemp[i]);
1153 for(i = 0; i < n_xbr_ch[chset]; i++) {
1154 anctemp[i] = get_bits(&s->gb, 3);
1155 if(anctemp[i] < 1) {
1156 av_log(s->avctx, AV_LOG_ERROR, "DTS-XBR: SYNC ERROR\n");
1157 return AVERROR_INVALIDDATA;
1161 /* generate scale factors */
1162 for(i = 0; i < n_xbr_ch[chset]; i++) {
1163 const uint32_t *scale_table;
1166 if (s->scalefactor_huffman[chan_base+i] == 6) {
1167 scale_table = ff_dca_scale_factor_quant7;
1169 scale_table = ff_dca_scale_factor_quant6;
1174 for(j = 0; j < active_bands[chset][i]; j++) {
1175 if(abits_high[i][j] > 0) {
1176 scale_table_high[i][j][0] =
1177 scale_table[get_bits(&s->gb, nbits)];
1179 if(xbr_tmode && s->transition_mode[i][j]) {
1180 scale_table_high[i][j][1] =
1181 scale_table[get_bits(&s->gb, nbits)];
1188 /* decode audio array for this block */
1189 for(i = 0; i < n_xbr_ch[chset]; i++) {
1190 for(j = 0; j < active_bands[chset][i]; j++) {
1191 const int xbr_abits = abits_high[i][j];
1192 const float quant_step_size = ff_dca_lossless_quant_d[xbr_abits];
1193 const int sfi = xbr_tmode && s->transition_mode[i][j] && subsubframe >= s->transition_mode[i][j];
1194 const float rscale = quant_step_size * scale_table_high[i][j][sfi];
1195 float *subband_samples = s->subband_samples[k][chan_base+i][j];
1202 get_array(&s->gb, block, 8, xbr_abits - 3);
1204 int block_code1, block_code2, size, levels, err;
1206 size = abits_sizes[xbr_abits - 1];
1207 levels = abits_levels[xbr_abits - 1];
1209 block_code1 = get_bits(&s->gb, size);
1210 block_code2 = get_bits(&s->gb, size);
1211 err = decode_blockcodes(block_code1, block_code2,
1214 av_log(s->avctx, AV_LOG_ERROR,
1215 "ERROR: DTS-XBR: block code look-up failed\n");
1216 return AVERROR_INVALIDDATA;
1220 /* scale & sum into subband */
1221 for(l = 0; l < 8; l++)
1222 subband_samples[l] += (float)block[l] * rscale;
1226 /* check DSYNC marker */
1227 if(s->aspf || subsubframe == s->subsubframes[subframe] - 1) {
1228 if(get_bits(&s->gb, 16) != 0xffff) {
1229 av_log(s->avctx, AV_LOG_ERROR, "DTS-XBR: Didn't get subframe DSYNC\n");
1230 return AVERROR_INVALIDDATA;
1234 /* advance sub-sub-frame index */
1235 if(++subsubframe >= s->subsubframes[subframe]) {
1241 /* skip to next channel set */
1242 i = get_bits_count(&s->gb);
1243 if(start_posn + chset_fsize[chset] * 8 != i) {
1244 j = start_posn + chset_fsize[chset] * 8 - i;
1246 av_log(s->avctx, AV_LOG_ERROR, "DTS-XBR: end of channel set,"
1247 " skipping further than expected (%d bits)\n", j);
1248 skip_bits_long(&s->gb, j);
1256 /* parse initial header for XXCH and dump details */
1257 int ff_dca_xxch_decode_frame(DCAContext *s)
1259 int hdr_size, spkmsk_bits, num_chsets, core_spk, hdr_pos;
1260 int i, chset, base_channel, chstart, fsize[8];
1262 /* assume header word has already been parsed */
1263 hdr_pos = get_bits_count(&s->gb) - 32;
1264 hdr_size = get_bits(&s->gb, 6) + 1;
1265 /*chhdr_crc =*/ skip_bits1(&s->gb);
1266 spkmsk_bits = get_bits(&s->gb, 5) + 1;
1267 num_chsets = get_bits(&s->gb, 2) + 1;
1269 for (i = 0; i < num_chsets; i++)
1270 fsize[i] = get_bits(&s->gb, 14) + 1;
1272 core_spk = get_bits(&s->gb, spkmsk_bits);
1273 s->xxch_core_spkmask = core_spk;
1274 s->xxch_nbits_spk_mask = spkmsk_bits;
1275 s->xxch_dmix_embedded = 0;
1277 /* skip to the end of the header */
1278 i = get_bits_count(&s->gb);
1279 if (hdr_pos + hdr_size * 8 > i)
1280 skip_bits_long(&s->gb, hdr_pos + hdr_size * 8 - i);
1282 for (chset = 0; chset < num_chsets; chset++) {
1283 chstart = get_bits_count(&s->gb);
1284 base_channel = s->prim_channels;
1285 s->xxch_chset = chset;
1287 /* XXCH and Core headers differ, see 6.4.2 "XXCH Channel Set Header" vs.
