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/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 /// number of PCM samples contained in each frame
135 int access_unit_size;
136 /// next power of two above the number of samples in each frame
137 int access_unit_size_pow2;
139 SubStream substream[MAX_SUBSTREAMS];
142 int filter_changed[MAX_CHANNELS][NUM_FILTERS];
144 int8_t noise_buffer[MAX_BLOCKSIZE_POW2];
145 int8_t bypassed_lsbs[MAX_BLOCKSIZE][MAX_CHANNELS];
146 int32_t sample_buffer[MAX_BLOCKSIZE][MAX_CHANNELS];
151 static const uint64_t thd_channel_order[] = {
152 AV_CH_FRONT_LEFT, AV_CH_FRONT_RIGHT, // LR
153 AV_CH_FRONT_CENTER, // C
154 AV_CH_LOW_FREQUENCY, // LFE
155 AV_CH_SIDE_LEFT, AV_CH_SIDE_RIGHT, // LRs
156 AV_CH_TOP_FRONT_LEFT, AV_CH_TOP_FRONT_RIGHT, // LRvh
157 AV_CH_FRONT_LEFT_OF_CENTER, AV_CH_FRONT_RIGHT_OF_CENTER, // LRc
158 AV_CH_BACK_LEFT, AV_CH_BACK_RIGHT, // LRrs
159 AV_CH_BACK_CENTER, // Cs
160 AV_CH_TOP_CENTER, // Ts
161 AV_CH_SURROUND_DIRECT_LEFT, AV_CH_SURROUND_DIRECT_RIGHT, // LRsd
162 AV_CH_WIDE_LEFT, AV_CH_WIDE_RIGHT, // LRw
163 AV_CH_TOP_FRONT_CENTER, // Cvh
164 AV_CH_LOW_FREQUENCY_2, // LFE2
167 static uint64_t thd_channel_layout_extract_channel(uint64_t channel_layout,
172 if (av_get_channel_layout_nb_channels(channel_layout) <= index)
175 for (i = 0; i < FF_ARRAY_ELEMS(thd_channel_order); i++)
176 if (channel_layout & thd_channel_order[i] && !index--)
177 return thd_channel_order[i];
181 static VLC huff_vlc[3];
183 /** Initialize static data, constant between all invocations of the codec. */
185 static av_cold void init_static(void)
187 if (!huff_vlc[0].bits) {
188 INIT_VLC_STATIC(&huff_vlc[0], VLC_BITS, 18,
189 &ff_mlp_huffman_tables[0][0][1], 2, 1,
190 &ff_mlp_huffman_tables[0][0][0], 2, 1, 512);
191 INIT_VLC_STATIC(&huff_vlc[1], VLC_BITS, 16,
192 &ff_mlp_huffman_tables[1][0][1], 2, 1,
193 &ff_mlp_huffman_tables[1][0][0], 2, 1, 512);
194 INIT_VLC_STATIC(&huff_vlc[2], VLC_BITS, 15,
195 &ff_mlp_huffman_tables[2][0][1], 2, 1,
196 &ff_mlp_huffman_tables[2][0][0], 2, 1, 512);
202 static inline int32_t calculate_sign_huff(MLPDecodeContext *m,
203 unsigned int substr, unsigned int ch)
205 SubStream *s = &m->substream[substr];
206 ChannelParams *cp = &s->channel_params[ch];
207 int lsb_bits = cp->huff_lsbs - s->quant_step_size[ch];
208 int sign_shift = lsb_bits + (cp->codebook ? 2 - cp->codebook : -1);
209 int32_t sign_huff_offset = cp->huff_offset;
211 if (cp->codebook > 0)
212 sign_huff_offset -= 7 << lsb_bits;
215 sign_huff_offset -= 1 << sign_shift;
217 return sign_huff_offset;
220 /** Read a sample, consisting of either, both or neither of entropy-coded MSBs
223 static inline int read_huff_channels(MLPDecodeContext *m, GetBitContext *gbp,
224 unsigned int substr, unsigned int pos)
226 SubStream *s = &m->substream[substr];
227 unsigned int mat, channel;
229 for (mat = 0; mat < s->num_primitive_matrices; mat++)
230 if (s->lsb_bypass[mat])
231 m->bypassed_lsbs[pos + s->blockpos][mat] = get_bits1(gbp);
233 for (channel = s->min_channel; channel <= s->max_channel; channel++) {
234 ChannelParams *cp = &s->channel_params[channel];
235 int codebook = cp->codebook;
236 int quant_step_size = s->quant_step_size[channel];
237 int lsb_bits = cp->huff_lsbs - quant_step_size;
241 result = get_vlc2(gbp, huff_vlc[codebook-1].table,
242 VLC_BITS, (9 + VLC_BITS - 1) / VLC_BITS);
245 return AVERROR_INVALIDDATA;
248 result = (result << lsb_bits) + get_bits(gbp, lsb_bits);
250 result += cp->sign_huff_offset;
251 result <<= quant_step_size;
253 m->sample_buffer[pos + s->blockpos][channel] = result;
259 static av_cold int mlp_decode_init(AVCodecContext *avctx)
261 MLPDecodeContext *m = avctx->priv_data;
266 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
267 m->substream[substr].lossless_check_data = 0xffffffff;
268 ff_mlpdsp_init(&m->dsp);
273 /** Read a major sync info header - contains high level information about
274 * the stream - sample rate, channel arrangement etc. Most of this
275 * information is not actually necessary for decoding, only for playback.
