projects / libav.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Cosmetics: align
[libav.git] / libavcodec / mlpdec.c
CommitLineData
b517af05
RP		 1/*
2 * MLP decoder
3 * Copyright (c) 2007-2008 Ian Caulfield
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22/**
23 * @file mlpdec.c
24 * MLP decoder
25 */
26
27#include "avcodec.h"
28#include "libavutil/intreadwrite.h"
29#include "bitstream.h"
30#include "libavutil/crc.h"
31#include "parser.h"
32#include "mlp_parser.h"
33
34/** Maximum number of channels that can be decoded. */
35#define MAX_CHANNELS 16
36
37/** Maximum number of matrices used in decoding. Most streams have one matrix
38 * per output channel, but some rematrix a channel (usually 0) more than once.
39 */
40
41#define MAX_MATRICES 15
42
43/** Maximum number of substreams that can be decoded. This could also be set
44 * higher, but again I haven't seen any examples with more than two. */
45#define MAX_SUBSTREAMS 2
46
47/** Maximum sample frequency seen in files. */
48#define MAX_SAMPLERATE 192000
49
50/** The maximum number of audio samples within one access unit. */
51#define MAX_BLOCKSIZE (40 * (MAX_SAMPLERATE / 48000))
52/** The next power of two greater than MAX_BLOCKSIZE. */
53#define MAX_BLOCKSIZE_POW2 (64 * (MAX_SAMPLERATE / 48000))
54
55/** Number of allowed filters. */
56#define NUM_FILTERS 2
57
58/** The maximum number of taps in either the IIR or FIR filter.
59 * I believe MLP actually specifies the maximum order for IIR filters as four,
60 * and that the sum of the orders of both filters must be <= 8. */
61#define MAX_FILTER_ORDER 8
62
63/** Number of bits used for VLC lookup - longest huffman code is 9. */
64#define VLC_BITS 9
65
66
67static const char* sample_message =
68 "Please file a bug report following the instructions at "
69 "http://ffmpeg.mplayerhq.hu/bugreports.html and include "
70 "a sample of this file.";
71
72typedef struct SubStream {
73 //! Set if a valid restart header has been read. Otherwise the substream can not be decoded.
74 uint8_t restart_seen;
75
76 //@{
77 /** restart header data */
78 //! The type of noise to be used in the rematrix stage.
79 uint16_t noise_type;
80
81 //! The index of the first channel coded in this substream.
82 uint8_t min_channel;
83 //! The index of the last channel coded in this substream.
84 uint8_t max_channel;
85 //! The number of channels input into the rematrix stage.
86 uint8_t max_matrix_channel;
87
88 //! The left shift applied to random noise in 0x31ea substreams.
89 uint8_t noise_shift;
90 //! The current seed value for the pseudorandom noise generator(s).
91 uint32_t noisegen_seed;
92
93 //! Set if the substream contains extra info to check the size of VLC blocks.
94 uint8_t data_check_present;
95
96 //! Bitmask of which parameter sets are conveyed in a decoding parameter block.
97 uint8_t param_presence_flags;
98#define PARAM_BLOCKSIZE (1 << 7)
99#define PARAM_MATRIX (1 << 6)
100#define PARAM_OUTSHIFT (1 << 5)
101#define PARAM_QUANTSTEP (1 << 4)
102#define PARAM_FIR (1 << 3)
103#define PARAM_IIR (1 << 2)
104#define PARAM_HUFFOFFSET (1 << 1)
105 //@}
106
107 //@{
108 /** matrix data */
109
110 //! Number of matrices to be applied.
111 uint8_t num_primitive_matrices;
112
113 //! matrix output channel
114 uint8_t matrix_out_ch[MAX_MATRICES];
115
116 //! Whether the LSBs of the matrix output are encoded in the bitstream.
117 uint8_t lsb_bypass[MAX_MATRICES];
118 //! Matrix coefficients, stored as 2.14 fixed point.
119 int32_t matrix_coeff[MAX_MATRICES][MAX_CHANNELS+2];
120 //! Left shift to apply to noise values in 0x31eb substreams.
121 uint8_t matrix_noise_shift[MAX_MATRICES];
122 //@}
123
124 //! Left shift to apply to huffman-decoded residuals.
125 uint8_t quant_step_size[MAX_CHANNELS];
126
127 //! Number of PCM samples in current audio block.
128 uint16_t blocksize;
129 //! Number of PCM samples decoded so far in this frame.
130 uint16_t blockpos;
131
132 //! Left shift to apply to decoded PCM values to get final 24-bit output.
133 int8_t output_shift[MAX_CHANNELS];
134
135 //! Running XOR of all output samples.
136 int32_t lossless_check_data;
137
138} SubStream;
139
140typedef struct MLPDecodeContext {
141 AVCodecContext *avctx;
142
143 //! Set if a valid major sync block has been read. Otherwise no decoding is possible.
144 uint8_t params_valid;
145
146 //! Number of substreams contained within this stream.
147 uint8_t num_substreams;
148
149 //! Index of the last substream to decode - further substreams are skipped.
150 uint8_t max_decoded_substream;
151
152 //! Number of PCM samples contained in each frame.
153 int access_unit_size;
154 //! Next power of two above the number of samples in each frame.
155 int access_unit_size_pow2;
156
157 SubStream substream[MAX_SUBSTREAMS];
158
159 //@{
160 /** filter data */
161#define FIR 0
162#define IIR 1
163 //! Number of taps in filter.
164 uint8_t filter_order[MAX_CHANNELS][NUM_FILTERS];
165 //! Right shift to apply to output of filter.
166 uint8_t filter_shift[MAX_CHANNELS][NUM_FILTERS];
167
168 int32_t filter_coeff[MAX_CHANNELS][NUM_FILTERS][MAX_FILTER_ORDER];
169 int32_t filter_state[MAX_CHANNELS][NUM_FILTERS][MAX_FILTER_ORDER];
170 //@}
171
172 //@{
173 /** sample data coding infomation */
174 //! Offset to apply to residual values.
