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
31 #include "libavutil/intreadwrite.h"
33 #include "libavutil/crc.h"
35 #include "mlp_parser.h"
38 /** number of bits used for VLC lookup - longest Huffman code is 9 */
42 static const char* sample_message =
43 "Please file a bug report following the instructions at "
44 "http://ffmpeg.org/bugreports.html and include "
45 "a sample of this file.";
47 typedef struct SubStream {
48 /// Set if a valid restart header has been read. Otherwise the substream cannot be decoded.
52 /** restart header data */
53 /// The type of noise to be used in the rematrix stage.
56 /// The index of the first channel coded in this substream.
58 /// The index of the last channel coded in this substream.
60 /// The number of channels input into the rematrix stage.
61 uint8_t max_matrix_channel;
62 /// For each channel output by the matrix, the output channel to map it to
63 uint8_t ch_assign[MAX_CHANNELS];
65 /// Channel coding parameters for channels in the substream
66 ChannelParams channel_params[MAX_CHANNELS];
68 /// The left shift applied to random noise in 0x31ea substreams.
70 /// The current seed value for the pseudorandom noise generator(s).
71 uint32_t noisegen_seed;
73 /// Set if the substream contains extra info to check the size of VLC blocks.
74 uint8_t data_check_present;
76 /// Bitmask of which parameter sets are conveyed in a decoding parameter block.
77 uint8_t param_presence_flags;
78 #define PARAM_BLOCKSIZE (1 << 7)
79 #define PARAM_MATRIX (1 << 6)
80 #define PARAM_OUTSHIFT (1 << 5)
81 #define PARAM_QUANTSTEP (1 << 4)
82 #define PARAM_FIR (1 << 3)
83 #define PARAM_IIR (1 << 2)
84 #define PARAM_HUFFOFFSET (1 << 1)
85 #define PARAM_PRESENCE (1 << 0)
91 /// Number of matrices to be applied.
92 uint8_t num_primitive_matrices;
94 /// matrix output channel
95 uint8_t matrix_out_ch[MAX_MATRICES];
97 /// Whether the LSBs of the matrix output are encoded in the bitstream.
98 uint8_t lsb_bypass[MAX_MATRICES];
99 /// Matrix coefficients, stored as 2.14 fixed point.
100 int32_t matrix_coeff[MAX_MATRICES][MAX_CHANNELS];
101 /// Left shift to apply to noise values in 0x31eb substreams.
102 uint8_t matrix_noise_shift[MAX_MATRICES];
105 /// Left shift to apply to Huffman-decoded residuals.
106 uint8_t quant_step_size[MAX_CHANNELS];
108 /// number of PCM samples in current audio block
110 /// Number of PCM samples decoded so far in this frame.
113 /// Left shift to apply to decoded PCM values to get final 24-bit output.
114 int8_t output_shift[MAX_CHANNELS];
116 /// Running XOR of all output samples.
117 int32_t lossless_check_data;
121 typedef struct MLPDecodeContext {
122 AVCodecContext *avctx;
125 /// Current access unit being read has a major sync.
126 int is_major_sync_unit;
128 /// Set if a valid major sync block has been read. Otherwise no decoding is possible.
129 uint8_t params_valid;
131 /// Number of substreams contained within this stream.
132 uint8_t num_substreams;
134 /// Index of the last substream to decode - further substreams are skipped.
135 uint8_t max_decoded_substream;
137 /// Stream needs channel reordering to comply with FFmpeg's channel order
138 uint8_t needs_reordering;
140 /// number of PCM samples contained in each frame
141 int access_unit_size;
142 /// next power of two above the number of samples in each frame
143 int access_unit_size_pow2;
145 SubStream substream[MAX_SUBSTREAMS];
148 int filter_changed[MAX_CHANNELS][NUM_FILTERS];
150 int8_t noise_buffer[MAX_BLOCKSIZE_POW2];
151 int8_t bypassed_lsbs[MAX_BLOCKSIZE][MAX_CHANNELS];
152 int32_t sample_buffer[MAX_BLOCKSIZE][MAX_CHANNELS];
157 static VLC huff_vlc[3];
159 /** Initialize static data, constant between all invocations of the codec. */
161 static av_cold void init_static(void)
163 if (!huff_vlc[0].bits) {
164 INIT_VLC_STATIC(&huff_vlc[0], VLC_BITS, 18,
