Add channel layout support to the AC-3 encoder.
[FFMpeg-mirror/lagarith.git] / libavcodec / mlpdec.c
blobbf490abbd84bbe7544c3d1229cf22a550f7ab191
1 /*
2 * MLP decoder
3 * Copyright (c) 2007-2008 Ian Caulfield
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 /**
23 * @file libavcodec/mlpdec.c
24 * MLP decoder
27 #include <stdint.h>
29 #include "avcodec.h"
30 #include "libavutil/intreadwrite.h"
31 #include "get_bits.h"
32 #include "libavutil/crc.h"
33 #include "parser.h"
34 #include "mlp_parser.h"
35 #include "mlp.h"
37 /** number of bits used for VLC lookup - longest Huffman code is 9 */
38 #define VLC_BITS 9
41 static const char* sample_message =
42 "Please file a bug report following the instructions at "
43 "http://ffmpeg.org/bugreports.html and include "
44 "a sample of this file.";
46 typedef struct SubStream {
47 //! Set if a valid restart header has been read. Otherwise the substream cannot be decoded.
48 uint8_t restart_seen;
50 //@{
51 /** restart header data */
52 //! The type of noise to be used in the rematrix stage.
53 uint16_t noise_type;
55 //! The index of the first channel coded in this substream.
56 uint8_t min_channel;
57 //! The index of the last channel coded in this substream.
58 uint8_t max_channel;
59 //! The number of channels input into the rematrix stage.
60 uint8_t max_matrix_channel;
61 //! For each channel output by the matrix, the output channel to map it to
62 uint8_t ch_assign[MAX_CHANNELS];
64 //! The left shift applied to random noise in 0x31ea substreams.
65 uint8_t noise_shift;
66 //! The current seed value for the pseudorandom noise generator(s).
67 uint32_t noisegen_seed;
69 //! Set if the substream contains extra info to check the size of VLC blocks.
70 uint8_t data_check_present;
72 //! Bitmask of which parameter sets are conveyed in a decoding parameter block.
73 uint8_t param_presence_flags;
74 #define PARAM_BLOCKSIZE (1 << 7)
75 #define PARAM_MATRIX (1 << 6)
76 #define PARAM_OUTSHIFT (1 << 5)
77 #define PARAM_QUANTSTEP (1 << 4)
78 #define PARAM_FIR (1 << 3)
79 #define PARAM_IIR (1 << 2)
80 #define PARAM_HUFFOFFSET (1 << 1)
81 #define PARAM_PRESENCE (1 << 0)
82 //@}
84 //@{
85 /** matrix data */
87 //! Number of matrices to be applied.
88 uint8_t num_primitive_matrices;
90 //! matrix output channel
91 uint8_t matrix_out_ch[MAX_MATRICES];
93 //! Whether the LSBs of the matrix output are encoded in the bitstream.
94 uint8_t lsb_bypass[MAX_MATRICES];
95 //! Matrix coefficients, stored as 2.14 fixed point.
96 int32_t matrix_coeff[MAX_MATRICES][MAX_CHANNELS+2];
97 //! Left shift to apply to noise values in 0x31eb substreams.
98 uint8_t matrix_noise_shift[MAX_MATRICES];
99 //@}
101 //! Left shift to apply to Huffman-decoded residuals.
102 uint8_t quant_step_size[MAX_CHANNELS];
104 //! number of PCM samples in current audio block
105 uint16_t blocksize;
106 //! Number of PCM samples decoded so far in this frame.
107 uint16_t blockpos;
109 //! Left shift to apply to decoded PCM values to get final 24-bit output.
110 int8_t output_shift[MAX_CHANNELS];
112 //! Running XOR of all output samples.
113 int32_t lossless_check_data;
115 } SubStream;
117 typedef struct MLPDecodeContext {
118 AVCodecContext *avctx;
120 //! Current access unit being read has a major sync.
121 int is_major_sync_unit;
123 //! Set if a valid major sync block has been read. Otherwise no decoding is possible.
124 uint8_t params_valid;
126 //! Number of substreams contained within this stream.
127 uint8_t num_substreams;
129 //! Index of the last substream to decode - further substreams are skipped.
