3 * Copyright (c) 2007-2008 Ian Caulfield
5 * This file is part of FFmpeg.
7 * FFmpeg 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 * FFmpeg 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 FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
30 #include "libavutil/intreadwrite.h"
31 #include "libavutil/channel_layout.h"
34 #include "libavutil/crc.h"
36 #include "mlp_parser.h"
40 /** number of bits used for VLC lookup - longest Huffman code is 9 */
43 typedef struct SubStream {
44 /// Set if a valid restart header has been read. Otherwise the substream cannot be decoded.
48 /** restart header data */
49 /// The type of noise to be used in the rematrix stage.
52 /// The index of the first channel coded in this substream.
54 /// The index of the last channel coded in this substream.
56 /// The number of channels input into the rematrix stage.
57 uint8_t max_matrix_channel;
58 /// For each channel output by the matrix, the output channel to map it to
59 uint8_t ch_assign[MAX_CHANNELS];
60 /// The channel layout for this substream
63 /// Channel coding parameters for channels in the substream
64 ChannelParams channel_params[MAX_CHANNELS];
66 /// The left shift applied to random noise in 0x31ea substreams.
68 /// The current seed value for the pseudorandom noise generator(s).
69 uint32_t noisegen_seed;
71 /// Set if the substream contains extra info to check the size of VLC blocks.
72 uint8_t data_check_present;
74 /// Bitmask of which parameter sets are conveyed in a decoding parameter block.
75 uint8_t param_presence_flags;
76 #define PARAM_BLOCKSIZE (1 << 7)
77 #define PARAM_MATRIX (1 << 6)
78 #define PARAM_OUTSHIFT (1 << 5)
79 #define PARAM_QUANTSTEP (1 << 4)
80 #define PARAM_FIR (1 << 3)
81 #define PARAM_IIR (1 << 2)
82 #define PARAM_HUFFOFFSET (1 << 1)
83 #define PARAM_PRESENCE (1 << 0)
89 /// Number of matrices to be applied.
90 uint8_t num_primitive_matrices;
92 /// matrix output channel
93 uint8_t matrix_out_ch[MAX_MATRICES];
95 /// Whether the LSBs of the matrix output are encoded in the bitstream.
96 uint8_t lsb_bypass[MAX_MATRICES];
97 /// Matrix coefficients, stored as 2.14 fixed point.
98 int32_t matrix_coeff[MAX_MATRICES][MAX_CHANNELS];
99 /// Left shift to apply to noise values in 0x31eb substreams.
100 uint8_t matrix_noise_shift[MAX_MATRICES];
103 /// Left shift to apply to Huffman-decoded residuals.
104 uint8_t quant_step_size[MAX_CHANNELS];
106 /// number of PCM samples in current audio block
108 /// Number of PCM samples decoded so far in this frame.
111 /// Left shift to apply to decoded PCM values to get final 24-bit output.
112 int8_t output_shift[MAX_CHANNELS];
114 /// Running XOR of all output samples.
115 int32_t lossless_check_data;
119 typedef struct MLPDecodeContext {
120 AVCodecContext *avctx;
122 /// Current access unit being read has a major sync.
123 int is_major_sync_unit;
125 /// Set if a valid major sync block has been read. Otherwise no decoding is possible.
126 uint8_t params_valid;
128 /// Number of substreams contained within this stream.
129 uint8_t num_substreams;
131 /// Index of the last substream to decode - further substreams are skipped.
132 uint8_t max_decoded_substream;
134 /// Stream needs channel reordering to comply with FFmpeg's channel order
135 uint8_t needs_reordering;
137 /// number of PCM samples contained in each frame
138 int access_unit_size;
139 /// next power of two above the number of samples in each frame
140 int access_unit_size_pow2;
142 SubStream substream[MAX_SUBSTREAMS];
145 int filter_changed[MAX_CHANNELS][NUM_FILTERS];
147 int8_t noise_buffer[MAX_BLOCKSIZE_POW2];
148 int8_t bypassed_lsbs[MAX_BLOCKSIZE][MAX_CHANNELS];
149 int32_t sample_buffer[MAX_BLOCKSIZE][MAX_CHANNELS];
154 static const uint64_t thd_channel_order[] = {
155 AV_CH_FRONT_LEFT, AV_CH_FRONT_RIGHT, // LR
156 AV_CH_FRONT_CENTER, // C
157 AV_CH_LOW_FREQUENCY, // LFE
158 AV_CH_SIDE_LEFT, AV_CH_SIDE_RIGHT, // LRs
159 AV_CH_TOP_FRONT_LEFT, AV_CH_TOP_FRONT_RIGHT, // LRvh
160 AV_CH_FRONT_LEFT_OF_CENTER, AV_CH_FRONT_RIGHT_OF_CENTER, // LRc
161 AV_CH_BACK_LEFT, AV_CH_BACK_RIGHT, // LRrs
162 AV_CH_BACK_CENTER, // Cs
163 AV_CH_TOP_CENTER, // Ts
164 AV_CH_SURROUND_DIRECT_LEFT, AV_CH_SURROUND_DIRECT_RIGHT, // LRsd
165 AV_CH_WIDE_LEFT, AV_CH_WIDE_RIGHT, // LRw
166 AV_CH_TOP_FRONT_CENTER, // Cvh
167 AV_CH_LOW_FREQUENCY_2, // LFE2
