2 * Copyright (C) 2016 foo86
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 #include "dca_syncwords.h"
27 static int get_linear(GetBitContext *gb, int n)
29 unsigned int v = get_bits_long(gb, n);
30 return (v >> 1) ^ -(v & 1);
33 static int get_rice_un(GetBitContext *gb, int k)
35 unsigned int v = get_unary(gb, 1, get_bits_left(gb));
36 return (v << k) | get_bits_long(gb, k);
39 static int get_rice(GetBitContext *gb, int k)
41 unsigned int v = get_rice_un(gb, k);
42 return (v >> 1) ^ -(v & 1);
45 static void get_array(GetBitContext *gb, int32_t *array, int size, int n)
49 for (i = 0; i < size; i++)
50 array[i] = get_bits(gb, n);
53 static void get_linear_array(GetBitContext *gb, int32_t *array, int size, int n)
58 memset(array, 0, sizeof(*array) * size);
59 else for (i = 0; i < size; i++)
60 array[i] = get_linear(gb, n);
63 static void get_rice_array(GetBitContext *gb, int32_t *array, int size, int k)
67 for (i = 0; i < size; i++)
68 array[i] = get_rice(gb, k);
71 static int parse_dmix_coeffs(DCAXllDecoder *s, DCAXllChSet *c)
73 // Size of downmix coefficient matrix
74 int m = c->primary_chset ? ff_dca_dmix_primary_nch[c->dmix_type] : c->hier_ofs;
75 int i, j, *coeff_ptr = c->dmix_coeff;
77 for (i = 0; i < m; i++) {
78 int code, sign, coeff, scale, scale_inv = 0;
81 // Downmix scale (only for non-primary channel sets)
82 if (!c->primary_chset) {
83 code = get_bits(&s->gb, 9);
84 sign = (code >> 8) - 1;
85 index = (code & 0xff) - FF_DCA_DMIXTABLE_OFFSET;
86 if (index >= FF_DCA_INV_DMIXTABLE_SIZE) {
87 av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL downmix scale index\n");
88 return AVERROR_INVALIDDATA;
90 scale = ff_dca_dmixtable[index + FF_DCA_DMIXTABLE_OFFSET];
91 scale_inv = ff_dca_inv_dmixtable[index];
92 c->dmix_scale[i] = (scale ^ sign) - sign;
93 c->dmix_scale_inv[i] = (scale_inv ^ sign) - sign;
96 // Downmix coefficients
97 for (j = 0; j < c->nchannels; j++) {
98 code = get_bits(&s->gb, 9);
99 sign = (code >> 8) - 1;
101 if (index >= FF_DCA_DMIXTABLE_SIZE) {
102 av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL downmix coefficient index\n");
103 return AVERROR_INVALIDDATA;
105 coeff = ff_dca_dmixtable[index];
106 if (!c->primary_chset)
107 // Multiply by |InvDmixScale| to get |UndoDmixScale|
108 coeff = mul16(scale_inv, coeff);
109 *coeff_ptr++ = (coeff ^ sign) - sign;
116 static int chs_parse_header(DCAXllDecoder *s, DCAXllChSet *c, DCAExssAsset *asset)
118 int i, j, k, ret, band, header_size, header_pos = get_bits_count(&s->gb);
119 DCAXllChSet *p = &s->chset[0];
122 // Size of channel set sub-header
123 header_size = get_bits(&s->gb, 10) + 1;
126 if (ff_dca_check_crc(s->avctx, &s->gb, header_pos, header_pos + header_size * 8)) {
127 av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL sub-header checksum\n");
128 return AVERROR_INVALIDDATA;
131 // Number of channels in the channel set
132 c->nchannels = get_bits(&s->gb, 4) + 1;
133 if (c->nchannels > DCA_XLL_CHANNELS_MAX) {
134 avpriv_request_sample(s->avctx, "%d XLL channels", c->nchannels);
135 return AVERROR_PATCHWELCOME;
139 c->residual_encode = get_bits(&s->gb, c->nchannels);
141 // PCM bit resolution
142 c->pcm_bit_res = get_bits(&s->gb, 5) + 1;
144 // Storage unit width
145 c->storage_bit_res = get_bits(&s->gb, 5) + 1;
146 if (c->storage_bit_res != 16 && c->storage_bit_res != 20 && c->storage_bit_res != 24) {
147 avpriv_request_sample(s->avctx, "%d-bit XLL storage resolution", c->storage_bit_res);
148 return AVERROR_PATCHWELCOME;
151 if (c->pcm_bit_res > c->storage_bit_res) {
152 av_log(s->avctx, AV_LOG_ERROR, "Invalid PCM bit resolution for XLL channel set (%d > %d)\n", c->pcm_bit_res, c->storage_bit_res);
153 return AVERROR_INVALIDDATA;
156 // Original sampling frequency
157 c->freq = ff_dca_sampling_freqs[get_bits(&s->gb, 4)];
158 if (c->freq > 192000) {
159 avpriv_request_sample(s->avctx, "%d Hz XLL sampling frequency", c->freq);
160 return AVERROR_PATCHWELCOME;
163 // Sampling frequency modifier
164 if (get_bits(&s->gb, 2)) {
165 avpriv_request_sample(s->avctx, "XLL sampling frequency modifier");
166 return AVERROR_PATCHWELCOME;
169 // Which replacement set this channel set is member of
170 if (get_bits(&s->gb, 2)) {
171 avpriv_request_sample(s->avctx, "XLL replacement set");
172 return AVERROR_PATCHWELCOME;
175 if (asset->one_to_one_map_ch_to_spkr) {
176 // Primary channel set flag
177 c->primary_chset = get_bits1(&s->gb);
178 if (c->primary_chset != (c == p)) {
179 av_log(s->avctx, AV_LOG_ERROR, "The first (and only) XLL channel set must be primary\n");
180 return AVERROR_INVALIDDATA;
183 // Downmix coefficients present in stream
184 c->dmix_coeffs_present = get_bits1(&s->gb);
186 // Downmix already performed by encoder
187 c->dmix_embedded = c->dmix_coeffs_present && get_bits1(&s->gb);
190 if (c->dmix_coeffs_present && c->primary_chset) {
191 c->dmix_type = get_bits(&s->gb, 3);
192 if (c->dmix_type >= DCA_DMIX_TYPE_COUNT) {
193 av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL primary channel set downmix type\n");
194 return AVERROR_INVALIDDATA;
198 // Whether the channel set is part of a hierarchy
199 c->hier_chset = get_bits1(&s->gb);
200 if (!c->hier_chset && s->nchsets != 1) {
201 avpriv_request_sample(s->avctx, "XLL channel set outside of hierarchy");
202 return AVERROR_PATCHWELCOME;
205 // Downmix coefficients
206 if (c->dmix_coeffs_present && (ret = parse_dmix_coeffs(s, c)) < 0)
209 // Channel mask enabled
210 if (!get_bits1(&s->gb)) {
211 avpriv_request_sample(s->avctx, "Disabled XLL channel mask");
212 return AVERROR_PATCHWELCOME;
215 // Channel mask for set
216 c->ch_mask = get_bits_long(&s->gb, s->ch_mask_nbits);
217 if (av_popcount(c->ch_mask) != c->nchannels) {
218 av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL channel mask\n");
219 return AVERROR_INVALIDDATA;
222 // Build the channel to speaker map
223 for (i = 0, j = 0; i < s->ch_mask_nbits; i++)
224 if (c->ch_mask & (1U << i))
225 c->ch_remap[j++] = i;
227 // Mapping coeffs present flag
228 if (c->nchannels != 2 || s->nchsets != 1 || get_bits1(&s->gb)) {
229 avpriv_request_sample(s->avctx, "Custom XLL channel to speaker mapping");
230 return AVERROR_PATCHWELCOME;
233 // Setup for LtRt decoding
234 c->primary_chset = 1;
235 c->dmix_coeffs_present = 0;
236 c->dmix_embedded = 0;
238 c->ch_mask = DCA_SPEAKER_LAYOUT_STEREO;
239 c->ch_remap[0] = DCA_SPEAKER_L;
240 c->ch_remap[1] = DCA_SPEAKER_R;
243 if (c->freq > 96000) {
244 // Extra frequency bands flag
245 if (get_bits1(&s->gb)) {
246 avpriv_request_sample(s->avctx, "Extra XLL frequency bands");
247 return AVERROR_PATCHWELCOME;
254 // Set the sampling frequency to that of the first frequency band.