175 int16_t huff_offset[MAX_CHANNELS];
176 //! Sign/rounding corrected version of huff_offset.
177 int32_t sign_huff_offset[MAX_CHANNELS];
178 //! Which VLC codebook to use to read residuals.
179 uint8_t codebook[MAX_CHANNELS];
180 //! Size of residual suffix not encoded using VLC.
181 uint8_t huff_lsbs[MAX_CHANNELS];
182 //@}
183
184 int8_t noise_buffer[MAX_BLOCKSIZE_POW2];
185 int8_t bypassed_lsbs[MAX_BLOCKSIZE][MAX_CHANNELS];
186 int32_t sample_buffer[MAX_BLOCKSIZE][MAX_CHANNELS+2];
187} MLPDecodeContext;
188
189/** Tables defining the huffman codes.
190 * There are three entropy coding methods used in MLP (four if you count
191 * "none" as a method). These use the same sequences for codes starting with
192 * 00 or 01, but have different codes starting with 1. */
193
194static const uint8_t huffman_tables[3][18][2] = {
195 { /* huffman table 0, -7 - +10 */
196 {0x01, 9}, {0x01, 8}, {0x01, 7}, {0x01, 6}, {0x01, 5}, {0x01, 4}, {0x01, 3},
197 {0x04, 3}, {0x05, 3}, {0x06, 3}, {0x07, 3},
198 {0x03, 3}, {0x05, 4}, {0x09, 5}, {0x11, 6}, {0x21, 7}, {0x41, 8}, {0x81, 9},
199 }, { /* huffman table 1, -7 - +8 */
200 {0x01, 9}, {0x01, 8}, {0x01, 7}, {0x01, 6}, {0x01, 5}, {0x01, 4}, {0x01, 3},
201 {0x02, 2}, {0x03, 2},
202 {0x03, 3}, {0x05, 4}, {0x09, 5}, {0x11, 6}, {0x21, 7}, {0x41, 8}, {0x81, 9},
203 }, { /* huffman table 2, -7 - +7 */
204 {0x01, 9}, {0x01, 8}, {0x01, 7}, {0x01, 6}, {0x01, 5}, {0x01, 4}, {0x01, 3},
205 {0x01, 1},
206 {0x03, 3}, {0x05, 4}, {0x09, 5}, {0x11, 6}, {0x21, 7}, {0x41, 8}, {0x81, 9},
207 }
208};
209
210static VLC huff_vlc[3];
211
212static int crc_init = 0;
213static AVCRC crc_63[1024];
214static AVCRC crc_1D[1024];
215
216
217/** Initialize static data, constant between all invocations of the codec. */
218
219static av_cold void init_static()
220{
221 INIT_VLC_STATIC(&huff_vlc[0], VLC_BITS, 18,
222 &huffman_tables[0][0][1], 2, 1,
223 &huffman_tables[0][0][0], 2, 1, 512);
224 INIT_VLC_STATIC(&huff_vlc[1], VLC_BITS, 16,
225 &huffman_tables[1][0][1], 2, 1,
226 &huffman_tables[1][0][0], 2, 1, 512);
227 INIT_VLC_STATIC(&huff_vlc[2], VLC_BITS, 15,
228 &huffman_tables[2][0][1], 2, 1,
229 &huffman_tables[2][0][0], 2, 1, 512);
230
231 if (!crc_init) {
232 av_crc_init(crc_63, 0, 8, 0x63, sizeof(crc_63));
233 av_crc_init(crc_1D, 0, 8, 0x1D, sizeof(crc_1D));
234 crc_init = 1;
235 }
236}
237
238
239/** MLP uses checksums that seem to be based on the standard CRC algorithm,
240 * but not (in implementation terms, the table lookup and XOR are reversed).
241 * We can implement this behavior using a standard av_crc on all but the
242 * last element, then XOR that with the last element. */
243
244static uint8_t mlp_checksum8(const uint8_t *buf, unsigned int buf_size)
245{
246 uint8_t checksum = av_crc(crc_63, 0x3c, buf, buf_size - 1); // crc_63[0xa2] == 0x3c
247 checksum ^= buf[buf_size-1];
248 return checksum;
249}
250
251/** Calculate an 8-bit checksum over a restart header -- a non-multiple-of-8
252 * number of bits, starting two bits into the first byte of buf. */
253
254static uint8_t mlp_restart_checksum(const uint8_t *buf, unsigned int bit_size)
255{
256 int i;
257 int num_bytes = (bit_size + 2) / 8;
258
259 int crc = crc_1D[buf[0] & 0x3f];
260 crc = av_crc(crc_1D, crc, buf + 1, num_bytes - 2);
261 crc ^= buf[num_bytes - 1];
262
263 for (i = 0; i < ((bit_size + 2) & 7); i++) {
264 crc <<= 1;
265 if (crc & 0x100)
266 crc ^= 0x11D;
267 crc ^= (buf[num_bytes] >> (7 - i)) & 1;
268 }
269
270 return crc;
271}
272
273static inline int32_t calculate_sign_huff(MLPDecodeContext *m,
274 unsigned int substr, unsigned int ch)
275{
276 SubStream *s = &m->substream[substr];
277 int lsb_bits = m->huff_lsbs[ch] - s->quant_step_size[ch];
278 int sign_shift = lsb_bits + (m->codebook[ch] ? 2 - m->codebook[ch] : -1);
279 int32_t sign_huff_offset = m->huff_offset[ch];
280
281 if (m->codebook[ch] > 0)
282 sign_huff_offset -= 7 << lsb_bits;
283
284 if (sign_shift >= 0)
285 sign_huff_offset -= 1 << sign_shift;
286
287 return sign_huff_offset;
288}
289
290/** Read a sample, consisting of either, both or neither of entropy-coded MSBs
291 * and plain LSBs. */
292
293static inline int read_huff_channels(MLPDecodeContext *m, GetBitContext *gbp,
