3 * Copyright (c) 2007-2008 Ian Caulfield
5 * This file is part of Libav.
7 * Libav is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * Libav is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
30 #include "libavutil/internal.h"
31 #include "libavutil/intreadwrite.h"
32 #include "libavutil/channel_layout.h"
35 #include "libavutil/crc.h"
37 #include "mlp_parser.h"
41 /** number of bits used for VLC lookup - longest Huffman code is 9 */
44 typedef struct SubStream {
45 /// Set if a valid restart header has been read. Otherwise the substream cannot be decoded.
49 /** restart header data */
50 /// The type of noise to be used in the rematrix stage.
53 /// The index of the first channel coded in this substream.
55 /// The index of the last channel coded in this substream.
57 /// The number of channels input into the rematrix stage.
58 uint8_t max_matrix_channel;
59 /// For each channel output by the matrix, the output channel to map it to
60 uint8_t ch_assign[MAX_CHANNELS];
61 /// The channel layout for this substream
64 /// Channel coding parameters for channels in the substream
65 ChannelParams channel_params[MAX_CHANNELS];
67 /// The left shift applied to random noise in 0x31ea substreams.
69 /// The current seed value for the pseudorandom noise generator(s).
70 uint32_t noisegen_seed;
72 /// Set if the substream contains extra info to check the size of VLC blocks.
73 uint8_t data_check_present;
75 /// Bitmask of which parameter sets are conveyed in a decoding parameter block.
76 uint8_t param_presence_flags;
77 #define PARAM_BLOCKSIZE (1 << 7)
78 #define PARAM_MATRIX (1 << 6)
79 #define PARAM_OUTSHIFT (1 << 5)
80 #define PARAM_QUANTSTEP (1 << 4)
81 #define PARAM_FIR (1 << 3)
82 #define PARAM_IIR (1 << 2)
83 #define PARAM_HUFFOFFSET (1 << 1)
84 #define PARAM_PRESENCE (1 << 0)
90 /// Number of matrices to be applied.
91 uint8_t num_primitive_matrices;
93 /// matrix output channel
94 uint8_t matrix_out_ch[MAX_MATRICES];
96 /// Whether the LSBs of the matrix output are encoded in the bitstream.
97 uint8_t lsb_bypass[MAX_MATRICES];
98 /// Matrix coefficients, stored as 2.14 fixed point.
99 int32_t matrix_coeff[MAX_MATRICES][MAX_CHANNELS];
100 /// Left shift to apply to noise values in 0x31eb substreams.
101 uint8_t matrix_noise_shift[MAX_MATRICES];
104 /// Left shift to apply to Huffman-decoded residuals.
105 uint8_t quant_step_size[MAX_CHANNELS];
107 /// number of PCM samples in current audio block
109 /// Number of PCM samples decoded so far in this frame.
112 /// Left shift to apply to decoded PCM values to get final 24-bit output.
113 int8_t output_shift[MAX_CHANNELS];
115 /// Running XOR of all output samples.
116 int32_t lossless_check_data;
120 typedef struct MLPDecodeContext {
121 AVCodecContext *avctx;
123 /// Current access unit being read has a major sync.
124 int is_major_sync_unit;
126 /// Set if a valid major sync block has been read. Otherwise no decoding is possible.
127 uint8_t params_valid;
129 /// Number of substreams contained within this stream.
130 uint8_t num_substreams;
132 /// Index of the last substream to decode - further substreams are skipped.
133 uint8_t max_decoded_substream;
135 /// number of PCM samples contained in each frame
136 int access_unit_size;
137 /// next power of two above the number of samples in each frame
138 int access_unit_size_pow2;
140 SubStream substream[MAX_SUBSTREAMS];
143 int filter_changed[MAX_CHANNELS][NUM_FILTERS];
145 int8_t noise_buffer[MAX_BLOCKSIZE_POW2];
146 int8_t bypassed_lsbs[MAX_BLOCKSIZE][MAX_CHANNELS];
147 int32_t sample_buffer[MAX_BLOCKSIZE][MAX_CHANNELS];
152 static const uint64_t thd_channel_order[] = {
153 AV_CH_FRONT_LEFT, AV_CH_FRONT_RIGHT, // LR
154 AV_CH_FRONT_CENTER, // C
155 AV_CH_LOW_FREQUENCY, // LFE
156 AV_CH_SIDE_LEFT, AV_CH_SIDE_RIGHT, // LRs
157 AV_CH_TOP_FRONT_LEFT, AV_CH_TOP_FRONT_RIGHT, // LRvh
158 AV_CH_FRONT_LEFT_OF_CENTER, AV_CH_FRONT_RIGHT_OF_CENTER, // LRc
159 AV_CH_BACK_LEFT, AV_CH_BACK_RIGHT, // LRrs
160 AV_CH_BACK_CENTER, // Cs
161 AV_CH_TOP_CENTER, // Ts
162 AV_CH_SURROUND_DIRECT_LEFT, AV_CH_SURROUND_DIRECT_RIGHT, // LRsd
163 AV_CH_WIDE_LEFT, AV_CH_WIDE_RIGHT, // LRw
164 AV_CH_TOP_FRONT_CENTER, // Cvh