170 static uint64_t thd_channel_layout_extract_channel(uint64_t channel_layout,
175 if (av_get_channel_layout_nb_channels(channel_layout) <= index)
178 for (i = 0; i < FF_ARRAY_ELEMS(thd_channel_order); i++)
179 if (channel_layout & thd_channel_order[i] && !index--)
180 return thd_channel_order[i];
184 static VLC huff_vlc[3];
186 /** Initialize static data, constant between all invocations of the codec. */
188 static av_cold void init_static(void)
190 if (!huff_vlc[0].bits) {
191 INIT_VLC_STATIC(&huff_vlc[0], VLC_BITS, 18,
192 &ff_mlp_huffman_tables[0][0][1], 2, 1,
193 &ff_mlp_huffman_tables[0][0][0], 2, 1, 512);
194 INIT_VLC_STATIC(&huff_vlc[1], VLC_BITS, 16,
195 &ff_mlp_huffman_tables[1][0][1], 2, 1,
196 &ff_mlp_huffman_tables[1][0][0], 2, 1, 512);
197 INIT_VLC_STATIC(&huff_vlc[2], VLC_BITS, 15,
198 &ff_mlp_huffman_tables[2][0][1], 2, 1,
199 &ff_mlp_huffman_tables[2][0][0], 2, 1, 512);
205 static inline int32_t calculate_sign_huff(MLPDecodeContext *m,
206 unsigned int substr, unsigned int ch)
208 SubStream *s = &m->substream[substr];
209 ChannelParams *cp = &s->channel_params[ch];
210 int lsb_bits = cp->huff_lsbs - s->quant_step_size[ch];
211 int sign_shift = lsb_bits + (cp->codebook ? 2 - cp->codebook : -1);
212 int32_t sign_huff_offset = cp->huff_offset;
214 if (cp->codebook > 0)
215 sign_huff_offset -= 7 << lsb_bits;
218 sign_huff_offset -= 1 << sign_shift;
220 return sign_huff_offset;
223 /** Read a sample, consisting of either, both or neither of entropy-coded MSBs
226 static inline int read_huff_channels(MLPDecodeContext *m, GetBitContext *gbp,
227 unsigned int substr, unsigned int pos)
229 SubStream *s = &m->substream[substr];
230 unsigned int mat, channel;
232 for (mat = 0; mat < s->num_primitive_matrices; mat++)
233 if (s->lsb_bypass[mat])
234 m->bypassed_lsbs[pos + s->blockpos][mat] = get_bits1(gbp);
236 for (channel = s->min_channel; channel <= s->max_channel; channel++) {
237 ChannelParams *cp = &s->channel_params[channel];
238 int codebook = cp->codebook;
239 int quant_step_size = s->quant_step_size[channel];
240 int lsb_bits = cp->huff_lsbs - quant_step_size;
244 result = get_vlc2(gbp, huff_vlc[codebook-1].table,
245 VLC_BITS, (9 + VLC_BITS - 1) / VLC_BITS);
248 return AVERROR_INVALIDDATA;
251 result = (result << lsb_bits) + get_bits(gbp, lsb_bits);
253 result += cp->sign_huff_offset;
254 result <<= quant_step_size;
256 m->sample_buffer[pos + s->blockpos][channel] = result;
262 static av_cold int mlp_decode_init(AVCodecContext *avctx)
264 MLPDecodeContext *m = avctx->priv_data;
269 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
270 m->substream[substr].lossless_check_data = 0xffffffff;
271 ff_mlpdsp_init(&m->dsp);
276 /** Read a major sync info header - contains high level information about
277 * the stream - sample rate, channel arrangement etc. Most of this
278 * information is not actually necessary for decoding, only for playback.
281 static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
286 if ((ret = ff_mlp_read_major_sync(m->avctx, &mh, gb)) != 0)
289 if (mh.group1_bits == 0) {
290 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown bits per sample\n");
291 return AVERROR_INVALIDDATA;
293 if (mh.group2_bits > mh.group1_bits) {
294 av_log(m->avctx, AV_LOG_ERROR,
295 "Channel group 2 cannot have more bits per sample than group 1.\n");
296 return AVERROR_INVALIDDATA;
299 if (mh.group2_samplerate && mh.group2_samplerate != mh.group1_samplerate) {
300 av_log(m->avctx, AV_LOG_ERROR,
301 "Channel groups with differing sample rates are not currently supported.\n");
302 return AVERROR_INVALIDDATA;
305 if (mh.group1_samplerate == 0) {
306 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown sampling rate\n");
307 return AVERROR_INVALIDDATA;
309 if (mh.group1_samplerate > MAX_SAMPLERATE) {
310 av_log(m->avctx, AV_LOG_ERROR,
311 "Sampling rate %d is greater than the supported maximum (%d).\n",
312 mh.group1_samplerate, MAX_SAMPLERATE);
313 return AVERROR_INVALIDDATA;
315 if (mh.access_unit_size > MAX_BLOCKSIZE) {
316 av_log(m->avctx, AV_LOG_ERROR,
317 "Block size %d is greater than the supported maximum (%d).\n",
318 mh.access_unit_size, MAX_BLOCKSIZE);
319 return AVERROR_INVALIDDATA;
321 if (mh.access_unit_size_pow2 > MAX_BLOCKSIZE_POW2) {
322 av_log(m->avctx, AV_LOG_ERROR,
323 "Block size pow2 %d is greater than the supported maximum (%d).\n",
