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 int min_channel, max_channel, max_matrix_channel;
393 const int std_max_matrix_channel = m->avctx->codec_id == AV_CODEC_ID_MLP
394 ? MAX_MATRIX_CHANNEL_MLP
395 : MAX_MATRIX_CHANNEL_TRUEHD;
397 sync_word = get_bits(gbp, 13);
399 if (sync_word != 0x31ea >> 1) {
400 av_log(m->avctx, AV_LOG_ERROR,
401 "restart header sync incorrect (got 0x%04x)\n", sync_word);
402 return AVERROR_INVALIDDATA;
405 s->noise_type = get_bits1(gbp);
407 if (m->avctx->codec_id == AV_CODEC_ID_MLP && s->noise_type) {
408 av_log(m->avctx, AV_LOG_ERROR, "MLP must have 0x31ea sync word.\n");
409 return AVERROR_INVALIDDATA;
412 skip_bits(gbp, 16); /* Output timestamp */
414 min_channel = get_bits(gbp, 4);
415 max_channel = get_bits(gbp, 4);
416 max_matrix_channel = get_bits(gbp, 4);
418 if (max_matrix_channel > std_max_matrix_channel) {
419 av_log(m->avctx, AV_LOG_ERROR,
420 "Max matrix channel cannot be greater than %d.\n",
422 return AVERROR_INVALIDDATA;
425 if (max_channel != max_matrix_channel) {
426 av_log(m->avctx, AV_LOG_ERROR,
427 "Max channel must be equal max matrix channel.\n");
428 return AVERROR_INVALIDDATA;
431 /* This should happen for TrueHD streams with >6 channels and MLP's noise
432 * type. It is not yet known if this is allowed. */
433 if (s->max_channel > MAX_MATRIX_CHANNEL_MLP && !s->noise_type) {
434 avpriv_request_sample(m->avctx,
435 "%d channels (more than the "
436 "maximum supported by the decoder)",
438 return AVERROR_PATCHWELCOME;
441 if (min_channel > max_channel) {
442 av_log(m->avctx, AV_LOG_ERROR,
443 "Substream min channel cannot be greater than max channel.\n");
444 return AVERROR_INVALIDDATA;
447 s->min_channel = min_channel;
448 s->max_channel = max_channel;
449 s->max_matrix_channel = max_matrix_channel;
451 #if FF_API_REQUEST_CHANNELS
452 if (m->avctx->request_channels > 0 &&
453 m->avctx->request_channels <= s->max_channel + 1 &&
454 m->max_decoded_substream > substr) {
455 av_log(m->avctx, AV_LOG_DEBUG,
456 "Extracting %d-channel downmix from substream %d. "
457 "Further substreams will be skipped.\n",
458 s->max_channel + 1, substr);
459 m->max_decoded_substream = substr;
462 if (m->avctx->request_channel_layout == s->ch_layout &&
463 m->max_decoded_substream > substr) {
464 av_log(m->avctx, AV_LOG_DEBUG,
465 "Extracting %d-channel downmix (0x%"PRIx64") from substream %d. "
466 "Further substreams will be skipped.\n",
467 s->max_channel + 1, s->ch_layout, substr);
468 m->max_decoded_substream = substr;
471 s->noise_shift = get_bits(gbp, 4);
472 s->noisegen_seed = get_bits(gbp, 23);
476 s->data_check_present = get_bits1(gbp);
477 lossless_check = get_bits(gbp, 8);
478 if (substr == m->max_decoded_substream
479 && s->lossless_check_data != 0xffffffff) {
480 tmp = xor_32_to_8(s->lossless_check_data);
481 if (tmp != lossless_check)
482 av_log(m->avctx, AV_LOG_WARNING,
483 "Lossless check failed - expected %02x, calculated %02x.\n",