165 &ff_mlp_huffman_tables[0][0][1], 2, 1,
166 &ff_mlp_huffman_tables[0][0][0], 2, 1, 512);
167 INIT_VLC_STATIC(&huff_vlc[1], VLC_BITS, 16,
168 &ff_mlp_huffman_tables[1][0][1], 2, 1,
169 &ff_mlp_huffman_tables[1][0][0], 2, 1, 512);
170 INIT_VLC_STATIC(&huff_vlc[2], VLC_BITS, 15,
171 &ff_mlp_huffman_tables[2][0][1], 2, 1,
172 &ff_mlp_huffman_tables[2][0][0], 2, 1, 512);
178 static inline int32_t calculate_sign_huff(MLPDecodeContext *m,
179 unsigned int substr, unsigned int ch)
181 SubStream *s = &m->substream[substr];
182 ChannelParams *cp = &s->channel_params[ch];
183 int lsb_bits = cp->huff_lsbs - s->quant_step_size[ch];
184 int sign_shift = lsb_bits + (cp->codebook ? 2 - cp->codebook : -1);
185 int32_t sign_huff_offset = cp->huff_offset;
187 if (cp->codebook > 0)
188 sign_huff_offset -= 7 << lsb_bits;
191 sign_huff_offset -= 1 << sign_shift;
193 return sign_huff_offset;
196 /** Read a sample, consisting of either, both or neither of entropy-coded MSBs
199 static inline int read_huff_channels(MLPDecodeContext *m, GetBitContext *gbp,
200 unsigned int substr, unsigned int pos)
202 SubStream *s = &m->substream[substr];
203 unsigned int mat, channel;
205 for (mat = 0; mat < s->num_primitive_matrices; mat++)
206 if (s->lsb_bypass[mat])
207 m->bypassed_lsbs[pos + s->blockpos][mat] = get_bits1(gbp);
209 for (channel = s->min_channel; channel <= s->max_channel; channel++) {
210 ChannelParams *cp = &s->channel_params[channel];
211 int codebook = cp->codebook;
212 int quant_step_size = s->quant_step_size[channel];
213 int lsb_bits = cp->huff_lsbs - quant_step_size;
217 result = get_vlc2(gbp, huff_vlc[codebook-1].table,
218 VLC_BITS, (9 + VLC_BITS - 1) / VLC_BITS);
221 return AVERROR_INVALIDDATA;
224 result = (result << lsb_bits) + get_bits(gbp, lsb_bits);
226 result += cp->sign_huff_offset;
227 result <<= quant_step_size;
229 m->sample_buffer[pos + s->blockpos][channel] = result;
235 static av_cold int mlp_decode_init(AVCodecContext *avctx)
237 MLPDecodeContext *m = avctx->priv_data;
242 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
243 m->substream[substr].lossless_check_data = 0xffffffff;
244 dsputil_init(&m->dsp, avctx);
246 avcodec_get_frame_defaults(&m->frame);
247 avctx->coded_frame = &m->frame;
252 /** Read a major sync info header - contains high level information about
253 * the stream - sample rate, channel arrangement etc. Most of this
254 * information is not actually necessary for decoding, only for playback.
257 static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
262 if ((ret = ff_mlp_read_major_sync(m->avctx, &mh, gb)) != 0)
265 if (mh.group1_bits == 0) {
266 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown bits per sample\n");
267 return AVERROR_INVALIDDATA;
269 if (mh.group2_bits > mh.group1_bits) {
270 av_log(m->avctx, AV_LOG_ERROR,
271 "Channel group 2 cannot have more bits per sample than group 1.\n");
272 return AVERROR_INVALIDDATA;
275 if (mh.group2_samplerate && mh.group2_samplerate != mh.group1_samplerate) {
276 av_log(m->avctx, AV_LOG_ERROR,
277 "Channel groups with differing sample rates are not currently supported.\n");
278 return AVERROR_INVALIDDATA;
281 if (mh.group1_samplerate == 0) {
282 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown sampling rate\n");
283 return AVERROR_INVALIDDATA;
285 if (mh.group1_samplerate > MAX_SAMPLERATE) {
286 av_log(m->avctx, AV_LOG_ERROR,
287 "Sampling rate %d is greater than the supported maximum (%d).\n",
288 mh.group1_samplerate, MAX_SAMPLERATE);
289 return AVERROR_INVALIDDATA;
291 if (mh.access_unit_size > MAX_BLOCKSIZE) {
292 av_log(m->avctx, AV_LOG_ERROR,
293 "Block size %d is greater than the supported maximum (%d).\n",
294 mh.access_unit_size, MAX_BLOCKSIZE);
295 return AVERROR_INVALIDDATA;
297 if (mh.access_unit_size_pow2 > MAX_BLOCKSIZE_POW2) {