255 // Frequency will be doubled again after bands assembly.
256 c->freq >>= c->nfreqbands - 1;
258 // Verify that all channel sets have the same audio characteristics
259 if (c != p && (c->nfreqbands != p->nfreqbands || c->freq != p->freq
260 || c->pcm_bit_res != p->pcm_bit_res
261 || c->storage_bit_res != p->storage_bit_res)) {
262 avpriv_request_sample(s->avctx, "Different XLL audio characteristics");
263 return AVERROR_PATCHWELCOME;
266 // Determine number of bits to read bit allocation coding parameter
267 if (c->storage_bit_res > 16)
269 else if (c->storage_bit_res > 8)
274 // Account for embedded downmix and decimator saturation
275 if ((s->nchsets > 1 || c->nfreqbands > 1) && c->nabits < 5)
278 for (band = 0, b = c->bands; band < c->nfreqbands; band++, b++) {
279 // Pairwise channel decorrelation
280 if ((b->decor_enabled = get_bits1(&s->gb)) && c->nchannels > 1) {
281 int ch_nbits = av_ceil_log2(c->nchannels);
283 // Original channel order
284 for (i = 0; i < c->nchannels; i++) {
285 b->orig_order[i] = get_bits(&s->gb, ch_nbits);
286 if (b->orig_order[i] >= c->nchannels) {
287 av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL original channel order\n");
288 return AVERROR_INVALIDDATA;
292 // Pairwise channel coefficients
293 for (i = 0; i < c->nchannels / 2; i++)
294 b->decor_coeff[i] = get_bits1(&s->gb) ? get_linear(&s->gb, 7) : 0;
296 for (i = 0; i < c->nchannels; i++)
297 b->orig_order[i] = i;
298 for (i = 0; i < c->nchannels / 2; i++)
299 b->decor_coeff[i] = 0;
302 // Adaptive predictor order
303 b->highest_pred_order = 0;
304 for (i = 0; i < c->nchannels; i++) {
305 b->adapt_pred_order[i] = get_bits(&s->gb, 4);
306 if (b->adapt_pred_order[i] > b->highest_pred_order)
307 b->highest_pred_order = b->adapt_pred_order[i];
309 if (b->highest_pred_order > s->nsegsamples) {
310 av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL adaptive predicition order\n");
311 return AVERROR_INVALIDDATA;
314 // Fixed predictor order
315 for (i = 0; i < c->nchannels; i++)
316 b->fixed_pred_order[i] = b->adapt_pred_order[i] ? 0 : get_bits(&s->gb, 2);
318 // Adaptive predictor quantized reflection coefficients
319 for (i = 0; i < c->nchannels; i++) {
320 for (j = 0; j < b->adapt_pred_order[i]; j++) {
321 k = get_linear(&s->gb, 8);
323 av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL reflection coefficient index\n");
324 return AVERROR_INVALIDDATA;
327 b->adapt_refl_coeff[i][j] = -(int)ff_dca_xll_refl_coeff[-k];
329 b->adapt_refl_coeff[i][j] = (int)ff_dca_xll_refl_coeff[ k];
333 // Downmix performed by encoder in extension frequency band
334 b->dmix_embedded = c->dmix_embedded && (band == 0 || get_bits1(&s->gb));
336 // MSB/LSB split flag in extension frequency band
337 if ((band == 0 && s->scalable_lsbs) || (band != 0 && get_bits1(&s->gb))) {
338 // Size of LSB section in any segment
339 b->lsb_section_size = get_bits_long(&s->gb, s->seg_size_nbits);
340 if (b->lsb_section_size < 0 || b->lsb_section_size > s->frame_size) {
341 av_log(s->avctx, AV_LOG_ERROR, "Invalid LSB section size\n");
342 return AVERROR_INVALIDDATA;
345 // Account for optional CRC bytes after LSB section
346 if (b->lsb_section_size && (s->band_crc_present > 2 ||
347 (band == 0 && s->band_crc_present > 1)))
348 b->lsb_section_size += 2;
350 // Number of bits to represent the samples in LSB part
351 for (i = 0; i < c->nchannels; i++) {
352 b->nscalablelsbs[i] = get_bits(&s->gb, 4);
353 if (b->nscalablelsbs[i] && !b->lsb_section_size) {
354 av_log(s->avctx, AV_LOG_ERROR, "LSB section missing with non-zero LSB width\n");
355 return AVERROR_INVALIDDATA;
359 b->lsb_section_size = 0;
360 for (i = 0; i < c->nchannels; i++)
361 b->nscalablelsbs[i] = 0;
364 // Scalable resolution flag in extension frequency band
365 if ((band == 0 && s->scalable_lsbs) || (band != 0 && get_bits1(&s->gb))) {
366 // Number of bits discarded by authoring
367 for (i = 0; i < c->nchannels; i++)
368 b->bit_width_adjust[i] = get_bits(&s->gb, 4);
370 for (i = 0; i < c->nchannels; i++)
371 b->bit_width_adjust[i] = 0;
377 // CRC16 of channel set sub-header
378 if (ff_dca_seek_bits(&s->gb, header_pos + header_size * 8)) {
379 av_log(s->avctx, AV_LOG_ERROR, "Read past end of XLL sub-header\n");
380 return AVERROR_INVALIDDATA;
386 static int chs_alloc_msb_band_data(DCAXllDecoder *s, DCAXllChSet *c)
388 int ndecisamples = c->nfreqbands > 1 ? DCA_XLL_DECI_HISTORY_MAX : 0;
389 int nchsamples = s->nframesamples + ndecisamples;
390 int i, j, nsamples = nchsamples * c->nchannels * c->nfreqbands;
393 // Reallocate MSB sample buffer
394 av_fast_malloc(&c->sample_buffer[0], &c->sample_size[0], nsamples * sizeof(int32_t));
395 if (!c->sample_buffer[0])