484 lossless_check, tmp);
489 memset(s->ch_assign, 0, sizeof(s->ch_assign));
491 for (ch = 0; ch <= s->max_matrix_channel; ch++) {
492 int ch_assign = get_bits(gbp, 6);
493 if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD) {
494 uint64_t channel = thd_channel_layout_extract_channel(s->ch_layout,
496 ch_assign = av_get_channel_layout_channel_index(s->ch_layout,
499 if (ch_assign > s->max_matrix_channel) {
500 avpriv_request_sample(m->avctx,
501 "Assignment of matrix channel %d to invalid output channel %d",
503 return AVERROR_PATCHWELCOME;
505 s->ch_assign[ch_assign] = ch;
508 checksum = ff_mlp_restart_checksum(buf, get_bits_count(gbp) - start_count);
510 if (checksum != get_bits(gbp, 8))
511 av_log(m->avctx, AV_LOG_ERROR, "restart header checksum error\n");
513 /* Set default decoding parameters. */
514 s->param_presence_flags = 0xff;
515 s->num_primitive_matrices = 0;
517 s->lossless_check_data = 0;
519 memset(s->output_shift , 0, sizeof(s->output_shift ));
520 memset(s->quant_step_size, 0, sizeof(s->quant_step_size));
522 for (ch = s->min_channel; ch <= s->max_channel; ch++) {
523 ChannelParams *cp = &s->channel_params[ch];
524 cp->filter_params[FIR].order = 0;
525 cp->filter_params[IIR].order = 0;
526 cp->filter_params[FIR].shift = 0;
527 cp->filter_params[IIR].shift = 0;
529 /* Default audio coding is 24-bit raw PCM. */
531 cp->sign_huff_offset = (-1) << 23;
536 if (substr == m->max_decoded_substream) {
537 m->avctx->channels = s->max_matrix_channel + 1;
538 m->avctx->channel_layout = s->ch_layout;
544 /** Read parameters for one of the prediction filters. */
546 static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
547 unsigned int substr, unsigned int channel,
550 SubStream *s = &m->substream[substr];
551 FilterParams *fp = &s->channel_params[channel].filter_params[filter];
552 const int max_order = filter ? MAX_IIR_ORDER : MAX_FIR_ORDER;
553 const char fchar = filter ? 'I' : 'F';
556 // Filter is 0 for FIR, 1 for IIR.
559 if (m->filter_changed[channel][filter]++ > 1) {
560 av_log(m->avctx, AV_LOG_ERROR, "Filters may change only once per access unit.\n");
561 return AVERROR_INVALIDDATA;
564 order = get_bits(gbp, 4);
565 if (order > max_order) {
566 av_log(m->avctx, AV_LOG_ERROR,
567 "%cIR filter order %d is greater than maximum %d.\n",
568 fchar, order, max_order);
569 return AVERROR_INVALIDDATA;
574 int32_t *fcoeff = s->channel_params[channel].coeff[filter];
575 int coeff_bits, coeff_shift;
577 fp->shift = get_bits(gbp, 4);
579 coeff_bits = get_bits(gbp, 5);
580 coeff_shift = get_bits(gbp, 3);
581 if (coeff_bits < 1 || coeff_bits > 16) {
582 av_log(m->avctx, AV_LOG_ERROR,
583 "%cIR filter coeff_bits must be between 1 and 16.\n",
585 return AVERROR_INVALIDDATA;
587 if (coeff_bits + coeff_shift > 16) {
588 av_log(m->avctx, AV_LOG_ERROR,
589 "Sum of coeff_bits and coeff_shift for %cIR filter must be 16 or less.\n",
591 return AVERROR_INVALIDDATA;
594 for (i = 0; i < order; i++)