324 mh.access_unit_size_pow2, MAX_BLOCKSIZE_POW2);
325 return AVERROR_INVALIDDATA;
328 if (mh.num_substreams == 0)
329 return AVERROR_INVALIDDATA;
330 if (m->avctx->codec_id == AV_CODEC_ID_MLP && mh.num_substreams > 2) {
331 av_log(m->avctx, AV_LOG_ERROR, "MLP only supports up to 2 substreams.\n");
332 return AVERROR_INVALIDDATA;
334 if (mh.num_substreams > MAX_SUBSTREAMS) {
335 avpriv_request_sample(m->avctx,
336 "%d substreams (more than the "
337 "maximum supported by the decoder)",
339 return AVERROR_PATCHWELCOME;
342 m->access_unit_size = mh.access_unit_size;
343 m->access_unit_size_pow2 = mh.access_unit_size_pow2;
345 m->num_substreams = mh.num_substreams;
346 m->max_decoded_substream = m->num_substreams - 1;
348 m->avctx->sample_rate = mh.group1_samplerate;
349 m->avctx->frame_size = mh.access_unit_size;
351 m->avctx->bits_per_raw_sample = mh.group1_bits;
352 if (mh.group1_bits > 16)
353 m->avctx->sample_fmt = AV_SAMPLE_FMT_S32;
355 m->avctx->sample_fmt = AV_SAMPLE_FMT_S16;
358 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
359 m->substream[substr].restart_seen = 0;
361 /* Set the layout for each substream. When there's more than one, the first
362 * substream is Stereo. Subsequent substreams' layouts are indicated in the
364 if (m->avctx->codec_id == AV_CODEC_ID_MLP) {
365 if ((substr = (mh.num_substreams > 1)))
366 m->substream[0].ch_layout = AV_CH_LAYOUT_STEREO;
367 m->substream[substr].ch_layout = mh.channel_layout_mlp;
369 if ((substr = (mh.num_substreams > 1)))
370 m->substream[0].ch_layout = AV_CH_LAYOUT_STEREO;
371 if (mh.num_substreams > 2)
372 if (mh.channel_layout_thd_stream2)
373 m->substream[2].ch_layout = mh.channel_layout_thd_stream2;
375 m->substream[2].ch_layout = mh.channel_layout_thd_stream1;
376 m->substream[substr].ch_layout = mh.channel_layout_thd_stream1;
378 if (m->avctx->channels<=2 && m->substream[substr].ch_layout == AV_CH_LAYOUT_MONO && m->max_decoded_substream == 1) {
379 av_log(m->avctx, AV_LOG_DEBUG, "Mono stream with 2 substreams, ignoring 2nd\n");
380 m->max_decoded_substream = 0;
381 if (m->avctx->channels==2)
382 m->avctx->channel_layout = AV_CH_LAYOUT_STEREO;
386 m->needs_reordering = mh.channel_arrangement >= 18 && mh.channel_arrangement <= 20;
391 /** Read a restart header from a block in a substream. This contains parameters
392 * required to decode the audio that do not change very often. Generally
393 * (always) present only in blocks following a major sync. */
395 static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
396 const uint8_t *buf, unsigned int substr)
398 SubStream *s = &m->substream[substr];
402 uint8_t lossless_check;
403 int start_count = get_bits_count(gbp);
404 const int max_matrix_channel = m->avctx->codec_id == AV_CODEC_ID_MLP
405 ? MAX_MATRIX_CHANNEL_MLP
406 : MAX_MATRIX_CHANNEL_TRUEHD;
407 int max_channel, min_channel, matrix_channel;
409 sync_word = get_bits(gbp, 13);
411 if (sync_word != 0x31ea >> 1) {
412 av_log(m->avctx, AV_LOG_ERROR,
413 "restart header sync incorrect (got 0x%04x)\n", sync_word);
414 return AVERROR_INVALIDDATA;
417 s->noise_type = get_bits1(gbp);
419 if (m->avctx->codec_id == AV_CODEC_ID_MLP && s->noise_type) {
420 av_log(m->avctx, AV_LOG_ERROR, "MLP must have 0x31ea sync word.\n");
421 return AVERROR_INVALIDDATA;
424 skip_bits(gbp, 16); /* Output timestamp */
426 min_channel = get_bits(gbp, 4);
427 max_channel = get_bits(gbp, 4);
428 matrix_channel = get_bits(gbp, 4);
430 if (matrix_channel > max_matrix_channel) {
431 av_log(m->avctx, AV_LOG_ERROR,
432 "Max matrix channel cannot be greater than %d.\n",
434 return AVERROR_INVALIDDATA;
437 if (max_channel != matrix_channel) {
438 av_log(m->avctx, AV_LOG_ERROR,
439 "Max channel must be equal max matrix channel.\n");
440 return AVERROR_INVALIDDATA;
443 /* This should happen for TrueHD streams with >6 channels and MLP's noise
444 * type. It is not yet known if this is allowed. */
445 if (max_channel > MAX_MATRIX_CHANNEL_MLP && !s->noise_type) {
446 avpriv_request_sample(m->avctx,
447 "%d channels (more than the "
448 "maximum supported by the decoder)",
450 return AVERROR_PATCHWELCOME;
453 if (min_channel > max_channel) {
454 av_log(m->avctx, AV_LOG_ERROR,
455 "Substream min channel cannot be greater than max channel.\n");
456 return AVERROR_INVALIDDATA;
459 s->min_channel = min_channel;
460 s->max_channel = max_channel;
461 s->max_matrix_channel = matrix_channel;