396 return AVERROR(ENOMEM);
398 ptr = c->sample_buffer[0] + ndecisamples;
399 for (i = 0; i < c->nfreqbands; i++) {
400 for (j = 0; j < c->nchannels; j++) {
401 c->bands[i].msb_sample_buffer[j] = ptr;
409 static int chs_alloc_lsb_band_data(DCAXllDecoder *s, DCAXllChSet *c)
411 int i, j, nsamples = 0;
414 // Determine number of frequency bands that have MSB/LSB split
415 for (i = 0; i < c->nfreqbands; i++)
416 if (c->bands[i].lsb_section_size)
417 nsamples += s->nframesamples * c->nchannels;
421 // Reallocate LSB sample buffer
422 av_fast_malloc(&c->sample_buffer[1], &c->sample_size[1], nsamples * sizeof(int32_t));
423 if (!c->sample_buffer[1])
424 return AVERROR(ENOMEM);
426 ptr = c->sample_buffer[1];
427 for (i = 0; i < c->nfreqbands; i++) {
428 if (c->bands[i].lsb_section_size) {
429 for (j = 0; j < c->nchannels; j++) {
430 c->bands[i].lsb_sample_buffer[j] = ptr;
431 ptr += s->nframesamples;
434 for (j = 0; j < c->nchannels; j++)
435 c->bands[i].lsb_sample_buffer[j] = NULL;
442 static int chs_parse_band_data(DCAXllDecoder *s, DCAXllChSet *c, int band, int seg, int band_data_end)
444 DCAXllBand *b = &c->bands[band];
447 // Start unpacking MSB portion of the segment
448 if (!(seg && get_bits1(&s->gb))) {
449 // Unpack segment type
450 // 0 - distinct coding parameters for each channel
451 // 1 - common coding parameters for all channels
452 c->seg_common = get_bits1(&s->gb);
454 // Determine number of coding parameters encoded in segment
455 k = c->seg_common ? 1 : c->nchannels;
457 // Unpack Rice coding parameters
458 for (i = 0; i < k; i++) {
459 // Unpack Rice coding flag
460 // 0 - linear code, 1 - Rice code
461 c->rice_code_flag[i] = get_bits1(&s->gb);
462 // Unpack Hybrid Rice coding flag
463 // 0 - Rice code, 1 - Hybrid Rice code
464 if (!c->seg_common && c->rice_code_flag[i] && get_bits1(&s->gb))
465 // Unpack binary code length for isolated samples
466 c->bitalloc_hybrid_linear[i] = get_bits(&s->gb, c->nabits) + 1;
468 // 0 indicates no Hybrid Rice coding
469 c->bitalloc_hybrid_linear[i] = 0;
472 // Unpack coding parameters
473 for (i = 0; i < k; i++) {
475 // Unpack coding parameter for part A of segment 0
476 c->bitalloc_part_a[i] = get_bits(&s->gb, c->nabits);
478 // Adjust for the linear code
479 if (!c->rice_code_flag[i] && c->bitalloc_part_a[i])
480 c->bitalloc_part_a[i]++;
483 c->nsamples_part_a[i] = b->adapt_pred_order[i];
485 c->nsamples_part_a[i] = b->highest_pred_order;
487 c->bitalloc_part_a[i] = 0;
488 c->nsamples_part_a[i] = 0;
491 // Unpack coding parameter for part B of segment
492 c->bitalloc_part_b[i] = get_bits(&s->gb, c->nabits);
494 // Adjust for the linear code
495 if (!c->rice_code_flag[i] && c->bitalloc_part_b[i])
496 c->bitalloc_part_b[i]++;
500 // Unpack entropy codes
501 for (i = 0; i < c->nchannels; i++) {
502 int32_t *part_a, *part_b;
505 // Select index of coding parameters
506 k = c->seg_common ? 0 : i;
508 // Slice the segment into parts A and B
509 part_a = b->msb_sample_buffer[i] + seg * s->nsegsamples;
510 part_b = part_a + c->nsamples_part_a[k];
511 nsamples_part_b = s->nsegsamples - c->nsamples_part_a[k];
513 if (get_bits_left(&s->gb) < 0)
514 return AVERROR_INVALIDDATA;
516 if (!c->rice_code_flag[k]) {
518 // Unpack all residuals of part A of segment 0
519 get_linear_array(&s->gb, part_a, c->nsamples_part_a[k],
520 c->bitalloc_part_a[k]);
522 // Unpack all residuals of part B of segment 0 and others
523 get_linear_array(&s->gb, part_b, nsamples_part_b,
524 c->bitalloc_part_b[k]);
527 // Unpack all residuals of part A of segment 0
528 get_rice_array(&s->gb, part_a, c->nsamples_part_a[k],
529 c->bitalloc_part_a[k]);
531 if (c->bitalloc_hybrid_linear[k]) {
533 // Unpack the number of isolated samples
534 int nisosamples = get_bits(&s->gb, s->nsegsamples_log2);
536 // Set all locations to 0
537 memset(part_b, 0, sizeof(*part_b) * nsamples_part_b);
539 // Extract the locations of isolated samples and flag by -1
540 for (j = 0; j < nisosamples; j++) {
541 int loc = get_bits(&s->gb, s->nsegsamples_log2);
542 if (loc >= nsamples_part_b) {
543 av_log(s->avctx, AV_LOG_ERROR, "Invalid isolated sample location\n");
544 return AVERROR_INVALIDDATA;
549 // Unpack all residuals of part B of segment 0 and others
550 for (j = 0; j < nsamples_part_b; j++) {
552 part_b[j] = get_linear(&s->gb, c->bitalloc_hybrid_linear[k]);
554 part_b[j] = get_rice(&s->gb, c->bitalloc_part_b[k]);
558 // Unpack all residuals of part B of segment 0 and others
559 get_rice_array(&s->gb, part_b, nsamples_part_b, c->bitalloc_part_b[k]);
564 // Unpack decimator history for frequency band 1
565 if (seg == 0 && band == 1) {
566 int nbits = get_bits(&s->gb, 5) + 1;
567 for (i = 0; i < c->nchannels; i++)