595 fcoeff[i] = get_sbits(gbp, coeff_bits) << coeff_shift;
597 if (get_bits1(gbp)) {
598 int state_bits, state_shift;
601 av_log(m->avctx, AV_LOG_ERROR,
602 "FIR filter has state data specified.\n");
603 return AVERROR_INVALIDDATA;
606 state_bits = get_bits(gbp, 4);
607 state_shift = get_bits(gbp, 4);
609 /* TODO: Check validity of state data. */
611 for (i = 0; i < order; i++)
612 fp->state[i] = get_sbits(gbp, state_bits) << state_shift;
619 /** Read parameters for primitive matrices. */
621 static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitContext *gbp)
623 SubStream *s = &m->substream[substr];
624 unsigned int mat, ch;
625 const int max_primitive_matrices = m->avctx->codec_id == AV_CODEC_ID_MLP
627 : MAX_MATRICES_TRUEHD;
629 if (m->matrix_changed++ > 1) {
630 av_log(m->avctx, AV_LOG_ERROR, "Matrices may change only once per access unit.\n");
631 return AVERROR_INVALIDDATA;
634 s->num_primitive_matrices = get_bits(gbp, 4);
636 if (s->num_primitive_matrices > max_primitive_matrices) {
637 av_log(m->avctx, AV_LOG_ERROR,
638 "Number of primitive matrices cannot be greater than %d.\n",
639 max_primitive_matrices);
640 return AVERROR_INVALIDDATA;
643 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
644 int frac_bits, max_chan;
645 s->matrix_out_ch[mat] = get_bits(gbp, 4);
646 frac_bits = get_bits(gbp, 4);
647 s->lsb_bypass [mat] = get_bits1(gbp);
649 if (s->matrix_out_ch[mat] > s->max_matrix_channel) {
650 av_log(m->avctx, AV_LOG_ERROR,
651 "Invalid channel %d specified as output from matrix.\n",
652 s->matrix_out_ch[mat]);
653 return AVERROR_INVALIDDATA;
655 if (frac_bits > 14) {
656 av_log(m->avctx, AV_LOG_ERROR,
657 "Too many fractional bits specified.\n");
658 return AVERROR_INVALIDDATA;
661 max_chan = s->max_matrix_channel;
665 for (ch = 0; ch <= max_chan; ch++) {
668 coeff_val = get_sbits(gbp, frac_bits + 2);
670 s->matrix_coeff[mat][ch] = coeff_val << (14 - frac_bits);
674 s->matrix_noise_shift[mat] = get_bits(gbp, 4);
676 s->matrix_noise_shift[mat] = 0;
682 /** Read channel parameters. */
684 static int read_channel_params(MLPDecodeContext *m, unsigned int substr,
685 GetBitContext *gbp, unsigned int ch)
687 SubStream *s = &m->substream[substr];
688 ChannelParams *cp = &s->channel_params[ch];
689 FilterParams *fir = &cp->filter_params[FIR];
690 FilterParams *iir = &cp->filter_params[IIR];
693 if (s->param_presence_flags & PARAM_FIR)
695 if ((ret = read_filter_params(m, gbp, substr, ch, FIR)) < 0)
698 if (s->param_presence_flags & PARAM_IIR)
700 if ((ret = read_filter_params(m, gbp, substr, ch, IIR)) < 0)
703 if (fir->order + iir->order > 8) {
704 av_log(m->avctx, AV_LOG_ERROR, "Total filter orders too high.\n");
705 return AVERROR_INVALIDDATA;
708 if (fir->order && iir->order &&
709 fir->shift != iir->shift) {
710 av_log(m->avctx, AV_LOG_ERROR,
711 "FIR and IIR filters must use the same precision.\n");
712 return AVERROR_INVALIDDATA;
714 /* The FIR and IIR filters must have the same precision.