294 unsigned int substr, unsigned int pos)
295{
296 SubStream *s = &m->substream[substr];
297 unsigned int mat, channel;
298
299 for (mat = 0; mat < s->num_primitive_matrices; mat++)
300 if (s->lsb_bypass[mat])
301 m->bypassed_lsbs[pos + s->blockpos][mat] = get_bits1(gbp);
302
303 for (channel = s->min_channel; channel <= s->max_channel; channel++) {
304 int codebook = m->codebook[channel];
305 int quant_step_size = s->quant_step_size[channel];
306 int lsb_bits = m->huff_lsbs[channel] - quant_step_size;
307 int result = 0;
308
309 if (codebook > 0)
310 result = get_vlc2(gbp, huff_vlc[codebook-1].table,
311 VLC_BITS, (9 + VLC_BITS - 1) / VLC_BITS);
312
313 if (result < 0)
314 return -1;
315
316 if (lsb_bits > 0)
317 result = (result << lsb_bits) + get_bits(gbp, lsb_bits);
318
319 result += m->sign_huff_offset[channel];
320 result <<= quant_step_size;
321
322 m->sample_buffer[pos + s->blockpos][channel] = result;
323 }
324
325 return 0;
326}
327
328static av_cold int mlp_decode_init(AVCodecContext *avctx)
329{
330 MLPDecodeContext *m = avctx->priv_data;
331 int substr;
332
333 init_static();
334 m->avctx = avctx;
335 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
336 m->substream[substr].lossless_check_data = 0xffffffff;
337 return 0;
338}
339
340/** Read a major sync info header - contains high level information about
341 * the stream - sample rate, channel arrangement etc. Most of this
342 * information is not actually necessary for decoding, only for playback.
343 */
344
345static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
346{
347 MLPHeaderInfo mh;
348 int substr;
349
350 if (ff_mlp_read_major_sync(m->avctx, &mh, gb) != 0)
351 return -1;
352
353 if (mh.group1_bits == 0) {
354 av_log(m->avctx, AV_LOG_ERROR, "Invalid/unknown bits per sample\n");
355 return -1;
356 }
357 if (mh.group2_bits > mh.group1_bits) {
358 av_log(m->avctx, AV_LOG_ERROR,
359 "Channel group 2 cannot have more bits per sample than group 1\n");
360 return -1;
361 }
362
363 if (mh.group2_samplerate && mh.group2_samplerate != mh.group1_samplerate) {
364 av_log(m->avctx, AV_LOG_ERROR,
365 "Channel groups with differing sample rates not currently supported\n");
366 return -1;
367 }
368
369 if (mh.group1_samplerate == 0) {
370 av_log(m->avctx, AV_LOG_ERROR, "Invalid/unknown sampling rate\n");
371 return -1;
372 }
373 if (mh.group1_samplerate > MAX_SAMPLERATE) {
374 av_log(m->avctx, AV_LOG_ERROR,
375 "Sampling rate %d is greater than maximum supported (%d)\n",
376 mh.group1_samplerate, MAX_SAMPLERATE);
377 return -1;
378 }
379 if (mh.access_unit_size > MAX_BLOCKSIZE) {
380 av_log(m->avctx, AV_LOG_ERROR,
381 "Block size %d is greater than maximum supported (%d)\n",
382 mh.access_unit_size, MAX_BLOCKSIZE);
383 return -1;
384 }
385 if (mh.access_unit_size_pow2 > MAX_BLOCKSIZE_POW2) {
386 av_log(m->avctx, AV_LOG_ERROR,
387 "Block size pow2 %d is greater than maximum supported (%d)\n",
388 mh.access_unit_size_pow2, MAX_BLOCKSIZE_POW2);
389 return -1;
390 }
391
392 if (mh.num_substreams == 0)
393 return -1;
394 if (mh.num_substreams > MAX_SUBSTREAMS) {
395 av_log(m->avctx, AV_LOG_ERROR,
396 "Number of substreams %d is more than maximum supported by "
397 "decoder. %s\n", mh.num_substreams, sample_message);
398 return -1;
399 }
400
401 m->access_unit_size = mh.access_unit_size;
402 m->access_unit_size_pow2 = mh.access_unit_size_pow2;
403
404 m->num_substreams = mh.num_substreams;
405 m->max_decoded_substream = m->num_substreams - 1;
406
407 m->avctx->sample_rate = mh.group1_samplerate;
408 m->avctx->frame_size = mh.access_unit_size;
409
410#ifdef CONFIG_AUDIO_NONSHORT
411 m->avctx->bits_per_sample = mh.group1_bits;
412 if (mh.group1_bits > 16) {
413 m->avctx->sample_fmt = SAMPLE_FMT_S32;
414 }
415#endif
416
417 m->params_valid = 1;
418 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
419 m->substream[substr].restart_seen = 0;
420
421 return 0;
422}
423
424/** Read a restart header from a block in a substream. This contains parameters
425 * required to decode the audio that do not change very often. Generally
426 * (always) present only in blocks following a major sync. */
427
428static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
429 const uint8_t *buf, unsigned int substr)
430{