568 for (j = 1; j < DCA_XLL_DECI_HISTORY_MAX; j++)
569 c->deci_history[i][j] = get_sbits_long(&s->gb, nbits);
572 // Start unpacking LSB portion of the segment
573 if (b->lsb_section_size) {
574 // Skip to the start of LSB portion
575 if (ff_dca_seek_bits(&s->gb, band_data_end - b->lsb_section_size * 8)) {
576 av_log(s->avctx, AV_LOG_ERROR, "Read past end of XLL band data\n");
577 return AVERROR_INVALIDDATA;
580 // Unpack all LSB parts of residuals of this segment
581 for (i = 0; i < c->nchannels; i++) {
582 if (b->nscalablelsbs[i]) {
584 b->lsb_sample_buffer[i] + seg * s->nsegsamples,
585 s->nsegsamples, b->nscalablelsbs[i]);
590 // Skip to the end of band data
591 if (ff_dca_seek_bits(&s->gb, band_data_end)) {
592 av_log(s->avctx, AV_LOG_ERROR, "Read past end of XLL band data\n");
593 return AVERROR_INVALIDDATA;
599 static av_cold void chs_clear_band_data(DCAXllDecoder *s, DCAXllChSet *c, int band, int seg)
601 DCAXllBand *b = &c->bands[band];
602 int i, offset, nsamples;
606 nsamples = s->nframesamples;
608 offset = seg * s->nsegsamples;
609 nsamples = s->nsegsamples;
612 for (i = 0; i < c->nchannels; i++) {
613 memset(b->msb_sample_buffer[i] + offset, 0, nsamples * sizeof(int32_t));
614 if (b->lsb_section_size)
615 memset(b->lsb_sample_buffer[i] + offset, 0, nsamples * sizeof(int32_t));
618 if (seg <= 0 && band)
619 memset(c->deci_history, 0, sizeof(c->deci_history));
622 memset(b->nscalablelsbs, 0, sizeof(b->nscalablelsbs));
623 memset(b->bit_width_adjust, 0, sizeof(b->bit_width_adjust));
627 static void chs_filter_band_data(DCAXllDecoder *s, DCAXllChSet *c, int band)
629 DCAXllBand *b = &c->bands[band];
630 int nsamples = s->nframesamples;
633 // Inverse adaptive or fixed prediction
634 for (i = 0; i < c->nchannels; i++) {
635 int32_t *buf = b->msb_sample_buffer[i];
636 int order = b->adapt_pred_order[i];
638 int coeff[DCA_XLL_ADAPT_PRED_ORDER_MAX];
639 // Conversion from reflection coefficients to direct form coefficients
640 for (j = 0; j < order; j++) {
641 int rc = b->adapt_refl_coeff[i][j];
642 for (k = 0; k < (j + 1) / 2; k++) {
643 int tmp1 = coeff[ k ];
644 int tmp2 = coeff[j - k - 1];
645 coeff[ k ] = tmp1 + mul16(rc, tmp2);
646 coeff[j - k - 1] = tmp2 + mul16(rc, tmp1);
650 // Inverse adaptive prediction
651 for (j = 0; j < nsamples - order; j++) {
653 for (k = 0; k < order; k++)
654 err += (int64_t)buf[j + k] * coeff[order - k - 1];
655 buf[j + k] -= (SUINT)clip23(norm16(err));
658 // Inverse fixed coefficient prediction
659 for (j = 0; j < b->fixed_pred_order[i]; j++)
660 for (k = 1; k < nsamples; k++)
661 buf[k] += (unsigned)buf[k - 1];
665 // Inverse pairwise channel decorrellation
666 if (b->decor_enabled) {
667 int32_t *tmp[DCA_XLL_CHANNELS_MAX];
669 for (i = 0; i < c->nchannels / 2; i++) {
670 int coeff = b->decor_coeff[i];
672 s->dcadsp->decor(b->msb_sample_buffer[i * 2 + 1],
673 b->msb_sample_buffer[i * 2 ],
678 // Reorder channel pointers to the original order
679 for (i = 0; i < c->nchannels; i++)
680 tmp[i] = b->msb_sample_buffer[i];
682 for (i = 0; i < c->nchannels; i++)
683 b->msb_sample_buffer[b->orig_order[i]] = tmp[i];
686 // Map output channel pointers for frequency band 0
687 if (c->nfreqbands == 1)
688 for (i = 0; i < c->nchannels; i++)
689 s->output_samples[c->ch_remap[i]] = b->msb_sample_buffer[i];
692 static int chs_get_lsb_width(DCAXllDecoder *s, DCAXllChSet *c, int band, int ch)
694 int adj = c->bands[band].bit_width_adjust[ch];
695 int shift = c->bands[band].nscalablelsbs[ch];
697 if (s->fixed_lsb_width)
698 shift = s->fixed_lsb_width;
699 else if (shift && adj)
707 static void chs_assemble_msbs_lsbs(DCAXllDecoder *s, DCAXllChSet *c, int band)
709 DCAXllBand *b = &c->bands[band];
710 int n, ch, nsamples = s->nframesamples;
712 for (ch = 0; ch < c->nchannels; ch++) {
713 int shift = chs_get_lsb_width(s, c, band, ch);
715 int32_t *msb = b->msb_sample_buffer[ch];
716 if (b->nscalablelsbs[ch]) {
717 int32_t *lsb = b->lsb_sample_buffer[ch];
718 int adj = b->bit_width_adjust[ch];
719 for (n = 0; n < nsamples; n++)
720 msb[n] = msb[n] * (SUINT)(1 << shift) + (lsb[n] << adj);
722 for (n = 0; n < nsamples; n++)
723 msb[n] = msb[n] * (SUINT)(1 << shift);
729 static int chs_assemble_freq_bands(DCAXllDecoder *s, DCAXllChSet *c)
731 int ch, nsamples = s->nframesamples;
734 av_assert1(c->nfreqbands > 1);
736 // Reallocate frequency band assembly buffer
737 av_fast_malloc(&c->sample_buffer[2], &c->sample_size[2],
738 2 * nsamples * c->nchannels * sizeof(int32_t));
739 if (!c->sample_buffer[2])
740 return AVERROR(ENOMEM);
742 // Assemble frequency bands 0 and 1
743 ptr = c->sample_buffer[2];
744 for (ch = 0; ch < c->nchannels; ch++) {