278 static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
283 if ((ret = ff_mlp_read_major_sync(m->avctx, &mh, gb)) != 0)
286 if (mh.group1_bits == 0) {
287 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown bits per sample\n");
288 return AVERROR_INVALIDDATA;
290 if (mh.group2_bits > mh.group1_bits) {
291 av_log(m->avctx, AV_LOG_ERROR,
292 "Channel group 2 cannot have more bits per sample than group 1.\n");
293 return AVERROR_INVALIDDATA;
296 if (mh.group2_samplerate && mh.group2_samplerate != mh.group1_samplerate) {
297 av_log(m->avctx, AV_LOG_ERROR,
298 "Channel groups with differing sample rates are not currently supported.\n");
299 return AVERROR_INVALIDDATA;
302 if (mh.group1_samplerate == 0) {
303 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown sampling rate\n");
304 return AVERROR_INVALIDDATA;
306 if (mh.group1_samplerate > MAX_SAMPLERATE) {
307 av_log(m->avctx, AV_LOG_ERROR,
308 "Sampling rate %d is greater than the supported maximum (%d).\n",
309 mh.group1_samplerate, MAX_SAMPLERATE);
310 return AVERROR_INVALIDDATA;
312 if (mh.access_unit_size > MAX_BLOCKSIZE) {
313 av_log(m->avctx, AV_LOG_ERROR,
314 "Block size %d is greater than the supported maximum (%d).\n",
315 mh.access_unit_size, MAX_BLOCKSIZE);
316 return AVERROR_INVALIDDATA;
318 if (mh.access_unit_size_pow2 > MAX_BLOCKSIZE_POW2) {
319 av_log(m->avctx, AV_LOG_ERROR,
320 "Block size pow2 %d is greater than the supported maximum (%d).\n",
321 mh.access_unit_size_pow2, MAX_BLOCKSIZE_POW2);
322 return AVERROR_INVALIDDATA;
325 if (mh.num_substreams == 0)
326 return AVERROR_INVALIDDATA;
327 if (m->avctx->codec_id == AV_CODEC_ID_MLP && mh.num_substreams > 2) {
328 av_log(m->avctx, AV_LOG_ERROR, "MLP only supports up to 2 substreams.\n");
329 return AVERROR_INVALIDDATA;
331 if (mh.num_substreams > MAX_SUBSTREAMS) {
332 avpriv_request_sample(m->avctx,
333 "%d substreams (more than the "
334 "maximum supported by the decoder)",
336 return AVERROR_PATCHWELCOME;
339 m->access_unit_size = mh.access_unit_size;
340 m->access_unit_size_pow2 = mh.access_unit_size_pow2;
342 m->num_substreams = mh.num_substreams;
343 m->max_decoded_substream = m->num_substreams - 1;
345 m->avctx->sample_rate = mh.group1_samplerate;
346 m->avctx->frame_size = mh.access_unit_size;
348 m->avctx->bits_per_raw_sample = mh.group1_bits;
349 if (mh.group1_bits > 16)
350 m->avctx->sample_fmt = AV_SAMPLE_FMT_S32;
352 m->avctx->sample_fmt = AV_SAMPLE_FMT_S16;
355 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
356 m->substream[substr].restart_seen = 0;
358 /* Set the layout for each substream. When there's more than one, the first
359 * substream is Stereo. Subsequent substreams' layouts are indicated in the
361 if (m->avctx->codec_id == AV_CODEC_ID_MLP) {
362 if ((substr = (mh.num_substreams > 1)))
363 m->substream[0].ch_layout = AV_CH_LAYOUT_STEREO;
364 m->substream[substr].ch_layout = mh.channel_layout_mlp;
366 if ((substr = (mh.num_substreams > 1)))
367 m->substream[0].ch_layout = AV_CH_LAYOUT_STEREO;
368 if (mh.num_substreams > 2)
369 if (mh.channel_layout_thd_stream2)
370 m->substream[2].ch_layout = mh.channel_layout_thd_stream2;
372 m->substream[2].ch_layout = mh.channel_layout_thd_stream1;
373 m->substream[substr].ch_layout = mh.channel_layout_thd_stream1;
379 /** Read a restart header from a block in a substream. This contains parameters
380 * required to decode the audio that do not change very often. Generally
381 * (always) present only in blocks following a major sync. */
383 static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
384 const uint8_t *buf, unsigned int substr)
386 SubStream *s = &m->substream[substr];
390 uint8_t lossless_check;
391 int start_count = get_bits_count(gbp);
392 const int max_matrix_channel = m->avctx->codec_id == AV_CODEC_ID_MLP
393 ? MAX_MATRIX_CHANNEL_MLP
394 : MAX_MATRIX_CHANNEL_TRUEHD;
396 sync_word = get_bits(gbp, 13);
398 if (sync_word != 0x31ea >> 1) {
399 av_log(m->avctx, AV_LOG_ERROR,
400 "restart header sync incorrect (got 0x%04x)\n", sync_word);
401 return AVERROR_INVALIDDATA;
404 s->noise_type = get_bits1(gbp);
406 if (m->avctx->codec_id == AV_CODEC_ID_MLP && s->noise_type) {
407 av_log(m->avctx, AV_LOG_ERROR, "MLP must have 0x31ea sync word.\n");
