91efaf36abb166f9ea63a37caf9cad45911d996d
[libav.git] / libavcodec / mlpdec.c
1 /*
2 * MLP decoder
3 * Copyright (c) 2007-2008 Ian Caulfield
4 *
5 * This file is part of Libav.
6 *
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.
11 *
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.
16 *
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
20 */
21
22 /**
23 * @file
24 * MLP decoder
25 */
26
27 #include <stdint.h>
28
29 #include "avcodec.h"
30 #include "dsputil.h"
31 #include "libavutil/intreadwrite.h"
32 #include "get_bits.h"
33 #include "libavutil/crc.h"
34 #include "parser.h"
35 #include "mlp_parser.h"
36 #include "mlp.h"
37
38 /** number of bits used for VLC lookup - longest Huffman code is 9 */
39 #define VLC_BITS 9
40
41 typedef struct SubStream {
42 /// Set if a valid restart header has been read. Otherwise the substream cannot be decoded.
43 uint8_t restart_seen;
44
45 //@{
46 /** restart header data */
47 /// The type of noise to be used in the rematrix stage.
48 uint16_t noise_type;
49
50 /// The index of the first channel coded in this substream.
51 uint8_t min_channel;
52 /// The index of the last channel coded in this substream.
53 uint8_t max_channel;
54 /// The number of channels input into the rematrix stage.
55 uint8_t max_matrix_channel;
56 /// For each channel output by the matrix, the output channel to map it to
57 uint8_t ch_assign[MAX_CHANNELS];
58
59 /// Channel coding parameters for channels in the substream
60 ChannelParams channel_params[MAX_CHANNELS];
61
62 /// The left shift applied to random noise in 0x31ea substreams.
63 uint8_t noise_shift;
64 /// The current seed value for the pseudorandom noise generator(s).
65 uint32_t noisegen_seed;
66
67 /// Set if the substream contains extra info to check the size of VLC blocks.
68 uint8_t data_check_present;
69
70 /// Bitmask of which parameter sets are conveyed in a decoding parameter block.
71 uint8_t param_presence_flags;
72 #define PARAM_BLOCKSIZE (1 << 7)
73 #define PARAM_MATRIX (1 << 6)
74 #define PARAM_OUTSHIFT (1 << 5)
75 #define PARAM_QUANTSTEP (1 << 4)
76 #define PARAM_FIR (1 << 3)
77 #define PARAM_IIR (1 << 2)
78 #define PARAM_HUFFOFFSET (1 << 1)
79 #define PARAM_PRESENCE (1 << 0)
80 //@}
81
82 //@{
83 /** matrix data */
84
85 /// Number of matrices to be applied.
86 uint8_t num_primitive_matrices;
87
88 /// matrix output channel
89 uint8_t matrix_out_ch[MAX_MATRICES];
90
91 /// Whether the LSBs of the matrix output are encoded in the bitstream.
92 uint8_t lsb_bypass[MAX_MATRICES];
93 /// Matrix coefficients, stored as 2.14 fixed point.
94 int32_t matrix_coeff[MAX_MATRICES][MAX_CHANNELS];
95 /// Left shift to apply to noise values in 0x31eb substreams.
96 uint8_t matrix_noise_shift[MAX_MATRICES];
97 //@}
98
99 /// Left shift to apply to Huffman-decoded residuals.
100 uint8_t quant_step_size[MAX_CHANNELS];
101
102 /// number of PCM samples in current audio block
103 uint16_t blocksize;
104 /// Number of PCM samples decoded so far in this frame.
105 uint16_t blockpos;
106
107 /// Left shift to apply to decoded PCM values to get final 24-bit output.
108 int8_t output_shift[MAX_CHANNELS];
109
110 /// Running XOR of all output samples.
111 int32_t lossless_check_data;
112
113 } SubStream;
114
115 typedef struct MLPDecodeContext {
116 AVCodecContext *avctx;
117 AVFrame frame;
118
119 /// Current access unit being read has a major sync.
120 int is_major_sync_unit;
121
122 /// Set if a valid major sync block has been read. Otherwise no decoding is possible.
123 uint8_t params_valid;
124
125 /// Number of substreams contained within this stream.
126 uint8_t num_substreams;
127
128 /// Index of the last substream to decode - further substreams are skipped.