130 uint8_t max_decoded_substream;
132 //! number of PCM samples contained in each frame
133 int access_unit_size;
134 //! next power of two above the number of samples in each frame
135 int access_unit_size_pow2;
137 SubStream substream[MAX_SUBSTREAMS];
139 ChannelParams channel_params[MAX_CHANNELS];
141 int matrix_changed;
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+2];
147 } MLPDecodeContext;
149 static VLC huff_vlc[3];
151 /** Initialize static data, constant between all invocations of the codec. */
153 static av_cold void init_static(void)
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);
165 ff_mlp_init_crc();
168 static inline int32_t calculate_sign_huff(MLPDecodeContext *m,
169 unsigned int substr, unsigned int ch)
171 ChannelParams *cp = &m->channel_params[ch];
172 SubStream *s = &m->substream[substr];
173 int lsb_bits = cp->huff_lsbs - s->quant_step_size[ch];
174 int sign_shift = lsb_bits + (cp->codebook ? 2 - cp->codebook : -1);
175 int32_t sign_huff_offset = cp->huff_offset;
177 if (cp->codebook > 0)
178 sign_huff_offset -= 7 << lsb_bits;
180 if (sign_shift >= 0)
181 sign_huff_offset -= 1 << sign_shift;
183 return sign_huff_offset;
186 /** Read a sample, consisting of either, both or neither of entropy-coded MSBs
187 * and plain LSBs. */
189 static inline int read_huff_channels(MLPDecodeContext *m, GetBitContext *gbp,
190 unsigned int substr, unsigned int pos)
192 SubStream *s = &m->substream[substr];
193 unsigned int mat, channel;
195 for (mat = 0; mat < s->num_primitive_matrices; mat++)
196 if (s->lsb_bypass[mat])
197 m->bypassed_lsbs[pos + s->blockpos][mat] = get_bits1(gbp);
199 for (channel = s->min_channel; channel <= s->max_channel; channel++) {
200 ChannelParams *cp = &m->channel_params[channel];
201 int codebook = cp->codebook;
202 int quant_step_size = s->quant_step_size[channel];
203 int lsb_bits = cp->huff_lsbs - quant_step_size;
204 int result = 0;
206 if (codebook > 0)
207 result = get_vlc2(gbp, huff_vlc[codebook-1].table,
208 VLC_BITS, (9 + VLC_BITS - 1) / VLC_BITS);
210 if (result < 0)
211 return -1;
213 if (lsb_bits > 0)
214 result = (result << lsb_bits) + get_bits(gbp, lsb_bits);
216 result += cp->sign_huff_offset;
217 result <<= quant_step_size;
219 m->sample_buffer[pos + s->blockpos][channel] = result;
222 return 0;
225 static av_cold int mlp_decode_init(AVCodecContext *avctx)
227 MLPDecodeContext *m = avctx->priv_data;
228 int substr;
230 init_static();
231 m->avctx = avctx;
232 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
233 m->substream[substr].lossless_check_data = 0xffffffff;
235 return 0;
238 /** Read a major sync info header - contains high level information about
239 * the stream - sample rate, channel arrangement etc. Most of this
240 * information is not actually necessary for decoding, only for playback.
243 static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
245 MLPHeaderInfo mh;
246 int substr;
248 if (ff_mlp_read_major_sync(m->avctx, &mh, gb) != 0)
249 return -1;
251 if (mh.group1_bits == 0) {
252 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown bits per sample\n");
253 return -1;
255 if (mh.group2_bits > mh.group1_bits) {
256 av_log(m->avctx, AV_LOG_ERROR,
257 "Channel group 2 cannot have more bits per sample than group 1.\n");
258 return -1;
261 if (mh.group2_samplerate && mh.group2_samplerate != mh.group1_samplerate) {
262 av_log(m->avctx, AV_LOG_ERROR,
263 "Channel groups with differing sample rates are not currently supported.\n");
264 return -1;
267 if (mh.group1_samplerate == 0) {
268 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown sampling rate\n");
269 return -1;
271 if (mh.group1_samplerate > MAX_SAMPLERATE) {
272 av_log(m->avctx, AV_LOG_ERROR,
273 "Sampling rate %d is greater than the supported maximum (%d).\n",
274 mh.group1_samplerate, MAX_SAMPLERATE);
275 return -1;
277 if (mh.access_unit_size > MAX_BLOCKSIZE) {
278 av_log(m->avctx, AV_LOG_ERROR,
279 "Block size %d is greater than the supported maximum (%d).\n",
280 mh.access_unit_size, MAX_BLOCKSIZE);
281 return -1;
283 if (mh.access_unit_size_pow2 > MAX_BLOCKSIZE_POW2) {
284 av_log(m->avctx, AV_LOG_ERROR,
285 "Block size pow2 %d is greater than the supported maximum (%d).\n",
286 mh.access_unit_size_pow2, MAX_BLOCKSIZE_POW2);
287 return -1;
290 if (mh.num_substreams == 0)
291 return -1;
292 if (m->avctx->codec_id == CODEC_ID_MLP && mh.num_substreams > 2) {
293 av_log(m->avctx, AV_LOG_ERROR, "MLP only supports up to 2 substreams.\n");
294 return -1;
296 if (mh.num_substreams > MAX_SUBSTREAMS) {