165 AV_CH_LOW_FREQUENCY_2, // LFE2
168 static uint64_t thd_channel_layout_extract_channel(uint64_t channel_layout,
173 if (av_get_channel_layout_nb_channels(channel_layout) <= index)
176 for (i = 0; i < FF_ARRAY_ELEMS(thd_channel_order); i++)
177 if (channel_layout & thd_channel_order[i] && !index--)
178 return thd_channel_order[i];
182 static VLC huff_vlc[3];
184 /** Initialize static data, constant between all invocations of the codec. */
186 static av_cold void init_static(void)
188 if (!huff_vlc[0].bits) {
189 INIT_VLC_STATIC(&huff_vlc[0], VLC_BITS, 18,
190 &ff_mlp_huffman_tables[0][0][1], 2, 1,
191 &ff_mlp_huffman_tables[0][0][0], 2, 1, 512);
192 INIT_VLC_STATIC(&huff_vlc[1], VLC_BITS, 16,
193 &ff_mlp_huffman_tables[1][0][1], 2, 1,
194 &ff_mlp_huffman_tables[1][0][0], 2, 1, 512);
195 INIT_VLC_STATIC(&huff_vlc[2], VLC_BITS, 15,
196 &ff_mlp_huffman_tables[2][0][1], 2, 1,
197 &ff_mlp_huffman_tables[2][0][0], 2, 1, 512);
203 static inline int32_t calculate_sign_huff(MLPDecodeContext *m,
204 unsigned int substr, unsigned int ch)
206 SubStream *s = &m->substream[substr];
207 ChannelParams *cp = &s->channel_params[ch];
208 int lsb_bits = cp->huff_lsbs - s->quant_step_size[ch];
209 int sign_shift = lsb_bits + (cp->codebook ? 2 - cp->codebook : -1);
210 int32_t sign_huff_offset = cp->huff_offset;
212 if (cp->codebook > 0)
213 sign_huff_offset -= 7 << lsb_bits;
216 sign_huff_offset -= 1 << sign_shift;
218 return sign_huff_offset;
221 /** Read a sample, consisting of either, both or neither of entropy-coded MSBs
224 static inline int read_huff_channels(MLPDecodeContext *m, GetBitContext *gbp,
225 unsigned int substr, unsigned int pos)
227 SubStream *s = &m->substream[substr];
228 unsigned int mat, channel;
230 for (mat = 0; mat < s->num_primitive_matrices; mat++)
231 if (s->lsb_bypass[mat])
232 m->bypassed_lsbs[pos + s->blockpos][mat] = get_bits1(gbp);
234 for (channel = s->min_channel; channel <= s->max_channel; channel++) {
235 ChannelParams *cp = &s->channel_params[channel];
236 int codebook = cp->codebook;
237 int quant_step_size = s->quant_step_size[channel];
238 int lsb_bits = cp->huff_lsbs - quant_step_size;
242 result = get_vlc2(gbp, huff_vlc[codebook-1].table,
243 VLC_BITS, (9 + VLC_BITS - 1) / VLC_BITS);
246 return AVERROR_INVALIDDATA;
249 result = (result << lsb_bits) + get_bits(gbp, lsb_bits);
251 result += cp->sign_huff_offset;
252 result <<= quant_step_size;
254 m->sample_buffer[pos + s->blockpos][channel] = result;
260 static av_cold int mlp_decode_init(AVCodecContext *avctx)
262 MLPDecodeContext *m = avctx->priv_data;
267 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
268 m->substream[substr].lossless_check_data = 0xffffffff;
269 ff_mlpdsp_init(&m->dsp);
274 /** Read a major sync info header - contains high level information about
275 * the stream - sample rate, channel arrangement etc. Most of this
276 * information is not actually necessary for decoding, only for playback.
279 static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
284 if ((ret = ff_mlp_read_major_sync(m->avctx, &mh, gb)) != 0)
287 if (mh.group1_bits == 0) {
288 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown bits per sample\n");
289 return AVERROR_INVALIDDATA;
291 if (mh.group2_bits > mh.group1_bits) {
292 av_log(m->avctx, AV_LOG_ERROR,
293 "Channel group 2 cannot have more bits per sample than group 1.\n");
294 return AVERROR_INVALIDDATA;
297 if (mh.group2_samplerate && mh.group2_samplerate != mh.group1_samplerate) {
298 av_log(m->avctx, AV_LOG_ERROR,
299 "Channel groups with differing sample rates are not currently supported.\n");
300 return AVERROR_INVALIDDATA;
303 if (mh.group1_samplerate == 0) {
304 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown sampling rate\n");
305 return AVERROR_INVALIDDATA;
307 if (mh.group1_samplerate > MAX_SAMPLERATE) {
308 av_log(m->avctx, AV_LOG_ERROR,
309 "Sampling rate %d is greater than the supported maximum (%d).\n",
310 mh.group1_samplerate, MAX_SAMPLERATE);
311 return AVERROR_INVALIDDATA;
313 if (mh.access_unit_size > MAX_BLOCKSIZE) {
314 av_log(m->avctx, AV_LOG_ERROR,
315 "Block size %d is greater than the supported maximum (%d).\n",
316 mh.access_unit_size, MAX_BLOCKSIZE);