129 uint8_t max_decoded_substream;
130
131 /// number of PCM samples contained in each frame
132 int access_unit_size;
133 /// next power of two above the number of samples in each frame
134 int access_unit_size_pow2;
135
136 SubStream substream[MAX_SUBSTREAMS];
137
138 int matrix_changed;
139 int filter_changed[MAX_CHANNELS][NUM_FILTERS];
140
141 int8_t noise_buffer[MAX_BLOCKSIZE_POW2];
142 int8_t bypassed_lsbs[MAX_BLOCKSIZE][MAX_CHANNELS];
143 int32_t sample_buffer[MAX_BLOCKSIZE][MAX_CHANNELS];
144
145 DSPContext dsp;
146 } MLPDecodeContext;
147
148 static VLC huff_vlc[3];
149
150 /** Initialize static data, constant between all invocations of the codec. */
151
152 static av_cold void init_static(void)
153 {
154 if (!huff_vlc[0].bits) {
155 INIT_VLC_STATIC(&huff_vlc[0], VLC_BITS, 18,
156 &ff_mlp_huffman_tables[0][0][1], 2, 1,
157 &ff_mlp_huffman_tables[0][0][0], 2, 1, 512);
158 INIT_VLC_STATIC(&huff_vlc[1], VLC_BITS, 16,
159 &ff_mlp_huffman_tables[1][0][1], 2, 1,
160 &ff_mlp_huffman_tables[1][0][0], 2, 1, 512);
161 INIT_VLC_STATIC(&huff_vlc[2], VLC_BITS, 15,
162 &ff_mlp_huffman_tables[2][0][1], 2, 1,
163 &ff_mlp_huffman_tables[2][0][0], 2, 1, 512);
164 }
165
166 ff_mlp_init_crc();
167 }
168
169 static inline int32_t calculate_sign_huff(MLPDecodeContext *m,
170 unsigned int substr, unsigned int ch)
171 {
172 SubStream *s = &m->substream[substr];
173 ChannelParams *cp = &s->channel_params[ch];
174 int lsb_bits = cp->huff_lsbs - s->quant_step_size[ch];
175 int sign_shift = lsb_bits + (cp->codebook ? 2 - cp->codebook : -1);
176 int32_t sign_huff_offset = cp->huff_offset;
177
178 if (cp->codebook > 0)
179 sign_huff_offset -= 7 << lsb_bits;
180
181 if (sign_shift >= 0)
182 sign_huff_offset -= 1 << sign_shift;
183
184 return sign_huff_offset;
185 }
186
187 /** Read a sample, consisting of either, both or neither of entropy-coded MSBs
188 * and plain LSBs. */
189
190 static inline int read_huff_channels(MLPDecodeContext *m, GetBitContext *gbp,
191 unsigned int substr, unsigned int pos)
192 {
193 SubStream *s = &m->substream[substr];
194 unsigned int mat, channel;
195
196 for (mat = 0; mat < s->num_primitive_matrices; mat++)
197 if (s->lsb_bypass[mat])
198 m->bypassed_lsbs[pos + s->blockpos][mat] = get_bits1(gbp);
199
200 for (channel = s->min_channel; channel <= s->max_channel; channel++) {
201 ChannelParams *cp = &s->channel_params[channel];
202 int codebook = cp->codebook;
203 int quant_step_size = s->quant_step_size[channel];
204 int lsb_bits = cp->huff_lsbs - quant_step_size;
205 int result = 0;
206
207 if (codebook > 0)
208 result = get_vlc2(gbp, huff_vlc[codebook-1].table,
209 VLC_BITS, (9 + VLC_BITS - 1) / VLC_BITS);
210
211 if (result < 0)
212 return AVERROR_INVALIDDATA;
213
214 if (lsb_bits > 0)
215 result = (result << lsb_bits) + get_bits(gbp, lsb_bits);
216
217 result += cp->sign_huff_offset;
218 result <<= quant_step_size;
219
220 m->sample_buffer[pos + s->blockpos][channel] = result;
221 }
222
223 return 0;
224 }
225
226 static av_cold int mlp_decode_init(AVCodecContext *avctx)
227 {
228 MLPDecodeContext *m = avctx->priv_data;
229 int substr;
230
231 init_static();
232 m->avctx = avctx;
233 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
234 m->substream[substr].lossless_check_data = 0xffffffff;
235 ff_dsputil_init(&m->dsp, avctx);
236
237 avcodec_get_frame_defaults(&m->frame);
238 avctx->coded_frame = &m->frame;
239
240 return 0;
241 }
242
243 /** Read a major sync info header - contains high level information about
244 * the stream - sample rate, channel arrangement etc. Most of this
245 * information is not actually necessary for decoding, only for playback.
246 */
247
248 static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
249 {
250 MLPHeaderInfo mh;
251 int substr, ret;
252
253 if ((ret = ff_mlp_read_major_sync(m->avctx, &mh, gb)) != 0)
254 return ret;
255
256 if (mh.group1_bits == 0) {
257 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown bits per sample\n");
258 return AVERROR_INVALIDDATA;
259 }
260 if (mh.group2_bits > mh.group1_bits) {
261 av_log(m->avctx, AV_LOG_ERROR,
262 "Channel group 2 cannot have more bits per sample than group 1.\n");
263 return AVERROR_INVALIDDATA;
264 }
265
266 if (mh.group2_samplerate && mh.group2_samplerate != mh.group1_samplerate) {
267 av_log(m->avctx, AV_LOG_ERROR,
268 "Channel groups with differing sample rates are not currently supported.\n");
269 return AVERROR_INVALIDDATA;
270 }
271
272 if (mh.group1_samplerate == 0) {
273 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown sampling rate\n");
274 return AVERROR_INVALIDDATA;
275 }
276 if (mh.group1_samplerate > MAX_SAMPLERATE) {
277 av_log(m->avctx, AV_LOG_ERROR,
278 "Sampling rate %d is greater than the supported maximum (%d).\n",
279 mh.group1_samplerate, MAX_SAMPLERATE);
280 return AVERROR_INVALIDDATA;
281 }
282 if (mh.access_unit_size > MAX_BLOCKSIZE) {
283 av_log(m->avctx, AV_LOG_ERROR,
284 "Block size %d is greater than the supported maximum (%d).\n",
285 mh.access_unit_size, MAX_BLOCKSIZE);
286 return AVERROR_INVALIDDATA;
287 }
288 if (mh.access_unit_size_pow2 > MAX_BLOCKSIZE_POW2) {
289 av_log(m->avctx, AV_LOG_ERROR,
290 "Block size pow2 %d is greater than the supported maximum (%d).\n",
291 mh.access_unit_size_pow2, MAX_BLOCKSIZE_POW2);
292 return AVERROR_INVALIDDATA;
293 }
294
295 if (mh.num_substreams == 0)
296 return AVERROR_INVALIDDATA;
297 if (m->avctx->codec_id == CODEC_ID_MLP && mh.num_substreams > 2) {
298 av_log(m->avctx, AV_LOG_ERROR, "MLP only supports up to 2 substreams.\n");
299 return AVERROR_INVALIDDATA;
300 }
301 if (mh.num_substreams > MAX_SUBSTREAMS) {
302 av_log_ask_for_sample(m->avctx,
303 "Number of substreams %d is larger than the maximum supported "
304 "by the decoder.\n", mh.num_substreams);