745 int32_t *band0 = c->bands[0].msb_sample_buffer[ch];
746 int32_t *band1 = c->bands[1].msb_sample_buffer[ch];
748 // Copy decimator history
749 memcpy(band0 - DCA_XLL_DECI_HISTORY_MAX,
750 c->deci_history[ch], sizeof(c->deci_history[0]));
753 s->dcadsp->assemble_freq_bands(ptr, band0, band1,
754 ff_dca_xll_band_coeff,
757 // Remap output channel pointer to assembly buffer
758 s->output_samples[c->ch_remap[ch]] = ptr;
765 static int parse_common_header(DCAXllDecoder *s)
767 int stream_ver, header_size, frame_size_nbits, nframesegs_log2;
769 // XLL extension sync word
770 if (get_bits_long(&s->gb, 32) != DCA_SYNCWORD_XLL) {
771 av_log(s->avctx, AV_LOG_VERBOSE, "Invalid XLL sync word\n");
772 return AVERROR(EAGAIN);
776 stream_ver = get_bits(&s->gb, 4) + 1;
777 if (stream_ver > 1) {
778 avpriv_request_sample(s->avctx, "XLL stream version %d", stream_ver);
779 return AVERROR_PATCHWELCOME;
782 // Lossless frame header length
783 header_size = get_bits(&s->gb, 8) + 1;
786 if (ff_dca_check_crc(s->avctx, &s->gb, 32, header_size * 8)) {
787 av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL common header checksum\n");
788 return AVERROR_INVALIDDATA;
791 // Number of bits used to read frame size
792 frame_size_nbits = get_bits(&s->gb, 5) + 1;
794 // Number of bytes in a lossless frame
795 s->frame_size = get_bits_long(&s->gb, frame_size_nbits);
796 if (s->frame_size < 0 || s->frame_size >= DCA_XLL_PBR_BUFFER_MAX) {
797 av_log(s->avctx, AV_LOG_ERROR, "Invalid XLL frame size (%d bytes)\n", s->frame_size);
798 return AVERROR_INVALIDDATA;
802 // Number of channels sets per frame
803 s->nchsets = get_bits(&s->gb, 4) + 1;
804 if (s->nchsets > DCA_XLL_CHSETS_MAX) {
805 avpriv_request_sample(s->avctx, "%d XLL channel sets", s->nchsets);
806 return AVERROR_PATCHWELCOME;
809 // Number of segments per frame
810 nframesegs_log2 = get_bits(&s->gb, 4);
811 s->nframesegs = 1 << nframesegs_log2;
812 if (s->nframesegs > 1024) {
813 av_log(s->avctx, AV_LOG_ERROR, "Too many segments per XLL frame\n");
814 return AVERROR_INVALIDDATA;
817 // Samples in segment per one frequency band for the first channel set
818 // Maximum value is 256 for sampling frequencies <= 48 kHz
819 // Maximum value is 512 for sampling frequencies > 48 kHz
820 s->nsegsamples_log2 = get_bits(&s->gb, 4);
821 if (!s->nsegsamples_log2) {
822 av_log(s->avctx, AV_LOG_ERROR, "Too few samples per XLL segment\n");
823 return AVERROR_INVALIDDATA;
825 s->nsegsamples = 1 << s->nsegsamples_log2;
826 if (s->nsegsamples > 512) {
827 av_log(s->avctx, AV_LOG_ERROR, "Too many samples per XLL segment\n");
828 return AVERROR_INVALIDDATA;
831 // Samples in frame per one frequency band for the first channel set
832 s->nframesamples_log2 = s->nsegsamples_log2 + nframesegs_log2;
833 s->nframesamples = 1 << s->nframesamples_log2;
834 if (s->nframesamples > 65536) {
835 av_log(s->avctx, AV_LOG_ERROR, "Too many samples per XLL frame\n");
836 return AVERROR_INVALIDDATA;
839 // Number of bits used to read segment size
840 s->seg_size_nbits = get_bits(&s->gb, 5) + 1;
842 // Presence of CRC16 within each frequency band
843 // 0 - No CRC16 within band
844 // 1 - CRC16 placed at the end of MSB0
845 // 2 - CRC16 placed at the end of MSB0 and LSB0
846 // 3 - CRC16 placed at the end of MSB0 and LSB0 and other frequency bands
847 s->band_crc_present = get_bits(&s->gb, 2);
849 // MSB/LSB split flag
850 s->scalable_lsbs = get_bits1(&s->gb);
852 // Channel position mask
853 s->ch_mask_nbits = get_bits(&s->gb, 5) + 1;
856 if (s->scalable_lsbs)
857 s->fixed_lsb_width = get_bits(&s->gb, 4);
859 s->fixed_lsb_width = 0;
863 // Header CRC16 protection
864 if (ff_dca_seek_bits(&s->gb, header_size * 8)) {
865 av_log(s->avctx, AV_LOG_ERROR, "Read past end of XLL common header\n");
866 return AVERROR_INVALIDDATA;
872 static int is_hier_dmix_chset(DCAXllChSet *c)
874 return !c->primary_chset && c->dmix_embedded && c->hier_chset;
877 static DCAXllChSet *find_next_hier_dmix_chset(DCAXllDecoder *s, DCAXllChSet *c)
880 while (++c < &s->chset[s->nchsets])
881 if (is_hier_dmix_chset(c))
887 static void prescale_down_mix(DCAXllChSet *c, DCAXllChSet *o)
889 int i, j, *coeff_ptr = c->dmix_coeff;
891 for (i = 0; i < c->hier_ofs; i++) {
892 int scale = o->dmix_scale[i];
893 int scale_inv = o->dmix_scale_inv[i];
894 c->dmix_scale[i] = mul15(c->dmix_scale[i], scale);
895 c->dmix_scale_inv[i] = mul16(c->dmix_scale_inv[i], scale_inv);
896 for (j = 0; j < c->nchannels; j++) {
897 int coeff = mul16(*coeff_ptr, scale_inv);
898 *coeff_ptr++ = mul15(coeff, o->dmix_scale[c->hier_ofs + j]);
903 static int parse_sub_headers(DCAXllDecoder *s, DCAExssAsset *asset)
905 DCAContext *dca = s->avctx->priv_data;