1288 5.3.2 "Primary Audio Coding Header", DTS Spec 1.3.1 */
1289 dca_parse_audio_coding_header(s, base_channel, 1);
1291 /* decode channel data */
1292 for (i = 0; i < (s->sample_blocks / 8); i++) {
1293 if (dca_decode_block(s, base_channel, i)) {
1294 av_log(s->avctx, AV_LOG_ERROR,
1295 "Error decoding DTS-XXCH extension\n");
1300 /* skip to end of this section */
1301 i = get_bits_count(&s->gb);
1302 if (chstart + fsize[chset] * 8 > i)
1303 skip_bits_long(&s->gb, chstart + fsize[chset] * 8 - i);
1305 s->xxch_chset = num_chsets;
1310 static float dca_dmix_code(unsigned code)
1312 int sign = (code >> 8) - 1;
1314 return ((ff_dca_dmixtable[code] ^ sign) - sign) * (1.0 / (1 << 15));
1318 * Main frame decoding function
1319 * FIXME add arguments
1321 static int dca_decode_frame(AVCodecContext *avctx, void *data,
1322 int *got_frame_ptr, AVPacket *avpkt)
1324 AVFrame *frame = data;
1325 const uint8_t *buf = avpkt->data;
1326 int buf_size = avpkt->size;
1330 int num_core_channels = 0;
1332 float **samples_flt;
1335 DCAContext *s = avctx->priv_data;
1337 int channels, full_channels;
1349 s->dca_buffer_size = avpriv_dca_convert_bitstream(buf, buf_size, s->dca_buffer,
1350 DCA_MAX_FRAME_SIZE + DCA_MAX_EXSS_HEADER_SIZE);
1351 if (s->dca_buffer_size == AVERROR_INVALIDDATA) {
1352 av_log(avctx, AV_LOG_ERROR, "Not a valid DCA frame\n");
1353 return AVERROR_INVALIDDATA;
1356 if ((ret = dca_parse_frame_header(s)) < 0) {
1357 // seems like the frame is corrupt, try with the next one
1360 // set AVCodec values with parsed data
1361 avctx->sample_rate = s->sample_rate;
1362 avctx->bit_rate = s->bit_rate;
1364 s->profile = FF_PROFILE_DTS;
1366 for (i = 0; i < (s->sample_blocks / 8); i++) {
1367 if ((ret = dca_decode_block(s, 0, i))) {
1368 av_log(avctx, AV_LOG_ERROR, "error decoding block\n");
1373 /* record number of core channels incase less than max channels are requested */
1374 num_core_channels = s->prim_channels;
1376 if (s->prim_channels + !!s->lfe > 2 &&
1377 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
1378 /* Stereo downmix coefficients
1380 * The decoder can only downmix to 2-channel, so we need to ensure
1381 * embedded downmix coefficients are actually targeting 2-channel.