317 return AVERROR_INVALIDDATA;
319 if (mh.access_unit_size_pow2 > MAX_BLOCKSIZE_POW2) {
320 av_log(m->avctx, AV_LOG_ERROR,
321 "Block size pow2 %d is greater than the supported maximum (%d).\n",
322 mh.access_unit_size_pow2, MAX_BLOCKSIZE_POW2);
323 return AVERROR_INVALIDDATA;
326 if (mh.num_substreams == 0)
327 return AVERROR_INVALIDDATA;
328 if (m->avctx->codec_id == AV_CODEC_ID_MLP && mh.num_substreams > 2) {
329 av_log(m->avctx, AV_LOG_ERROR, "MLP only supports up to 2 substreams.\n");
330 return AVERROR_INVALIDDATA;
332 if (mh.num_substreams > MAX_SUBSTREAMS) {
333 avpriv_request_sample(m->avctx,
334 "%d substreams (more than the "
335 "maximum supported by the decoder)",
337 return AVERROR_PATCHWELCOME;
340 m->access_unit_size = mh.access_unit_size;
341 m->access_unit_size_pow2 = mh.access_unit_size_pow2;
343 m->num_substreams = mh.num_substreams;
344 m->max_decoded_substream = m->num_substreams - 1;
346 m->avctx->sample_rate = mh.group1_samplerate;
347 m->avctx->frame_size = mh.access_unit_size;
349 m->avctx->bits_per_raw_sample = mh.group1_bits;
350 if (mh.group1_bits > 16)
351 m->avctx->sample_fmt = AV_SAMPLE_FMT_S32;
353 m->avctx->sample_fmt = AV_SAMPLE_FMT_S16;
356 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
357 m->substream[substr].restart_seen = 0;
359 /* Set the layout for each substream. When there's more than one, the first
360 * substream is Stereo. Subsequent substreams' layouts are indicated in the
362 if (m->avctx->codec_id == AV_CODEC_ID_MLP) {
363 if ((substr = (mh.num_substreams > 1)))
364 m->substream[0].ch_layout = AV_CH_LAYOUT_STEREO;
365 m->substream[substr].ch_layout = mh.channel_layout_mlp;
367 if ((substr = (mh.num_substreams > 1)))
368 m->substream[0].ch_layout = AV_CH_LAYOUT_STEREO;
369 if (mh.num_substreams > 2)
370 if (mh.channel_layout_thd_stream2)
371 m->substream[2].ch_layout = mh.channel_layout_thd_stream2;
373 m->substream[2].ch_layout = mh.channel_layout_thd_stream1;
374 m->substream[substr].ch_layout = mh.channel_layout_thd_stream1;
380 /** Read a restart header from a block in a substream. This contains parameters
381 * required to decode the audio that do not change very often. Generally
382 * (always) present only in blocks following a major sync. */
384 static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
385 const uint8_t *buf, unsigned int substr)
387 SubStream *s = &m->substream[substr];
391 uint8_t lossless_check;
392 int start_count = get_bits_count(gbp);
393 int min_channel, max_channel, max_matrix_channel;
394 const int std_max_matrix_channel = m->avctx->codec_id == AV_CODEC_ID_MLP
395 ? MAX_MATRIX_CHANNEL_MLP
396 : MAX_MATRIX_CHANNEL_TRUEHD;
398 sync_word = get_bits(gbp, 13);
400 if (sync_word != 0x31ea >> 1) {
401 av_log(m->avctx, AV_LOG_ERROR,
402 "restart header sync incorrect (got 0x%04x)\n", sync_word);
403 return AVERROR_INVALIDDATA;
406 s->noise_type = get_bits1(gbp);
408 if (m->avctx->codec_id == AV_CODEC_ID_MLP && s->noise_type) {
409 av_log(m->avctx, AV_LOG_ERROR, "MLP must have 0x31ea sync word.\n");
410 return AVERROR_INVALIDDATA;
413 skip_bits(gbp, 16); /* Output timestamp */
415 min_channel = get_bits(gbp, 4);
416 max_channel = get_bits(gbp, 4);
417 max_matrix_channel = get_bits(gbp, 4);
419 if (max_matrix_channel > std_max_matrix_channel) {
420 av_log(m->avctx, AV_LOG_ERROR,
421 "Max matrix channel cannot be greater than %d.\n",
423 return AVERROR_INVALIDDATA;
426 if (max_channel != max_matrix_channel) {
427 av_log(m->avctx, AV_LOG_ERROR,
428 "Max channel must be equal max matrix channel.\n");
429 return AVERROR_INVALIDDATA;
432 /* This should happen for TrueHD streams with >6 channels and MLP's noise
433 * type. It is not yet known if this is allowed. */
434 if (s->max_channel > MAX_MATRIX_CHANNEL_MLP && !s->noise_type) {
435 avpriv_request_sample(m->avctx,
436 "%d channels (more than the "
437 "maximum supported by the decoder)",
439 return AVERROR_PATCHWELCOME;
442 if (min_channel > max_channel) {
443 av_log(m->avctx, AV_LOG_ERROR,
444 "Substream min channel cannot be greater than max channel.\n");
445 return AVERROR_INVALIDDATA;
448 s->min_channel = min_channel;
449 s->max_channel = max_channel;
450 s->max_matrix_channel = max_matrix_channel;