431 SubStream *s = &m->substream[substr];
432 unsigned int ch;
433 int sync_word, tmp;
434 uint8_t checksum;
435 uint8_t lossless_check;
436 int start_count = get_bits_count(gbp);
437
438 sync_word = get_bits(gbp, 13);
439
440 if (sync_word != 0x31ea >> 1) {
441 av_log(m->avctx, AV_LOG_ERROR,
442 "Restart header sync incorrect (got 0x%04x)\n", sync_word);
443 return -1;
444 }
445 s->noise_type = get_bits1(gbp);
446
447 skip_bits(gbp, 16); /* Output timestamp */
448
449 s->min_channel = get_bits(gbp, 4);
450 s->max_channel = get_bits(gbp, 4);
451 s->max_matrix_channel = get_bits(gbp, 4);
452
453 if (s->min_channel > s->max_channel) {
454 av_log(m->avctx, AV_LOG_ERROR,
455 "Substream min channel cannot be greater than max channel.\n");
456 return -1;
457 }
458
459 if (m->avctx->request_channels > 0
460 && s->max_channel + 1 >= m->avctx->request_channels
461 && substr < m->max_decoded_substream) {
462 av_log(m->avctx, AV_LOG_INFO,
463 "Extracting %d channel downmix from substream %d. "
464 "Further substreams will be skipped.\n",
465 s->max_channel + 1, substr);
466 m->max_decoded_substream = substr;
467 }
468
469 s->noise_shift = get_bits(gbp, 4);
470 s->noisegen_seed = get_bits(gbp, 23);
471
472 skip_bits(gbp, 19);
473
474 s->data_check_present = get_bits1(gbp);
475 lossless_check = get_bits(gbp, 8);
476 if (substr == m->max_decoded_substream
477 && s->lossless_check_data != 0xffffffff) {
478 tmp = s->lossless_check_data;
479 tmp ^= tmp >> 16;
480 tmp ^= tmp >> 8;
481 tmp &= 0xff;
482 if (tmp != lossless_check)
483 av_log(m->avctx, AV_LOG_WARNING,
484 "Lossless check failed - expected %02x, calculated %02x\n",
485 lossless_check, tmp);
486 else
487 dprintf(m->avctx, "Lossless check passed for substream %d (%x)\n",
488 substr, tmp);
489 }
490
491 skip_bits(gbp, 16);
492
493 for (ch = 0; ch <= s->max_matrix_channel; ch++) {
494 int ch_assign = get_bits(gbp, 6);
495 dprintf(m->avctx, "ch_assign[%d][%d] = %d\n", substr, ch,
496 ch_assign);
497 if (ch_assign != ch) {
498 av_log(m->avctx, AV_LOG_ERROR,
499 "Non 1:1 channel assignments are used in this stream. %s\n",
500 sample_message);
501 return -1;
502 }
503 }
504
505 checksum = mlp_restart_checksum(buf, get_bits_count(gbp) - start_count);
506
507 if (checksum != get_bits(gbp, 8))
508 av_log(m->avctx, AV_LOG_ERROR, "Restart header checksum error\n");
509
510 /* Set default decoding parameters */
511 s->param_presence_flags = 0xff;
512 s->num_primitive_matrices = 0;
513 s->blocksize = 8;
514 s->lossless_check_data = 0;
515
516 memset(s->output_shift , 0, sizeof(s->output_shift ));
517 memset(s->quant_step_size, 0, sizeof(s->quant_step_size));
518
519 for (ch = s->min_channel; ch <= s->max_channel; ch++) {
520 m->filter_order[ch][FIR] = 0;
521 m->filter_order[ch][IIR] = 0;
522 m->filter_shift[ch][FIR] = 0;
523 m->filter_shift[ch][IIR] = 0;
524
525 /* Default audio coding is 24-bit raw PCM */
526 m->huff_offset [ch] = 0;
527 m->sign_huff_offset[ch] = (-1) << 23;
528 m->codebook [ch] = 0;
529 m->huff_lsbs [ch] = 24;
530 }
531
532 if (substr == m->max_decoded_substream) {
533 m->avctx->channels = s->max_channel + 1;
534 }
535
536 return 0;
537}
538
539/** Read parameters for one of the prediction filters. */
540
541static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
542 unsigned int channel, unsigned int filter)
543{
544 const char fchar = filter ? 'I' : 'F';
545 int i, order;
546
547 // filter is 0 for FIR, 1 for IIR
548 assert(filter < 2);
549
550 order = get_bits(gbp, 4);
551 if (order > MAX_FILTER_ORDER) {
552 av_log(m->avctx, AV_LOG_ERROR,
553 "%cIR filter order %d is greater than maximum %d\n",
554 fchar, order, MAX_FILTER_ORDER);
555 return -1;
556 }
557 m->filter_order[channel][filter] = order;
558
559 if (order > 0) {
560 int coeff_bits, coeff_shift;
561
562 m->filter_shift[channel][filter] = get_bits(gbp, 4);
563
564 coeff_bits = get_bits(gbp, 5);
565 coeff_shift = get_bits(gbp, 3);
566 if (coeff_bits < 1 || coeff_bits > 16) {
567 av_log(m->avctx, AV_LOG_ERROR,
568 "%cIR filter coeff_bits must be between 1 and 16\n",
569 fchar);
570 return -1;
571 }
572 if (coeff_bits + coeff_shift > 16) {
573 av_log(m->avctx, AV_LOG_ERROR,
574 "Sum of coeff_bits and coeff_shift for %cIR filter must be 16 or less\n",