408 return AVERROR_INVALIDDATA;
411 skip_bits(gbp, 16); /* Output timestamp */
413 s->min_channel = get_bits(gbp, 4);
414 s->max_channel = get_bits(gbp, 4);
415 s->max_matrix_channel = get_bits(gbp, 4);
417 if (s->max_matrix_channel > max_matrix_channel) {
418 av_log(m->avctx, AV_LOG_ERROR,
419 "Max matrix channel cannot be greater than %d.\n",
421 return AVERROR_INVALIDDATA;
424 if (s->max_channel != s->max_matrix_channel) {
425 av_log(m->avctx, AV_LOG_ERROR,
426 "Max channel must be equal max matrix channel.\n");
427 return AVERROR_INVALIDDATA;
430 /* This should happen for TrueHD streams with >6 channels and MLP's noise
431 * type. It is not yet known if this is allowed. */
432 if (s->max_channel > MAX_MATRIX_CHANNEL_MLP && !s->noise_type) {
433 avpriv_request_sample(m->avctx,
434 "%d channels (more than the "
435 "maximum supported by the decoder)",
437 return AVERROR_PATCHWELCOME;
440 if (s->min_channel > s->max_channel) {
441 av_log(m->avctx, AV_LOG_ERROR,
442 "Substream min channel cannot be greater than max channel.\n");
443 return AVERROR_INVALIDDATA;
446 #if FF_API_REQUEST_CHANNELS
447 if (m->avctx->request_channels > 0 &&
448 m->avctx->request_channels <= s->max_channel + 1 &&
449 m->max_decoded_substream > substr) {
450 av_log(m->avctx, AV_LOG_DEBUG,
451 "Extracting %d-channel downmix from substream %d. "
452 "Further substreams will be skipped.\n",
453 s->max_channel + 1, substr);
454 m->max_decoded_substream = substr;
457 if (m->avctx->request_channel_layout == s->ch_layout &&
458 m->max_decoded_substream > substr) {
459 av_log(m->avctx, AV_LOG_DEBUG,
460 "Extracting %d-channel downmix (0x%"PRIx64") from substream %d. "
461 "Further substreams will be skipped.\n",
462 s->max_channel + 1, s->ch_layout, substr);
463 m->max_decoded_substream = substr;
466 s->noise_shift = get_bits(gbp, 4);
467 s->noisegen_seed = get_bits(gbp, 23);
471 s->data_check_present = get_bits1(gbp);
472 lossless_check = get_bits(gbp, 8);
473 if (substr == m->max_decoded_substream
474 && s->lossless_check_data != 0xffffffff) {
475 tmp = xor_32_to_8(s->lossless_check_data);
476 if (tmp != lossless_check)
477 av_log(m->avctx, AV_LOG_WARNING,
478 "Lossless check failed - expected %02x, calculated %02x.\n",
479 lossless_check, tmp);
484 memset(s->ch_assign, 0, sizeof(s->ch_assign));
486 for (ch = 0; ch <= s->max_matrix_channel; ch++) {
487 int ch_assign = get_bits(gbp, 6);
488 if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD) {
489 uint64_t channel = thd_channel_layout_extract_channel(s->ch_layout,
491 ch_assign = av_get_channel_layout_channel_index(s->ch_layout,
494 if (ch_assign > s->max_matrix_channel) {
495 avpriv_request_sample(m->avctx,
496 "Assignment of matrix channel %d to invalid output channel %d",
498 return AVERROR_PATCHWELCOME;
500 s->ch_assign[ch_assign] = ch;
503 checksum = ff_mlp_restart_checksum(buf, get_bits_count(gbp) - start_count);
505 if (checksum != get_bits(gbp, 8))
506 av_log(m->avctx, AV_LOG_ERROR, "restart header checksum error\n");
508 /* Set default decoding parameters. */
509 s->param_presence_flags = 0xff;
510 s->num_primitive_matrices = 0;
512 s->lossless_check_data = 0;
514 memset(s->output_shift , 0, sizeof(s->output_shift ));
515 memset(s->quant_step_size, 0, sizeof(s->quant_step_size));
517 for (ch = s->min_channel; ch <= s->max_channel; ch++) {
518 ChannelParams *cp = &s->channel_params[ch];
519 cp->filter_params[FIR].order = 0;
520 cp->filter_params[IIR].order = 0;
521 cp->filter_params[FIR].shift = 0;
522 cp->filter_params[IIR].shift = 0;
524 /* Default audio coding is 24-bit raw PCM. */
526 cp->sign_huff_offset = (-1) << 23;
531 if (substr == m->max_decoded_substream) {
532 m->avctx->channels = s->max_matrix_channel + 1;
533 m->avctx->channel_layout = s->ch_layout;
539 /** Read parameters for one of the prediction filters. */
541 static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
542 unsigned int substr, unsigned int channel,
545 SubStream *s = &m->substream[substr];
546 FilterParams *fp = &s->channel_params[channel].filter_params[filter];
547 const int max_order = filter ? MAX_IIR_ORDER : MAX_FIR_ORDER;
548 const char fchar = filter ? 'I' : 'F';
551 // Filter is 0 for FIR, 1 for IIR.