305 return AVERROR_PATCHWELCOME;
306 }
307
308 m->access_unit_size = mh.access_unit_size;
309 m->access_unit_size_pow2 = mh.access_unit_size_pow2;
310
311 m->num_substreams = mh.num_substreams;
312 m->max_decoded_substream = m->num_substreams - 1;
313
314 m->avctx->sample_rate = mh.group1_samplerate;
315 m->avctx->frame_size = mh.access_unit_size;
316
317 m->avctx->bits_per_raw_sample = mh.group1_bits;
318 if (mh.group1_bits > 16)
319 m->avctx->sample_fmt = AV_SAMPLE_FMT_S32;
320 else
321 m->avctx->sample_fmt = AV_SAMPLE_FMT_S16;
322
323 m->params_valid = 1;
324 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
325 m->substream[substr].restart_seen = 0;
326
327 return 0;
328 }
329
330 /** Read a restart header from a block in a substream. This contains parameters
331 * required to decode the audio that do not change very often. Generally
332 * (always) present only in blocks following a major sync. */
333
334 static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
335 const uint8_t *buf, unsigned int substr)
336 {
337 SubStream *s = &m->substream[substr];
338 unsigned int ch;
339 int sync_word, tmp;
340 uint8_t checksum;
341 uint8_t lossless_check;
342 int start_count = get_bits_count(gbp);
343 const int max_matrix_channel = m->avctx->codec_id == CODEC_ID_MLP
344 ? MAX_MATRIX_CHANNEL_MLP
345 : MAX_MATRIX_CHANNEL_TRUEHD;
346
347 sync_word = get_bits(gbp, 13);
348
349 if (sync_word != 0x31ea >> 1) {
350 av_log(m->avctx, AV_LOG_ERROR,
351 "restart header sync incorrect (got 0x%04x)\n", sync_word);
352 return AVERROR_INVALIDDATA;
353 }
354
355 s->noise_type = get_bits1(gbp);
356
357 if (m->avctx->codec_id == CODEC_ID_MLP && s->noise_type) {
358 av_log(m->avctx, AV_LOG_ERROR, "MLP must have 0x31ea sync word.\n");
359 return AVERROR_INVALIDDATA;
360 }
361
362 skip_bits(gbp, 16); /* Output timestamp */
363
364 s->min_channel = get_bits(gbp, 4);
365 s->max_channel = get_bits(gbp, 4);
366 s->max_matrix_channel = get_bits(gbp, 4);
367
368 if (s->max_matrix_channel > max_matrix_channel) {
369 av_log(m->avctx, AV_LOG_ERROR,
370 "Max matrix channel cannot be greater than %d.\n",
371 max_matrix_channel);
372 return AVERROR_INVALIDDATA;
373 }
374
375 if (s->max_channel != s->max_matrix_channel) {
376 av_log(m->avctx, AV_LOG_ERROR,
377 "Max channel must be equal max matrix channel.\n");
378 return AVERROR_INVALIDDATA;
379 }
380
381 /* This should happen for TrueHD streams with >6 channels and MLP's noise
382 * type. It is not yet known if this is allowed. */
383 if (s->max_channel > MAX_MATRIX_CHANNEL_MLP && !s->noise_type) {
384 av_log_ask_for_sample(m->avctx,
385 "Number of channels %d is larger than the maximum supported "
386 "by the decoder.\n", s->max_channel + 2);
387 return AVERROR_PATCHWELCOME;
388 }
389
390 if (s->min_channel > s->max_channel) {
391 av_log(m->avctx, AV_LOG_ERROR,
392 "Substream min channel cannot be greater than max channel.\n");
393 return AVERROR_INVALIDDATA;
394 }
395
396 if (m->avctx->request_channels > 0
397 && s->max_channel + 1 >= m->avctx->request_channels
398 && substr < m->max_decoded_substream) {
399 av_log(m->avctx, AV_LOG_DEBUG,
400 "Extracting %d channel downmix from substream %d. "
401 "Further substreams will be skipped.\n",
402 s->max_channel + 1, substr);
403 m->max_decoded_substream = substr;
404 }
405
406 s->noise_shift = get_bits(gbp, 4);
407 s->noisegen_seed = get_bits(gbp, 23);
408
409 skip_bits(gbp, 19);
410
411 s->data_check_present = get_bits1(gbp);
412 lossless_check = get_bits(gbp, 8);
413 if (substr == m->max_decoded_substream
414 && s->lossless_check_data != 0xffffffff) {
415 tmp = xor_32_to_8(s->lossless_check_data);
416 if (tmp != lossless_check)
417 av_log(m->avctx, AV_LOG_WARNING,
418 "Lossless check failed - expected %02x, calculated %02x.\n",
419 lossless_check, tmp);
420 }
421
422 skip_bits(gbp, 16);
423
424 memset(s->ch_assign, 0, sizeof(s->ch_assign));
425
426 for (ch = 0; ch <= s->max_matrix_channel; ch++) {
427 int ch_assign = get_bits(gbp, 6);
428 if (ch_assign > s->max_matrix_channel) {
429 av_log_ask_for_sample(m->avctx,
430 "Assignment of matrix channel %d to invalid output channel %d.\n",
431 ch, ch_assign);
432 return AVERROR_PATCHWELCOME;
433 }
434 s->ch_assign[ch_assign] = ch;
435 }
436
437 checksum = ff_mlp_restart_checksum(buf, get_bits_count(gbp) - start_count);
438
439 if (checksum != get_bits(gbp, 8))
440 av_log(m->avctx, AV_LOG_ERROR, "restart header checksum error\n");
441
442 /* Set default decoding parameters. */
443 s->param_presence_flags = 0xff;
444 s->num_primitive_matrices = 0;
445 s->blocksize = 8;
446 s->lossless_check_data = 0;
447
448 memset(s->output_shift , 0, sizeof(s->output_shift ));
449 memset(s->quant_step_size, 0, sizeof(s->quant_step_size));
450
451 for (ch = s->min_channel; ch <= s->max_channel; ch++) {
452 ChannelParams *cp = &s->channel_params[ch];
453 cp->filter_params[FIR].order = 0;
454 cp->filter_params[IIR].order = 0;
455 cp->filter_params[FIR].shift = 0;
456 cp->filter_params[IIR].shift = 0;
457
458 /* Default audio coding is 24-bit raw PCM. */
459 cp->huff_offset = 0;
460 cp->sign_huff_offset = (-1) << 23;
461 cp->codebook = 0;
462 cp->huff_lsbs = 24;
463 }
464
465 if (substr == m->max_decoded_substream)
466 m->avctx->channels = s->max_matrix_channel + 1;
467
468 return 0;
469 }
470
471 /** Read parameters for one of the prediction filters. */
472
473 static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
474 unsigned int substr, unsigned int channel,
475 unsigned int filter)
476 {
477 SubStream *s = &m->substream[substr];
478 FilterParams *fp = &s->channel_params[channel].filter_params[filter];
479 const int max_order = filter ? MAX_IIR_ORDER : MAX_FIR_ORDER;
480 const char fchar = filter ? 'I' : 'F';
481 int i, order;
482
483 // Filter is 0 for FIR, 1 for IIR.