452 #if FF_API_REQUEST_CHANNELS
453 FF_DISABLE_DEPRECATION_WARNINGS
454 if (m->avctx->request_channels > 0 &&
455 m->avctx->request_channels <= s->max_channel + 1 &&
456 m->max_decoded_substream > substr) {
457 av_log(m->avctx, AV_LOG_DEBUG,
458 "Extracting %d-channel downmix from substream %d. "
459 "Further substreams will be skipped.\n",
460 s->max_channel + 1, substr);
461 m->max_decoded_substream = substr;
463 FF_ENABLE_DEPRECATION_WARNINGS
465 if (m->avctx->request_channel_layout == s->ch_layout &&
466 m->max_decoded_substream > substr) {
467 av_log(m->avctx, AV_LOG_DEBUG,
468 "Extracting %d-channel downmix (0x%"PRIx64") from substream %d. "
469 "Further substreams will be skipped.\n",
470 s->max_channel + 1, s->ch_layout, substr);
471 m->max_decoded_substream = substr;
474 s->noise_shift = get_bits(gbp, 4);
475 s->noisegen_seed = get_bits(gbp, 23);
479 s->data_check_present = get_bits1(gbp);
480 lossless_check = get_bits(gbp, 8);
481 if (substr == m->max_decoded_substream
482 && s->lossless_check_data != 0xffffffff) {
483 tmp = xor_32_to_8(s->lossless_check_data);
484 if (tmp != lossless_check)
485 av_log(m->avctx, AV_LOG_WARNING,
486 "Lossless check failed - expected %02x, calculated %02x.\n",
487 lossless_check, tmp);
492 memset(s->ch_assign, 0, sizeof(s->ch_assign));
494 for (ch = 0; ch <= s->max_matrix_channel; ch++) {
495 int ch_assign = get_bits(gbp, 6);
496 if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD) {
497 uint64_t channel = thd_channel_layout_extract_channel(s->ch_layout,
499 ch_assign = av_get_channel_layout_channel_index(s->ch_layout,
502 if (ch_assign > s->max_matrix_channel) {
503 avpriv_request_sample(m->avctx,
504 "Assignment of matrix channel %d to invalid output channel %d",
506 return AVERROR_PATCHWELCOME;
508 s->ch_assign[ch_assign] = ch;
511 checksum = ff_mlp_restart_checksum(buf, get_bits_count(gbp) - start_count);
513 if (checksum != get_bits(gbp, 8))
514 av_log(m->avctx, AV_LOG_ERROR, "restart header checksum error\n");
516 /* Set default decoding parameters. */
517 s->param_presence_flags = 0xff;
518 s->num_primitive_matrices = 0;
520 s->lossless_check_data = 0;
522 memset(s->output_shift , 0, sizeof(s->output_shift ));
523 memset(s->quant_step_size, 0, sizeof(s->quant_step_size));
525 for (ch = s->min_channel; ch <= s->max_channel; ch++) {
526 ChannelParams *cp = &s->channel_params[ch];
527 cp->filter_params[FIR].order = 0;
528 cp->filter_params[IIR].order = 0;
529 cp->filter_params[FIR].shift = 0;
530 cp->filter_params[IIR].shift = 0;
532 /* Default audio coding is 24-bit raw PCM. */
534 cp->sign_huff_offset = (-1) << 23;
539 if (substr == m->max_decoded_substream) {
540 m->avctx->channels = s->max_matrix_channel + 1;
541 m->avctx->channel_layout = s->ch_layout;
547 /** Read parameters for one of the prediction filters. */
549 static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
550 unsigned int substr, unsigned int channel,
553 SubStream *s = &m->substream[substr];
554 FilterParams *fp = &s->channel_params[channel].filter_params[filter];
555 const int max_order = filter ? MAX_IIR_ORDER : MAX_FIR_ORDER;
556 const char fchar = filter ? 'I' : 'F';
559 // Filter is 0 for FIR, 1 for IIR.
562 if (m->filter_changed[channel][filter]++ > 1) {
563 av_log(m->avctx, AV_LOG_ERROR, "Filters may change only once per access unit.\n");
564 return AVERROR_INVALIDDATA;
567 order = get_bits(gbp, 4);
568 if (order > max_order) {
569 av_log(m->avctx, AV_LOG_ERROR,
570 "%cIR filter order %d is greater than maximum %d.\n",
571 fchar, order, max_order);
572 return AVERROR_INVALIDDATA;
577 int32_t *fcoeff = s->channel_params[channel].coeff[filter];
578 int coeff_bits, coeff_shift;
580 fp->shift = get_bits(gbp, 4);
582 coeff_bits = get_bits(gbp, 5);
583 coeff_shift = get_bits(gbp, 3);
584 if (coeff_bits < 1 || coeff_bits > 16) {
585 av_log(m->avctx, AV_LOG_ERROR,
586 "%cIR filter coeff_bits must be between 1 and 16.\n",
588 return AVERROR_INVALIDDATA;
590 if (coeff_bits + coeff_shift > 16) {
591 av_log(m->avctx, AV_LOG_ERROR,
592 "Sum of coeff_bits and coeff_shift for %cIR filter must be 16 or less.\n",
594 return AVERROR_INVALIDDATA;
597 for (i = 0; i < order; i++)