715 * To simplify the filtering code, only the precision of the
716 * FIR filter is considered. If only the IIR filter is employed,
717 * the FIR filter precision is set to that of the IIR filter, so
718 * that the filtering code can use it. */
719 if (!fir->order && iir->order)
720 fir->shift = iir->shift;
722 if (s->param_presence_flags & PARAM_HUFFOFFSET)
724 cp->huff_offset = get_sbits(gbp, 15);
726 cp->codebook = get_bits(gbp, 2);
727 cp->huff_lsbs = get_bits(gbp, 5);
729 if (cp->huff_lsbs > 24) {
730 av_log(m->avctx, AV_LOG_ERROR, "Invalid huff_lsbs.\n");
731 return AVERROR_INVALIDDATA;
734 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
739 /** Read decoding parameters that change more often than those in the restart
742 static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
745 SubStream *s = &m->substream[substr];
749 if (s->param_presence_flags & PARAM_PRESENCE)
751 s->param_presence_flags = get_bits(gbp, 8);
753 if (s->param_presence_flags & PARAM_BLOCKSIZE)
754 if (get_bits1(gbp)) {
755 s->blocksize = get_bits(gbp, 9);
756 if (s->blocksize < 8 || s->blocksize > m->access_unit_size) {
757 av_log(m->avctx, AV_LOG_ERROR, "Invalid blocksize.");
759 return AVERROR_INVALIDDATA;
763 if (s->param_presence_flags & PARAM_MATRIX)
765 if ((ret = read_matrix_params(m, substr, gbp)) < 0)
768 if (s->param_presence_flags & PARAM_OUTSHIFT)
770 for (ch = 0; ch <= s->max_matrix_channel; ch++)
771 s->output_shift[ch] = get_sbits(gbp, 4);
773 if (s->param_presence_flags & PARAM_QUANTSTEP)
775 for (ch = 0; ch <= s->max_channel; ch++) {
776 ChannelParams *cp = &s->channel_params[ch];
778 s->quant_step_size[ch] = get_bits(gbp, 4);
780 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
783 for (ch = s->min_channel; ch <= s->max_channel; ch++)
785 if ((ret = read_channel_params(m, substr, gbp, ch)) < 0)
791 #define MSB_MASK(bits) (-1u << bits)
793 /** Generate PCM samples using the prediction filters and residual values
794 * read from the data stream, and update the filter state. */
796 static void filter_channel(MLPDecodeContext *m, unsigned int substr,
797 unsigned int channel)
799 SubStream *s = &m->substream[substr];
800 const int32_t *fircoeff = s->channel_params[channel].coeff[FIR];
801 int32_t state_buffer[NUM_FILTERS][MAX_BLOCKSIZE + MAX_FIR_ORDER];
802 int32_t *firbuf = state_buffer[FIR] + MAX_BLOCKSIZE;
803 int32_t *iirbuf = state_buffer[IIR] + MAX_BLOCKSIZE;
804 FilterParams *fir = &s->channel_params[channel].filter_params[FIR];
805 FilterParams *iir = &s->channel_params[channel].filter_params[IIR];
806 unsigned int filter_shift = fir->shift;
807 int32_t mask = MSB_MASK(s->quant_step_size[channel]);
809 memcpy(firbuf, fir->state, MAX_FIR_ORDER * sizeof(int32_t));