554 if (m->filter_changed[channel][filter]++ > 1) {
555 av_log(m->avctx, AV_LOG_ERROR, "Filters may change only once per access unit.\n");
556 return AVERROR_INVALIDDATA;
559 order = get_bits(gbp, 4);
560 if (order > max_order) {
561 av_log(m->avctx, AV_LOG_ERROR,
562 "%cIR filter order %d is greater than maximum %d.\n",
563 fchar, order, max_order);
564 return AVERROR_INVALIDDATA;
569 int32_t *fcoeff = s->channel_params[channel].coeff[filter];
570 int coeff_bits, coeff_shift;
572 fp->shift = get_bits(gbp, 4);
574 coeff_bits = get_bits(gbp, 5);
575 coeff_shift = get_bits(gbp, 3);
576 if (coeff_bits < 1 || coeff_bits > 16) {
577 av_log(m->avctx, AV_LOG_ERROR,
578 "%cIR filter coeff_bits must be between 1 and 16.\n",
580 return AVERROR_INVALIDDATA;
582 if (coeff_bits + coeff_shift > 16) {
583 av_log(m->avctx, AV_LOG_ERROR,
584 "Sum of coeff_bits and coeff_shift for %cIR filter must be 16 or less.\n",
586 return AVERROR_INVALIDDATA;
589 for (i = 0; i < order; i++)
590 fcoeff[i] = get_sbits(gbp, coeff_bits) << coeff_shift;
592 if (get_bits1(gbp)) {
593 int state_bits, state_shift;
596 av_log(m->avctx, AV_LOG_ERROR,
597 "FIR filter has state data specified.\n");
598 return AVERROR_INVALIDDATA;
601 state_bits = get_bits(gbp, 4);
602 state_shift = get_bits(gbp, 4);
604 /* TODO: Check validity of state data. */
606 for (i = 0; i < order; i++)
607 fp->state[i] = get_sbits(gbp, state_bits) << state_shift;
614 /** Read parameters for primitive matrices. */
616 static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitContext *gbp)
618 SubStream *s = &m->substream[substr];
619 unsigned int mat, ch;
620 const int max_primitive_matrices = m->avctx->codec_id == AV_CODEC_ID_MLP
622 : MAX_MATRICES_TRUEHD;
624 if (m->matrix_changed++ > 1) {
625 av_log(m->avctx, AV_LOG_ERROR, "Matrices may change only once per access unit.\n");
626 return AVERROR_INVALIDDATA;
629 s->num_primitive_matrices = get_bits(gbp, 4);
631 if (s->num_primitive_matrices > max_primitive_matrices) {
632 av_log(m->avctx, AV_LOG_ERROR,
633 "Number of primitive matrices cannot be greater than %d.\n",
634 max_primitive_matrices);
635 return AVERROR_INVALIDDATA;
638 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
639 int frac_bits, max_chan;
640 s->matrix_out_ch[mat] = get_bits(gbp, 4);
641 frac_bits = get_bits(gbp, 4);
642 s->lsb_bypass [mat] = get_bits1(gbp);
644 if (s->matrix_out_ch[mat] > s->max_matrix_channel) {
645 av_log(m->avctx, AV_LOG_ERROR,
646 "Invalid channel %d specified as output from matrix.\n",
647 s->matrix_out_ch[mat]);
648 return AVERROR_INVALIDDATA;
650 if (frac_bits > 14) {
651 av_log(m->avctx, AV_LOG_ERROR,
652 "Too many fractional bits specified.\n");
653 return AVERROR_INVALIDDATA;
656 max_chan = s->max_matrix_channel;
660 for (ch = 0; ch <= max_chan; ch++) {
663 coeff_val = get_sbits(gbp, frac_bits + 2);
665 s->matrix_coeff[mat][ch] = coeff_val << (14 - frac_bits);
669 s->matrix_noise_shift[mat] = get_bits(gbp, 4);
671 s->matrix_noise_shift[mat] = 0;
677 /** Read channel parameters. */
679 static int read_channel_params(MLPDecodeContext *m, unsigned int substr,
680 GetBitContext *gbp, unsigned int ch)
682 SubStream *s = &m->substream[substr];
683 ChannelParams *cp = &s->channel_params[ch];
684 FilterParams *fir = &cp->filter_params[FIR];
685 FilterParams *iir = &cp->filter_params[IIR];
688 if (s->param_presence_flags & PARAM_FIR)
690 if ((ret = read_filter_params(m, gbp, substr, ch, FIR)) < 0)
693 if (s->param_presence_flags & PARAM_IIR)
695 if ((ret = read_filter_params(m, gbp, substr, ch, IIR)) < 0)
698 if (fir->order + iir->order > 8) {
699 av_log(m->avctx, AV_LOG_ERROR, "Total filter orders too high.\n");
700 return AVERROR_INVALIDDATA;
703 if (fir->order && iir->order &&
704 fir->shift != iir->shift) {
705 av_log(m->avctx, AV_LOG_ERROR,
706 "FIR and IIR filters must use the same precision.\n");
707 return AVERROR_INVALIDDATA;
709 /* The FIR and IIR filters must have the same precision.