1383 if (s->core_downmix && (s->core_downmix_amode == DCA_STEREO ||
1384 s->core_downmix_amode == DCA_STEREO_TOTAL)) {
1385 for (i = 0; i < num_core_channels + !!s->lfe; i++) {
1386 /* Range checked earlier */
1387 s->downmix_coef[i][0] = dca_dmix_code(s->core_downmix_codes[i][0]);
1388 s->downmix_coef[i][1] = dca_dmix_code(s->core_downmix_codes[i][1]);
1390 s->output = s->core_downmix_amode;
1392 int am = s->amode & DCA_CHANNEL_MASK;
1393 if (am >= FF_ARRAY_ELEMS(ff_dca_default_coeffs)) {
1394 av_log(s->avctx, AV_LOG_ERROR,
1395 "Invalid channel mode %d\n", am);
1396 return AVERROR_INVALIDDATA;
1398 if (num_core_channels + !!s->lfe >
1399 FF_ARRAY_ELEMS(ff_dca_default_coeffs[0])) {
1400 avpriv_request_sample(s->avctx, "Downmixing %d channels",
1401 s->prim_channels + !!s->lfe);
1402 return AVERROR_PATCHWELCOME;
1404 for (i = 0; i < num_core_channels + !!s->lfe; i++) {
1405 s->downmix_coef[i][0] = ff_dca_default_coeffs[am][i][0];
1406 s->downmix_coef[i][1] = ff_dca_default_coeffs[am][i][1];
1409 av_dlog(s->avctx, "Stereo downmix coeffs:\n");
1410 for (i = 0; i < num_core_channels + !!s->lfe; i++) {
1411 av_dlog(s->avctx, "L, input channel %d = %f\n", i,
1412 s->downmix_coef[i][0]);
1413 av_dlog(s->avctx, "R, input channel %d = %f\n", i,
1414 s->downmix_coef[i][1]);
1416 av_dlog(s->avctx, "\n");
1420 s->core_ext_mask = ff_dca_ext_audio_descr_mask[s->ext_descr];
1422 s->core_ext_mask = 0;
1424 core_ss_end = FFMIN(s->frame_size, s->dca_buffer_size) * 8;
1426 /* only scan for extensions if ext_descr was unknown or indicated a
1427 * supported XCh extension */
1428 if (s->core_ext_mask < 0 || s->core_ext_mask & (DCA_EXT_XCH | DCA_EXT_XXCH)) {
1429 /* if ext_descr was unknown, clear s->core_ext_mask so that the
1430 * extensions scan can fill it up */
1431 s->core_ext_mask = FFMAX(s->core_ext_mask, 0);
1433 /* extensions start at 32-bit boundaries into bitstream */
1434 skip_bits_long(&s->gb, (-get_bits_count(&s->gb)) & 31);
1436 while (core_ss_end - get_bits_count(&s->gb) >= 32) {
1437 uint32_t bits = get_bits_long(&s->gb, 32);
1441 int ext_amode, xch_fsize;
1443 s->xch_base_channel = s->prim_channels;
1445 /* validate sync word using XCHFSIZE field */
1446 xch_fsize = show_bits(&s->gb, 10);
1447 if ((s->frame_size != (get_bits_count(&s->gb) >> 3) - 4 + xch_fsize) &&
1448 (s->frame_size != (get_bits_count(&s->gb) >> 3) - 4 + xch_fsize + 1))
1451 /* skip length-to-end-of-frame field for the moment */
1452 skip_bits(&s->gb, 10);
1454 s->core_ext_mask |= DCA_EXT_XCH;
1456 /* extension amode(number of channels in extension) should be 1 */
1457 /* AFAIK XCh is not used for more channels */