810 memcpy(iirbuf, iir->state, MAX_IIR_ORDER * sizeof(int32_t));
812 m->dsp.mlp_filter_channel(firbuf, fircoeff,
813 fir->order, iir->order,
814 filter_shift, mask, s->blocksize,
815 &m->sample_buffer[s->blockpos][channel]);
817 memcpy(fir->state, firbuf - s->blocksize, MAX_FIR_ORDER * sizeof(int32_t));
818 memcpy(iir->state, iirbuf - s->blocksize, MAX_IIR_ORDER * sizeof(int32_t));
821 /** Read a block of PCM residual data (or actual if no filtering active). */
823 static int read_block_data(MLPDecodeContext *m, GetBitContext *gbp,
826 SubStream *s = &m->substream[substr];
827 unsigned int i, ch, expected_stream_pos = 0;
830 if (s->data_check_present) {
831 expected_stream_pos = get_bits_count(gbp);
832 expected_stream_pos += get_bits(gbp, 16);
833 avpriv_request_sample(m->avctx,
834 "Substreams with VLC block size check info");
837 if (s->blockpos + s->blocksize > m->access_unit_size) {
838 av_log(m->avctx, AV_LOG_ERROR, "too many audio samples in frame\n");
839 return AVERROR_INVALIDDATA;
842 memset(&m->bypassed_lsbs[s->blockpos][0], 0,
843 s->blocksize * sizeof(m->bypassed_lsbs[0]));
845 for (i = 0; i < s->blocksize; i++)
846 if ((ret = read_huff_channels(m, gbp, substr, i)) < 0)
849 for (ch = s->min_channel; ch <= s->max_channel; ch++)
850 filter_channel(m, substr, ch);
852 s->blockpos += s->blocksize;
854 if (s->data_check_present) {
855 if (get_bits_count(gbp) != expected_stream_pos)
856 av_log(m->avctx, AV_LOG_ERROR, "block data length mismatch\n");
863 /** Data table used for TrueHD noise generation function. */
865 static const int8_t noise_table[256] = {
866 30, 51, 22, 54, 3, 7, -4, 38, 14, 55, 46, 81, 22, 58, -3, 2,
867 52, 31, -7, 51, 15, 44, 74, 30, 85, -17, 10, 33, 18, 80, 28, 62,
868 10, 32, 23, 69, 72, 26, 35, 17, 73, 60, 8, 56, 2, 6, -2, -5,
869 51, 4, 11, 50, 66, 76, 21, 44, 33, 47, 1, 26, 64, 48, 57, 40,
870 38, 16, -10, -28, 92, 22, -18, 29, -10, 5, -13, 49, 19, 24, 70, 34,
871 61, 48, 30, 14, -6, 25, 58, 33, 42, 60, 67, 17, 54, 17, 22, 30,
872 67, 44, -9, 50, -11, 43, 40, 32, 59, 82, 13, 49, -14, 55, 60, 36,
873 48, 49, 31, 47, 15, 12, 4, 65, 1, 23, 29, 39, 45, -2, 84, 69,
874 0, 72, 37, 57, 27, 41, -15, -16, 35, 31, 14, 61, 24, 0, 27, 24,
875 16, 41, 55, 34, 53, 9, 56, 12, 25, 29, 53, 5, 20, -20, -8, 20,
876 13, 28, -3, 78, 38, 16, 11, 62, 46, 29, 21, 24, 46, 65, 43, -23,
877 89, 18, 74, 21, 38, -12, 19, 12, -19, 8, 15, 33, 4, 57, 9, -8,
878 36, 35, 26, 28, 7, 83, 63, 79, 75, 11, 3, 87, 37, 47, 34, 40,
879 39, 19, 20, 42, 27, 34, 39, 77, 13, 42, 59, 64, 45, -1, 32, 37,
880 45, -5, 53, -6, 7, 36, 50, 23, 6, 32, 9, -21, 18, 71, 27, 52,
881 -25, 31, 35, 42, -1, 68, 63, 52, 26, 43, 66, 37, 41, 25, 40, 70,
884 /** Noise generation functions.