710 * To simplify the filtering code, only the precision of the
711 * FIR filter is considered. If only the IIR filter is employed,
712 * the FIR filter precision is set to that of the IIR filter, so
713 * that the filtering code can use it. */
714 if (!fir->order && iir->order)
715 fir->shift = iir->shift;
717 if (s->param_presence_flags & PARAM_HUFFOFFSET)
719 cp->huff_offset = get_sbits(gbp, 15);
721 cp->codebook = get_bits(gbp, 2);
722 cp->huff_lsbs = get_bits(gbp, 5);
724 if (cp->huff_lsbs > 24) {
725 av_log(m->avctx, AV_LOG_ERROR, "Invalid huff_lsbs.\n");
726 return AVERROR_INVALIDDATA;
729 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
734 /** Read decoding parameters that change more often than those in the restart
737 static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
740 SubStream *s = &m->substream[substr];
744 if (s->param_presence_flags & PARAM_PRESENCE)
746 s->param_presence_flags = get_bits(gbp, 8);
748 if (s->param_presence_flags & PARAM_BLOCKSIZE)
749 if (get_bits1(gbp)) {
750 s->blocksize = get_bits(gbp, 9);
751 if (s->blocksize < 8 || s->blocksize > m->access_unit_size) {
752 av_log(m->avctx, AV_LOG_ERROR, "Invalid blocksize.");
754 return AVERROR_INVALIDDATA;
758 if (s->param_presence_flags & PARAM_MATRIX)
760 if ((ret = read_matrix_params(m, substr, gbp)) < 0)
763 if (s->param_presence_flags & PARAM_OUTSHIFT)
765 for (ch = 0; ch <= s->max_matrix_channel; ch++)
766 s->output_shift[ch] = get_sbits(gbp, 4);
768 if (s->param_presence_flags & PARAM_QUANTSTEP)
770 for (ch = 0; ch <= s->max_channel; ch++) {
771 ChannelParams *cp = &s->channel_params[ch];
773 s->quant_step_size[ch] = get_bits(gbp, 4);
775 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
778 for (ch = s->min_channel; ch <= s->max_channel; ch++)
780 if ((ret = read_channel_params(m, substr, gbp, ch)) < 0)
786 #define MSB_MASK(bits) (-1u << bits)
788 /** Generate PCM samples using the prediction filters and residual values
789 * read from the data stream, and update the filter state. */
791 static void filter_channel(MLPDecodeContext *m, unsigned int substr,
792 unsigned int channel)
794 SubStream *s = &m->substream[substr];
795 const int32_t *fircoeff = s->channel_params[channel].coeff[FIR];
796 int32_t state_buffer[NUM_FILTERS][MAX_BLOCKSIZE + MAX_FIR_ORDER];
797 int32_t *firbuf = state_buffer[FIR] + MAX_BLOCKSIZE;
798 int32_t *iirbuf = state_buffer[IIR] + MAX_BLOCKSIZE;
799 FilterParams *fir = &s->channel_params[channel].filter_params[FIR];
800 FilterParams *iir = &s->channel_params[channel].filter_params[IIR];
801 unsigned int filter_shift = fir->shift;
802 int32_t mask = MSB_MASK(s->quant_step_size[channel]);
804 memcpy(firbuf, fir->state, MAX_FIR_ORDER * sizeof(int32_t));
805 memcpy(iirbuf, iir->state, MAX_IIR_ORDER * sizeof(int32_t));
807 m->dsp.mlp_filter_channel(firbuf, fircoeff,
808 fir->order, iir->order,
809 filter_shift, mask, s->blocksize,
810 &m->sample_buffer[s->blockpos][channel]);
812 memcpy(fir->state, firbuf - s->blocksize, MAX_FIR_ORDER * sizeof(int32_t));
813 memcpy(iir->state, iirbuf - s->blocksize, MAX_IIR_ORDER * sizeof(int32_t));
816 /** Read a block of PCM residual data (or actual if no filtering active). */
818 static int read_block_data(MLPDecodeContext *m, GetBitContext *gbp,
821 SubStream *s = &m->substream[substr];
822 unsigned int i, ch, expected_stream_pos = 0;
825 if (s->data_check_present) {
826 expected_stream_pos = get_bits_count(gbp);
827 expected_stream_pos += get_bits(gbp, 16);
828 avpriv_request_sample(m->avctx,
829 "Substreams with VLC block size check info");
832 if (s->blockpos + s->blocksize > m->access_unit_size) {
833 av_log(m->avctx, AV_LOG_ERROR, "too many audio samples in frame\n");
834 return AVERROR_INVALIDDATA;
837 memset(&m->bypassed_lsbs[s->blockpos][0], 0,