484 assert(filter < 2);
485
486 if (m->filter_changed[channel][filter]++ > 1) {
487 av_log(m->avctx, AV_LOG_ERROR, "Filters may change only once per access unit.\n");
488 return AVERROR_INVALIDDATA;
489 }
490
491 order = get_bits(gbp, 4);
492 if (order > max_order) {
493 av_log(m->avctx, AV_LOG_ERROR,
494 "%cIR filter order %d is greater than maximum %d.\n",
495 fchar, order, max_order);
496 return AVERROR_INVALIDDATA;
497 }
498 fp->order = order;
499
500 if (order > 0) {
501 int32_t *fcoeff = s->channel_params[channel].coeff[filter];
502 int coeff_bits, coeff_shift;
503
504 fp->shift = get_bits(gbp, 4);
505
506 coeff_bits = get_bits(gbp, 5);
507 coeff_shift = get_bits(gbp, 3);
508 if (coeff_bits < 1 || coeff_bits > 16) {
509 av_log(m->avctx, AV_LOG_ERROR,
510 "%cIR filter coeff_bits must be between 1 and 16.\n",
511 fchar);
512 return AVERROR_INVALIDDATA;
513 }
514 if (coeff_bits + coeff_shift > 16) {
515 av_log(m->avctx, AV_LOG_ERROR,
516 "Sum of coeff_bits and coeff_shift for %cIR filter must be 16 or less.\n",
517 fchar);
518 return AVERROR_INVALIDDATA;
519 }
520
521 for (i = 0; i < order; i++)
522 fcoeff[i] = get_sbits(gbp, coeff_bits) << coeff_shift;
523
524 if (get_bits1(gbp)) {
525 int state_bits, state_shift;
526
527 if (filter == FIR) {
528 av_log(m->avctx, AV_LOG_ERROR,
529 "FIR filter has state data specified.\n");
530 return AVERROR_INVALIDDATA;
531 }
532
533 state_bits = get_bits(gbp, 4);
534 state_shift = get_bits(gbp, 4);
535
536 /* TODO: Check validity of state data. */
537
538 for (i = 0; i < order; i++)
539 fp->state[i] = get_sbits(gbp, state_bits) << state_shift;
540 }
541 }
542
543 return 0;
544 }
545
546 /** Read parameters for primitive matrices. */
547
548 static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitContext *gbp)
549 {
550 SubStream *s = &m->substream[substr];
551 unsigned int mat, ch;
552 const int max_primitive_matrices = m->avctx->codec_id == CODEC_ID_MLP
553 ? MAX_MATRICES_MLP
554 : MAX_MATRICES_TRUEHD;
555
556 if (m->matrix_changed++ > 1) {
557 av_log(m->avctx, AV_LOG_ERROR, "Matrices may change only once per access unit.\n");
558 return AVERROR_INVALIDDATA;
559 }
560
561 s->num_primitive_matrices = get_bits(gbp, 4);
562
563 if (s->num_primitive_matrices > max_primitive_matrices) {
564 av_log(m->avctx, AV_LOG_ERROR,
565 "Number of primitive matrices cannot be greater than %d.\n",
566 max_primitive_matrices);
567 return AVERROR_INVALIDDATA;
568 }
569
570 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
571 int frac_bits, max_chan;
572 s->matrix_out_ch[mat] = get_bits(gbp, 4);
573 frac_bits = get_bits(gbp, 4);
574 s->lsb_bypass [mat] = get_bits1(gbp);
575
576 if (s->matrix_out_ch[mat] > s->max_matrix_channel) {
577 av_log(m->avctx, AV_LOG_ERROR,
578 "Invalid channel %d specified as output from matrix.\n",
579 s->matrix_out_ch[mat]);
580 return AVERROR_INVALIDDATA;
581 }
582 if (frac_bits > 14) {
583 av_log(m->avctx, AV_LOG_ERROR,
584 "Too many fractional bits specified.\n");
585 return AVERROR_INVALIDDATA;
586 }
587
588 max_chan = s->max_matrix_channel;
589 if (!s->noise_type)
590 max_chan+=2;
591
592 for (ch = 0; ch <= max_chan; ch++) {
593 int coeff_val = 0;
594 if (get_bits1(gbp))
595 coeff_val = get_sbits(gbp, frac_bits + 2);
596
597 s->matrix_coeff[mat][ch] = coeff_val << (14 - frac_bits);
598 }
599
600 if (s->noise_type)
601 s->matrix_noise_shift[mat] = get_bits(gbp, 4);
602 else
603 s->matrix_noise_shift[mat] = 0;
604 }
605
606 return 0;
607 }
608
609 /** Read channel parameters. */
610
611 static int read_channel_params(MLPDecodeContext *m, unsigned int substr,
612 GetBitContext *gbp, unsigned int ch)
613 {
614 SubStream *s = &m->substream[substr];
615 ChannelParams *cp = &s->channel_params[ch];
616 FilterParams *fir = &cp->filter_params[FIR];
617 FilterParams *iir = &cp->filter_params[IIR];
618 int ret;
619
620 if (s->param_presence_flags & PARAM_FIR)
621 if (get_bits1(gbp))
622 if ((ret = read_filter_params(m, gbp, substr, ch, FIR)) < 0)
623 return ret;
624
625 if (s->param_presence_flags & PARAM_IIR)
626 if (get_bits1(gbp))
627 if ((ret = read_filter_params(m, gbp, substr, ch, IIR)) < 0)
628 return ret;
629
630 if (fir->order + iir->order > 8) {
631 av_log(m->avctx, AV_LOG_ERROR, "Total filter orders too high.\n");
632 return AVERROR_INVALIDDATA;
633 }
634
635 if (fir->order && iir->order &&
636 fir->shift != iir->shift) {
637 av_log(m->avctx, AV_LOG_ERROR,
638 "FIR and IIR filters must use the same precision.\n");
639 return AVERROR_INVALIDDATA;
640 }
641 /* The FIR and IIR filters must have the same precision.