1458 if ((ext_amode = get_bits(&s->gb, 4)) != 1) {
1459 av_log(avctx, AV_LOG_ERROR,
1460 "XCh extension amode %d not supported!\n",
1465 if (s->xch_base_channel < 2) {
1466 avpriv_request_sample(avctx, "XCh with fewer than 2 base channels");
1470 /* much like core primary audio coding header */
1471 dca_parse_audio_coding_header(s, s->xch_base_channel, 0);
1473 for (i = 0; i < (s->sample_blocks / 8); i++)
1474 if ((ret = dca_decode_block(s, s->xch_base_channel, i))) {
1475 av_log(avctx, AV_LOG_ERROR, "error decoding XCh extension\n");
1483 /* XXCh: extended channels */
1484 /* usually found either in core or HD part in DTS-HD HRA streams,
1485 * but not in DTS-ES which contains XCh extensions instead */
1486 s->core_ext_mask |= DCA_EXT_XXCH;
1487 ff_dca_xxch_decode_frame(s);
1491 int fsize96 = show_bits(&s->gb, 12) + 1;
1492 if (s->frame_size != (get_bits_count(&s->gb) >> 3) - 4 + fsize96)
1495 av_log(avctx, AV_LOG_DEBUG, "X96 extension found at %d bits\n",
1496 get_bits_count(&s->gb));
1497 skip_bits(&s->gb, 12);
1498 av_log(avctx, AV_LOG_DEBUG, "FSIZE96 = %d bytes\n", fsize96);
1499 av_log(avctx, AV_LOG_DEBUG, "REVNO = %d\n", get_bits(&s->gb, 4));
1501 s->core_ext_mask |= DCA_EXT_X96;
1506 skip_bits_long(&s->gb, (-get_bits_count(&s->gb)) & 31);
1509 /* no supported extensions, skip the rest of the core substream */
1510 skip_bits_long(&s->gb, core_ss_end - get_bits_count(&s->gb));
1513 if (s->core_ext_mask & DCA_EXT_X96)
1514 s->profile = FF_PROFILE_DTS_96_24;
1515 else if (s->core_ext_mask & (DCA_EXT_XCH | DCA_EXT_XXCH))
1516 s->profile = FF_PROFILE_DTS_ES;
1518 /* check for ExSS (HD part) */
1519 if (s->dca_buffer_size - s->frame_size > 32 &&
1520 get_bits_long(&s->gb, 32) == DCA_HD_MARKER)
1521 ff_dca_exss_parse_header(s);
1523 avctx->profile = s->profile;
1525 full_channels = channels = s->prim_channels + !!s->lfe;
1527 /* If we have XXCH then the channel layout is managed differently */
1528 /* note that XLL will also have another way to do things */
1529 if (!(s->core_ext_mask & DCA_EXT_XXCH)
1530 || (s->core_ext_mask & DCA_EXT_XXCH && avctx->request_channels > 0
1531 && avctx->request_channels
1532 < num_core_channels + !!s->lfe + s->xxch_chset_nch[0]))
1533 { /* xxx should also do MA extensions */
1534 if (s->amode < 16) {
1535 avctx->channel_layout = ff_dca_core_channel_layout[s->amode];
1537 if (s->prim_channels + !!s->lfe > 2 &&
1538 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
1540 * Neither the core's auxiliary data nor our default tables contain
1541 * downmix coefficients for the additional channel coded in the XCh
1542 * extension, so when we're doing a Stereo downmix, don't decode it.