838 s->blocksize * sizeof(m->bypassed_lsbs[0]));
840 for (i = 0; i < s->blocksize; i++)
841 if ((ret = read_huff_channels(m, gbp, substr, i)) < 0)
844 for (ch = s->min_channel; ch <= s->max_channel; ch++)
845 filter_channel(m, substr, ch);
847 s->blockpos += s->blocksize;
849 if (s->data_check_present) {
850 if (get_bits_count(gbp) != expected_stream_pos)
851 av_log(m->avctx, AV_LOG_ERROR, "block data length mismatch\n");
858 /** Data table used for TrueHD noise generation function. */
860 static const int8_t noise_table[256] = {
861 30, 51, 22, 54, 3, 7, -4, 38, 14, 55, 46, 81, 22, 58, -3, 2,
862 52, 31, -7, 51, 15, 44, 74, 30, 85, -17, 10, 33, 18, 80, 28, 62,
863 10, 32, 23, 69, 72, 26, 35, 17, 73, 60, 8, 56, 2, 6, -2, -5,
864 51, 4, 11, 50, 66, 76, 21, 44, 33, 47, 1, 26, 64, 48, 57, 40,
865 38, 16, -10, -28, 92, 22, -18, 29, -10, 5, -13, 49, 19, 24, 70, 34,
866 61, 48, 30, 14, -6, 25, 58, 33, 42, 60, 67, 17, 54, 17, 22, 30,
867 67, 44, -9, 50, -11, 43, 40, 32, 59, 82, 13, 49, -14, 55, 60, 36,
868 48, 49, 31, 47, 15, 12, 4, 65, 1, 23, 29, 39, 45, -2, 84, 69,
869 0, 72, 37, 57, 27, 41, -15, -16, 35, 31, 14, 61, 24, 0, 27, 24,
870 16, 41, 55, 34, 53, 9, 56, 12, 25, 29, 53, 5, 20, -20, -8, 20,
871 13, 28, -3, 78, 38, 16, 11, 62, 46, 29, 21, 24, 46, 65, 43, -23,
872 89, 18, 74, 21, 38, -12, 19, 12, -19, 8, 15, 33, 4, 57, 9, -8,
873 36, 35, 26, 28, 7, 83, 63, 79, 75, 11, 3, 87, 37, 47, 34, 40,
874 39, 19, 20, 42, 27, 34, 39, 77, 13, 42, 59, 64, 45, -1, 32, 37,
875 45, -5, 53, -6, 7, 36, 50, 23, 6, 32, 9, -21, 18, 71, 27, 52,
876 -25, 31, 35, 42, -1, 68, 63, 52, 26, 43, 66, 37, 41, 25, 40, 70,
879 /** Noise generation functions.
880 * I'm not sure what these are for - they seem to be some kind of pseudorandom
881 * sequence generators, used to generate noise data which is used when the
882 * channels are rematrixed. I'm not sure if they provide a practical benefit
883 * to compression, or just obfuscate the decoder. Are they for some kind of
886 /** Generate two channels of noise, used in the matrix when
887 * restart sync word == 0x31ea. */
889 static void generate_2_noise_channels(MLPDecodeContext *m, unsigned int substr)
891 SubStream *s = &m->substream[substr];
893 uint32_t seed = s->noisegen_seed;
894 unsigned int maxchan = s->max_matrix_channel;
896 for (i = 0; i < s->blockpos; i++) {
897 uint16_t seed_shr7 = seed >> 7;
898 m->sample_buffer[i][maxchan+1] = ((int8_t)(seed >> 15)) << s->noise_shift;
899 m->sample_buffer[i][maxchan+2] = ((int8_t) seed_shr7) << s->noise_shift;
901 seed = (seed << 16) ^ seed_shr7 ^ (seed_shr7 << 5);
904 s->noisegen_seed = seed;
907 /** Generate a block of noise, used when restart sync word == 0x31eb. */
909 static void fill_noise_buffer(MLPDecodeContext *m, unsigned int substr)
911 SubStream *s = &m->substream[substr];
913 uint32_t seed = s->noisegen_seed;
915 for (i = 0; i < m->access_unit_size_pow2; i++) {
916 uint8_t seed_shr15 = seed >> 15;
917 m->noise_buffer[i] = noise_table[seed_shr15];
918 seed = (seed << 8) ^ seed_shr15 ^ (seed_shr15 << 5);
921 s->noisegen_seed = seed;
925 /** Apply the channel matrices in turn to reconstruct the original audio
928 static void rematrix_channels(MLPDecodeContext *m, unsigned int substr)
930 SubStream *s = &m->substream[substr];
931 unsigned int mat, src_ch, i;
932 unsigned int maxchan;
934 maxchan = s->max_matrix_channel;
935 if (!s->noise_type) {
936 generate_2_noise_channels(m, substr);
939 fill_noise_buffer(m, substr);
942 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
943 int matrix_noise_shift = s->matrix_noise_shift[mat];
944 unsigned int dest_ch = s->matrix_out_ch[mat];
945 int32_t mask = MSB_MASK(s->quant_step_size[dest_ch]);
946 int32_t *coeffs = s->matrix_coeff[mat];
947 int index = s->num_primitive_matrices - mat;
948 int index2 = 2 * index + 1;