642 * To simplify the filtering code, only the precision of the
643 * FIR filter is considered. If only the IIR filter is employed,
644 * the FIR filter precision is set to that of the IIR filter, so
645 * that the filtering code can use it. */
646 if (!fir->order && iir->order)
647 fir->shift = iir->shift;
648
649 if (s->param_presence_flags & PARAM_HUFFOFFSET)
650 if (get_bits1(gbp))
651 cp->huff_offset = get_sbits(gbp, 15);
652
653 cp->codebook = get_bits(gbp, 2);
654 cp->huff_lsbs = get_bits(gbp, 5);
655
656 if (cp->huff_lsbs > 24) {
657 av_log(m->avctx, AV_LOG_ERROR, "Invalid huff_lsbs.\n");
658 return AVERROR_INVALIDDATA;
659 }
660
661 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
662
663 return 0;
664 }
665
666 /** Read decoding parameters that change more often than those in the restart
667 * header. */
668
669 static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
670 unsigned int substr)
671 {
672 SubStream *s = &m->substream[substr];
673 unsigned int ch;
674 int ret;
675
676 if (s->param_presence_flags & PARAM_PRESENCE)
677 if (get_bits1(gbp))
678 s->param_presence_flags = get_bits(gbp, 8);
679
680 if (s->param_presence_flags & PARAM_BLOCKSIZE)
681 if (get_bits1(gbp)) {
682 s->blocksize = get_bits(gbp, 9);
683 if (s->blocksize < 8 || s->blocksize > m->access_unit_size) {
684 av_log(m->avctx, AV_LOG_ERROR, "Invalid blocksize.");
685 s->blocksize = 0;
686 return AVERROR_INVALIDDATA;
687 }
688 }
689
690 if (s->param_presence_flags & PARAM_MATRIX)
691 if (get_bits1(gbp))
692 if ((ret = read_matrix_params(m, substr, gbp)) < 0)
693 return ret;
694
695 if (s->param_presence_flags & PARAM_OUTSHIFT)
696 if (get_bits1(gbp))
697 for (ch = 0; ch <= s->max_matrix_channel; ch++)
698 s->output_shift[ch] = get_sbits(gbp, 4);
699
700 if (s->param_presence_flags & PARAM_QUANTSTEP)
701 if (get_bits1(gbp))
702 for (ch = 0; ch <= s->max_channel; ch++) {
703 ChannelParams *cp = &s->channel_params[ch];
704
705 s->quant_step_size[ch] = get_bits(gbp, 4);
706
707 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
708 }
709
710 for (ch = s->min_channel; ch <= s->max_channel; ch++)
711 if (get_bits1(gbp))
712 if ((ret = read_channel_params(m, substr, gbp, ch)) < 0)
713 return ret;
714
715 return 0;
716 }
717
718 #define MSB_MASK(bits) (-1u << bits)
719
720 /** Generate PCM samples using the prediction filters and residual values
721 * read from the data stream, and update the filter state. */
722
723 static void filter_channel(MLPDecodeContext *m, unsigned int substr,
724 unsigned int channel)
725 {
726 SubStream *s = &m->substream[substr];
727 const int32_t *fircoeff = s->channel_params[channel].coeff[FIR];
728 int32_t state_buffer[NUM_FILTERS][MAX_BLOCKSIZE + MAX_FIR_ORDER];
729 int32_t *firbuf = state_buffer[FIR] + MAX_BLOCKSIZE;
730 int32_t *iirbuf = state_buffer[IIR] + MAX_BLOCKSIZE;
731 FilterParams *fir = &s->channel_params[channel].filter_params[FIR];
732 FilterParams *iir = &s->channel_params[channel].filter_params[IIR];
733 unsigned int filter_shift = fir->shift;
734 int32_t mask = MSB_MASK(s->quant_step_size[channel]);
735
736 memcpy(firbuf, fir->state, MAX_FIR_ORDER * sizeof(int32_t));
737 memcpy(iirbuf, iir->state, MAX_IIR_ORDER * sizeof(int32_t));
738
739 m->dsp.mlp_filter_channel(firbuf, fircoeff,
740 fir->order, iir->order,
741 filter_shift, mask, s->blocksize,
742 &m->sample_buffer[s->blockpos][channel]);
743
744 memcpy(fir->state, firbuf - s->blocksize, MAX_FIR_ORDER * sizeof(int32_t));
745 memcpy(iir->state, iirbuf - s->blocksize, MAX_IIR_ORDER * sizeof(int32_t));
746 }
747
748 /** Read a block of PCM residual data (or actual if no filtering active). */
749
750 static int read_block_data(MLPDecodeContext *m, GetBitContext *gbp,
751 unsigned int substr)
752 {
753 SubStream *s = &m->substream[substr];
754 unsigned int i, ch, expected_stream_pos = 0;
755 int ret;
756
757 if (s->data_check_present) {
758 expected_stream_pos = get_bits_count(gbp);
759 expected_stream_pos += get_bits(gbp, 16);
760 av_log_ask_for_sample(m->avctx, "This file contains some features "
761 "we have not tested yet.\n");
762 }
763
764 if (s->blockpos + s->blocksize > m->access_unit_size) {
765 av_log(m->avctx, AV_LOG_ERROR, "too many audio samples in frame\n");
766 return AVERROR_INVALIDDATA;
767 }
768