463 #if FF_API_REQUEST_CHANNELS
464 if (m->avctx->request_channels > 0 &&
465 m->avctx->request_channels <= s->max_channel + 1 &&
466 m->max_decoded_substream > substr) {
467 av_log(m->avctx, AV_LOG_DEBUG,
468 "Extracting %d-channel downmix from substream %d. "
469 "Further substreams will be skipped.\n",
470 s->max_channel + 1, substr);
471 m->max_decoded_substream = substr;
474 if (m->avctx->request_channel_layout == s->ch_layout &&
475 m->max_decoded_substream > substr) {
476 av_log(m->avctx, AV_LOG_DEBUG,
477 "Extracting %d-channel downmix (0x%"PRIx64") from substream %d. "
478 "Further substreams will be skipped.\n",
479 s->max_channel + 1, s->ch_layout, substr);
480 m->max_decoded_substream = substr;
483 s->noise_shift = get_bits(gbp, 4);
484 s->noisegen_seed = get_bits(gbp, 23);
488 s->data_check_present = get_bits1(gbp);
489 lossless_check = get_bits(gbp, 8);
490 if (substr == m->max_decoded_substream
491 && s->lossless_check_data != 0xffffffff) {
492 tmp = xor_32_to_8(s->lossless_check_data);
493 if (tmp != lossless_check)
494 av_log(m->avctx, AV_LOG_WARNING,
495 "Lossless check failed - expected %02x, calculated %02x.\n",
496 lossless_check, tmp);
501 memset(s->ch_assign, 0, sizeof(s->ch_assign));
503 for (ch = 0; ch <= s->max_matrix_channel; ch++) {
504 int ch_assign = get_bits(gbp, 6);
505 if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD) {
506 uint64_t channel = thd_channel_layout_extract_channel(s->ch_layout,
508 ch_assign = av_get_channel_layout_channel_index(s->ch_layout,
511 if ((unsigned)ch_assign > s->max_matrix_channel) {
512 avpriv_request_sample(m->avctx,
513 "Assignment of matrix channel %d to invalid output channel %d",
515 return AVERROR_PATCHWELCOME;
517 s->ch_assign[ch_assign] = ch;
520 checksum = ff_mlp_restart_checksum(buf, get_bits_count(gbp) - start_count);
522 if (checksum != get_bits(gbp, 8))
523 av_log(m->avctx, AV_LOG_ERROR, "restart header checksum error\n");
525 /* Set default decoding parameters. */
526 s->param_presence_flags = 0xff;
527 s->num_primitive_matrices = 0;
529 s->lossless_check_data = 0;
531 memset(s->output_shift , 0, sizeof(s->output_shift ));
532 memset(s->quant_step_size, 0, sizeof(s->quant_step_size));
534 for (ch = s->min_channel; ch <= s->max_channel; ch++) {
535 ChannelParams *cp = &s->channel_params[ch];
536 cp->filter_params[FIR].order = 0;
537 cp->filter_params[IIR].order = 0;
538 cp->filter_params[FIR].shift = 0;
539 cp->filter_params[IIR].shift = 0;
541 /* Default audio coding is 24-bit raw PCM. */
543 cp->sign_huff_offset = (-1) << 23;
548 if (substr == m->max_decoded_substream) {
549 m->avctx->channels = s->max_matrix_channel + 1;
550 m->avctx->channel_layout = s->ch_layout;
552 if (m->avctx->codec_id == AV_CODEC_ID_MLP && m->needs_reordering) {
553 if (m->avctx->channel_layout == (AV_CH_LAYOUT_QUAD|AV_CH_LOW_FREQUENCY) ||
554 m->avctx->channel_layout == AV_CH_LAYOUT_5POINT0_BACK) {
555 int i = s->ch_assign[4];
556 s->ch_assign[4] = s->ch_assign[3];
557 s->ch_assign[3] = s->ch_assign[2];
559 } else if (m->avctx->channel_layout == AV_CH_LAYOUT_5POINT1_BACK) {
560 FFSWAP(int, s->ch_assign[2], s->ch_assign[4]);
561 FFSWAP(int, s->ch_assign[3], s->ch_assign[5]);
570 /** Read parameters for one of the prediction filters. */
572 static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
573 unsigned int substr, unsigned int channel,
576 SubStream *s = &m->substream[substr];
577 FilterParams *fp = &s->channel_params[channel].filter_params[filter];
578 const int max_order = filter ? MAX_IIR_ORDER : MAX_FIR_ORDER;
579 const char fchar = filter ? 'I' : 'F';
582 // Filter is 0 for FIR, 1 for IIR.
583 av_assert0(filter < 2);
585 if (m->filter_changed[channel][filter]++ > 1) {
586 av_log(m->avctx, AV_LOG_ERROR, "Filters may change only once per access unit.\n");
587 return AVERROR_INVALIDDATA;
590 order = get_bits(gbp, 4);
591 if (order > max_order) {
592 av_log(m->avctx, AV_LOG_ERROR,
593 "%cIR filter order %d is greater than maximum %d.\n",
594 fchar, order, max_order);
595 return AVERROR_INVALIDDATA;
600 int32_t *fcoeff = s->channel_params[channel].coeff[filter];
601 int coeff_bits, coeff_shift;
603 fp->shift = get_bits(gbp, 4);
605 coeff_bits = get_bits(gbp, 5);
606 coeff_shift = get_bits(gbp, 3);
607 if (coeff_bits < 1 || coeff_bits > 16) {
608 av_log(m->avctx, AV_LOG_ERROR,