950 /* TODO: DSPContext? */
952 for (i = 0; i < s->blockpos; i++) {
953 int32_t bypassed_lsb = m->bypassed_lsbs[i][mat];
954 int32_t *samples = m->sample_buffer[i];
957 for (src_ch = 0; src_ch <= maxchan; src_ch++)
958 accum += (int64_t) samples[src_ch] * coeffs[src_ch];
960 if (matrix_noise_shift) {
961 index &= m->access_unit_size_pow2 - 1;
962 accum += m->noise_buffer[index] << (matrix_noise_shift + 7);
966 samples[dest_ch] = ((accum >> 14) & mask) + bypassed_lsb;
971 /** Write the audio data into the output buffer. */
973 static int output_data(MLPDecodeContext *m, unsigned int substr,
974 AVFrame *frame, int *got_frame_ptr)
976 AVCodecContext *avctx = m->avctx;
977 SubStream *s = &m->substream[substr];
978 unsigned int i, out_ch = 0;
982 int is32 = (m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
984 if (m->avctx->channels != s->max_matrix_channel + 1) {
985 av_log(m->avctx, AV_LOG_ERROR, "channel count mismatch\n");
986 return AVERROR_INVALIDDATA;
990 av_log(avctx, AV_LOG_ERROR, "No samples to output.\n");
991 return AVERROR_INVALIDDATA;
994 /* get output buffer */
995 frame->nb_samples = s->blockpos;
996 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
997 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1000 data_32 = (int32_t *)frame->data[0];
1001 data_16 = (int16_t *)frame->data[0];
1003 for (i = 0; i < s->blockpos; i++) {
1004 for (out_ch = 0; out_ch <= s->max_matrix_channel; out_ch++) {
1005 int mat_ch = s->ch_assign[out_ch];
1006 int32_t sample = m->sample_buffer[i][mat_ch]
1007 << s->output_shift[mat_ch];
1008 s->lossless_check_data ^= (sample & 0xffffff) << mat_ch;
1009 if (is32) *data_32++ = sample << 8;
1010 else *data_16++ = sample >> 8;
1019 /** Read an access unit from the stream.
1020 * @return negative on error, 0 if not enough data is present in the input stream,
1021 * otherwise the number of bytes consumed. */
1023 static int read_access_unit(AVCodecContext *avctx, void* data,
1024 int *got_frame_ptr, AVPacket *avpkt)
1026 const uint8_t *buf = avpkt->data;
1027 int buf_size = avpkt->size;
1028 MLPDecodeContext *m = avctx->priv_data;
1030 unsigned int length, substr;
1031 unsigned int substream_start;
1032 unsigned int header_size = 4;
1033 unsigned int substr_header_size = 0;
1034 uint8_t substream_parity_present[MAX_SUBSTREAMS];
1035 uint16_t substream_data_len[MAX_SUBSTREAMS];
1036 uint8_t parity_bits;
1042 length = (AV_RB16(buf) & 0xfff) * 2;
1044 if (length < 4 || length > buf_size)
1045 return AVERROR_INVALIDDATA;
1047 init_get_bits(&gb, (buf + 4), (length - 4) * 8);
1049 m->is_major_sync_unit = 0;
1050 if (show_bits_long(&gb, 31) == (0xf8726fba >> 1)) {
1051 if (read_major_sync(m, &gb) < 0)
1053 m->is_major_sync_unit = 1;
1057 if (!m->params_valid) {
1058 av_log(m->avctx, AV_LOG_WARNING,
1059 "Stream parameters not seen; skipping frame.\n");
1064 substream_start = 0;
1066 for (substr = 0; substr < m->num_substreams; substr++) {
1067 int extraword_present, checkdata_present, end, nonrestart_substr;
1069 extraword_present = get_bits1(&gb);
1070 nonrestart_substr = get_bits1(&gb);
1071 checkdata_present = get_bits1(&gb);
1074 end = get_bits(&gb, 12) * 2;
1076 substr_header_size += 2;
1078 if (extraword_present) {
1079 if (m->avctx->codec_id == AV_CODEC_ID_MLP) {
1080 av_log(m->avctx, AV_LOG_ERROR, "There must be no extraword for MLP.\n");
1084 substr_header_size += 2;
1087 if (!(nonrestart_substr ^ m->is_major_sync_unit)) {
1088 av_log(m->avctx, AV_LOG_ERROR, "Invalid nonrestart_substr.\n");
1092 if (end + header_size + substr_header_size > length) {
1093 av_log(m->avctx, AV_LOG_ERROR,
1094 "Indicated length of substream %d data goes off end of "
1095 "packet.\n", substr);
1096 end = length - header_size - substr_header_size;
1099 if (end < substream_start) {
1100 av_log(avctx, AV_LOG_ERROR,