1547 #if FF_API_REQUEST_CHANNELS
1548 FF_DISABLE_DEPRECATION_WARNINGS
1549 if (s->xch_present && !s->xch_disable &&
1550 (!avctx->request_channels ||
1551 avctx->request_channels > num_core_channels + !!s->lfe)) {
1552 FF_ENABLE_DEPRECATION_WARNINGS
1554 if (s->xch_present && !s->xch_disable) {
1556 if (avctx->channel_layout & AV_CH_BACK_CENTER) {
1557 avpriv_request_sample(avctx, "XCh with Back center channel");
1558 return AVERROR_INVALIDDATA;
1560 avctx->channel_layout |= AV_CH_BACK_CENTER;
1562 avctx->channel_layout |= AV_CH_LOW_FREQUENCY;
1563 s->channel_order_tab = ff_dca_channel_reorder_lfe_xch[s->amode];
1565 s->channel_order_tab = ff_dca_channel_reorder_nolfe_xch[s->amode];
1567 if (s->channel_order_tab[s->xch_base_channel] < 0)
1568 return AVERROR_INVALIDDATA;
1570 channels = num_core_channels + !!s->lfe;
1571 s->xch_present = 0; /* disable further xch processing */
1573 avctx->channel_layout |= AV_CH_LOW_FREQUENCY;
1574 s->channel_order_tab = ff_dca_channel_reorder_lfe[s->amode];
1576 s->channel_order_tab = ff_dca_channel_reorder_nolfe[s->amode];
1579 if (channels > !!s->lfe &&
1580 s->channel_order_tab[channels - 1 - !!s->lfe] < 0)
1581 return AVERROR_INVALIDDATA;
1583 if (av_get_channel_layout_nb_channels(avctx->channel_layout) != channels) {
1584 av_log(avctx, AV_LOG_ERROR, "Number of channels %d mismatches layout %d\n", channels, av_get_channel_layout_nb_channels(avctx->channel_layout));
1585 return AVERROR_INVALIDDATA;
1588 if (num_core_channels + !!s->lfe > 2 &&
1589 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO) {
1591 s->output = s->prim_channels == 2 ? s->amode : DCA_STEREO;
1592 avctx->channel_layout = AV_CH_LAYOUT_STEREO;
1594 else if (avctx->request_channel_layout & AV_CH_LAYOUT_NATIVE) {
1595 static const int8_t dca_channel_order_native[9] = { 0, 1, 2, 3, 4, 5, 6, 7, 8 };
1596 s->channel_order_tab = dca_channel_order_native;
1598 s->lfe_index = ff_dca_lfe_index[s->amode];
1600 av_log(avctx, AV_LOG_ERROR,
1601 "Non standard configuration %d !\n", s->amode);
1602 return AVERROR_INVALIDDATA;
1605 s->xxch_dmix_embedded = 0;
1607 /* we only get here if an XXCH channel set can be added to the mix */
1608 channel_mask = s->xxch_core_spkmask;
1610 if (avctx->request_channels > 0
1611 && avctx->request_channels < s->prim_channels) {
1612 channels = num_core_channels + !!s->lfe;
1613 for (i = 0; i < s->xxch_chset && channels + s->xxch_chset_nch[i]
1614 <= avctx->request_channels; i++) {
1615 channels += s->xxch_chset_nch[i];
1616 channel_mask |= s->xxch_spk_masks[i];
1619 channels = s->prim_channels + !!s->lfe;
1620 for (i = 0; i < s->xxch_chset; i++) {
1621 channel_mask |= s->xxch_spk_masks[i];
1625 /* Given the DTS spec'ed channel mask, generate an avcodec version */
1627 for (i = 0; i < s->xxch_nbits_spk_mask; ++i) {
1628 if (channel_mask & (1 << i)) {
1629 channel_layout |= ff_dca_map_xxch_to_native[i];
1633 /* make sure that we have managed to get equivalent dts/avcodec channel
1634 * masks in some sense -- unfortunately some channels could overlap */
1635 if (av_popcount(channel_mask) != av_popcount(channel_layout)) {
1636 av_log(avctx, AV_LOG_DEBUG,
1637 "DTS-XXCH: Inconsistent avcodec/dts channel layouts\n");
1638 return AVERROR_INVALIDDATA;
1641 avctx->channel_layout = channel_layout;
1643 if (!(avctx->request_channel_layout & AV_CH_LAYOUT_NATIVE)) {
1644 /* Estimate DTS --> avcodec ordering table */
1645 for (chset = -1, j = 0; chset < s->xxch_chset; ++chset) {
1646 mask = chset >= 0 ? s->xxch_spk_masks[chset]
1647 : s->xxch_core_spkmask;
1648 for (i = 0; i < s->xxch_nbits_spk_mask; i++) {
1649 if (mask & ~(DCA_XXCH_LFE1 | DCA_XXCH_LFE2) & (1 << i)) {
1650 lavc = ff_dca_map_xxch_to_native[i];
1651 posn = av_popcount(channel_layout & (lavc - 1));