769 memset(&m->bypassed_lsbs[s->blockpos][0], 0,
770 s->blocksize * sizeof(m->bypassed_lsbs[0]));
771
772 for (i = 0; i < s->blocksize; i++)
773 if ((ret = read_huff_channels(m, gbp, substr, i)) < 0)
774 return ret;
775
776 for (ch = s->min_channel; ch <= s->max_channel; ch++)
777 filter_channel(m, substr, ch);
778
779 s->blockpos += s->blocksize;
780
781 if (s->data_check_present) {
782 if (get_bits_count(gbp) != expected_stream_pos)
783 av_log(m->avctx, AV_LOG_ERROR, "block data length mismatch\n");
784 skip_bits(gbp, 8);
785 }
786
787 return 0;
788 }
789
790 /** Data table used for TrueHD noise generation function. */
791
792 static const int8_t noise_table[256] = {
793 30, 51, 22, 54, 3, 7, -4, 38, 14, 55, 46, 81, 22, 58, -3, 2,
794 52, 31, -7, 51, 15, 44, 74, 30, 85, -17, 10, 33, 18, 80, 28, 62,
795 10, 32, 23, 69, 72, 26, 35, 17, 73, 60, 8, 56, 2, 6, -2, -5,
796 51, 4, 11, 50, 66, 76, 21, 44, 33, 47, 1, 26, 64, 48, 57, 40,
797 38, 16, -10, -28, 92, 22, -18, 29, -10, 5, -13, 49, 19, 24, 70, 34,
798 61, 48, 30, 14, -6, 25, 58, 33, 42, 60, 67, 17, 54, 17, 22, 30,
799 67, 44, -9, 50, -11, 43, 40, 32, 59, 82, 13, 49, -14, 55, 60, 36,
800 48, 49, 31, 47, 15, 12, 4, 65, 1, 23, 29, 39, 45, -2, 84, 69,
801 0, 72, 37, 57, 27, 41, -15, -16, 35, 31, 14, 61, 24, 0, 27, 24,
802 16, 41, 55, 34, 53, 9, 56, 12, 25, 29, 53, 5, 20, -20, -8, 20,
803 13, 28, -3, 78, 38, 16, 11, 62, 46, 29, 21, 24, 46, 65, 43, -23,
804 89, 18, 74, 21, 38, -12, 19, 12, -19, 8, 15, 33, 4, 57, 9, -8,
805 36, 35, 26, 28, 7, 83, 63, 79, 75, 11, 3, 87, 37, 47, 34, 40,
806 39, 19, 20, 42, 27, 34, 39, 77, 13, 42, 59, 64, 45, -1, 32, 37,
807 45, -5, 53, -6, 7, 36, 50, 23, 6, 32, 9, -21, 18, 71, 27, 52,
808 -25, 31, 35, 42, -1, 68, 63, 52, 26, 43, 66, 37, 41, 25, 40, 70,
809 };
810
811 /** Noise generation functions.
812 * I'm not sure what these are for - they seem to be some kind of pseudorandom
813 * sequence generators, used to generate noise data which is used when the
814 * channels are rematrixed. I'm not sure if they provide a practical benefit
815 * to compression, or just obfuscate the decoder. Are they for some kind of
816 * dithering? */
817
818 /** Generate two channels of noise, used in the matrix when
819 * restart sync word == 0x31ea. */
820
821 static void generate_2_noise_channels(MLPDecodeContext *m, unsigned int substr)
822 {
823 SubStream *s = &m->substream[substr];
824 unsigned int i;
825 uint32_t seed = s->noisegen_seed;
826 unsigned int maxchan = s->max_matrix_channel;
827
828 for (i = 0; i < s->blockpos; i++) {
829 uint16_t seed_shr7 = seed >> 7;
830 m->sample_buffer[i][maxchan+1] = ((int8_t)(seed >> 15)) << s->noise_shift;
831 m->sample_buffer[i][maxchan+2] = ((int8_t) seed_shr7) << s->noise_shift;
832
833 seed = (seed << 16) ^ seed_shr7 ^ (seed_shr7 << 5);
834 }
835
836 s->noisegen_seed = seed;
837 }
838
839 /** Generate a block of noise, used when restart sync word == 0x31eb. */
840
841 static void fill_noise_buffer(MLPDecodeContext *m, unsigned int substr)
842 {
843 SubStream *s = &m->substream[substr];
844 unsigned int i;
845 uint32_t seed = s->noisegen_seed;
846
847 for (i = 0; i < m->access_unit_size_pow2; i++) {
848 uint8_t seed_shr15 = seed >> 15;
849 m->noise_buffer[i] = noise_table[seed_shr15];
850 seed = (seed << 8) ^ seed_shr15 ^ (seed_shr15 << 5);
851 }
852
853 s->noisegen_seed = seed;
854 }
855
856
857 /** Apply the channel matrices in turn to reconstruct the original audio
858 * samples. */
859
860 static void rematrix_channels(MLPDecodeContext *m, unsigned int substr)
861 {
862 SubStream *s = &m->substream[substr];
863 unsigned int mat, src_ch, i;
864 unsigned int maxchan;
865
866 maxchan = s->max_matrix_channel;
867 if (!s->noise_type) {
868 generate_2_noise_channels(m, substr);
869 maxchan += 2;
870 } else {
871 fill_noise_buffer(m, substr);
872 }
873
874 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
875 int matrix_noise_shift = s->matrix_noise_shift[mat];
876 unsigned int dest_ch = s->matrix_out_ch[mat];
877 int32_t mask = MSB_MASK(s->quant_step_size[dest_ch]);
878 int32_t *coeffs = s->matrix_coeff[mat];
879 int index = s->num_primitive_matrices - mat;
880 int index2 = 2 * index + 1;
881
882 /* TODO: DSPContext? */
883
884 for (i = 0; i < s->blockpos; i++) {
885 int32_t bypassed_lsb = m->bypassed_lsbs[i][mat];