297 av_log(m->avctx, AV_LOG_ERROR,
298 "Number of substreams %d is larger than the maximum supported "
299 "by the decoder. %s\n", mh.num_substreams, sample_message);
300 return -1;
303 m->access_unit_size = mh.access_unit_size;
304 m->access_unit_size_pow2 = mh.access_unit_size_pow2;
306 m->num_substreams = mh.num_substreams;
307 m->max_decoded_substream = m->num_substreams - 1;
309 m->avctx->sample_rate = mh.group1_samplerate;
310 m->avctx->frame_size = mh.access_unit_size;
312 m->avctx->bits_per_raw_sample = mh.group1_bits;
313 if (mh.group1_bits > 16)
314 m->avctx->sample_fmt = SAMPLE_FMT_S32;
315 else
316 m->avctx->sample_fmt = SAMPLE_FMT_S16;
318 m->params_valid = 1;
319 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
320 m->substream[substr].restart_seen = 0;
322 return 0;
325 /** Read a restart header from a block in a substream. This contains parameters
326 * required to decode the audio that do not change very often. Generally
327 * (always) present only in blocks following a major sync. */
329 static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
330 const uint8_t *buf, unsigned int substr)
332 SubStream *s = &m->substream[substr];
333 unsigned int ch;
334 int sync_word, tmp;
335 uint8_t checksum;
336 uint8_t lossless_check;
337 int start_count = get_bits_count(gbp);
339 sync_word = get_bits(gbp, 13);
341 if (sync_word != 0x31ea >> 1) {
342 av_log(m->avctx, AV_LOG_ERROR,
343 "restart header sync incorrect (got 0x%04x)\n", sync_word);
344 return -1;
346 s->noise_type = get_bits1(gbp);
348 skip_bits(gbp, 16); /* Output timestamp */
350 s->min_channel = get_bits(gbp, 4);
351 s->max_channel = get_bits(gbp, 4);
352 s->max_matrix_channel = get_bits(gbp, 4);
354 if (s->min_channel > s->max_channel) {
355 av_log(m->avctx, AV_LOG_ERROR,
356 "Substream min channel cannot be greater than max channel.\n");
357 return -1;
360 if (m->avctx->request_channels > 0
361 && s->max_channel + 1 >= m->avctx->request_channels
362 && substr < m->max_decoded_substream) {
363 av_log(m->avctx, AV_LOG_INFO,
364 "Extracting %d channel downmix from substream %d. "
365 "Further substreams will be skipped.\n",
366 s->max_channel + 1, substr);
367 m->max_decoded_substream = substr;
370 s->noise_shift = get_bits(gbp, 4);
371 s->noisegen_seed = get_bits(gbp, 23);
373 skip_bits(gbp, 19);
375 s->data_check_present = get_bits1(gbp);
376 lossless_check = get_bits(gbp, 8);
377 if (substr == m->max_decoded_substream
378 && s->lossless_check_data != 0xffffffff) {
379 tmp = xor_32_to_8(s->lossless_check_data);
380 if (tmp != lossless_check)
381 av_log(m->avctx, AV_LOG_WARNING,
382 "Lossless check failed - expected %02x, calculated %02x.\n",
383 lossless_check, tmp);
386 skip_bits(gbp, 16);
388 memset(s->ch_assign, 0, sizeof(s->ch_assign));
390 for (ch = 0; ch <= s->max_matrix_channel; ch++) {
391 int ch_assign = get_bits(gbp, 6);
392 if (ch_assign > s->max_matrix_channel) {
393 av_log(m->avctx, AV_LOG_ERROR,
394 "Assignment of matrix channel %d to invalid output channel %d. %s\n",
395 ch, ch_assign, sample_message);
396 return -1;
398 s->ch_assign[ch_assign] = ch;
401 checksum = ff_mlp_restart_checksum(buf, get_bits_count(gbp) - start_count);
403 if (checksum != get_bits(gbp, 8))
404 av_log(m->avctx, AV_LOG_ERROR, "restart header checksum error\n");
406 /* Set default decoding parameters. */
407 s->param_presence_flags = 0xff;
408 s->num_primitive_matrices = 0;
409 s->blocksize = 8;
410 s->lossless_check_data = 0;
412 memset(s->output_shift , 0, sizeof(s->output_shift ));
413 memset(s->quant_step_size, 0, sizeof(s->quant_step_size));
415 for (ch = s->min_channel; ch <= s->max_channel; ch++) {
416 ChannelParams *cp = &m->channel_params[ch];
417 cp->filter_params[FIR].order = 0;
418 cp->filter_params[IIR].order = 0;
419 cp->filter_params[FIR].shift = 0;
420 cp->filter_params[IIR].shift = 0;
422 /* Default audio coding is 24-bit raw PCM. */
423 cp->huff_offset = 0;
424 cp->sign_huff_offset = (-1) << 23;
425 cp->codebook = 0;
426 cp->huff_lsbs = 24;
429 if (substr == m->max_decoded_substream)
430 m->avctx->channels = s->max_matrix_channel + 1;
432 return 0;
435 /** Read parameters for one of the prediction filters. */
437 static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
438 unsigned int channel, unsigned int filter)
440 FilterParams *fp = &m->channel_params[channel].filter_params[filter];
441 const int max_order = filter ? MAX_IIR_ORDER : MAX_FIR_ORDER;
442 const char fchar = filter ? 'I' : 'F';
443 int i, order;
445 // Filter is 0 for FIR, 1 for IIR.