609 "%cIR filter coeff_bits must be between 1 and 16.\n",
611 return AVERROR_INVALIDDATA;
613 if (coeff_bits + coeff_shift > 16) {
614 av_log(m->avctx, AV_LOG_ERROR,
615 "Sum of coeff_bits and coeff_shift for %cIR filter must be 16 or less.\n",
617 return AVERROR_INVALIDDATA;
620 for (i = 0; i < order; i++)
621 fcoeff[i] = get_sbits(gbp, coeff_bits) << coeff_shift;
623 if (get_bits1(gbp)) {
624 int state_bits, state_shift;
627 av_log(m->avctx, AV_LOG_ERROR,
628 "FIR filter has state data specified.\n");
629 return AVERROR_INVALIDDATA;
632 state_bits = get_bits(gbp, 4);
633 state_shift = get_bits(gbp, 4);
635 /* TODO: Check validity of state data. */
637 for (i = 0; i < order; i++)
638 fp->state[i] = state_bits ? get_sbits(gbp, state_bits) << state_shift : 0;
645 /** Read parameters for primitive matrices. */
647 static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitContext *gbp)
649 SubStream *s = &m->substream[substr];
650 unsigned int mat, ch;
651 const int max_primitive_matrices = m->avctx->codec_id == AV_CODEC_ID_MLP
653 : MAX_MATRICES_TRUEHD;
655 if (m->matrix_changed++ > 1) {
656 av_log(m->avctx, AV_LOG_ERROR, "Matrices may change only once per access unit.\n");
657 return AVERROR_INVALIDDATA;
660 s->num_primitive_matrices = get_bits(gbp, 4);
662 if (s->num_primitive_matrices > max_primitive_matrices) {
663 av_log(m->avctx, AV_LOG_ERROR,
664 "Number of primitive matrices cannot be greater than %d.\n",
665 max_primitive_matrices);
666 return AVERROR_INVALIDDATA;
669 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
670 int frac_bits, max_chan;
671 s->matrix_out_ch[mat] = get_bits(gbp, 4);
672 frac_bits = get_bits(gbp, 4);
673 s->lsb_bypass [mat] = get_bits1(gbp);
675 if (s->matrix_out_ch[mat] > s->max_matrix_channel) {
676 av_log(m->avctx, AV_LOG_ERROR,
677 "Invalid channel %d specified as output from matrix.\n",
678 s->matrix_out_ch[mat]);
679 return AVERROR_INVALIDDATA;
681 if (frac_bits > 14) {
682 av_log(m->avctx, AV_LOG_ERROR,
683 "Too many fractional bits specified.\n");
684 return AVERROR_INVALIDDATA;
687 max_chan = s->max_matrix_channel;
691 for (ch = 0; ch <= max_chan; ch++) {
694 coeff_val = get_sbits(gbp, frac_bits + 2);
696 s->matrix_coeff[mat][ch] = coeff_val << (14 - frac_bits);
700 s->matrix_noise_shift[mat] = get_bits(gbp, 4);
702 s->matrix_noise_shift[mat] = 0;
708 /** Read channel parameters. */
710 static int read_channel_params(MLPDecodeContext *m, unsigned int substr,
711 GetBitContext *gbp, unsigned int ch)
713 SubStream *s = &m->substream[substr];
714 ChannelParams *cp = &s->channel_params[ch];
715 FilterParams *fir = &cp->filter_params[FIR];
716 FilterParams *iir = &cp->filter_params[IIR];
719 if (s->param_presence_flags & PARAM_FIR)
721 if ((ret = read_filter_params(m, gbp, substr, ch, FIR)) < 0)
724 if (s->param_presence_flags & PARAM_IIR)
726 if ((ret = read_filter_params(m, gbp, substr, ch, IIR)) < 0)
729 if (fir->order + iir->order > 8) {
730 av_log(m->avctx, AV_LOG_ERROR, "Total filter orders too high.\n");
731 return AVERROR_INVALIDDATA;
734 if (fir->order && iir->order &&
735 fir->shift != iir->shift) {
736 av_log(m->avctx, AV_LOG_ERROR,
737 "FIR and IIR filters must use the same precision.\n");
738 return AVERROR_INVALIDDATA;
740 /* The FIR and IIR filters must have the same precision.
741 * To simplify the filtering code, only the precision of the
742 * FIR filter is considered. If only the IIR filter is employed,
743 * the FIR filter precision is set to that of the IIR filter, so
744 * that the filtering code can use it. */
745 if (!fir->order && iir->order)
746 fir->shift = iir->shift;
748 if (s->param_presence_flags & PARAM_HUFFOFFSET)
750 cp->huff_offset = get_sbits(gbp, 15);
752 cp->codebook = get_bits(gbp, 2);
753 cp->huff_lsbs = get_bits(gbp, 5);
755 if (cp->huff_lsbs > 24) {
756 av_log(m->avctx, AV_LOG_ERROR, "Invalid huff_lsbs.\n");
758 return AVERROR_INVALIDDATA;
761 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
766 /** Read decoding parameters that change more often than those in the restart
769 static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
772 SubStream *s = &m->substream[substr];
776 if (s->param_presence_flags & PARAM_PRESENCE)
778 s->param_presence_flags = get_bits(gbp, 8);
780 if (s->param_presence_flags & PARAM_BLOCKSIZE)
781 if (get_bits1(gbp)) {
782 s->blocksize = get_bits(gbp, 9);