298 av_log(m->avctx, AV_LOG_ERROR,
299 "Block size pow2 %d is greater than the supported maximum (%d).\n",
300 mh.access_unit_size_pow2, MAX_BLOCKSIZE_POW2);
301 return AVERROR_INVALIDDATA;
304 if (mh.num_substreams == 0)
305 return AVERROR_INVALIDDATA;
306 if (m->avctx->codec_id == CODEC_ID_MLP && mh.num_substreams > 2) {
307 av_log(m->avctx, AV_LOG_ERROR, "MLP only supports up to 2 substreams.\n");
308 return AVERROR_INVALIDDATA;
310 if (mh.num_substreams > MAX_SUBSTREAMS) {
311 av_log(m->avctx, AV_LOG_ERROR,
312 "Number of substreams %d is larger than the maximum supported "
313 "by the decoder. %s\n", mh.num_substreams, sample_message);
314 return AVERROR_INVALIDDATA;
317 m->access_unit_size = mh.access_unit_size;
318 m->access_unit_size_pow2 = mh.access_unit_size_pow2;
320 m->num_substreams = mh.num_substreams;
321 m->max_decoded_substream = m->num_substreams - 1;
323 m->avctx->sample_rate = mh.group1_samplerate;
324 m->avctx->frame_size = mh.access_unit_size;
326 m->avctx->bits_per_raw_sample = mh.group1_bits;
327 if (mh.group1_bits > 16)
328 m->avctx->sample_fmt = AV_SAMPLE_FMT_S32;
330 m->avctx->sample_fmt = AV_SAMPLE_FMT_S16;
333 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
334 m->substream[substr].restart_seen = 0;
336 if (mh.stream_type == 0xbb) {
338 m->avctx->channel_layout = ff_mlp_layout[mh.channels_mlp];
339 } else { /* mh.stream_type == 0xba */
341 if (mh.channels_thd_stream2) {
342 m->avctx->channel_layout = ff_truehd_layout(mh.channels_thd_stream2);
344 m->avctx->channel_layout = ff_truehd_layout(mh.channels_thd_stream1);
346 if (m->avctx->channels &&
347 !m->avctx->request_channels && !m->avctx->request_channel_layout &&
348 av_get_channel_layout_nb_channels(m->avctx->channel_layout) != m->avctx->channels) {
349 m->avctx->channel_layout = 0;
350 av_log_ask_for_sample(m->avctx, "Unknown channel layout.");
354 m->needs_reordering = mh.channels_mlp >= 18 && mh.channels_mlp <= 20;
359 /** Read a restart header from a block in a substream. This contains parameters
360 * required to decode the audio that do not change very often. Generally
361 * (always) present only in blocks following a major sync. */
363 static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
364 const uint8_t *buf, unsigned int substr)
366 SubStream *s = &m->substream[substr];
370 uint8_t lossless_check;
371 int start_count = get_bits_count(gbp);
372 const int max_matrix_channel = m->avctx->codec_id == CODEC_ID_MLP
373 ? MAX_MATRIX_CHANNEL_MLP
374 : MAX_MATRIX_CHANNEL_TRUEHD;
376 sync_word = get_bits(gbp, 13);
378 if (sync_word != 0x31ea >> 1) {
379 av_log(m->avctx, AV_LOG_ERROR,
380 "restart header sync incorrect (got 0x%04x)\n", sync_word);
381 return AVERROR_INVALIDDATA;
384 s->noise_type = get_bits1(gbp);
386 if (m->avctx->codec_id == CODEC_ID_MLP && s->noise_type) {
387 av_log(m->avctx, AV_LOG_ERROR, "MLP must have 0x31ea sync word.\n");
388 return AVERROR_INVALIDDATA;
391 skip_bits(gbp, 16); /* Output timestamp */
393 s->min_channel = get_bits(gbp, 4);
394 s->max_channel = get_bits(gbp, 4);
395 s->max_matrix_channel = get_bits(gbp, 4);
397 if (s->max_matrix_channel > max_matrix_channel) {
398 av_log(m->avctx, AV_LOG_ERROR,
399 "Max matrix channel cannot be greater than %d.\n",
401 return AVERROR_INVALIDDATA;
404 if (s->max_channel != s->max_matrix_channel) {
405 av_log(m->avctx, AV_LOG_ERROR,
406 "Max channel must be equal max matrix channel.\n");
407 return AVERROR_INVALIDDATA;
410 /* This should happen for TrueHD streams with >6 channels and MLP's noise
411 * type. It is not yet known if this is allowed. */
412 if (s->max_channel > MAX_MATRIX_CHANNEL_MLP && !s->noise_type) {
413 av_log(m->avctx, AV_LOG_ERROR,
414 "Number of channels %d is larger than the maximum supported "
415 "by the decoder. %s\n", s->max_channel+2, sample_message);