885 * I'm not sure what these are for - they seem to be some kind of pseudorandom
886 * sequence generators, used to generate noise data which is used when the
887 * channels are rematrixed. I'm not sure if they provide a practical benefit
888 * to compression, or just obfuscate the decoder. Are they for some kind of
891 /** Generate two channels of noise, used in the matrix when
892 * restart sync word == 0x31ea. */
894 static void generate_2_noise_channels(MLPDecodeContext *m, unsigned int substr)
896 SubStream *s = &m->substream[substr];
898 uint32_t seed = s->noisegen_seed;
899 unsigned int maxchan = s->max_matrix_channel;
901 for (i = 0; i < s->blockpos; i++) {
902 uint16_t seed_shr7 = seed >> 7;
903 m->sample_buffer[i][maxchan+1] = ((int8_t)(seed >> 15)) << s->noise_shift;
904 m->sample_buffer[i][maxchan+2] = ((int8_t) seed_shr7) << s->noise_shift;
906 seed = (seed << 16) ^ seed_shr7 ^ (seed_shr7 << 5);
909 s->noisegen_seed = seed;
912 /** Generate a block of noise, used when restart sync word == 0x31eb. */
914 static void fill_noise_buffer(MLPDecodeContext *m, unsigned int substr)
916 SubStream *s = &m->substream[substr];
918 uint32_t seed = s->noisegen_seed;
920 for (i = 0; i < m->access_unit_size_pow2; i++) {
921 uint8_t seed_shr15 = seed >> 15;
922 m->noise_buffer[i] = noise_table[seed_shr15];
923 seed = (seed << 8) ^ seed_shr15 ^ (seed_shr15 << 5);
926 s->noisegen_seed = seed;
930 /** Apply the channel matrices in turn to reconstruct the original audio
933 static void rematrix_channels(MLPDecodeContext *m, unsigned int substr)
935 SubStream *s = &m->substream[substr];
936 unsigned int mat, src_ch, i;
937 unsigned int maxchan;
939 maxchan = s->max_matrix_channel;
940 if (!s->noise_type) {
941 generate_2_noise_channels(m, substr);
944 fill_noise_buffer(m, substr);
947 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
948 int matrix_noise_shift = s->matrix_noise_shift[mat];
949 unsigned int dest_ch = s->matrix_out_ch[mat];
950 int32_t mask = MSB_MASK(s->quant_step_size[dest_ch]);
951 int32_t *coeffs = s->matrix_coeff[mat];
952 int index = s->num_primitive_matrices - mat;
953 int index2 = 2 * index + 1;
955 /* TODO: DSPContext? */
957 for (i = 0; i < s->blockpos; i++) {
958 int32_t bypassed_lsb = m->bypassed_lsbs[i][mat];
959 int32_t *samples = m->sample_buffer[i];
962 for (src_ch = 0; src_ch <= maxchan; src_ch++)
963 accum += (int64_t) samples[src_ch] * coeffs[src_ch];
965 if (matrix_noise_shift) {
966 index &= m->access_unit_size_pow2 - 1;
967 accum += m->noise_buffer[index] << (matrix_noise_shift + 7);
971 samples[dest_ch] = ((accum >> 14) & mask) + bypassed_lsb;
976 /** Write the audio data into the output buffer. */
978 static int output_data(MLPDecodeContext *m, unsigned int substr,
979 AVFrame *frame, int *got_frame_ptr)
981 AVCodecContext *avctx = m->avctx;
982 SubStream *s = &m->substream[substr];
983 unsigned int i, out_ch = 0;
987 int is32 = (m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
989 if (m->avctx->channels != s->max_matrix_channel + 1) {
990 av_log(m->avctx, AV_LOG_ERROR, "channel count mismatch\n");
991 return AVERROR_INVALIDDATA;
995 av_log(avctx, AV_LOG_ERROR, "No samples to output.\n");
996 return AVERROR_INVALIDDATA;
999 /* get output buffer */
1000 frame->nb_samples = s->blockpos;
1001 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1002 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1005 data_32 = (int32_t *)frame->data[0];
1006 data_16 = (int16_t *)frame->data[0];
1008 for (i = 0; i < s->blockpos; i++) {
1009 for (out_ch = 0; out_ch <= s->max_matrix_channel; out_ch++) {
1010 int mat_ch = s->ch_assign[out_ch];
1011 int32_t sample = m->sample_buffer[i][mat_ch]
1012 << s->output_shift[mat_ch];
1013 s->lossless_check_data ^= (sample & 0xffffff) << mat_ch;
1014 if (is32) *data_32++ = sample << 8;
1015 else *data_16++ = sample >> 8;
1024 /** Read an access unit from the stream.