783 if (s->blocksize < 8 || s->blocksize > m->access_unit_size) {
784 av_log(m->avctx, AV_LOG_ERROR, "Invalid blocksize.\n");
786 return AVERROR_INVALIDDATA;
790 if (s->param_presence_flags & PARAM_MATRIX)
792 if ((ret = read_matrix_params(m, substr, gbp)) < 0)
795 if (s->param_presence_flags & PARAM_OUTSHIFT)
797 for (ch = 0; ch <= s->max_matrix_channel; ch++)
798 s->output_shift[ch] = get_sbits(gbp, 4);
800 if (s->param_presence_flags & PARAM_QUANTSTEP)
802 for (ch = 0; ch <= s->max_channel; ch++) {
803 ChannelParams *cp = &s->channel_params[ch];
805 s->quant_step_size[ch] = get_bits(gbp, 4);
807 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
810 for (ch = s->min_channel; ch <= s->max_channel; ch++)
812 if ((ret = read_channel_params(m, substr, gbp, ch)) < 0)
818 #define MSB_MASK(bits) (-1u << bits)
820 /** Generate PCM samples using the prediction filters and residual values
821 * read from the data stream, and update the filter state. */
823 static void filter_channel(MLPDecodeContext *m, unsigned int substr,
824 unsigned int channel)
826 SubStream *s = &m->substream[substr];
827 const int32_t *fircoeff = s->channel_params[channel].coeff[FIR];
828 int32_t state_buffer[NUM_FILTERS][MAX_BLOCKSIZE + MAX_FIR_ORDER];
829 int32_t *firbuf = state_buffer[FIR] + MAX_BLOCKSIZE;
830 int32_t *iirbuf = state_buffer[IIR] + MAX_BLOCKSIZE;
831 FilterParams *fir = &s->channel_params[channel].filter_params[FIR];
832 FilterParams *iir = &s->channel_params[channel].filter_params[IIR];
833 unsigned int filter_shift = fir->shift;
834 int32_t mask = MSB_MASK(s->quant_step_size[channel]);
836 memcpy(firbuf, fir->state, MAX_FIR_ORDER * sizeof(int32_t));
837 memcpy(iirbuf, iir->state, MAX_IIR_ORDER * sizeof(int32_t));
839 m->dsp.mlp_filter_channel(firbuf, fircoeff,
840 fir->order, iir->order,
841 filter_shift, mask, s->blocksize,
842 &m->sample_buffer[s->blockpos][channel]);
844 memcpy(fir->state, firbuf - s->blocksize, MAX_FIR_ORDER * sizeof(int32_t));
845 memcpy(iir->state, iirbuf - s->blocksize, MAX_IIR_ORDER * sizeof(int32_t));
848 /** Read a block of PCM residual data (or actual if no filtering active). */
850 static int read_block_data(MLPDecodeContext *m, GetBitContext *gbp,
853 SubStream *s = &m->substream[substr];
854 unsigned int i, ch, expected_stream_pos = 0;
857 if (s->data_check_present) {
858 expected_stream_pos = get_bits_count(gbp);
859 expected_stream_pos += get_bits(gbp, 16);
860 avpriv_request_sample(m->avctx,
861 "Substreams with VLC block size check info");
864 if (s->blockpos + s->blocksize > m->access_unit_size) {
865 av_log(m->avctx, AV_LOG_ERROR, "too many audio samples in frame\n");
866 return AVERROR_INVALIDDATA;
869 memset(&m->bypassed_lsbs[s->blockpos][0], 0,
870 s->blocksize * sizeof(m->bypassed_lsbs[0]));
872 for (i = 0; i < s->blocksize; i++)
873 if ((ret = read_huff_channels(m, gbp, substr, i)) < 0)
876 for (ch = s->min_channel; ch <= s->max_channel; ch++)
877 filter_channel(m, substr, ch);
879 s->blockpos += s->blocksize;
881 if (s->data_check_present) {
882 if (get_bits_count(gbp) != expected_stream_pos)
883 av_log(m->avctx, AV_LOG_ERROR, "block data length mismatch\n");
890 /** Data table used for TrueHD noise generation function. */
892 static const int8_t noise_table[256] = {
893 30, 51, 22, 54, 3, 7, -4, 38, 14, 55, 46, 81, 22, 58, -3, 2,
894 52, 31, -7, 51, 15, 44, 74, 30, 85, -17, 10, 33, 18, 80, 28, 62,
895 10, 32, 23, 69, 72, 26, 35, 17, 73, 60, 8, 56, 2, 6, -2, -5,
896 51, 4, 11, 50, 66, 76, 21, 44, 33, 47, 1, 26, 64, 48, 57, 40,
897 38, 16, -10, -28, 92, 22, -18, 29, -10, 5, -13, 49, 19, 24, 70, 34,
898 61, 48, 30, 14, -6, 25, 58, 33, 42, 60, 67, 17, 54, 17, 22, 30,
899 67, 44, -9, 50, -11, 43, 40, 32, 59, 82, 13, 49, -14, 55, 60, 36,
900 48, 49, 31, 47, 15, 12, 4, 65, 1, 23, 29, 39, 45, -2, 84, 69,
901 0, 72, 37, 57, 27, 41, -15, -16, 35, 31, 14, 61, 24, 0, 27, 24,
902 16, 41, 55, 34, 53, 9, 56, 12, 25, 29, 53, 5, 20, -20, -8, 20,
903 13, 28, -3, 78, 38, 16, 11, 62, 46, 29, 21, 24, 46, 65, 43, -23,
904 89, 18, 74, 21, 38, -12, 19, 12, -19, 8, 15, 33, 4, 57, 9, -8,
905 36, 35, 26, 28, 7, 83, 63, 79, 75, 11, 3, 87, 37, 47, 34, 40,
906 39, 19, 20, 42, 27, 34, 39, 77, 13, 42, 59, 64, 45, -1, 32, 37,
907 45, -5, 53, -6, 7, 36, 50, 23, 6, 32, 9, -21, 18, 71, 27, 52,
908 -25, 31, 35, 42, -1, 68, 63, 52, 26, 43, 66, 37, 41, 25, 40, 70,
911 /** Noise generation functions.