909 // Parse channel set headers
913 for (i = 0, c = s->chset; i < s->nchsets; i++, c++) {
914 c->hier_ofs = s->nchannels;
915 if ((ret = chs_parse_header(s, c, asset)) < 0)
917 if (c->nfreqbands > s->nfreqbands)
918 s->nfreqbands = c->nfreqbands;
920 s->nchannels += c->nchannels;
921 if (c->residual_encode != (1 << c->nchannels) - 1)
925 // Pre-scale downmixing coefficients for all non-primary channel sets
926 for (i = s->nchsets - 1, c = &s->chset[i]; i > 0; i--, c--) {
927 if (is_hier_dmix_chset(c)) {
928 DCAXllChSet *o = find_next_hier_dmix_chset(s, c);
930 prescale_down_mix(c, o);
934 // Determine number of active channel sets to decode
935 switch (dca->request_channel_layout) {
936 case DCA_SPEAKER_LAYOUT_STEREO:
937 s->nactivechsets = 1;
939 case DCA_SPEAKER_LAYOUT_5POINT0:
940 case DCA_SPEAKER_LAYOUT_5POINT1:
941 s->nactivechsets = (s->chset[0].nchannels < 5 && s->nchsets > 1) ? 2 : 1;
944 s->nactivechsets = s->nchsets;
951 static int parse_navi_table(DCAXllDecoder *s)
953 int chs, seg, band, navi_nb, navi_pos, *navi_ptr;
956 // Determine size of NAVI table
957 navi_nb = s->nfreqbands * s->nframesegs * s->nchsets;
958 if (navi_nb > 1024) {
959 av_log(s->avctx, AV_LOG_ERROR, "Too many NAVI entries (%d)\n", navi_nb);
960 return AVERROR_INVALIDDATA;
963 // Reallocate NAVI table
964 av_fast_malloc(&s->navi, &s->navi_size, navi_nb * sizeof(*s->navi));
966 return AVERROR(ENOMEM);
969 navi_pos = get_bits_count(&s->gb);
971 for (band = 0; band < s->nfreqbands; band++) {
972 for (seg = 0; seg < s->nframesegs; seg++) {
973 for (chs = 0, c = s->chset; chs < s->nchsets; chs++, c++) {
975 if (c->nfreqbands > band) {
976 size = get_bits_long(&s->gb, s->seg_size_nbits);
977 if (size < 0 || size >= s->frame_size) {
978 av_log(s->avctx, AV_LOG_ERROR, "Invalid NAVI segment size (%d bytes)\n", size);
979 return AVERROR_INVALIDDATA;
990 skip_bits(&s->gb, -get_bits_count(&s->gb) & 7);
991 skip_bits(&s->gb, 16);
994 if (ff_dca_check_crc(s->avctx, &s->gb, navi_pos, get_bits_count(&s->gb))) {
995 av_log(s->avctx, AV_LOG_ERROR, "Invalid NAVI checksum\n");
996 return AVERROR_INVALIDDATA;
1002 static int parse_band_data(DCAXllDecoder *s)
1004 int ret, chs, seg, band, navi_pos, *navi_ptr;
1007 for (chs = 0, c = s->chset; chs < s->nactivechsets; chs++, c++) {
1008 if ((ret = chs_alloc_msb_band_data(s, c)) < 0)
1010 if ((ret = chs_alloc_lsb_band_data(s, c)) < 0)
1014 navi_pos = get_bits_count(&s->gb);
1016 for (band = 0; band < s->nfreqbands; band++) {
1017 for (seg = 0; seg < s->nframesegs; seg++) {
1018 for (chs = 0, c = s->chset; chs < s->nchsets; chs++, c++) {
1019 if (c->nfreqbands > band) {
1020 navi_pos += *navi_ptr * 8;
1021 if (navi_pos > s->gb.size_in_bits) {
1022 av_log(s->avctx, AV_LOG_ERROR, "Invalid NAVI position\n");
1023 return AVERROR_INVALIDDATA;
1025 if (chs < s->nactivechsets &&
1026 (ret = chs_parse_band_data(s, c, band, seg, navi_pos)) < 0) {
1027 if (s->avctx->err_recognition & AV_EF_EXPLODE)
1029 chs_clear_band_data(s, c, band, seg);
1031 skip_bits_long(&s->gb, navi_pos - get_bits_count(&s->gb));
1041 static int parse_frame(DCAXllDecoder *s, uint8_t *data, int size, DCAExssAsset *asset)
1045 if ((ret = init_get_bits8(&s->gb, data, size)) < 0)
1047 if ((ret = parse_common_header(s)) < 0)
1049 if ((ret = parse_sub_headers(s, asset)) < 0)
1051 if ((ret = parse_navi_table(s)) < 0)
1053 if ((ret = parse_band_data(s)) < 0)
1055 if (ff_dca_seek_bits(&s->gb, s->frame_size * 8)) {
1056 av_log(s->avctx, AV_LOG_ERROR, "Read past end of XLL frame\n");
1057 return AVERROR_INVALIDDATA;
1062 static void clear_pbr(DCAXllDecoder *s)
1068 static int copy_to_pbr(DCAXllDecoder *s, uint8_t *data, int size, int delay)
1070 if (size > DCA_XLL_PBR_BUFFER_MAX)
1071 return AVERROR(ENOSPC);
1073 if (!s->pbr_buffer && !(s->pbr_buffer = av_malloc(DCA_XLL_PBR_BUFFER_MAX + AV_INPUT_BUFFER_PADDING_SIZE)))
1074 return AVERROR(ENOMEM);
1076 memcpy(s->pbr_buffer, data, size);
1077 s->pbr_length = size;
1078 s->pbr_delay = delay;
1082 static int parse_frame_no_pbr(DCAXllDecoder *s, uint8_t *data, int size, DCAExssAsset *asset)
1084 int ret = parse_frame(s, data, size, asset);
1086 // If XLL packet data didn't start with a sync word, we must have jumped
1087 // right into the middle of PBR smoothing period
1088 if (ret == AVERROR(EAGAIN) && asset->xll_sync_present && asset->xll_sync_offset < size) {
1089 // Skip to the next sync word in this packet
1090 data += asset->xll_sync_offset;
1091 size -= asset->xll_sync_offset;
1093 // If decoding delay is set, put the frame into PBR buffer and return
1094 // failure code. Higher level decoder is expected to switch to lossy
1095 // core decoding or mute its output until decoding delay expires.