575 fchar);
576 return -1;
577 }
578
579 for (i = 0; i < order; i++)
580 m->filter_coeff[channel][filter][i] =
581 get_sbits(gbp, coeff_bits) << coeff_shift;
582
583 if (get_bits1(gbp)) {
584 int state_bits, state_shift;
585
586 if (filter == FIR) {
587 av_log(m->avctx, AV_LOG_ERROR,
588 "FIR filter has state data specified\n");
589 return -1;
590 }
591
592 state_bits = get_bits(gbp, 4);
593 state_shift = get_bits(gbp, 4);
594
595 /* TODO: check validity of state data */
596
597 for (i = 0; i < order; i++)
598 m->filter_state[channel][filter][i] =
599 get_sbits(gbp, state_bits) << state_shift;
600 }
601 }
602
603 return 0;
604}
605
606/** Read decoding parameters that change more often than those in the restart
607 * header. */
608
609static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
610 unsigned int substr)
611{
612 SubStream *s = &m->substream[substr];
613 unsigned int mat, ch;
614
615 if (get_bits1(gbp))
616 s->param_presence_flags = get_bits(gbp, 8);
617
618 if (s->param_presence_flags & PARAM_BLOCKSIZE)
619 if (get_bits1(gbp)) {
620 s->blocksize = get_bits(gbp, 9);
621 if (s->blocksize > MAX_BLOCKSIZE) {
622 av_log(m->avctx, AV_LOG_ERROR, "Block size too large\n");
623 s->blocksize = 0;
624 return -1;
625 }
626 }
627
628 if (s->param_presence_flags & PARAM_MATRIX)
629 if (get_bits1(gbp)) {
630 s->num_primitive_matrices = get_bits(gbp, 4);
631
632 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
633 int frac_bits, max_chan;
634 s->matrix_out_ch[mat] = get_bits(gbp, 4);
635 frac_bits = get_bits(gbp, 4);
636 s->lsb_bypass [mat] = get_bits1(gbp);
637
638 if (s->matrix_out_ch[mat] > s->max_channel) {
639 av_log(m->avctx, AV_LOG_ERROR,
640 "Invalid channel %d specified as output from matrix\n",
641 s->matrix_out_ch[mat]);
642 return -1;
643 }
644 if (frac_bits > 14) {
645 av_log(m->avctx, AV_LOG_ERROR,
646 "Too many fractional bits specified\n");
647 return -1;
648 }
649
650 max_chan = s->max_matrix_channel;
651 if (!s->noise_type)
652 max_chan+=2;
653
654 for (ch = 0; ch <= max_chan; ch++) {
655 int coeff_val = 0;
656 if (get_bits1(gbp))
657 coeff_val = get_sbits(gbp, frac_bits + 2);
658
659 s->matrix_coeff[mat][ch] = coeff_val << (14 - frac_bits);
660 }
661
662 if (s->noise_type)
663 s->matrix_noise_shift[mat] = get_bits(gbp, 4);
664 else
665 s->matrix_noise_shift[mat] = 0;
666 }
667 }
668
669 if (s->param_presence_flags & PARAM_OUTSHIFT)
670 if (get_bits1(gbp))
671 for (ch = 0; ch <= s->max_matrix_channel; ch++) {
672 s->output_shift[ch] = get_bits(gbp, 4);
673 dprintf(m->avctx, "output shift[%d] = %d\n",
674 ch, s->output_shift[ch]);
675 /* TODO: validate */
676 }
677
678 if (s->param_presence_flags & PARAM_QUANTSTEP)
679 if (get_bits1(gbp))
680 for (ch = 0; ch <= s->max_channel; ch++) {
681 s->quant_step_size[ch] = get_bits(gbp, 4);
682 /* TODO: validate */
683
684 m->sign_huff_offset[ch] = calculate_sign_huff(m, substr, ch);
685 }
686
687 for (ch = s->min_channel; ch <= s->max_channel; ch++)
688 if (get_bits1(gbp)) {
689 if (s->param_presence_flags & PARAM_FIR)
690 if (get_bits1(gbp))
691 if (read_filter_params(m, gbp, ch, FIR) < 0)
692 return -1;
693
694 if (s->param_presence_flags & PARAM_IIR)
695 if (get_bits1(gbp))
696 if (read_filter_params(m, gbp, ch, IIR) < 0)
697 return -1;
698
699 if (m->filter_order[ch][FIR] && m->filter_order[ch][IIR] &&
700 m->filter_shift[ch][FIR] != m->filter_shift[ch][IIR]) {
701 av_log(m->avctx, AV_LOG_ERROR,
702 "FIR and IIR filters must use same precision\n");
703 return -1;
704 }
705 /* The FIR and IIR filters must have the same precision.
706 * To simplify the filtering code, only the precision of the
707 * FIR filter is considered. If only the IIR filter is employed,
708 * the FIR filter precision is set to that of the IIR filter, so
709 * that the filtering code can use it. */
710 if (!m->filter_order[ch][FIR] && m->filter_order[ch][IIR])
711 m->filter_shift[ch][FIR] = m->filter_shift[ch][IIR];
712
713 if (s->param_presence_flags & PARAM_HUFFOFFSET)
714 if (get_bits1(gbp))
715 m->huff_offset[ch] = get_sbits(gbp, 15);
716
717 m->codebook [ch] = get_bits(gbp, 2);
718 m->huff_lsbs[ch] = get_bits(gbp, 5);
719
720 m->sign_huff_offset[ch] = calculate_sign_huff(m, substr, ch);