446 assert(filter < 2);
448 if (m->filter_changed[channel][filter]++ > 1) {
449 av_log(m->avctx, AV_LOG_ERROR, "Filters may change only once per access unit.\n");
450 return -1;
453 order = get_bits(gbp, 4);
454 if (order > max_order) {
455 av_log(m->avctx, AV_LOG_ERROR,
456 "%cIR filter order %d is greater than maximum %d.\n",
457 fchar, order, max_order);
458 return -1;
460 fp->order = order;
462 if (order > 0) {
463 int coeff_bits, coeff_shift;
465 fp->shift = get_bits(gbp, 4);
467 coeff_bits = get_bits(gbp, 5);
468 coeff_shift = get_bits(gbp, 3);
469 if (coeff_bits < 1 || coeff_bits > 16) {
470 av_log(m->avctx, AV_LOG_ERROR,
471 "%cIR filter coeff_bits must be between 1 and 16.\n",
472 fchar);
473 return -1;
475 if (coeff_bits + coeff_shift > 16) {
476 av_log(m->avctx, AV_LOG_ERROR,
477 "Sum of coeff_bits and coeff_shift for %cIR filter must be 16 or less.\n",
478 fchar);
479 return -1;
482 for (i = 0; i < order; i++)
483 fp->coeff[i] = get_sbits(gbp, coeff_bits) << coeff_shift;
485 if (get_bits1(gbp)) {
486 int state_bits, state_shift;
488 if (filter == FIR) {
489 av_log(m->avctx, AV_LOG_ERROR,
490 "FIR filter has state data specified.\n");
491 return -1;
494 state_bits = get_bits(gbp, 4);
495 state_shift = get_bits(gbp, 4);
497 /* TODO: Check validity of state data. */
499 for (i = 0; i < order; i++)
500 fp->state[i] = get_sbits(gbp, state_bits) << state_shift;
504 return 0;
507 /** Read parameters for primitive matrices. */
509 static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitContext *gbp)
511 SubStream *s = &m->substream[substr];
512 unsigned int mat, ch;
514 if (m->matrix_changed++ > 1) {
515 av_log(m->avctx, AV_LOG_ERROR, "Matrices may change only once per access unit.\n");
516 return -1;
519 s->num_primitive_matrices = get_bits(gbp, 4);
521 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
522 int frac_bits, max_chan;
523 s->matrix_out_ch[mat] = get_bits(gbp, 4);
524 frac_bits = get_bits(gbp, 4);
525 s->lsb_bypass [mat] = get_bits1(gbp);
527 if (s->matrix_out_ch[mat] > s->max_matrix_channel) {
528 av_log(m->avctx, AV_LOG_ERROR,
529 "Invalid channel %d specified as output from matrix.\n",
530 s->matrix_out_ch[mat]);
531 return -1;
533 if (frac_bits > 14) {
534 av_log(m->avctx, AV_LOG_ERROR,
535 "Too many fractional bits specified.\n");
536 return -1;
539 max_chan = s->max_matrix_channel;
540 if (!s->noise_type)
541 max_chan+=2;
543 for (ch = 0; ch <= max_chan; ch++) {
544 int coeff_val = 0;
545 if (get_bits1(gbp))
546 coeff_val = get_sbits(gbp, frac_bits + 2);
548 s->matrix_coeff[mat][ch] = coeff_val << (14 - frac_bits);
551 if (s->noise_type)
552 s->matrix_noise_shift[mat] = get_bits(gbp, 4);
553 else
554 s->matrix_noise_shift[mat] = 0;
557 return 0;
560 /** Read channel parameters. */
562 static int read_channel_params(MLPDecodeContext *m, unsigned int substr,
563 GetBitContext *gbp, unsigned int ch)
565 ChannelParams *cp = &m->channel_params[ch];
566 FilterParams *fir = &cp->filter_params[FIR];
567 FilterParams *iir = &cp->filter_params[IIR];
568 SubStream *s = &m->substream[substr];
570 if (s->param_presence_flags & PARAM_FIR)
571 if (get_bits1(gbp))
572 if (read_filter_params(m, gbp, ch, FIR) < 0)
573 return -1;
575 if (s->param_presence_flags & PARAM_IIR)
576 if (get_bits1(gbp))
577 if (read_filter_params(m, gbp, ch, IIR) < 0)
578 return -1;
580 if (fir->order + iir->order > 8) {
581 av_log(m->avctx, AV_LOG_ERROR, "Total filter orders too high.\n");
582 return -1;
585 if (fir->order && iir->order &&
586 fir->shift != iir->shift) {
587 av_log(m->avctx, AV_LOG_ERROR,
588 "FIR and IIR filters must use the same precision.\n");
589 return -1;
591 /* The FIR and IIR filters must have the same precision.