1096 if (asset->xll_delay_nframes > 0) {
1097 if ((ret = copy_to_pbr(s, data, size, asset->xll_delay_nframes)) < 0)
1099 return AVERROR(EAGAIN);
1102 // No decoding delay, just parse the frame in place
1103 ret = parse_frame(s, data, size, asset);
1109 if (s->frame_size > size)
1110 return AVERROR(EINVAL);
1112 // If the XLL decoder didn't consume full packet, start PBR smoothing period
1113 if (s->frame_size < size)
1114 if ((ret = copy_to_pbr(s, data + s->frame_size, size - s->frame_size, 0)) < 0)
1120 static int parse_frame_pbr(DCAXllDecoder *s, uint8_t *data, int size, DCAExssAsset *asset)
1124 if (size > DCA_XLL_PBR_BUFFER_MAX - s->pbr_length) {
1125 ret = AVERROR(ENOSPC);
1129 memcpy(s->pbr_buffer + s->pbr_length, data, size);
1130 s->pbr_length += size;
1132 // Respect decoding delay after synchronization error
1133 if (s->pbr_delay > 0 && --s->pbr_delay)
1134 return AVERROR(EAGAIN);
1136 if ((ret = parse_frame(s, s->pbr_buffer, s->pbr_length, asset)) < 0)
1139 if (s->frame_size > s->pbr_length) {
1140 ret = AVERROR(EINVAL);
1144 if (s->frame_size == s->pbr_length) {
1145 // End of PBR smoothing period
1148 s->pbr_length -= s->frame_size;
1149 memmove(s->pbr_buffer, s->pbr_buffer + s->frame_size, s->pbr_length);
1155 // For now, throw out all PBR state on failure.
1156 // Perhaps we can be smarter and try to resync somehow.
1161 int ff_dca_xll_parse(DCAXllDecoder *s, uint8_t *data, DCAExssAsset *asset)
1165 if (s->hd_stream_id != asset->hd_stream_id) {
1167 s->hd_stream_id = asset->hd_stream_id;
1171 ret = parse_frame_pbr(s, data + asset->xll_offset, asset->xll_size, asset);
1173 ret = parse_frame_no_pbr(s, data + asset->xll_offset, asset->xll_size, asset);
1178 static void undo_down_mix(DCAXllDecoder *s, DCAXllChSet *o, int band)
1180 int i, j, k, nchannels = 0, *coeff_ptr = o->dmix_coeff;
1183 for (i = 0, c = s->chset; i < s->nactivechsets; i++, c++) {
1187 av_assert1(band < c->nfreqbands);
1188 for (j = 0; j < c->nchannels; j++) {
1189 for (k = 0; k < o->nchannels; k++) {
1190 int coeff = *coeff_ptr++;
1192 s->dcadsp->dmix_sub(c->bands[band].msb_sample_buffer[j],
1193 o->bands[band].msb_sample_buffer[k],
1194 coeff, s->nframesamples);
1196 s->dcadsp->dmix_sub(c->deci_history[j],
1198 coeff, DCA_XLL_DECI_HISTORY_MAX);
1203 nchannels += c->nchannels;
1204 if (nchannels >= o->hier_ofs)
1209 static void scale_down_mix(DCAXllDecoder *s, DCAXllChSet *o, int band)
1211 int i, j, nchannels = 0;
1214 for (i = 0, c = s->chset; i < s->nactivechsets; i++, c++) {
1218 av_assert1(band < c->nfreqbands);
1219 for (j = 0; j < c->nchannels; j++) {
1220 int scale = o->dmix_scale[nchannels++];
1221 if (scale != (1 << 15)) {
1222 s->dcadsp->dmix_scale(c->bands[band].msb_sample_buffer[j],
1223 scale, s->nframesamples);
1225 s->dcadsp->dmix_scale(c->deci_history[j],
1226 scale, DCA_XLL_DECI_HISTORY_MAX);
1230 if (nchannels >= o->hier_ofs)
1235 // Clear all band data and replace non-residual encoded channels with lossy
1237 static av_cold void force_lossy_output(DCAXllDecoder *s, DCAXllChSet *c)
1239 DCAContext *dca = s->avctx->priv_data;
1242 for (band = 0; band < c->nfreqbands; band++)
1243 chs_clear_band_data(s, c, band, -1);
1245 for (ch = 0; ch < c->nchannels; ch++) {
1246 if (!(c->residual_encode & (1 << ch)))
1248 if (ff_dca_core_map_spkr(&dca->core, c->ch_remap[ch]) < 0)
1250 c->residual_encode &= ~(1 << ch);
1254 static int combine_residual_frame(DCAXllDecoder *s, DCAXllChSet *c)
1256 DCAContext *dca = s->avctx->priv_data;
1257 int ch, nsamples = s->nframesamples;
1260 // Verify that core is compatible
1261 if (!(dca->packet & DCA_PACKET_CORE)) {
1262 av_log(s->avctx, AV_LOG_ERROR, "Residual encoded channels are present without core\n");
1263 return AVERROR(EINVAL);
1266 if (c->freq != dca->core.output_rate) {
1267 av_log(s->avctx, AV_LOG_WARNING, "Sample rate mismatch between core (%d Hz) and XLL (%d Hz)\n", dca->core.output_rate, c->freq);
1268 return AVERROR_INVALIDDATA;
1271 if (nsamples != dca->core.npcmsamples) {
1272 av_log(s->avctx, AV_LOG_WARNING, "Number of samples per frame mismatch between core (%d) and XLL (%d)\n", dca->core.npcmsamples, nsamples);
1273 return AVERROR_INVALIDDATA;
1276 // See if this channel set is downmixed and find the next channel set in
1277 // hierarchy. If downmixed, undo core pre-scaling before combining with
1278 // residual (residual is not scaled).