912 * I'm not sure what these are for - they seem to be some kind of pseudorandom
913 * sequence generators, used to generate noise data which is used when the
914 * channels are rematrixed. I'm not sure if they provide a practical benefit
915 * to compression, or just obfuscate the decoder. Are they for some kind of
918 /** Generate two channels of noise, used in the matrix when
919 * restart sync word == 0x31ea. */
921 static void generate_2_noise_channels(MLPDecodeContext *m, unsigned int substr)
923 SubStream *s = &m->substream[substr];
925 uint32_t seed = s->noisegen_seed;
926 unsigned int maxchan = s->max_matrix_channel;
928 for (i = 0; i < s->blockpos; i++) {
929 uint16_t seed_shr7 = seed >> 7;
930 m->sample_buffer[i][maxchan+1] = ((int8_t)(seed >> 15)) << s->noise_shift;
931 m->sample_buffer[i][maxchan+2] = ((int8_t) seed_shr7) << s->noise_shift;
933 seed = (seed << 16) ^ seed_shr7 ^ (seed_shr7 << 5);
936 s->noisegen_seed = seed;
939 /** Generate a block of noise, used when restart sync word == 0x31eb. */
941 static void fill_noise_buffer(MLPDecodeContext *m, unsigned int substr)
943 SubStream *s = &m->substream[substr];
945 uint32_t seed = s->noisegen_seed;
947 for (i = 0; i < m->access_unit_size_pow2; i++) {
948 uint8_t seed_shr15 = seed >> 15;
949 m->noise_buffer[i] = noise_table[seed_shr15];
950 seed = (seed << 8) ^ seed_shr15 ^ (seed_shr15 << 5);
953 s->noisegen_seed = seed;
957 /** Apply the channel matrices in turn to reconstruct the original audio
960 static void rematrix_channels(MLPDecodeContext *m, unsigned int substr)
962 SubStream *s = &m->substream[substr];
963 unsigned int mat, src_ch, i;
964 unsigned int maxchan;
966 maxchan = s->max_matrix_channel;
967 if (!s->noise_type) {
968 generate_2_noise_channels(m, substr);
971 fill_noise_buffer(m, substr);
974 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
975 int matrix_noise_shift = s->matrix_noise_shift[mat];
976 unsigned int dest_ch = s->matrix_out_ch[mat];
977 int32_t mask = MSB_MASK(s->quant_step_size[dest_ch]);
978 int32_t *coeffs = s->matrix_coeff[mat];
979 int index = s->num_primitive_matrices - mat;
980 int index2 = 2 * index + 1;
982 /* TODO: DSPContext? */
984 for (i = 0; i < s->blockpos; i++) {
985 int32_t bypassed_lsb = m->bypassed_lsbs[i][mat];
986 int32_t *samples = m->sample_buffer[i];
989 for (src_ch = 0; src_ch <= maxchan; src_ch++)
990 accum += (int64_t) samples[src_ch] * coeffs[src_ch];
992 if (matrix_noise_shift) {
993 index &= m->access_unit_size_pow2 - 1;
994 accum += m->noise_buffer[index] << (matrix_noise_shift + 7);
998 samples[dest_ch] = ((accum >> 14) & mask) + bypassed_lsb;
1003 /** Write the audio data into the output buffer. */
1005 static int output_data(MLPDecodeContext *m, unsigned int substr,
1006 AVFrame *frame, int *got_frame_ptr)
1008 AVCodecContext *avctx = m->avctx;
1009 SubStream *s = &m->substream[substr];
1010 unsigned int i, out_ch = 0;
1014 int is32 = (m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
1016 if (m->avctx->channels != s->max_matrix_channel + 1) {
1017 av_log(m->avctx, AV_LOG_ERROR, "channel count mismatch\n");
1018 return AVERROR_INVALIDDATA;
1022 av_log(avctx, AV_LOG_ERROR, "No samples to output.\n");
1023 return AVERROR_INVALIDDATA;
1026 /* get output buffer */
1027 frame->nb_samples = s->blockpos;
1028 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1030 data_32 = (int32_t *)frame->data[0];
1031 data_16 = (int16_t *)frame->data[0];
1033 for (i = 0; i < s->blockpos; i++) {
1034 for (out_ch = 0; out_ch <= s->max_matrix_channel; out_ch++) {
1035 int mat_ch = s->ch_assign[out_ch];
1036 int32_t sample = m->sample_buffer[i][mat_ch]
1037 << s->output_shift[mat_ch];
1038 s->lossless_check_data ^= (sample & 0xffffff) << mat_ch;
1039 if (is32) *data_32++ = sample << 8;
1040 else *data_16++ = sample >> 8;
1049 /** Read an access unit from the stream.
1050 * @return negative on error, 0 if not enough data is present in the input stream,
1051 * otherwise the number of bytes consumed. */
1053 static int read_access_unit(AVCodecContext *avctx, void* data,
1054 int *got_frame_ptr, AVPacket *avpkt)
1056 const uint8_t *buf = avpkt->data;
1057 int buf_size = avpkt->size;
1058 MLPDecodeContext *m = avctx->priv_data;
1060 unsigned int length, substr;
1061 unsigned int substream_start;
1062 unsigned int header_size = 4;
1063 unsigned int substr_header_size = 0;
1064 uint8_t substream_parity_present[MAX_SUBSTREAMS];
1065 uint16_t substream_data_len[MAX_SUBSTREAMS];
1066 uint8_t parity_bits;
1072 length = (AV_RB16(buf) & 0xfff) * 2;
1074 if (length < 4 || length > buf_size)
1075 return AVERROR_INVALIDDATA;
1077 init_get_bits(&gb, (buf + 4), (length - 4) * 8);
1079 m->is_major_sync_unit = 0;
1080 if (show_bits_long(&gb, 31) == (0xf8726fba >> 1)) {
1081 if (read_major_sync(m, &gb) < 0)
1083 m->is_major_sync_unit = 1;
1087 if (!m->params_valid) {