1101 "Indicated end offset of substream %d data "
1102 "is smaller than calculated start offset.\n",
1107 if (substr > m->max_decoded_substream)
1110 substream_parity_present[substr] = checkdata_present;
1111 substream_data_len[substr] = end - substream_start;
1112 substream_start = end;
1115 parity_bits = ff_mlp_calculate_parity(buf, 4);
1116 parity_bits ^= ff_mlp_calculate_parity(buf + header_size, substr_header_size);
1118 if ((((parity_bits >> 4) ^ parity_bits) & 0xF) != 0xF) {
1119 av_log(avctx, AV_LOG_ERROR, "Parity check failed.\n");
1123 buf += header_size + substr_header_size;
1125 for (substr = 0; substr <= m->max_decoded_substream; substr++) {
1126 SubStream *s = &m->substream[substr];
1127 init_get_bits(&gb, buf, substream_data_len[substr] * 8);
1129 m->matrix_changed = 0;
1130 memset(m->filter_changed, 0, sizeof(m->filter_changed));
1134 if (get_bits1(&gb)) {
1135 if (get_bits1(&gb)) {
1136 /* A restart header should be present. */
1137 if (read_restart_header(m, &gb, buf, substr) < 0)
1139 s->restart_seen = 1;
1142 if (!s->restart_seen)
1144 if (read_decoding_params(m, &gb, substr) < 0)
1148 if (!s->restart_seen)
1151 if ((ret = read_block_data(m, &gb, substr)) < 0)
1154 if (get_bits_count(&gb) >= substream_data_len[substr] * 8)
1155 goto substream_length_mismatch;
1157 } while (!get_bits1(&gb));
1159 skip_bits(&gb, (-get_bits_count(&gb)) & 15);
1161 if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 32) {
1164 if (get_bits(&gb, 16) != 0xD234)
1165 return AVERROR_INVALIDDATA;
1167 shorten_by = get_bits(&gb, 16);
1168 if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD && shorten_by & 0x2000)
1169 s->blockpos -= FFMIN(shorten_by & 0x1FFF, s->blockpos);
1170 else if (m->avctx->codec_id == AV_CODEC_ID_MLP && shorten_by != 0xD234)
1171 return AVERROR_INVALIDDATA;
1173 if (substr == m->max_decoded_substream)
1174 av_log(m->avctx, AV_LOG_INFO, "End of stream indicated.\n");
1177 if (substream_parity_present[substr]) {
1178 uint8_t parity, checksum;
1180 if (substream_data_len[substr] * 8 - get_bits_count(&gb) != 16)
1181 goto substream_length_mismatch;
1183 parity = ff_mlp_calculate_parity(buf, substream_data_len[substr] - 2);
1184 checksum = ff_mlp_checksum8 (buf, substream_data_len[substr] - 2);
1186 if ((get_bits(&gb, 8) ^ parity) != 0xa9 )
1187 av_log(m->avctx, AV_LOG_ERROR, "Substream %d parity check failed.\n", substr);
1188 if ( get_bits(&gb, 8) != checksum)
1189 av_log(m->avctx, AV_LOG_ERROR, "Substream %d checksum failed.\n" , substr);
1192 if (substream_data_len[substr] * 8 != get_bits_count(&gb))
1193 goto substream_length_mismatch;
1196 if (!s->restart_seen)
1197 av_log(m->avctx, AV_LOG_ERROR,
1198 "No restart header present in substream %d.\n", substr);
1200 buf += substream_data_len[substr];
1203 rematrix_channels(m, m->max_decoded_substream);
1205 if ((ret = output_data(m, m->max_decoded_substream, data, got_frame_ptr)) < 0)
1210 substream_length_mismatch:
1211 av_log(m->avctx, AV_LOG_ERROR, "substream %d length mismatch\n", substr);
1212 return AVERROR_INVALIDDATA;
1215 m->params_valid = 0;
1216 return AVERROR_INVALIDDATA;
1219 AVCodec ff_mlp_decoder = {
1221 .type = AVMEDIA_TYPE_AUDIO,
1222 .id = AV_CODEC_ID_MLP,
1223 .priv_data_size = sizeof(MLPDecodeContext),
1224 .init = mlp_decode_init,
1225 .decode = read_access_unit,
1226 .capabilities = CODEC_CAP_DR1,
1227 .long_name = NULL_IF_CONFIG_SMALL("MLP (Meridian Lossless Packing)"),
1230 #if CONFIG_TRUEHD_DECODER
1231 AVCodec ff_truehd_decoder = {
1233 .type = AVMEDIA_TYPE_AUDIO,
1234 .id = AV_CODEC_ID_TRUEHD,
1235 .priv_data_size = sizeof(MLPDecodeContext),
1236 .init = mlp_decode_init,
1237 .decode = read_access_unit,
1238 .capabilities = CODEC_CAP_DR1,
1239 .long_name = NULL_IF_CONFIG_SMALL("TrueHD"),
1241 #endif /* CONFIG_TRUEHD_DECODER */