721
722 /* TODO: validate */
723 }
724
725 return 0;
726}
727
728#define MSB_MASK(bits) (-1u << bits)
729
730/** Generate PCM samples using the prediction filters and residual values
731 * read from the data stream, and update the filter state. */
732
733static void filter_channel(MLPDecodeContext *m, unsigned int substr,
734 unsigned int channel)
735{
736 SubStream *s = &m->substream[substr];
737 int32_t filter_state_buffer[NUM_FILTERS][MAX_BLOCKSIZE + MAX_FILTER_ORDER];
738 unsigned int filter_shift = m->filter_shift[channel][FIR];
739 int32_t mask = MSB_MASK(s->quant_step_size[channel]);
740 int index = MAX_BLOCKSIZE;
741 int j, i;
742
743 for (j = 0; j < NUM_FILTERS; j++) {
744 memcpy(& filter_state_buffer [j][MAX_BLOCKSIZE],
745 &m->filter_state[channel][j][0],
746 MAX_FILTER_ORDER * sizeof(int32_t));
747 }
748
749 for (i = 0; i < s->blocksize; i++) {
750 int32_t residual = m->sample_buffer[i + s->blockpos][channel];
751 unsigned int order;
752 int64_t accum = 0;
753 int32_t result;
754
755 /* TODO: Move this code to DSPContext? */
756
757 for (j = 0; j < NUM_FILTERS; j++)
758 for (order = 0; order < m->filter_order[channel][j]; order++)
759 accum += (int64_t)filter_state_buffer[j][index + order] *
760 m->filter_coeff[channel][j][order];
761
762 accum = accum >> filter_shift;
763 result = (accum + residual) & mask;
764
765 --index;
766
767 filter_state_buffer[FIR][index] = result;
768 filter_state_buffer[IIR][index] = result - accum;
769
770 m->sample_buffer[i + s->blockpos][channel] = result;
771 }
772
773 for (j = 0; j < NUM_FILTERS; j++) {
774 memcpy(&m->filter_state[channel][j][0],
775 & filter_state_buffer [j][index],
776 MAX_FILTER_ORDER * sizeof(int32_t));
777 }
778}
779
780/** Read a block of PCM residual data (or actual if no filtering active). */
781
782static int read_block_data(MLPDecodeContext *m, GetBitContext *gbp,
783 unsigned int substr)
784{
785 SubStream *s = &m->substream[substr];
786 unsigned int i, ch, expected_stream_pos = 0;
787
788 if (s->data_check_present) {
789 expected_stream_pos = get_bits_count(gbp);
790 expected_stream_pos += get_bits(gbp, 16);
791 av_log(m->avctx, AV_LOG_WARNING, "This file contains some features "
792 "we have not tested yet. %s\n", sample_message);
793 }
794
795 if (s->blockpos + s->blocksize > m->access_unit_size) {
796 av_log(m->avctx, AV_LOG_ERROR, "Too many audio samples in frame\n");
797 return -1;
798 }
799
800 memset(&m->bypassed_lsbs[s->blockpos][0], 0,
801 s->blocksize * sizeof(m->bypassed_lsbs[0]));
802
803 for (i = 0; i < s->blocksize; i++) {
804 if (read_huff_channels(m, gbp, substr, i) < 0)
805 return -1;
806 }
807
808 for (ch = s->min_channel; ch <= s->max_channel; ch++) {
809 filter_channel(m, substr, ch);
810 }
811
812 s->blockpos += s->blocksize;
813
814 if (s->data_check_present) {
815 if (get_bits_count(gbp) != expected_stream_pos)
816 av_log(m->avctx, AV_LOG_ERROR, "Block data length mismatch\n");
817 skip_bits(gbp, 8);
818 }
819
820 return 0;
821}
822
823/** Data table used for TrueHD noise generation function */
824
825static const int8_t noise_table[256] = {
826 30, 51, 22, 54, 3, 7, -4, 38, 14, 55, 46, 81, 22, 58, -3, 2,
827 52, 31, -7, 51, 15, 44, 74, 30, 85, -17, 10, 33, 18, 80, 28, 62,
828 10, 32, 23, 69, 72, 26, 35, 17, 73, 60, 8, 56, 2, 6, -2, -5,
829 51, 4, 11, 50, 66, 76, 21, 44, 33, 47, 1, 26, 64, 48, 57, 40,
830 38, 16, -10, -28, 92, 22, -18, 29, -10, 5, -13, 49, 19, 24, 70, 34,
831 61, 48, 30, 14, -6, 25, 58, 33, 42, 60, 67, 17, 54, 17, 22, 30,
832 67, 44, -9, 50, -11, 43, 40, 32, 59, 82, 13, 49, -14, 55, 60, 36,
833 48, 49, 31, 47, 15, 12, 4, 65, 1, 23, 29, 39, 45, -2, 84, 69,
834 0, 72, 37, 57, 27, 41, -15, -16, 35, 31, 14, 61, 24, 0, 27, 24,
835 16, 41, 55, 34, 53, 9, 56, 12, 25, 29, 53, 5, 20, -20, -8, 20,
836 13, 28, -3, 78, 38, 16, 11, 62, 46, 29, 21, 24, 46, 65, 43, -23,
837 89, 18, 74, 21, 38, -12, 19, 12, -19, 8, 15, 33, 4, 57, 9, -8,
838 36, 35, 26, 28, 7, 83, 63, 79, 75, 11, 3, 87, 37, 47, 34, 40,
839 39, 19, 20, 42, 27, 34, 39, 77, 13, 42, 59, 64, 45, -1, 32, 37,
840 45, -5, 53, -6, 7, 36, 50, 23, 6, 32, 9, -21, 18, 71, 27, 52,
841 -25, 31, 35, 42, -1, 68, 63, 52, 26, 43, 66, 37, 41, 25, 40, 70,
842};
843
844/** Noise generation functions.