1652 s->xxch_order_tab[j++] = posn;
1658 s->lfe_index = av_popcount(channel_layout & (AV_CH_LOW_FREQUENCY-1));
1659 } else { /* native ordering */
1660 for (i = 0; i < channels; i++)
1661 s->xxch_order_tab[i] = i;
1663 s->lfe_index = channels - 1;
1666 s->channel_order_tab = s->xxch_order_tab;
1669 if (avctx->channels != channels) {
1670 if (avctx->channels)
1671 av_log(avctx, AV_LOG_INFO, "Number of channels changed in DCA decoder (%d -> %d)\n", avctx->channels, channels);
1672 avctx->channels = channels;
1675 /* get output buffer */
1676 frame->nb_samples = 256 * (s->sample_blocks / 8);
1677 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1679 samples_flt = (float **) frame->extended_data;
1681 /* allocate buffer for extra channels if downmixing */
1682 if (avctx->channels < full_channels) {
1683 ret = av_samples_get_buffer_size(NULL, full_channels - channels,
1685 avctx->sample_fmt, 0);
1689 av_fast_malloc(&s->extra_channels_buffer,
1690 &s->extra_channels_buffer_size, ret);
1691 if (!s->extra_channels_buffer)
1692 return AVERROR(ENOMEM);
1694 ret = av_samples_fill_arrays((uint8_t **) s->extra_channels, NULL,
1695 s->extra_channels_buffer,
1696 full_channels - channels,
1697 frame->nb_samples, avctx->sample_fmt, 0);
1702 /* filter to get final output */
1703 for (i = 0; i < (s->sample_blocks / 8); i++) {
1706 for (ch = 0; ch < channels; ch++)
1707 s->samples_chanptr[ch] = samples_flt[ch] + i * 256;
1708 for (; ch < full_channels; ch++)
1709 s->samples_chanptr[ch] = s->extra_channels[ch - channels] + i * 256;
1711 dca_filter_channels(s, i);
1713 /* If this was marked as a DTS-ES stream we need to subtract back- */
1714 /* channel from SL & SR to remove matrixed back-channel signal */
1715 if ((s->source_pcm_res & 1) && s->xch_present) {
1716 float *back_chan = s->samples_chanptr[s->channel_order_tab[s->xch_base_channel]];
1717 float *lt_chan = s->samples_chanptr[s->channel_order_tab[s->xch_base_channel - 2]];
1718 float *rt_chan = s->samples_chanptr[s->channel_order_tab[s->xch_base_channel - 1]];
1719 s->fdsp->vector_fmac_scalar(lt_chan, back_chan, -M_SQRT1_2, 256);
1720 s->fdsp->vector_fmac_scalar(rt_chan, back_chan, -M_SQRT1_2, 256);
1723 /* If stream contains XXCH, we might need to undo an embedded downmix */
1724 if (s->xxch_dmix_embedded) {
1725 /* Loop over channel sets in turn */
1726 ch = num_core_channels;
1727 for (chset = 0; chset < s->xxch_chset; chset++) {
1728 endch = ch + s->xxch_chset_nch[chset];
1729 mask = s->xxch_dmix_embedded;
1732 for (j = ch; j < endch; j++) {
1733 if (mask & (1 << j)) { /* this channel has been mixed-out */
1734 src_chan = s->samples_chanptr[s->channel_order_tab[j]];
1735 for (k = 0; k < endch; k++) {
1736 achan = s->channel_order_tab[k];
1737 scale = s->xxch_dmix_coeff[j][k];
1739 dst_chan = s->samples_chanptr[achan];
1740 s->fdsp->vector_fmac_scalar(dst_chan, src_chan,
1747 /* if a downmix has been embedded then undo the pre-scaling */
1748 if ((mask & (1 << ch)) && s->xxch_dmix_sf[chset] != 1.0f) {
1749 scale = s->xxch_dmix_sf[chset];
1751 for (j = 0; j < ch; j++) {
1752 src_chan = s->samples_chanptr[s->channel_order_tab[j]];
1753 for (k = 0; k < 256; k++)
1754 src_chan[k] *= scale;
1757 /* LFE channel is always part of core, scale if it exists */
1759 src_chan = s->samples_chanptr[s->lfe_index];
1760 for (k = 0; k < 256; k++)
1761 src_chan[k] *= scale;
1771 /* update lfe history */
1772 lfe_samples = 2 * s->lfe * (s->sample_blocks / 8);
1773 for (i = 0; i < 2 * s->lfe * 4; i++)
1774 s->lfe_data[i] = s->lfe_data[i + lfe_samples];
1778 * DCA_STEREO_TOTAL (Lt/Rt) is equivalent to Dolby Surround */
1779 ret = ff_side_data_update_matrix_encoding(frame,
1780 (s->output & ~DCA_LFE) == DCA_STEREO_TOTAL ?