886 int32_t *samples = m->sample_buffer[i];
887 int64_t accum = 0;
888
889 for (src_ch = 0; src_ch <= maxchan; src_ch++)
890 accum += (int64_t) samples[src_ch] * coeffs[src_ch];
891
892 if (matrix_noise_shift) {
893 index &= m->access_unit_size_pow2 - 1;
894 accum += m->noise_buffer[index] << (matrix_noise_shift + 7);
895 index += index2;
896 }
897
898 samples[dest_ch] = ((accum >> 14) & mask) + bypassed_lsb;
899 }
900 }
901 }
902
903 /** Write the audio data into the output buffer. */
904
905 static int output_data(MLPDecodeContext *m, unsigned int substr,
906 void *data, int *got_frame_ptr)
907 {
908 AVCodecContext *avctx = m->avctx;
909 SubStream *s = &m->substream[substr];
910 unsigned int i, out_ch = 0;
911 int32_t *data_32;
912 int16_t *data_16;
913 int ret;
914 int is32 = (m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
915
916 if (m->avctx->channels != s->max_matrix_channel + 1) {
917 av_log(m->avctx, AV_LOG_ERROR, "channel count mismatch\n");
918 return AVERROR_INVALIDDATA;
919 }
920
921 /* get output buffer */
922 m->frame.nb_samples = s->blockpos;
923 if ((ret = avctx->get_buffer(avctx, &m->frame)) < 0) {
924 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
925 return ret;
926 }
927 data_32 = (int32_t *)m->frame.data[0];
928 data_16 = (int16_t *)m->frame.data[0];
929
930 for (i = 0; i < s->blockpos; i++) {
931 for (out_ch = 0; out_ch <= s->max_matrix_channel; out_ch++) {
932 int mat_ch = s->ch_assign[out_ch];
933 int32_t sample = m->sample_buffer[i][mat_ch]
934 << s->output_shift[mat_ch];
935 s->lossless_check_data ^= (sample & 0xffffff) << mat_ch;
936 if (is32) *data_32++ = sample << 8;
937 else *data_16++ = sample >> 8;
938 }
939 }
940
941 *got_frame_ptr = 1;
942 *(AVFrame *)data = m->frame;
943
944 return 0;
945 }
946
947 /** Read an access unit from the stream.
948 * @return negative on error, 0 if not enough data is present in the input stream,
949 * otherwise the number of bytes consumed. */
950
951 static int read_access_unit(AVCodecContext *avctx, void* data,
952 int *got_frame_ptr, AVPacket *avpkt)
953 {
954 const uint8_t *buf = avpkt->data;
955 int buf_size = avpkt->size;
956 MLPDecodeContext *m = avctx->priv_data;
957 GetBitContext gb;
958 unsigned int length, substr;
959 unsigned int substream_start;
960 unsigned int header_size = 4;
961 unsigned int substr_header_size = 0;
962 uint8_t substream_parity_present[MAX_SUBSTREAMS];
963 uint16_t substream_data_len[MAX_SUBSTREAMS];
964 uint8_t parity_bits;
965 int ret;
966
967 if (buf_size < 4)
968 return 0;
969
970 length = (AV_RB16(buf) & 0xfff) * 2;
971
972 if (length < 4 || length > buf_size)
973 return AVERROR_INVALIDDATA;
974
975 init_get_bits(&gb, (buf + 4), (length - 4) * 8);
976
977 m->is_major_sync_unit = 0;
978 if (show_bits_long(&gb, 31) == (0xf8726fba >> 1)) {
979 if (read_major_sync(m, &gb) < 0)
980 goto error;
981 m->is_major_sync_unit = 1;
982 header_size += 28;
983 }
984
985 if (!m->params_valid) {
986 av_log(m->avctx, AV_LOG_WARNING,
987 "Stream parameters not seen; skipping frame.\n");
988 *got_frame_ptr = 0;
989 return length;
990 }
991
992 substream_start = 0;
993
994 for (substr = 0; substr < m->num_substreams; substr++) {
995 int extraword_present, checkdata_present, end, nonrestart_substr;
996
997 extraword_present = get_bits1(&gb);
998 nonrestart_substr = get_bits1(&gb);
999 checkdata_present = get_bits1(&gb);
1000 skip_bits1(&gb);
1001
1002 end = get_bits(&gb, 12) * 2;
1003
1004 substr_header_size += 2;
1005
1006 if (extraword_present) {
1007 if (m->avctx->codec_id == CODEC_ID_MLP) {
1008 av_log(m->avctx, AV_LOG_ERROR, "There must be no extraword for MLP.\n");
1009 goto error;
1010 }
1011 skip_bits(&gb, 16);
1012 substr_header_size += 2;
1013 }
1014
1015 if (!(nonrestart_substr ^ m->is_major_sync_unit)) {
1016 av_log(m->avctx, AV_LOG_ERROR, "Invalid nonrestart_substr.\n");
1017 goto error;
1018 }
1019
1020 if (end + header_size + substr_header_size > length) {
1021 av_log(m->avctx, AV_LOG_ERROR,
1022 "Indicated length of substream %d data goes off end of "
1023 "packet.\n", substr);
1024 end = length - header_size - substr_header_size;
1025 }
1026
1027 if (end < substream_start) {
1028 av_log(avctx, AV_LOG_ERROR,
1029 "Indicated end offset of substream %d data "