592 * To simplify the filtering code, only the precision of the
593 * FIR filter is considered. If only the IIR filter is employed,
594 * the FIR filter precision is set to that of the IIR filter, so
595 * that the filtering code can use it. */
596 if (!fir->order && iir->order)
597 fir->shift = iir->shift;
599 if (s->param_presence_flags & PARAM_HUFFOFFSET)
600 if (get_bits1(gbp))
601 cp->huff_offset = get_sbits(gbp, 15);
603 cp->codebook = get_bits(gbp, 2);
604 cp->huff_lsbs = get_bits(gbp, 5);
606 if (cp->huff_lsbs > 24) {
607 av_log(m->avctx, AV_LOG_ERROR, "Invalid huff_lsbs.\n");
608 return -1;
611 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
613 return 0;
616 /** Read decoding parameters that change more often than those in the restart
617 * header. */
619 static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
620 unsigned int substr)
622 SubStream *s = &m->substream[substr];
623 unsigned int ch;
625 if (s->param_presence_flags & PARAM_PRESENCE)
626 if (get_bits1(gbp))
627 s->param_presence_flags = get_bits(gbp, 8);
629 if (s->param_presence_flags & PARAM_BLOCKSIZE)
630 if (get_bits1(gbp)) {
631 s->blocksize = get_bits(gbp, 9);
632 if (s->blocksize < 8 || s->blocksize > m->access_unit_size) {
633 av_log(m->avctx, AV_LOG_ERROR, "Invalid blocksize.");
634 s->blocksize = 0;
635 return -1;
639 if (s->param_presence_flags & PARAM_MATRIX)
640 if (get_bits1(gbp))
641 if (read_matrix_params(m, substr, gbp) < 0)
642 return -1;
644 if (s->param_presence_flags & PARAM_OUTSHIFT)
645 if (get_bits1(gbp))
646 for (ch = 0; ch <= s->max_matrix_channel; ch++)
647 s->output_shift[ch] = get_sbits(gbp, 4);
649 if (s->param_presence_flags & PARAM_QUANTSTEP)
650 if (get_bits1(gbp))
651 for (ch = 0; ch <= s->max_channel; ch++) {
652 ChannelParams *cp = &m->channel_params[ch];
654 s->quant_step_size[ch] = get_bits(gbp, 4);
656 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
659 for (ch = s->min_channel; ch <= s->max_channel; ch++)
660 if (get_bits1(gbp))
661 if (read_channel_params(m, substr, gbp, ch) < 0)
662 return -1;
664 return 0;
667 #define MSB_MASK(bits) (-1u << bits)
669 /** Generate PCM samples using the prediction filters and residual values
670 * read from the data stream, and update the filter state. */
672 static void filter_channel(MLPDecodeContext *m, unsigned int substr,
673 unsigned int channel)
675 SubStream *s = &m->substream[substr];
676 int32_t firbuf[MAX_BLOCKSIZE + MAX_FIR_ORDER];
677 int32_t iirbuf[MAX_BLOCKSIZE + MAX_IIR_ORDER];
678 FilterParams *fir = &m->channel_params[channel].filter_params[FIR];
679 FilterParams *iir = &m->channel_params[channel].filter_params[IIR];
680 unsigned int filter_shift = fir->shift;
681 int32_t mask = MSB_MASK(s->quant_step_size[channel]);
682 int index = MAX_BLOCKSIZE;
683 int i;
685 memcpy(&firbuf[index], fir->state, MAX_FIR_ORDER * sizeof(int32_t));
686 memcpy(&iirbuf[index], iir->state, MAX_IIR_ORDER * sizeof(int32_t));
688 for (i = 0; i < s->blocksize; i++) {
689 int32_t residual = m->sample_buffer[i + s->blockpos][channel];
690 unsigned int order;
691 int64_t accum = 0;
692 int32_t result;
694 /* TODO: Move this code to DSPContext? */
696 for (order = 0; order < fir->order; order++)
697 accum += (int64_t) firbuf[index + order] * fir->coeff[order];
698 for (order = 0; order < iir->order; order++)
699 accum += (int64_t) iirbuf[index + order] * iir->coeff[order];
701 accum = accum >> filter_shift;
702 result = (accum + residual) & mask;
704 --index;
706 firbuf[index] = result;
707 iirbuf[index] = result - accum;
709 m->sample_buffer[i + s->blockpos][channel] = result;
712 memcpy(fir->state, &firbuf[index], MAX_FIR_ORDER * sizeof(int32_t));
713 memcpy(iir->state, &iirbuf[index], MAX_IIR_ORDER * sizeof(int32_t));
716 /** Read a block of PCM residual data (or actual if no filtering active). */
718 static int read_block_data(MLPDecodeContext *m, GetBitContext *gbp,
719 unsigned int substr)
721 SubStream *s = &m->substream[substr];
722 unsigned int i, ch, expected_stream_pos = 0;
724 if (s->data_check_present) {