1781 AV_MATRIX_ENCODING_DOLBY : AV_MATRIX_ENCODING_NONE);
1791 * DCA initialization
1793 * @param avctx pointer to the AVCodecContext
1796 static av_cold int dca_decode_init(AVCodecContext *avctx)
1798 DCAContext *s = avctx->priv_data;
1803 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & CODEC_FLAG_BITEXACT);
1805 return AVERROR(ENOMEM);
1807 ff_mdct_init(&s->imdct, 6, 1, 1.0);
1808 ff_synth_filter_init(&s->synth);
1809 ff_dcadsp_init(&s->dcadsp);
1810 ff_fmt_convert_init(&s->fmt_conv, avctx);
1812 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1814 /* allow downmixing to stereo */
1815 #if FF_API_REQUEST_CHANNELS
1816 FF_DISABLE_DEPRECATION_WARNINGS
1817 if (avctx->request_channels == 2)
1818 avctx->request_channel_layout = AV_CH_LAYOUT_STEREO;
1819 FF_ENABLE_DEPRECATION_WARNINGS
1821 if (avctx->channels > 2 &&
1822 avctx->request_channel_layout == AV_CH_LAYOUT_STEREO)
1823 avctx->channels = 2;
1828 static av_cold int dca_decode_end(AVCodecContext *avctx)
1830 DCAContext *s = avctx->priv_data;
1831 ff_mdct_end(&s->imdct);
1832 av_freep(&s->extra_channels_buffer);
1837 static const AVProfile profiles[] = {
1838 { FF_PROFILE_DTS, "DTS" },
1839 { FF_PROFILE_DTS_ES, "DTS-ES" },
1840 { FF_PROFILE_DTS_96_24, "DTS 96/24" },
1841 { FF_PROFILE_DTS_HD_HRA, "DTS-HD HRA" },
1842 { FF_PROFILE_DTS_HD_MA, "DTS-HD MA" },
1843 { FF_PROFILE_UNKNOWN },
1846 static const AVOption options[] = {
1847 { "disable_xch", "disable decoding of the XCh extension", offsetof(DCAContext, xch_disable), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM },
1851 static const AVClass dca_decoder_class = {
1852 .class_name = "DCA decoder",
1853 .item_name = av_default_item_name,
1855 .version = LIBAVUTIL_VERSION_INT,
1856 .category = AV_CLASS_CATEGORY_DECODER,
1859 AVCodec ff_dca_decoder = {
1861 .long_name = NULL_IF_CONFIG_SMALL("DCA (DTS Coherent Acoustics)"),
1862 .type = AVMEDIA_TYPE_AUDIO,
1863 .id = AV_CODEC_ID_DTS,
1864 .priv_data_size = sizeof(DCAContext),
1865 .init = dca_decode_init,
1866 .decode = dca_decode_frame,
1867 .close = dca_decode_end,
1868 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
1869 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1870 AV_SAMPLE_FMT_NONE },
1871 .profiles = NULL_IF_CONFIG_SMALL(profiles),
1872 .priv_class = &dca_decoder_class,