1025 * @return negative on error, 0 if not enough data is present in the input stream,
1026 * otherwise the number of bytes consumed. */
1028 static int read_access_unit(AVCodecContext *avctx, void* data,
1029 int *got_frame_ptr, AVPacket *avpkt)
1031 const uint8_t *buf = avpkt->data;
1032 int buf_size = avpkt->size;
1033 MLPDecodeContext *m = avctx->priv_data;
1035 unsigned int length, substr;
1036 unsigned int substream_start;
1037 unsigned int header_size = 4;
1038 unsigned int substr_header_size = 0;
1039 uint8_t substream_parity_present[MAX_SUBSTREAMS];
1040 uint16_t substream_data_len[MAX_SUBSTREAMS];
1041 uint8_t parity_bits;
1047 length = (AV_RB16(buf) & 0xfff) * 2;
1049 if (length < 4 || length > buf_size)
1050 return AVERROR_INVALIDDATA;
1052 init_get_bits(&gb, (buf + 4), (length - 4) * 8);
1054 m->is_major_sync_unit = 0;
1055 if (show_bits_long(&gb, 31) == (0xf8726fba >> 1)) {
1056 if (read_major_sync(m, &gb) < 0)
1058 m->is_major_sync_unit = 1;
1062 if (!m->params_valid) {
1063 av_log(m->avctx, AV_LOG_WARNING,
1064 "Stream parameters not seen; skipping frame.\n");
1069 substream_start = 0;
1071 for (substr = 0; substr < m->num_substreams; substr++) {
1072 int extraword_present, checkdata_present, end, nonrestart_substr;
1074 extraword_present = get_bits1(&gb);
1075 nonrestart_substr = get_bits1(&gb);
1076 checkdata_present = get_bits1(&gb);
1079 end = get_bits(&gb, 12) * 2;
1081 substr_header_size += 2;
1083 if (extraword_present) {
1084 if (m->avctx->codec_id == AV_CODEC_ID_MLP) {
1085 av_log(m->avctx, AV_LOG_ERROR, "There must be no extraword for MLP.\n");
1089 substr_header_size += 2;
1092 if (!(nonrestart_substr ^ m->is_major_sync_unit)) {
1093 av_log(m->avctx, AV_LOG_ERROR, "Invalid nonrestart_substr.\n");
1097 if (end + header_size + substr_header_size > length) {
1098 av_log(m->avctx, AV_LOG_ERROR,
1099 "Indicated length of substream %d data goes off end of "
1100 "packet.\n", substr);
1101 end = length - header_size - substr_header_size;
1104 if (end < substream_start) {
1105 av_log(avctx, AV_LOG_ERROR,
1106 "Indicated end offset of substream %d data "
1107 "is smaller than calculated start offset.\n",
1112 if (substr > m->max_decoded_substream)
1115 substream_parity_present[substr] = checkdata_present;
1116 substream_data_len[substr] = end - substream_start;
1117 substream_start = end;
1120 parity_bits = ff_mlp_calculate_parity(buf, 4);
1121 parity_bits ^= ff_mlp_calculate_parity(buf + header_size, substr_header_size);
1123 if ((((parity_bits >> 4) ^ parity_bits) & 0xF) != 0xF) {
1124 av_log(avctx, AV_LOG_ERROR, "Parity check failed.\n");
1128 buf += header_size + substr_header_size;
1130 for (substr = 0; substr <= m->max_decoded_substream; substr++) {
1131 SubStream *s = &m->substream[substr];
1132 init_get_bits(&gb, buf, substream_data_len[substr] * 8);
1134 m->matrix_changed = 0;
1135 memset(m->filter_changed, 0, sizeof(m->filter_changed));