725 expected_stream_pos = get_bits_count(gbp);
726 expected_stream_pos += get_bits(gbp, 16);
727 av_log(m->avctx, AV_LOG_WARNING, "This file contains some features "
728 "we have not tested yet. %s\n", sample_message);
731 if (s->blockpos + s->blocksize > m->access_unit_size) {
732 av_log(m->avctx, AV_LOG_ERROR, "too many audio samples in frame\n");
733 return -1;
736 memset(&m->bypassed_lsbs[s->blockpos][0], 0,
737 s->blocksize * sizeof(m->bypassed_lsbs[0]));
739 for (i = 0; i < s->blocksize; i++)
740 if (read_huff_channels(m, gbp, substr, i) < 0)
741 return -1;
743 for (ch = s->min_channel; ch <= s->max_channel; ch++)
744 filter_channel(m, substr, ch);
746 s->blockpos += s->blocksize;
748 if (s->data_check_present) {
749 if (get_bits_count(gbp) != expected_stream_pos)
750 av_log(m->avctx, AV_LOG_ERROR, "block data length mismatch\n");
751 skip_bits(gbp, 8);
754 return 0;
757 /** Data table used for TrueHD noise generation function. */
759 static const int8_t noise_table[256] = {
760 30, 51, 22, 54, 3, 7, -4, 38, 14, 55, 46, 81, 22, 58, -3, 2,
761 52, 31, -7, 51, 15, 44, 74, 30, 85, -17, 10, 33, 18, 80, 28, 62,
762 10, 32, 23, 69, 72, 26, 35, 17, 73, 60, 8, 56, 2, 6, -2, -5,
763 51, 4, 11, 50, 66, 76, 21, 44, 33, 47, 1, 26, 64, 48, 57, 40,
764 38, 16, -10, -28, 92, 22, -18, 29, -10, 5, -13, 49, 19, 24, 70, 34,
765 61, 48, 30, 14, -6, 25, 58, 33, 42, 60, 67, 17, 54, 17, 22, 30,
766 67, 44, -9, 50, -11, 43, 40, 32, 59, 82, 13, 49, -14, 55, 60, 36,
767 48, 49, 31, 47, 15, 12, 4, 65, 1, 23, 29, 39, 45, -2, 84, 69,
768 0, 72, 37, 57, 27, 41, -15, -16, 35, 31, 14, 61, 24, 0, 27, 24,
769 16, 41, 55, 34, 53, 9, 56, 12, 25, 29, 53, 5, 20, -20, -8, 20,
770 13, 28, -3, 78, 38, 16, 11, 62, 46, 29, 21, 24, 46, 65, 43, -23,
771 89, 18, 74, 21, 38, -12, 19, 12, -19, 8, 15, 33, 4, 57, 9, -8,
772 36, 35, 26, 28, 7, 83, 63, 79, 75, 11, 3, 87, 37, 47, 34, 40,
773 39, 19, 20, 42, 27, 34, 39, 77, 13, 42, 59, 64, 45, -1, 32, 37,
774 45, -5, 53, -6, 7, 36, 50, 23, 6, 32, 9, -21, 18, 71, 27, 52,
775 -25, 31, 35, 42, -1, 68, 63, 52, 26, 43, 66, 37, 41, 25, 40, 70,
778 /** Noise generation functions.
779 * I'm not sure what these are for - they seem to be some kind of pseudorandom
780 * sequence generators, used to generate noise data which is used when the
781 * channels are rematrixed. I'm not sure if they provide a practical benefit
782 * to compression, or just obfuscate the decoder. Are they for some kind of
783 * dithering? */
785 /** Generate two channels of noise, used in the matrix when
786 * restart sync word == 0x31ea. */
788 static void generate_2_noise_channels(MLPDecodeContext *m, unsigned int substr)
790 SubStream *s = &m->substream[substr];
791 unsigned int i;
792 uint32_t seed = s->noisegen_seed;
793 unsigned int maxchan = s->max_matrix_channel;
795 for (i = 0; i < s->blockpos; i++) {
796 uint16_t seed_shr7 = seed >> 7;
797 m->sample_buffer[i][maxchan+1] = ((int8_t)(seed >> 15)) << s->noise_shift;
798 m->sample_buffer[i][maxchan+2] = ((int8_t) seed_shr7) << s->noise_shift;
800 seed = (seed << 16) ^ seed_shr7 ^ (seed_shr7 << 5);
803 s->noisegen_seed = seed;
806 /** Generate a block of noise, used when restart sync word == 0x31eb. */
808 static void fill_noise_buffer(MLPDecodeContext *m, unsigned int substr)
810 SubStream *s = &m->substream[substr];
811 unsigned int i;
812 uint32_t seed = s->noisegen_seed;
814 for (i = 0; i < m->access_unit_size_pow2; i++) {
815 uint8_t seed_shr15 = seed >> 15;
816 m->noise_buffer[i] = noise_table[seed_shr15];
817 seed = (seed << 8) ^ seed_shr15 ^ (seed_shr15 << 5);
820 s->noisegen_seed = seed;
824 /** Apply the channel matrices in turn to reconstruct the original audio
825 * samples. */
827 static void rematrix_channels(MLPDecodeContext *m, unsigned int substr)
829 SubStream *s = &m->substream[substr];
830 unsigned int mat, src_ch, i;
831 unsigned int maxchan;
833 maxchan = s->max_matrix_channel;
834 if (!s->noise_type) {
835 generate_2_noise_channels(m, substr);
836 maxchan += 2;
837 } else {
838 fill_noise_buffer(m, substr);
841 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