845 * I'm not sure what these are for - they seem to be some kind of pseudorandom
846 * sequence generators, used to generate noise data which is used when the
847 * channels are rematrixed. I'm not sure if they provide a practical benefit
848 * to compression, or just obfuscate the decoder. Are they for some kind of
849 * dithering? */
850
851/** Generate two channels of noise, used in the matrix when
852 * restart sync word == 0x31ea. */
853
854static void generate_2_noise_channels(MLPDecodeContext *m, unsigned int substr)
855{
856 SubStream *s = &m->substream[substr];
857 unsigned int i;
858 uint32_t seed = s->noisegen_seed;
859 unsigned int maxchan = s->max_matrix_channel;
860
861 for (i = 0; i < s->blockpos; i++) {
862 uint16_t seed_shr7 = seed >> 7;
863 m->sample_buffer[i][maxchan+1] = ((int8_t)(seed >> 15)) << s->noise_shift;
864 m->sample_buffer[i][maxchan+2] = ((int8_t) seed_shr7) << s->noise_shift;
865
866 seed = (seed << 16) ^ seed_shr7 ^ (seed_shr7 << 5);
867 }
868
869 s->noisegen_seed = seed;
870}
871
872/** Generate a block of noise, used when restart sync word == 0x31eb. */
873
874static void fill_noise_buffer(MLPDecodeContext *m, unsigned int substr)
875{
876 SubStream *s = &m->substream[substr];
877 unsigned int i;
878 uint32_t seed = s->noisegen_seed;
879
880 for (i = 0; i < m->access_unit_size_pow2; i++) {
881 uint8_t seed_shr15 = seed >> 15;
882 m->noise_buffer[i] = noise_table[seed_shr15];
883 seed = (seed << 8) ^ seed_shr15 ^ (seed_shr15 << 5);
884 }
885
886 s->noisegen_seed = seed;
887}
888
889
890/** Apply the channel matrices in turn to reconstruct the original audio
891 * samples. */
892
893static void rematrix_channels(MLPDecodeContext *m, unsigned int substr)
894{
895 SubStream *s = &m->substream[substr];
896 unsigned int mat, src_ch, i;
897 unsigned int maxchan;
898
899 maxchan = s->max_matrix_channel;
900 if (!s->noise_type) {
901 generate_2_noise_channels(m, substr);
902 maxchan += 2;
903 } else {
904 fill_noise_buffer(m, substr);
905 }
906
907 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
908 int matrix_noise_shift = s->matrix_noise_shift[mat];
909 unsigned int dest_ch = s->matrix_out_ch[mat];
910 int32_t mask = MSB_MASK(s->quant_step_size[dest_ch]);
911
912 /* TODO: DSPContext? */
913
914 for (i = 0; i < s->blockpos; i++) {
915 int64_t accum = 0;
916 for (src_ch = 0; src_ch <= maxchan; src_ch++) {
917 accum += (int64_t)m->sample_buffer[i][src_ch]
918 * s->matrix_coeff[mat][src_ch];
919 }
920 if (matrix_noise_shift) {
921 uint32_t index = s->num_primitive_matrices - mat;
922 index = (i * (index * 2 + 1) + index) & (m->access_unit_size_pow2 - 1);
923 accum += m->noise_buffer[index] << (matrix_noise_shift + 7);
924 }
925 m->sample_buffer[i][dest_ch] = ((accum >> 14) & mask)
926 + m->bypassed_lsbs[i][mat];
927 }
928 }
929}
930
931/** Write the audio data into the output buffer. */
932
933static int output_data_internal(MLPDecodeContext *m, unsigned int substr,
934 uint8_t *data, unsigned int *data_size, int is32)
935{
936 SubStream *s = &m->substream[substr];
937 unsigned int i, ch = 0;
938 int32_t *data_32 = (int32_t*) data;
939 int16_t *data_16 = (int16_t*) data;
940
941 if (*data_size < (s->max_channel + 1) * s->blockpos * (is32 ? 4 : 2))
942 return -1;
943
944 for (i = 0; i < s->blockpos; i++) {
945 for (ch = 0; ch <= s->max_channel; ch++) {
946 int32_t sample = m->sample_buffer[i][ch] << s->output_shift[ch];
947 s->lossless_check_data ^= (sample & 0xffffff) << ch;
948 if (is32) *data_32++ = sample << 8;
949 else *data_16++ = sample >> 8;
950 }
951 }
952
953 *data_size = i * ch * (is32 ? 4 : 2);
954
955 return 0;
956}
957
958static int output_data(MLPDecodeContext *m, unsigned int substr,
959 uint8_t *data, unsigned int *data_size)
960{
961 if (m->avctx->sample_fmt == SAMPLE_FMT_S32)
962 return output_data_internal(m, substr, data, data_size, 1);
963 else
964 return output_data_internal(m, substr, data, data_size, 0);
965}
966
967
968/** XOR together all the bytes of a buffer.
969 * Does this belong in dspcontext? */
970
971static uint8_t calculate_parity(const uint8_t *buf, unsigned int buf_size)
972{
973 uint32_t scratch = 0;
974 const uint8_t *buf_end = buf + buf_size;
975
976 for (; buf < buf_end - 3; buf += 4)
977 scratch ^= *((const uint32_t*)buf);
978
979 scratch ^= scratch >> 16;
980 scratch ^= scratch >> 8;
981
982 for (; buf < buf_end; buf++)
983 scratch ^= *buf;
984
985 return scratch;
986}
987
988/** Read an access unit from the stream.