1139 if (get_bits1(&gb)) {
1140 if (get_bits1(&gb)) {
1141 /* A restart header should be present. */
1142 if (read_restart_header(m, &gb, buf, substr) < 0)
1144 s->restart_seen = 1;
1147 if (!s->restart_seen)
1149 if (read_decoding_params(m, &gb, substr) < 0)
1153 if (!s->restart_seen)
1156 if ((ret = read_block_data(m, &gb, substr)) < 0)
1159 if (get_bits_count(&gb) >= substream_data_len[substr] * 8)
1160 goto substream_length_mismatch;
1162 } while (!get_bits1(&gb));
1164 skip_bits(&gb, (-get_bits_count(&gb)) & 15);
1166 if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 32) {
1169 if (get_bits(&gb, 16) != 0xD234)
1170 return AVERROR_INVALIDDATA;
1172 shorten_by = get_bits(&gb, 16);
1173 if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD && shorten_by & 0x2000)
1174 s->blockpos -= FFMIN(shorten_by & 0x1FFF, s->blockpos);
1175 else if (m->avctx->codec_id == AV_CODEC_ID_MLP && shorten_by != 0xD234)
1176 return AVERROR_INVALIDDATA;
1178 if (substr == m->max_decoded_substream)
1179 av_log(m->avctx, AV_LOG_INFO, "End of stream indicated.\n");
1182 if (substream_parity_present[substr]) {
1183 uint8_t parity, checksum;
1185 if (substream_data_len[substr] * 8 - get_bits_count(&gb) != 16)
1186 goto substream_length_mismatch;
1188 parity = ff_mlp_calculate_parity(buf, substream_data_len[substr] - 2);
1189 checksum = ff_mlp_checksum8 (buf, substream_data_len[substr] - 2);
1191 if ((get_bits(&gb, 8) ^ parity) != 0xa9 )
1192 av_log(m->avctx, AV_LOG_ERROR, "Substream %d parity check failed.\n", substr);
1193 if ( get_bits(&gb, 8) != checksum)
1194 av_log(m->avctx, AV_LOG_ERROR, "Substream %d checksum failed.\n" , substr);
1197 if (substream_data_len[substr] * 8 != get_bits_count(&gb))
1198 goto substream_length_mismatch;
1201 if (!s->restart_seen)
1202 av_log(m->avctx, AV_LOG_ERROR,
1203 "No restart header present in substream %d.\n", substr);
1205 buf += substream_data_len[substr];
1208 rematrix_channels(m, m->max_decoded_substream);
1210 if ((ret = output_data(m, m->max_decoded_substream, data, got_frame_ptr)) < 0)
1215 substream_length_mismatch:
1216 av_log(m->avctx, AV_LOG_ERROR, "substream %d length mismatch\n", substr);
1217 return AVERROR_INVALIDDATA;
1220 m->params_valid = 0;
1221 return AVERROR_INVALIDDATA;
1224 AVCodec ff_mlp_decoder = {
1226 .type = AVMEDIA_TYPE_AUDIO,
1227 .id = AV_CODEC_ID_MLP,
1228 .priv_data_size = sizeof(MLPDecodeContext),
1229 .init = mlp_decode_init,
1230 .decode = read_access_unit,
1231 .capabilities = CODEC_CAP_DR1,
1232 .long_name = NULL_IF_CONFIG_SMALL("MLP (Meridian Lossless Packing)"),
1235 #if CONFIG_TRUEHD_DECODER
1236 AVCodec ff_truehd_decoder = {
1238 .type = AVMEDIA_TYPE_AUDIO,
1239 .id = AV_CODEC_ID_TRUEHD,
1240 .priv_data_size = sizeof(MLPDecodeContext),
1241 .init = mlp_decode_init,
1242 .decode = read_access_unit,
1243 .capabilities = CODEC_CAP_DR1,
1244 .long_name = NULL_IF_CONFIG_SMALL("TrueHD"),
1246 #endif /* CONFIG_TRUEHD_DECODER */