842 int matrix_noise_shift = s->matrix_noise_shift[mat];
843 unsigned int dest_ch = s->matrix_out_ch[mat];
844 int32_t mask = MSB_MASK(s->quant_step_size[dest_ch]);
845 int32_t *coeffs = s->matrix_coeff[mat];
846 int index = s->num_primitive_matrices - mat;
847 int index2 = 2 * index + 1;
849 /* TODO: DSPContext? */
851 for (i = 0; i < s->blockpos; i++) {
852 int32_t bypassed_lsb = m->bypassed_lsbs[i][mat];
853 int32_t *samples = m->sample_buffer[i];
854 int64_t accum = 0;
856 for (src_ch = 0; src_ch <= maxchan; src_ch++)
857 accum += (int64_t) samples[src_ch] * coeffs[src_ch];
859 if (matrix_noise_shift) {
860 index &= m->access_unit_size_pow2 - 1;
861 accum += m->noise_buffer[index] << (matrix_noise_shift + 7);
862 index += index2;
865 samples[dest_ch] = ((accum >> 14) & mask) + bypassed_lsb;
870 /** Write the audio data into the output buffer. */
872 static int output_data_internal(MLPDecodeContext *m, unsigned int substr,
873 uint8_t *data, unsigned int *data_size, int is32)
875 SubStream *s = &m->substream[substr];
876 unsigned int i, out_ch = 0;
877 int32_t *data_32 = (int32_t*) data;
878 int16_t *data_16 = (int16_t*) data;
880 if (*data_size < (s->max_channel + 1) * s->blockpos * (is32 ? 4 : 2))
881 return -1;
883 for (i = 0; i < s->blockpos; i++) {
884 for (out_ch = 0; out_ch <= s->max_matrix_channel; out_ch++) {
885 int mat_ch = s->ch_assign[out_ch];
886 int32_t sample = m->sample_buffer[i][mat_ch]
887 << s->output_shift[mat_ch];
888 s->lossless_check_data ^= (sample & 0xffffff) << mat_ch;
889 if (is32) *data_32++ = sample << 8;
890 else *data_16++ = sample >> 8;
894 *data_size = i * out_ch * (is32 ? 4 : 2);
896 return 0;
899 static int output_data(MLPDecodeContext *m, unsigned int substr,
900 uint8_t *data, unsigned int *data_size)
902 if (m->avctx->sample_fmt == SAMPLE_FMT_S32)
903 return output_data_internal(m, substr, data, data_size, 1);
904 else
905 return output_data_internal(m, substr, data, data_size, 0);
909 /** Read an access unit from the stream.
910 * Returns < 0 on error, 0 if not enough data is present in the input stream
911 * otherwise returns the number of bytes consumed. */
913 static int read_access_unit(AVCodecContext *avctx, void* data, int *data_size,
914 AVPacket *avpkt)
916 const uint8_t *buf = avpkt->data;
917 int buf_size = avpkt->size;
918 MLPDecodeContext *m = avctx->priv_data;
919 GetBitContext gb;
920 unsigned int length, substr;
921 unsigned int substream_start;
922 unsigned int header_size = 4;
923 unsigned int substr_header_size = 0;
924 uint8_t substream_parity_present[MAX_SUBSTREAMS];
925 uint16_t substream_data_len[MAX_SUBSTREAMS];
926 uint8_t parity_bits;
928 if (buf_size < 4)
929 return 0;
931 length = (AV_RB16(buf) & 0xfff) * 2;
933 if (length > buf_size)
934 return -1;
936 init_get_bits(&gb, (buf + 4), (length - 4) * 8);
938 m->is_major_sync_unit = 0;
939 if (show_bits_long(&gb, 31) == (0xf8726fba >> 1)) {
940 if (read_major_sync(m, &gb) < 0)
941 goto error;
942 m->is_major_sync_unit = 1;
943 header_size += 28;
946 if (!m->params_valid) {
947 av_log(m->avctx, AV_LOG_WARNING,
948 "Stream parameters not seen; skipping frame.\n");
949 *data_size = 0;
950 return length;
953 substream_start = 0;
955 for (substr = 0; substr < m->num_substreams; substr++) {
956 int extraword_present, checkdata_present, end, nonrestart_substr;
958 extraword_present = get_bits1(&gb);
959 nonrestart_substr = get_bits1(&gb);
960 checkdata_present = get_bits1(&gb);
961 skip_bits1(&gb);
963 end = get_bits(&gb, 12) * 2;
965 substr_header_size += 2;
967 if (extraword_present) {
968 if (m->avctx->codec_id == CODEC_ID_MLP) {
969 av_log(m->avctx, AV_LOG_ERROR, "There must be no extraword for MLP.\n");
970 goto error;
972 skip_bits(&gb, 16);
973 substr_header_size += 2;
976 if (!(nonrestart_substr ^ m->is_major_sync_unit)) {
977 av_log(m->avctx, AV_LOG_ERROR, "Invalid nonrestart_substr.\n");
978 goto error;
981 if (end + header_size + substr_header_size > length) {
982 av_log(m->avctx, AV_LOG_ERROR,
983 "Indicated length of substream %d data goes off end of "