598 fcoeff[i] = get_sbits(gbp, coeff_bits) << coeff_shift;
600 if (get_bits1(gbp)) {
601 int state_bits, state_shift;
604 av_log(m->avctx, AV_LOG_ERROR,
605 "FIR filter has state data specified.\n");
606 return AVERROR_INVALIDDATA;
609 state_bits = get_bits(gbp, 4);
610 state_shift = get_bits(gbp, 4);
612 /* TODO: Check validity of state data. */
614 for (i = 0; i < order; i++)
615 fp->state[i] = get_sbits(gbp, state_bits) << state_shift;
622 /** Read parameters for primitive matrices. */
624 static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitContext *gbp)
626 SubStream *s = &m->substream[substr];
627 unsigned int mat, ch;
628 const int max_primitive_matrices = m->avctx->codec_id == AV_CODEC_ID_MLP
630 : MAX_MATRICES_TRUEHD;
632 if (m->matrix_changed++ > 1) {
633 av_log(m->avctx, AV_LOG_ERROR, "Matrices may change only once per access unit.\n");
634 return AVERROR_INVALIDDATA;
637 s->num_primitive_matrices = get_bits(gbp, 4);
639 if (s->num_primitive_matrices > max_primitive_matrices) {
640 av_log(m->avctx, AV_LOG_ERROR,
641 "Number of primitive matrices cannot be greater than %d.\n",
642 max_primitive_matrices);
643 return AVERROR_INVALIDDATA;
646 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
647 int frac_bits, max_chan;
648 s->matrix_out_ch[mat] = get_bits(gbp, 4);
649 frac_bits = get_bits(gbp, 4);
650 s->lsb_bypass [mat] = get_bits1(gbp);
652 if (s->matrix_out_ch[mat] > s->max_matrix_channel) {
653 av_log(m->avctx, AV_LOG_ERROR,
654 "Invalid channel %d specified as output from matrix.\n",
655 s->matrix_out_ch[mat]);
656 return AVERROR_INVALIDDATA;
658 if (frac_bits > 14) {
659 av_log(m->avctx, AV_LOG_ERROR,
660 "Too many fractional bits specified.\n");
661 return AVERROR_INVALIDDATA;
664 max_chan = s->max_matrix_channel;
668 for (ch = 0; ch <= max_chan; ch++) {
671 coeff_val = get_sbits(gbp, frac_bits + 2);
673 s->matrix_coeff[mat][ch] = coeff_val << (14 - frac_bits);
677 s->matrix_noise_shift[mat] = get_bits(gbp, 4);
679 s->matrix_noise_shift[mat] = 0;
685 /** Read channel parameters. */
687 static int read_channel_params(MLPDecodeContext *m, unsigned int substr,
688 GetBitContext *gbp, unsigned int ch)
690 SubStream *s = &m->substream[substr];
691 ChannelParams *cp = &s->channel_params[ch];
692 FilterParams *fir = &cp->filter_params[FIR];
693 FilterParams *iir = &cp->filter_params[IIR];
696 if (s->param_presence_flags & PARAM_FIR)
698 if ((ret = read_filter_params(m, gbp, substr, ch, FIR)) < 0)
701 if (s->param_presence_flags & PARAM_IIR)
703 if ((ret = read_filter_params(m, gbp, substr, ch, IIR)) < 0)
706 if (fir->order + iir->order > 8) {
707 av_log(m->avctx, AV_LOG_ERROR, "Total filter orders too high.\n");
708 return AVERROR_INVALIDDATA;
711 if (fir->order && iir->order &&
712 fir->shift != iir->shift) {
713 av_log(m->avctx, AV_LOG_ERROR,
714 "FIR and IIR filters must use the same precision.\n");
715 return AVERROR_INVALIDDATA;
717 /* The FIR and IIR filters must have the same precision.
718 * To simplify the filtering code, only the precision of the
719 * FIR filter is considered. If only the IIR filter is employed,
720 * the FIR filter precision is set to that of the IIR filter, so
721 * that the filtering code can use it. */
722 if (!fir->order && iir->order)
723 fir->shift = iir->shift;
725 if (s->param_presence_flags & PARAM_HUFFOFFSET)
727 cp->huff_offset = get_sbits(gbp, 15);
729 cp->codebook = get_bits(gbp, 2);
730 cp->huff_lsbs = get_bits(gbp, 5);
732 if (cp->huff_lsbs > 24) {
733 av_log(m->avctx, AV_LOG_ERROR, "Invalid huff_lsbs.\n");
734 return AVERROR_INVALIDDATA;
737 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
742 /** Read decoding parameters that change more often than those in the restart
745 static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
748 SubStream *s = &m->substream[substr];
752 if (s->param_presence_flags & PARAM_PRESENCE)
754 s->param_presence_flags = get_bits(gbp, 8);
756 if (s->param_presence_flags & PARAM_BLOCKSIZE)
757 if (get_bits1(gbp)) {
758 s->blocksize = get_bits(gbp, 9);
759 if (s->blocksize < 8 || s->blocksize > m->access_unit_size) {
760 av_log(m->avctx, AV_LOG_ERROR, "Invalid blocksize.");
762 return AVERROR_INVALIDDATA;
766 if (s->param_presence_flags & PARAM_MATRIX)