1030 "is smaller than calculated start offset.\n",
1031 substr);
1032 goto error;
1033 }
1034
1035 if (substr > m->max_decoded_substream)
1036 continue;
1037
1038 substream_parity_present[substr] = checkdata_present;
1039 substream_data_len[substr] = end - substream_start;
1040 substream_start = end;
1041 }
1042
1043 parity_bits = ff_mlp_calculate_parity(buf, 4);
1044 parity_bits ^= ff_mlp_calculate_parity(buf + header_size, substr_header_size);
1045
1046 if ((((parity_bits >> 4) ^ parity_bits) & 0xF) != 0xF) {
1047 av_log(avctx, AV_LOG_ERROR, "Parity check failed.\n");
1048 goto error;
1049 }
1050
1051 buf += header_size + substr_header_size;
1052
1053 for (substr = 0; substr <= m->max_decoded_substream; substr++) {
1054 SubStream *s = &m->substream[substr];
1055 init_get_bits(&gb, buf, substream_data_len[substr] * 8);
1056
1057 m->matrix_changed = 0;
1058 memset(m->filter_changed, 0, sizeof(m->filter_changed));
1059
1060 s->blockpos = 0;
1061 do {
1062 if (get_bits1(&gb)) {
1063 if (get_bits1(&gb)) {
1064 /* A restart header should be present. */
1065 if (read_restart_header(m, &gb, buf, substr) < 0)
1066 goto next_substr;
1067 s->restart_seen = 1;
1068 }
1069
1070 if (!s->restart_seen)
1071 goto next_substr;
1072 if (read_decoding_params(m, &gb, substr) < 0)
1073 goto next_substr;
1074 }
1075
1076 if (!s->restart_seen)
1077 goto next_substr;
1078
1079 if ((ret = read_block_data(m, &gb, substr)) < 0)
1080 return ret;
1081
1082 if (get_bits_count(&gb) >= substream_data_len[substr] * 8)
1083 goto substream_length_mismatch;
1084
1085 } while (!get_bits1(&gb));
1086
1087 skip_bits(&gb, (-get_bits_count(&gb)) & 15);
1088
1089 if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 32) {
1090 int shorten_by;
1091
1092 if (get_bits(&gb, 16) != 0xD234)
1093 return AVERROR_INVALIDDATA;
1094
1095 shorten_by = get_bits(&gb, 16);
1096 if (m->avctx->codec_id == CODEC_ID_TRUEHD && shorten_by & 0x2000)
1097 s->blockpos -= FFMIN(shorten_by & 0x1FFF, s->blockpos);
1098 else if (m->avctx->codec_id == CODEC_ID_MLP && shorten_by != 0xD234)
1099 return AVERROR_INVALIDDATA;
1100
1101 if (substr == m->max_decoded_substream)
1102 av_log(m->avctx, AV_LOG_INFO, "End of stream indicated.\n");
1103 }
1104
1105 if (substream_parity_present[substr]) {
1106 uint8_t parity, checksum;
1107
1108 if (substream_data_len[substr] * 8 - get_bits_count(&gb) != 16)
1109 goto substream_length_mismatch;
1110
1111 parity = ff_mlp_calculate_parity(buf, substream_data_len[substr] - 2);
1112 checksum = ff_mlp_checksum8 (buf, substream_data_len[substr] - 2);
1113
1114 if ((get_bits(&gb, 8) ^ parity) != 0xa9 )
1115 av_log(m->avctx, AV_LOG_ERROR, "Substream %d parity check failed.\n", substr);
1116 if ( get_bits(&gb, 8) != checksum)
1117 av_log(m->avctx, AV_LOG_ERROR, "Substream %d checksum failed.\n" , substr);
1118 }
1119
1120 if (substream_data_len[substr] * 8 != get_bits_count(&gb))
1121 goto substream_length_mismatch;
1122
1123 next_substr:
1124 if (!s->restart_seen)
1125 av_log(m->avctx, AV_LOG_ERROR,
1126 "No restart header present in substream %d.\n", substr);
1127
1128 buf += substream_data_len[substr];
1129 }
1130
1131 rematrix_channels(m, m->max_decoded_substream);
1132
1133 if ((ret = output_data(m, m->max_decoded_substream, data, got_frame_ptr)) < 0)
1134 return ret;
1135
1136 return length;
1137
1138 substream_length_mismatch:
1139 av_log(m->avctx, AV_LOG_ERROR, "substream %d length mismatch\n", substr);
1140 return AVERROR_INVALIDDATA;
1141
1142 error:
1143 m->params_valid = 0;
1144 return AVERROR_INVALIDDATA;
1145 }
1146
1147 AVCodec ff_mlp_decoder = {
1148 .name = "mlp",
1149 .type = AVMEDIA_TYPE_AUDIO,
1150 .id = CODEC_ID_MLP,
1151 .priv_data_size = sizeof(MLPDecodeContext),
1152 .init = mlp_decode_init,
1153 .decode = read_access_unit,
1154 .capabilities = CODEC_CAP_DR1,
1155 .long_name = NULL_IF_CONFIG_SMALL("MLP (Meridian Lossless Packing)"),
1156 };
1157
1158 #if CONFIG_TRUEHD_DECODER
1159 AVCodec ff_truehd_decoder = {
1160 .name = "truehd",
1161 .type = AVMEDIA_TYPE_AUDIO,
1162 .id = CODEC_ID_TRUEHD,
1163 .priv_data_size = sizeof(MLPDecodeContext),
1164 .init = mlp_decode_init,
1165 .decode = read_access_unit,
1166 .capabilities = CODEC_CAP_DR1,
1167 .long_name = NULL_IF_CONFIG_SMALL("TrueHD"),
1168 };
1169 #endif /* CONFIG_TRUEHD_DECODER */