1279 o = find_next_hier_dmix_chset(s, c);
1281 // Reduce core bit width and combine with residual
1282 for (ch = 0; ch < c->nchannels; ch++) {
1283 int n, spkr, shift, round;
1286 if (c->residual_encode & (1 << ch))
1289 // Map this channel to core speaker
1290 spkr = ff_dca_core_map_spkr(&dca->core, c->ch_remap[ch]);
1292 av_log(s->avctx, AV_LOG_WARNING, "Residual encoded channel (%d) references unavailable core channel\n", c->ch_remap[ch]);
1293 return AVERROR_INVALIDDATA;
1296 // Account for LSB width
1297 shift = 24 - c->pcm_bit_res + chs_get_lsb_width(s, c, 0, ch);
1299 av_log(s->avctx, AV_LOG_WARNING, "Invalid core shift (%d bits)\n", shift);
1300 return AVERROR_INVALIDDATA;
1303 round = shift > 0 ? 1 << (shift - 1) : 0;
1305 src = dca->core.output_samples[spkr];
1306 dst = c->bands[0].msb_sample_buffer[ch];
1308 // Undo embedded core downmix pre-scaling
1309 int scale_inv = o->dmix_scale_inv[c->hier_ofs + ch];
1310 for (n = 0; n < nsamples; n++)
1311 dst[n] += (SUINT)clip23((mul16(src[n], scale_inv) + round) >> shift);
1313 // No downmix scaling
1314 for (n = 0; n < nsamples; n++)
1315 dst[n] += (unsigned)((src[n] + round) >> shift);
1322 int ff_dca_xll_filter_frame(DCAXllDecoder *s, AVFrame *frame)
1324 AVCodecContext *avctx = s->avctx;
1325 DCAContext *dca = avctx->priv_data;
1326 DCAExssAsset *asset = &dca->exss.assets[0];
1327 DCAXllChSet *p = &s->chset[0], *c;
1328 enum AVMatrixEncoding matrix_encoding = AV_MATRIX_ENCODING_NONE;
1329 int i, j, k, ret, shift, nsamples, request_mask;
1330 int ch_remap[DCA_SPEAKER_COUNT];
1332 // Force lossy downmixed output during recovery
1333 if (dca->packet & DCA_PACKET_RECOVERY) {
1334 for (i = 0, c = s->chset; i < s->nchsets; i++, c++) {
1335 if (i < s->nactivechsets)
1336 force_lossy_output(s, c);
1338 if (!c->primary_chset)
1339 c->dmix_embedded = 0;
1342 s->scalable_lsbs = 0;
1343 s->fixed_lsb_width = 0;
1346 // Filter frequency bands for active channel sets
1348 for (i = 0, c = s->chset; i < s->nactivechsets; i++, c++) {
1349 chs_filter_band_data(s, c, 0);
1351 if (c->residual_encode != (1 << c->nchannels) - 1
1352 && (ret = combine_residual_frame(s, c)) < 0)
1355 if (s->scalable_lsbs)
1356 chs_assemble_msbs_lsbs(s, c, 0);
1358 if (c->nfreqbands > 1) {
1359 chs_filter_band_data(s, c, 1);
1360 chs_assemble_msbs_lsbs(s, c, 1);
1363 s->output_mask |= c->ch_mask;
1366 // Undo hierarchial downmix and/or apply scaling
1367 for (i = 1, c = &s->chset[1]; i < s->nchsets; i++, c++) {
1368 if (!is_hier_dmix_chset(c))
1371 if (i >= s->nactivechsets) {
1372 for (j = 0; j < c->nfreqbands; j++)
1373 if (c->bands[j].dmix_embedded)
1374 scale_down_mix(s, c, j);
1378 for (j = 0; j < c->nfreqbands; j++)
1379 if (c->bands[j].dmix_embedded)
1380 undo_down_mix(s, c, j);
1383 // Assemble frequency bands for active channel sets
1384 if (s->nfreqbands > 1) {
1385 for (i = 0; i < s->nactivechsets; i++)
1386 if ((ret = chs_assemble_freq_bands(s, &s->chset[i])) < 0)
1390 // Normalize to regular 5.1 layout if downmixing
1391 if (dca->request_channel_layout) {
1392 if (s->output_mask & DCA_SPEAKER_MASK_Lss) {
1393 s->output_samples[DCA_SPEAKER_Ls] = s->output_samples[DCA_SPEAKER_Lss];
1394 s->output_mask = (s->output_mask & ~DCA_SPEAKER_MASK_Lss) | DCA_SPEAKER_MASK_Ls;
1396 if (s->output_mask & DCA_SPEAKER_MASK_Rss) {
1397 s->output_samples[DCA_SPEAKER_Rs] = s->output_samples[DCA_SPEAKER_Rss];
1398 s->output_mask = (s->output_mask & ~DCA_SPEAKER_MASK_Rss) | DCA_SPEAKER_MASK_Rs;
1402 // Handle downmixing to stereo request
1403 if (dca->request_channel_layout == DCA_SPEAKER_LAYOUT_STEREO
1404 && DCA_HAS_STEREO(s->output_mask) && p->dmix_embedded
1405 && (p->dmix_type == DCA_DMIX_TYPE_LoRo ||
1406 p->dmix_type == DCA_DMIX_TYPE_LtRt))
1407 request_mask = DCA_SPEAKER_LAYOUT_STEREO;
1409 request_mask = s->output_mask;
1410 if (!ff_dca_set_channel_layout(avctx, ch_remap, request_mask))
1411 return AVERROR(EINVAL);
1413 avctx->sample_rate = p->freq << (s->nfreqbands - 1);
1415 switch (p->storage_bit_res) {
1417 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
1418 shift = 16 - p->pcm_bit_res;
1422 avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
1423 shift = 24 - p->pcm_bit_res;
1426 return AVERROR(EINVAL);
1429 avctx->bits_per_raw_sample = p->storage_bit_res;
1430 avctx->profile = FF_PROFILE_DTS_HD_MA;
1431 avctx->bit_rate = 0;
1433 frame->nb_samples = nsamples = s->nframesamples << (s->nfreqbands - 1);
1434 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1437 // Downmix primary channel set to stereo
1438 if (request_mask != s->output_mask) {
1439 ff_dca_downmix_to_stereo_fixed(s->dcadsp, s->output_samples,
1440 p->dmix_coeff, nsamples,
1444 for (i = 0; i < avctx->channels; i++) {
1445 int32_t *samples = s->output_samples[ch_remap[i]];
1446 if (frame->format == AV_SAMPLE_FMT_S16P) {
1447 int16_t *plane = (int16_t *)frame->extended_data[i];
1448 for (k = 0; k < nsamples; k++)
1449 plane[k] = av_clip_int16(samples[k] * (SUINT)(1 << shift));
1451 int32_t *plane = (int32_t *)frame->extended_data[i];
1452 for (k = 0; k < nsamples; k++)
1453 plane[k] = clip23(samples[k] * (SUINT)(1 << shift)) * (1 << 8);
1457 if (!asset->one_to_one_map_ch_to_spkr) {
1458 if (asset->representation_type == DCA_REPR_TYPE_LtRt)
1459 matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
1460 else if (asset->representation_type == DCA_REPR_TYPE_LhRh)
1461 matrix_encoding = AV_MATRIX_ENCODING_DOLBYHEADPHONE;
1462 } else if (request_mask != s->output_mask && p->dmix_type == DCA_DMIX_TYPE_LtRt) {
1463 matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
1465 if ((ret = ff_side_data_update_matrix_encoding(frame, matrix_encoding)) < 0)
1471 av_cold void ff_dca_xll_flush(DCAXllDecoder *s)
1476 av_cold void ff_dca_xll_close(DCAXllDecoder *s)
1481 for (i = 0, c = s->chset; i < DCA_XLL_CHSETS_MAX; i++, c++) {
1482 for (j = 0; j < DCA_XLL_SAMPLE_BUFFERS_MAX; j++) {
1483 av_freep(&c->sample_buffer[j]);
1484 c->sample_size[j] = 0;
1491 av_freep(&s->pbr_buffer);