768 if ((ret = read_matrix_params(m, substr, gbp)) < 0)
771 if (s->param_presence_flags & PARAM_OUTSHIFT)
773 for (ch = 0; ch <= s->max_matrix_channel; ch++)
774 s->output_shift[ch] = get_sbits(gbp, 4);
776 if (s->param_presence_flags & PARAM_QUANTSTEP)
778 for (ch = 0; ch <= s->max_channel; ch++) {
779 ChannelParams *cp = &s->channel_params[ch];
781 s->quant_step_size[ch] = get_bits(gbp, 4);
783 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
786 for (ch = s->min_channel; ch <= s->max_channel; ch++)
788 if ((ret = read_channel_params(m, substr, gbp, ch)) < 0)
794 #define MSB_MASK(bits) (-1u << bits)
796 /** Generate PCM samples using the prediction filters and residual values
797 * read from the data stream, and update the filter state. */
799 static void filter_channel(MLPDecodeContext *m, unsigned int substr,
800 unsigned int channel)
802 SubStream *s = &m->substream[substr];
803 const int32_t *fircoeff = s->channel_params[channel].coeff[FIR];
804 int32_t state_buffer[NUM_FILTERS][MAX_BLOCKSIZE + MAX_FIR_ORDER];
805 int32_t *firbuf = state_buffer[FIR] + MAX_BLOCKSIZE;
806 int32_t *iirbuf = state_buffer[IIR] + MAX_BLOCKSIZE;
807 FilterParams *fir = &s->channel_params[channel].filter_params[FIR];
808 FilterParams *iir = &s->channel_params[channel].filter_params[IIR];
809 unsigned int filter_shift = fir->shift;
810 int32_t mask = MSB_MASK(s->quant_step_size[channel]);
812 memcpy(firbuf, fir->state, MAX_FIR_ORDER * sizeof(int32_t));
813 memcpy(iirbuf, iir->state, MAX_IIR_ORDER * sizeof(int32_t));
815 m->dsp.mlp_filter_channel(firbuf, fircoeff,
816 fir->order, iir->order,
817 filter_shift, mask, s->blocksize,
818 &m->sample_buffer[s->blockpos][channel]);
820 memcpy(fir->state, firbuf - s->blocksize, MAX_FIR_ORDER * sizeof(int32_t));
821 memcpy(iir->state, iirbuf - s->blocksize, MAX_IIR_ORDER * sizeof(int32_t));
824 /** Read a block of PCM residual data (or actual if no filtering active). */
826 static int read_block_data(MLPDecodeContext *m, GetBitContext *gbp,
829 SubStream *s = &m->substream[substr];
830 unsigned int i, ch, expected_stream_pos = 0;
833 if (s->data_check_present) {
834 expected_stream_pos = get_bits_count(gbp);
835 expected_stream_pos += get_bits(gbp, 16);
836 avpriv_request_sample(m->avctx,
837 "Substreams with VLC block size check info");
840 if (s->blockpos + s->blocksize > m->access_unit_size) {
841 av_log(m->avctx, AV_LOG_ERROR, "too many audio samples in frame\n");
842 return AVERROR_INVALIDDATA;
845 memset(&m->bypassed_lsbs[s->blockpos][0], 0,
846 s->blocksize * sizeof(m->bypassed_lsbs[0]));
848 for (i = 0; i < s->blocksize; i++)
849 if ((ret = read_huff_channels(m, gbp, substr, i)) < 0)
852 for (ch = s->min_channel; ch <= s->max_channel; ch++)
853 filter_channel(m, substr, ch);
855 s->blockpos += s->blocksize;
857 if (s->data_check_present) {
858 if (get_bits_count(gbp) != expected_stream_pos)
859 av_log(m->avctx, AV_LOG_ERROR, "block data length mismatch\n");
866 /** Data table used for TrueHD noise generation function. */
868 static const int8_t noise_table[256] = {
869 30, 51, 22, 54, 3, 7, -4, 38, 14, 55, 46, 81, 22, 58, -3, 2,
870 52, 31, -7, 51, 15, 44, 74, 30, 85, -17, 10, 33, 18, 80, 28, 62,
871 10, 32, 23, 69, 72, 26, 35, 17, 73, 60, 8, 56, 2, 6, -2, -5,
872 51, 4, 11, 50, 66, 76, 21, 44, 33, 47, 1, 26, 64, 48, 57, 40,
873 38, 16, -10, -28, 92, 22, -18, 29, -10, 5, -13, 49, 19, 24, 70, 34,
874 61, 48, 30, 14, -6, 25, 58, 33, 42, 60, 67, 17, 54, 17, 22, 30,
875 67, 44, -9, 50, -11, 43, 40, 32, 59, 82, 13, 49, -14, 55, 60, 36,
876 48, 49, 31, 47, 15, 12, 4, 65, 1, 23, 29, 39, 45, -2, 84, 69,
877 0, 72, 37, 57, 27, 41, -15, -16, 35, 31, 14, 61, 24, 0, 27, 24,
878 16, 41, 55, 34, 53, 9, 56, 12, 25, 29, 53, 5, 20, -20, -8, 20,
879 13, 28, -3, 78, 38, 16, 11, 62, 46, 29, 21, 24, 46, 65, 43, -23,
880 89, 18, 74, 21, 38, -12, 19, 12, -19, 8, 15, 33, 4, 57, 9, -8,
881 36, 35, 26, 28, 7, 83, 63, 79, 75, 11, 3, 87, 37, 47, 34, 40,
882 39, 19, 20, 42, 27, 34, 39, 77, 13, 42, 59, 64, 45, -1, 32, 37,
883 45, -5, 53, -6, 7, 36, 50, 23, 6, 32, 9, -21, 18, 71, 27, 52,
884 -25, 31, 35, 42, -1, 68, 63, 52, 26, 43, 66, 37, 41, 25, 40, 70,
887 /** Noise generation functions.