989 * Returns < 0 on error, 0 if not enough data is present in the input stream
990 * otherwise returns the number of bytes consumed. */
991
992static int read_access_unit(AVCodecContext *avctx, void* data, int *data_size,
993 const uint8_t *buf, int buf_size)
994{
995 MLPDecodeContext *m = avctx->priv_data;
996 GetBitContext gb;
997 unsigned int length, substr;
998 unsigned int substream_start;
999 unsigned int header_size = 4;
1000 unsigned int substr_header_size = 0;
1001 uint8_t substream_parity_present[MAX_SUBSTREAMS];
1002 uint16_t substream_data_len[MAX_SUBSTREAMS];
1003 uint8_t parity_bits;
1004
1005 if (buf_size < 4)
1006 return 0;
1007
1008 length = (AV_RB16(buf) & 0xfff) * 2;
1009
1010 if (length > buf_size)
1011 return -1;
1012
1013 init_get_bits(&gb, (buf + 4), (length - 4) * 8);
1014
1015 if (show_bits_long(&gb, 31) == (0xf8726fba >> 1)) {
1016 dprintf(m->avctx, "Found major sync\n");
1017 if (read_major_sync(m, &gb) < 0)
1018 goto error;
1019 header_size += 28;
1020 }
1021
1022 if (!m->params_valid) {
1023 av_log(m->avctx, AV_LOG_WARNING,
1024 "Stream parameters not seen; skipping frame\n");
1025 *data_size = 0;
1026 return length;
1027 }
1028
1029 substream_start = 0;
1030
1031 for (substr = 0; substr < m->num_substreams; substr++) {
1032 int extraword_present, checkdata_present, end;
1033
1034 extraword_present = get_bits1(&gb);
1035 skip_bits1(&gb);
1036 checkdata_present = get_bits1(&gb);
1037 skip_bits1(&gb);
1038
1039 end = get_bits(&gb, 12) * 2;
1040
1041 substr_header_size += 2;
1042
1043 if (extraword_present) {
1044 skip_bits(&gb, 16);
1045 substr_header_size += 2;
1046 }
1047
1048 if (end + header_size + substr_header_size > length) {
1049 av_log(m->avctx, AV_LOG_ERROR,
1050 "Indicated length of substream %d data goes off end of "
1051 "packet.\n", substr);
1052 end = length - header_size - substr_header_size;
1053 }
1054
1055 if (end < substream_start) {
1056 av_log(avctx, AV_LOG_ERROR,
1057 "Indicated end offset of substream %d data "
1058 "is smaller than calculated start offset.\n",
1059 substr);
1060 goto error;
1061 }
1062
1063 if (substr > m->max_decoded_substream)
1064 continue;
1065
1066 substream_parity_present[substr] = checkdata_present;
1067 substream_data_len[substr] = end - substream_start;
1068 substream_start = end;
1069 }
1070
1071 parity_bits = calculate_parity(buf, 4);
1072 parity_bits ^= calculate_parity(buf + header_size, substr_header_size);
1073
1074 if ((((parity_bits >> 4) ^ parity_bits) & 0xF) != 0xF) {
1075 av_log(avctx, AV_LOG_ERROR, "Parity check failed.\n");
1076 goto error;
1077 }
1078
1079 buf += header_size + substr_header_size;
1080
1081 for (substr = 0; substr <= m->max_decoded_substream; substr++) {
1082 SubStream *s = &m->substream[substr];
1083 init_get_bits(&gb, buf, substream_data_len[substr] * 8);
1084
1085 s->blockpos = 0;
1086 do {
1087 if (get_bits1(&gb)) {
1088 if (get_bits1(&gb)) {
1089 /* A restart header should be present */
1090 if (read_restart_header(m, &gb, buf, substr) < 0)
1091 goto next_substr;
1092 s->restart_seen = 1;
1093 }
1094
1095 if (!s->restart_seen) {
1096 av_log(m->avctx, AV_LOG_ERROR,
1097 "No restart header present in substream %d.\n",
1098 substr);
1099 goto next_substr;
1100 }
1101
1102 if (read_decoding_params(m, &gb, substr) < 0)
1103 goto next_substr;
1104 }
1105
1106 if (!s->restart_seen) {
1107 av_log(m->avctx, AV_LOG_ERROR,
1108 "No restart header present in substream %d.\n",
1109 substr);
1110 goto next_substr;
1111 }
1112
1113 if (read_block_data(m, &gb, substr) < 0)
1114 return -1;
1115
1116 } while ((get_bits_count(&gb) < substream_data_len[substr] * 8)
1117 && get_bits1(&gb) == 0);
1118
1119 skip_bits(&gb, (-get_bits_count(&gb)) & 15);
1120 if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 48 &&
1121 (show_bits_long(&gb, 32) == 0xd234d234 ||
1122 show_bits_long(&gb, 20) == 0xd234e)) {
1123 skip_bits(&gb, 18);
1124 if (substr == m->max_decoded_substream)
1125 av_log(m->avctx, AV_LOG_INFO, "End of stream indicated\n");
1126
1127 if (get_bits1(&gb)) {
1128 int shorten_by = get_bits(&gb, 13);
1129 shorten_by = FFMIN(shorten_by, s->blockpos);
1130 s->blockpos -= shorten_by;
1131 } else
1132 skip_bits(&gb, 13);
1133 }
1134 if (substream_parity_present[substr]) {
1135 uint8_t parity, checksum;
1136
1137 parity = calculate_parity(buf, substream_data_len[substr] - 2);
1138 if ((parity ^ get_bits(&gb, 8)) != 0xa9)
1139 av_log(m->avctx, AV_LOG_ERROR,
1140 "Substream %d parity check failed\n", substr);
1141
1142 checksum = mlp_checksum8(buf, substream_data_len[substr] - 2);
1143 if (checksum != get_bits(&gb, 8))
1144 av_log(m->avctx, AV_LOG_ERROR, "Substream %d checksum failed\n",
1145 substr);
1146 }
1147 if (substream_data_len[substr] * 8 != get_bits_count(&gb)) {
1148 av_log(m->avctx, AV_LOG_ERROR, "Substream %d length mismatch.\n",
1149 substr);
1150 return -1;
1151 }
1152
1153next_substr:
1154 buf += substream_data_len[substr];
1155 }
1156
1157 rematrix_channels(m, m->max_decoded_substream);
1158
1159 if (output_data(m, m->max_decoded_substream, data, data_size) < 0)
1160 return -1;
1161
1162 return length;
1163
1164error:
1165 m->params_valid = 0;
1166 return -1;
1167}
1168
1169AVCodec mlp_decoder = {
1170 "mlp",
1171 CODEC_TYPE_AUDIO,
1172 CODEC_ID_MLP,
1173 sizeof(MLPDecodeContext),
1174 mlp_decode_init,
1175 NULL,
1176 NULL,
1177 read_access_unit,
1178 .long_name = NULL_IF_CONFIG_SMALL("Meridian Lossless Packing"),
1179};
1180
Libav master