984 "packet.\n", substr);
985 end = length - header_size - substr_header_size;
988 if (end < substream_start) {
989 av_log(avctx, AV_LOG_ERROR,
990 "Indicated end offset of substream %d data "
991 "is smaller than calculated start offset.\n",
992 substr);
993 goto error;
996 if (substr > m->max_decoded_substream)
997 continue;
999 substream_parity_present[substr] = checkdata_present;
1000 substream_data_len[substr] = end - substream_start;
1001 substream_start = end;
1004 parity_bits = ff_mlp_calculate_parity(buf, 4);
1005 parity_bits ^= ff_mlp_calculate_parity(buf + header_size, substr_header_size);
1007 if ((((parity_bits >> 4) ^ parity_bits) & 0xF) != 0xF) {
1008 av_log(avctx, AV_LOG_ERROR, "Parity check failed.\n");
1009 goto error;
1012 buf += header_size + substr_header_size;
1014 for (substr = 0; substr <= m->max_decoded_substream; substr++) {
1015 SubStream *s = &m->substream[substr];
1016 init_get_bits(&gb, buf, substream_data_len[substr] * 8);
1018 m->matrix_changed = 0;
1019 memset(m->filter_changed, 0, sizeof(m->filter_changed));
1021 s->blockpos = 0;
1022 do {
1023 if (get_bits1(&gb)) {
1024 if (get_bits1(&gb)) {
1025 /* A restart header should be present. */
1026 if (read_restart_header(m, &gb, buf, substr) < 0)
1027 goto next_substr;
1028 s->restart_seen = 1;
1031 if (!s->restart_seen)
1032 goto next_substr;
1033 if (read_decoding_params(m, &gb, substr) < 0)
1034 goto next_substr;
1037 if (!s->restart_seen)
1038 goto next_substr;
1040 if (read_block_data(m, &gb, substr) < 0)
1041 return -1;
1043 if (get_bits_count(&gb) >= substream_data_len[substr] * 8)
1044 goto substream_length_mismatch;
1046 } while (!get_bits1(&gb));
1048 skip_bits(&gb, (-get_bits_count(&gb)) & 15);
1050 if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 32) {
1051 int shorten_by;
1053 if (get_bits(&gb, 16) != 0xD234)
1054 return -1;
1056 shorten_by = get_bits(&gb, 16);
1057 if (m->avctx->codec_id == CODEC_ID_TRUEHD && shorten_by & 0x2000)
1058 s->blockpos -= FFMIN(shorten_by & 0x1FFF, s->blockpos);
1059 else if (m->avctx->codec_id == CODEC_ID_MLP && shorten_by != 0xD234)
1060 return -1;
1062 if (substr == m->max_decoded_substream)
1063 av_log(m->avctx, AV_LOG_INFO, "End of stream indicated.\n");
1066 if (substream_parity_present[substr]) {
1067 uint8_t parity, checksum;
1069 if (substream_data_len[substr] * 8 - get_bits_count(&gb) != 16)
1070 goto substream_length_mismatch;
1072 parity = ff_mlp_calculate_parity(buf, substream_data_len[substr] - 2);
1073 checksum = ff_mlp_checksum8 (buf, substream_data_len[substr] - 2);
1075 if ((get_bits(&gb, 8) ^ parity) != 0xa9 )
1076 av_log(m->avctx, AV_LOG_ERROR, "Substream %d parity check failed.\n", substr);
1077 if ( get_bits(&gb, 8) != checksum)
1078 av_log(m->avctx, AV_LOG_ERROR, "Substream %d checksum failed.\n" , substr);
1081 if (substream_data_len[substr] * 8 != get_bits_count(&gb))
1082 goto substream_length_mismatch;
1084 next_substr:
1085 if (!s->restart_seen)
1086 av_log(m->avctx, AV_LOG_ERROR,
1087 "No restart header present in substream %d.\n", substr);
1089 buf += substream_data_len[substr];
1092 rematrix_channels(m, m->max_decoded_substream);
1094 if (output_data(m, m->max_decoded_substream, data, data_size) < 0)
1095 return -1;
1097 return length;
1099 substream_length_mismatch:
1100 av_log(m->avctx, AV_LOG_ERROR, "substream %d length mismatch\n", substr);
1101 return -1;
1103 error:
1104 m->params_valid = 0;
1105 return -1;
1108 #if CONFIG_MLP_DECODER
1109 AVCodec mlp_decoder = {
1110 "mlp",
1111 CODEC_TYPE_AUDIO,
1112 CODEC_ID_MLP,
1113 sizeof(MLPDecodeContext),
1114 mlp_decode_init,
1115 NULL,
1116 NULL,
1117 read_access_unit,
1118 .long_name = NULL_IF_CONFIG_SMALL("MLP (Meridian Lossless Packing)"),
1120 #endif /* CONFIG_MLP_DECODER */
1122 #if CONFIG_TRUEHD_DECODER
1123 AVCodec truehd_decoder = {
1124 "truehd",
1125 CODEC_TYPE_AUDIO,
1126 CODEC_ID_TRUEHD,
1127 sizeof(MLPDecodeContext),
1128 mlp_decode_init,
1129 NULL,
1130 NULL,
1131 read_access_unit,
1132 .long_name = NULL_IF_CONFIG_SMALL("TrueHD"),
1134 #endif /* CONFIG_TRUEHD_DECODER */