888 * I'm not sure what these are for - they seem to be some kind of pseudorandom
889 * sequence generators, used to generate noise data which is used when the
890 * channels are rematrixed. I'm not sure if they provide a practical benefit
891 * to compression, or just obfuscate the decoder. Are they for some kind of
894 /** Generate two channels of noise, used in the matrix when
895 * restart sync word == 0x31ea. */
897 static void generate_2_noise_channels(MLPDecodeContext *m, unsigned int substr)
899 SubStream *s = &m->substream[substr];
901 uint32_t seed = s->noisegen_seed;
902 unsigned int maxchan = s->max_matrix_channel;
904 for (i = 0; i < s->blockpos; i++) {
905 uint16_t seed_shr7 = seed >> 7;
906 m->sample_buffer[i][maxchan+1] = ((int8_t)(seed >> 15)) << s->noise_shift;
907 m->sample_buffer[i][maxchan+2] = ((int8_t) seed_shr7) << s->noise_shift;
909 seed = (seed << 16) ^ seed_shr7 ^ (seed_shr7 << 5);
912 s->noisegen_seed = seed;
915 /** Generate a block of noise, used when restart sync word == 0x31eb. */
917 static void fill_noise_buffer(MLPDecodeContext *m, unsigned int substr)
919 SubStream *s = &m->substream[substr];
921 uint32_t seed = s->noisegen_seed;
923 for (i = 0; i < m->access_unit_size_pow2; i++) {
924 uint8_t seed_shr15 = seed >> 15;
925 m->noise_buffer[i] = noise_table[seed_shr15];
926 seed = (seed << 8) ^ seed_shr15 ^ (seed_shr15 << 5);
929 s->noisegen_seed = seed;
933 /** Apply the channel matrices in turn to reconstruct the original audio
936 static void rematrix_channels(MLPDecodeContext *m, unsigned int substr)
938 SubStream *s = &m->substream[substr];
939 unsigned int mat, src_ch, i;
940 unsigned int maxchan;
942 maxchan = s->max_matrix_channel;
943 if (!s->noise_type) {
944 generate_2_noise_channels(m, substr);
947 fill_noise_buffer(m, substr);
950 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
951 int matrix_noise_shift = s->matrix_noise_shift[mat];
952 unsigned int dest_ch = s->matrix_out_ch[mat];
953 int32_t mask = MSB_MASK(s->quant_step_size[dest_ch]);
954 int32_t *coeffs = s->matrix_coeff[mat];
955 int index = s->num_primitive_matrices - mat;
956 int index2 = 2 * index + 1;
958 /* TODO: DSPContext? */
960 for (i = 0; i < s->blockpos; i++) {
961 int32_t bypassed_lsb = m->bypassed_lsbs[i][mat];
962 int32_t *samples = m->sample_buffer[i];
965 for (src_ch = 0; src_ch <= maxchan; src_ch++)
966 accum += (int64_t) samples[src_ch] * coeffs[src_ch];
968 if (matrix_noise_shift) {
969 index &= m->access_unit_size_pow2 - 1;
970 accum += m->noise_buffer[index] << (matrix_noise_shift + 7);
974 samples[dest_ch] = ((accum >> 14) & mask) + bypassed_lsb;
979 /** Write the audio data into the output buffer. */
981 static int output_data(MLPDecodeContext *m, unsigned int substr,
982 AVFrame *frame, int *got_frame_ptr)
984 AVCodecContext *avctx = m->avctx;
985 SubStream *s = &m->substream[substr];
986 unsigned int i, out_ch = 0;
990 int is32 = (m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
992 if (m->avctx->channels != s->max_matrix_channel + 1) {
993 av_log(m->avctx, AV_LOG_ERROR, "channel count mismatch\n");
994 return AVERROR_INVALIDDATA;
998 av_log(avctx, AV_LOG_ERROR, "No samples to output.\n");
999 return AVERROR_INVALIDDATA;
1002 /* get output buffer */
1003 frame->nb_samples = s->blockpos;
1004 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1005 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1008 data_32 = (int32_t *)frame->data[0];
1009 data_16 = (int16_t *)frame->data[0];
1011 for (i = 0; i < s->blockpos; i++) {
1012 for (out_ch = 0; out_ch <= s->max_matrix_channel; out_ch++) {
1013 int mat_ch = s->ch_assign[out_ch];
1014 int32_t sample = m->sample_buffer[i][mat_ch]
1015 << s->output_shift[mat_ch];
1016 s->lossless_check_data ^= (sample & 0xffffff) << mat_ch;
1017 if (is32) *data_32++ = sample << 8;
1018 else *data_16++ = sample >> 8;
1027 /** Read an access unit from the stream.
1028 * @return negative on error, 0 if not enough data is present in the input stream,
1029 * otherwise the number of bytes consumed. */
1031 static int read_access_unit(AVCodecContext *avctx, void* data,
1032 int *got_frame_ptr, AVPacket *avpkt)
1034 const uint8_t *buf = avpkt->data;
1035 int buf_size = avpkt->size;
1036 MLPDecodeContext *m = avctx->priv_data;
1038 unsigned int length, substr;
1039 unsigned int substream_start;
1040 unsigned int header_size = 4;
1041 unsigned int substr_header_size = 0;
1042 uint8_t substream_parity_present[MAX_SUBSTREAMS];
1043 uint16_t substream_data_len[MAX_SUBSTREAMS];
1044 uint8_t parity_bits;
1050 length = (AV_RB16(buf) & 0xfff) * 2;
1052 if (length < 4 || length > buf_size)
1053 return AVERROR_INVALIDDATA;
1055 init_get_bits(&gb, (buf + 4), (length - 4) * 8);
1057 m->is_major_sync_unit = 0;
1058 if (show_bits_long(&gb, 31) == (0xf8726fba >> 1)) {
1059 if (read_major_sync(m, &gb) < 0)
1061 m->is_major_sync_unit = 1;