1088 av_log(m->avctx, AV_LOG_WARNING,
1089 "Stream parameters not seen; skipping frame.\n");
1094 substream_start = 0;
1096 for (substr = 0; substr < m->num_substreams; substr++) {
1097 int extraword_present, checkdata_present, end, nonrestart_substr;
1099 extraword_present = get_bits1(&gb);
1100 nonrestart_substr = get_bits1(&gb);
1101 checkdata_present = get_bits1(&gb);
1104 end = get_bits(&gb, 12) * 2;
1106 substr_header_size += 2;
1108 if (extraword_present) {
1109 if (m->avctx->codec_id == AV_CODEC_ID_MLP) {
1110 av_log(m->avctx, AV_LOG_ERROR, "There must be no extraword for MLP.\n");
1114 substr_header_size += 2;
1117 if (!(nonrestart_substr ^ m->is_major_sync_unit)) {
1118 av_log(m->avctx, AV_LOG_ERROR, "Invalid nonrestart_substr.\n");
1122 if (end + header_size + substr_header_size > length) {
1123 av_log(m->avctx, AV_LOG_ERROR,
1124 "Indicated length of substream %d data goes off end of "
1125 "packet.\n", substr);
1126 end = length - header_size - substr_header_size;
1129 if (end < substream_start) {
1130 av_log(avctx, AV_LOG_ERROR,
1131 "Indicated end offset of substream %d data "
1132 "is smaller than calculated start offset.\n",
1137 if (substr > m->max_decoded_substream)
1140 substream_parity_present[substr] = checkdata_present;
1141 substream_data_len[substr] = end - substream_start;
1142 substream_start = end;
1145 parity_bits = ff_mlp_calculate_parity(buf, 4);
1146 parity_bits ^= ff_mlp_calculate_parity(buf + header_size, substr_header_size);
1148 if ((((parity_bits >> 4) ^ parity_bits) & 0xF) != 0xF) {
1149 av_log(avctx, AV_LOG_ERROR, "Parity check failed.\n");
1153 buf += header_size + substr_header_size;
1155 for (substr = 0; substr <= m->max_decoded_substream; substr++) {
1156 SubStream *s = &m->substream[substr];
1157 init_get_bits(&gb, buf, substream_data_len[substr] * 8);
1159 m->matrix_changed = 0;
1160 memset(m->filter_changed, 0, sizeof(m->filter_changed));
1164 if (get_bits1(&gb)) {
1165 if (get_bits1(&gb)) {
1166 /* A restart header should be present. */
1167 if (read_restart_header(m, &gb, buf, substr) < 0)
1169 s->restart_seen = 1;
1172 if (!s->restart_seen)
1174 if (read_decoding_params(m, &gb, substr) < 0)
1178 if (!s->restart_seen)
1181 if ((ret = read_block_data(m, &gb, substr)) < 0)
1184 if (get_bits_count(&gb) >= substream_data_len[substr] * 8)
1185 goto substream_length_mismatch;
1187 } while (!get_bits1(&gb));
1189 skip_bits(&gb, (-get_bits_count(&gb)) & 15);
1191 if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 32) {
1194 if (get_bits(&gb, 16) != 0xD234)
1195 return AVERROR_INVALIDDATA;
1197 shorten_by = get_bits(&gb, 16);
1198 if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD && shorten_by & 0x2000)
1199 s->blockpos -= FFMIN(shorten_by & 0x1FFF, s->blockpos);
1200 else if (m->avctx->codec_id == AV_CODEC_ID_MLP && shorten_by != 0xD234)
1201 return AVERROR_INVALIDDATA;
1203 if (substr == m->max_decoded_substream)
1204 av_log(m->avctx, AV_LOG_INFO, "End of stream indicated.\n");
1207 if (substream_parity_present[substr]) {
1208 uint8_t parity, checksum;
1210 if (substream_data_len[substr] * 8 - get_bits_count(&gb) != 16)
1211 goto substream_length_mismatch;
1213 parity = ff_mlp_calculate_parity(buf, substream_data_len[substr] - 2);
1214 checksum = ff_mlp_checksum8 (buf, substream_data_len[substr] - 2);
1216 if ((get_bits(&gb, 8) ^ parity) != 0xa9 )
1217 av_log(m->avctx, AV_LOG_ERROR, "Substream %d parity check failed.\n", substr);
1218 if ( get_bits(&gb, 8) != checksum)
1219 av_log(m->avctx, AV_LOG_ERROR, "Substream %d checksum failed.\n" , substr);
1222 if (substream_data_len[substr] * 8 != get_bits_count(&gb))
1223 goto substream_length_mismatch;
1226 if (!s->restart_seen)
1227 av_log(m->avctx, AV_LOG_ERROR,
1228 "No restart header present in substream %d.\n", substr);
1230 buf += substream_data_len[substr];
1233 rematrix_channels(m, m->max_decoded_substream);
1235 if ((ret = output_data(m, m->max_decoded_substream, data, got_frame_ptr)) < 0)
1240 substream_length_mismatch:
1241 av_log(m->avctx, AV_LOG_ERROR, "substream %d length mismatch\n", substr);
1242 return AVERROR_INVALIDDATA;
1245 m->params_valid = 0;
1246 return AVERROR_INVALIDDATA;
1249 #if CONFIG_MLP_DECODER
1250 AVCodec ff_mlp_decoder = {
1252 .type = AVMEDIA_TYPE_AUDIO,
1253 .id = AV_CODEC_ID_MLP,
1254 .priv_data_size = sizeof(MLPDecodeContext),
1255 .init = mlp_decode_init,
1256 .decode = read_access_unit,
1257 .capabilities = CODEC_CAP_DR1,
1258 .long_name = NULL_IF_CONFIG_SMALL("MLP (Meridian Lossless Packing)"),
1261 #if CONFIG_TRUEHD_DECODER
1262 AVCodec ff_truehd_decoder = {
1264 .type = AVMEDIA_TYPE_AUDIO,
1265 .id = AV_CODEC_ID_TRUEHD,
1266 .priv_data_size = sizeof(MLPDecodeContext),
1267 .init = mlp_decode_init,
1268 .decode = read_access_unit,
1269 .capabilities = CODEC_CAP_DR1,
1270 .long_name = NULL_IF_CONFIG_SMALL("TrueHD"),
1272 #endif /* CONFIG_TRUEHD_DECODER */