1065 if (!m->params_valid) {
1066 av_log(m->avctx, AV_LOG_WARNING,
1067 "Stream parameters not seen; skipping frame.\n");
1072 substream_start = 0;
1074 for (substr = 0; substr < m->num_substreams; substr++) {
1075 int extraword_present, checkdata_present, end, nonrestart_substr;
1077 extraword_present = get_bits1(&gb);
1078 nonrestart_substr = get_bits1(&gb);
1079 checkdata_present = get_bits1(&gb);
1082 end = get_bits(&gb, 12) * 2;
1084 substr_header_size += 2;
1086 if (extraword_present) {
1087 if (m->avctx->codec_id == AV_CODEC_ID_MLP) {
1088 av_log(m->avctx, AV_LOG_ERROR, "There must be no extraword for MLP.\n");
1092 substr_header_size += 2;
1095 if (!(nonrestart_substr ^ m->is_major_sync_unit)) {
1096 av_log(m->avctx, AV_LOG_ERROR, "Invalid nonrestart_substr.\n");
1100 if (end + header_size + substr_header_size > length) {
1101 av_log(m->avctx, AV_LOG_ERROR,
1102 "Indicated length of substream %d data goes off end of "
1103 "packet.\n", substr);
1104 end = length - header_size - substr_header_size;
1107 if (end < substream_start) {
1108 av_log(avctx, AV_LOG_ERROR,
1109 "Indicated end offset of substream %d data "
1110 "is smaller than calculated start offset.\n",
1115 if (substr > m->max_decoded_substream)
1118 substream_parity_present[substr] = checkdata_present;
1119 substream_data_len[substr] = end - substream_start;
1120 substream_start = end;
1123 parity_bits = ff_mlp_calculate_parity(buf, 4);
1124 parity_bits ^= ff_mlp_calculate_parity(buf + header_size, substr_header_size);
1126 if ((((parity_bits >> 4) ^ parity_bits) & 0xF) != 0xF) {
1127 av_log(avctx, AV_LOG_ERROR, "Parity check failed.\n");
1131 buf += header_size + substr_header_size;
1133 for (substr = 0; substr <= m->max_decoded_substream; substr++) {
1134 SubStream *s = &m->substream[substr];
1135 init_get_bits(&gb, buf, substream_data_len[substr] * 8);
1137 m->matrix_changed = 0;
1138 memset(m->filter_changed, 0, sizeof(m->filter_changed));
1142 if (get_bits1(&gb)) {
1143 if (get_bits1(&gb)) {
1144 /* A restart header should be present. */
1145 if (read_restart_header(m, &gb, buf, substr) < 0)
1147 s->restart_seen = 1;
1150 if (!s->restart_seen)
1152 if (read_decoding_params(m, &gb, substr) < 0)
1156 if (!s->restart_seen)
1159 if ((ret = read_block_data(m, &gb, substr)) < 0)
1162 if (get_bits_count(&gb) >= substream_data_len[substr] * 8)
1163 goto substream_length_mismatch;
1165 } while (!get_bits1(&gb));
1167 skip_bits(&gb, (-get_bits_count(&gb)) & 15);
1169 if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 32) {
1172 if (get_bits(&gb, 16) != 0xD234)
1173 return AVERROR_INVALIDDATA;
1175 shorten_by = get_bits(&gb, 16);
1176 if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD && shorten_by & 0x2000)
1177 s->blockpos -= FFMIN(shorten_by & 0x1FFF, s->blockpos);
1178 else if (m->avctx->codec_id == AV_CODEC_ID_MLP && shorten_by != 0xD234)
1179 return AVERROR_INVALIDDATA;
1181 if (substr == m->max_decoded_substream)
1182 av_log(m->avctx, AV_LOG_INFO, "End of stream indicated.\n");
1185 if (substream_parity_present[substr]) {
1186 uint8_t parity, checksum;
1188 if (substream_data_len[substr] * 8 - get_bits_count(&gb) != 16)
1189 goto substream_length_mismatch;
1191 parity = ff_mlp_calculate_parity(buf, substream_data_len[substr] - 2);
1192 checksum = ff_mlp_checksum8 (buf, substream_data_len[substr] - 2);
1194 if ((get_bits(&gb, 8) ^ parity) != 0xa9 )
1195 av_log(m->avctx, AV_LOG_ERROR, "Substream %d parity check failed.\n", substr);
1196 if ( get_bits(&gb, 8) != checksum)
1197 av_log(m->avctx, AV_LOG_ERROR, "Substream %d checksum failed.\n" , substr);
1200 if (substream_data_len[substr] * 8 != get_bits_count(&gb))
1201 goto substream_length_mismatch;
1204 if (!s->restart_seen)
1205 av_log(m->avctx, AV_LOG_ERROR,
1206 "No restart header present in substream %d.\n", substr);
1208 buf += substream_data_len[substr];
1211 rematrix_channels(m, m->max_decoded_substream);
1213 if ((ret = output_data(m, m->max_decoded_substream, data, got_frame_ptr)) < 0)
1218 substream_length_mismatch:
1219 av_log(m->avctx, AV_LOG_ERROR, "substream %d length mismatch\n", substr);
1220 return AVERROR_INVALIDDATA;
1223 m->params_valid = 0;
1224 return AVERROR_INVALIDDATA;
1227 AVCodec ff_mlp_decoder = {
1229 .long_name = NULL_IF_CONFIG_SMALL("MLP (Meridian Lossless Packing)"),
1230 .type = AVMEDIA_TYPE_AUDIO,
1231 .id = AV_CODEC_ID_MLP,
1232 .priv_data_size = sizeof(MLPDecodeContext),
1233 .init = mlp_decode_init,
1234 .decode = read_access_unit,
1235 .capabilities = CODEC_CAP_DR1,
1238 #if CONFIG_TRUEHD_DECODER
1239 AVCodec ff_truehd_decoder = {
1241 .long_name = NULL_IF_CONFIG_SMALL("TrueHD"),
1242 .type = AVMEDIA_TYPE_AUDIO,
1243 .id = AV_CODEC_ID_TRUEHD,
1244 .priv_data_size = sizeof(MLPDecodeContext),
1245 .init = mlp_decode_init,
1246 .decode = read_access_unit,
1247 .capabilities = CODEC_CAP_DR1,
1249 #endif /* CONFIG_TRUEHD_DECODER */