416 return AVERROR_INVALIDDATA;
419 if (s->min_channel > s->max_channel) {
420 av_log(m->avctx, AV_LOG_ERROR,
421 "Substream min channel cannot be greater than max channel.\n");
422 return AVERROR_INVALIDDATA;
425 if (m->avctx->request_channels > 0
426 && s->max_channel + 1 >= m->avctx->request_channels
427 && substr < m->max_decoded_substream) {
428 av_log(m->avctx, AV_LOG_DEBUG,
429 "Extracting %d channel downmix from substream %d. "
430 "Further substreams will be skipped.\n",
431 s->max_channel + 1, substr);
432 m->max_decoded_substream = substr;
435 s->noise_shift = get_bits(gbp, 4);
436 s->noisegen_seed = get_bits(gbp, 23);
440 s->data_check_present = get_bits1(gbp);
441 lossless_check = get_bits(gbp, 8);
442 if (substr == m->max_decoded_substream
443 && s->lossless_check_data != 0xffffffff) {
444 tmp = xor_32_to_8(s->lossless_check_data);
445 if (tmp != lossless_check)
446 av_log(m->avctx, AV_LOG_WARNING,
447 "Lossless check failed - expected %02x, calculated %02x.\n",
448 lossless_check, tmp);
453 memset(s->ch_assign, 0, sizeof(s->ch_assign));
455 for (ch = 0; ch <= s->max_matrix_channel; ch++) {
456 int ch_assign = get_bits(gbp, 6);
457 if (ch_assign > s->max_matrix_channel) {
458 av_log(m->avctx, AV_LOG_ERROR,
459 "Assignment of matrix channel %d to invalid output channel %d. %s\n",
460 ch, ch_assign, sample_message);
461 return AVERROR_INVALIDDATA;
463 s->ch_assign[ch_assign] = ch;
466 if (m->avctx->codec_id == CODEC_ID_MLP && m->needs_reordering) {
467 if (m->avctx->channel_layout == (AV_CH_LAYOUT_QUAD|AV_CH_LOW_FREQUENCY) ||
468 m->avctx->channel_layout == AV_CH_LAYOUT_5POINT0_BACK) {
469 int i = s->ch_assign[4];
470 s->ch_assign[4] = s->ch_assign[3];
471 s->ch_assign[3] = s->ch_assign[2];
473 } else if (m->avctx->channel_layout == AV_CH_LAYOUT_5POINT1_BACK) {
474 FFSWAP(int, s->ch_assign[2], s->ch_assign[4]);
475 FFSWAP(int, s->ch_assign[3], s->ch_assign[5]);
478 if (m->avctx->codec_id == CODEC_ID_TRUEHD &&
479 (m->avctx->channel_layout == AV_CH_LAYOUT_7POINT1 ||
480 m->avctx->channel_layout == AV_CH_LAYOUT_7POINT1_WIDE)) {
481 FFSWAP(int, s->ch_assign[4], s->ch_assign[6]);
482 FFSWAP(int, s->ch_assign[5], s->ch_assign[7]);
483 } else if (m->avctx->codec_id == CODEC_ID_TRUEHD &&
484 (m->avctx->channel_layout == AV_CH_LAYOUT_6POINT1 ||
485 m->avctx->channel_layout == (AV_CH_LAYOUT_6POINT1 | AV_CH_TOP_CENTER) ||
486 m->avctx->channel_layout == (AV_CH_LAYOUT_6POINT1 | AV_CH_TOP_FRONT_CENTER))) {
487 int i = s->ch_assign[6];
488 s->ch_assign[6] = s->ch_assign[5];
489 s->ch_assign[5] = s->ch_assign[4];
493 checksum = ff_mlp_restart_checksum(buf, get_bits_count(gbp) - start_count);
495 if (checksum != get_bits(gbp, 8))
496 av_log(m->avctx, AV_LOG_ERROR, "restart header checksum error\n");
498 /* Set default decoding parameters. */
499 s->param_presence_flags = 0xff;
500 s->num_primitive_matrices = 0;
502 s->lossless_check_data = 0;
504 memset(s->output_shift , 0, sizeof(s->output_shift ));
505 memset(s->quant_step_size, 0, sizeof(s->quant_step_size));
507 for (ch = s->min_channel; ch <= s->max_channel; ch++) {
508 ChannelParams *cp = &s->channel_params[ch];
509 cp->filter_params[FIR].order = 0;
510 cp->filter_params[IIR].order = 0;
511 cp->filter_params[FIR].shift = 0;
512 cp->filter_params[IIR].shift = 0;
514 /* Default audio coding is 24-bit raw PCM. */
516 cp->sign_huff_offset = (-1) << 23;
521 if (substr == m->max_decoded_substream)
522 m->avctx->channels = s->max_matrix_channel + 1;
527 /** Read parameters for one of the prediction filters. */
529 static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
530 unsigned int substr, unsigned int channel,
533 SubStream *s = &m->substream[substr];
534 FilterParams *fp = &s->channel_params[channel].filter_params[filter];
535 const int max_order = filter ? MAX_IIR_ORDER : MAX_FIR_ORDER;
536 const char fchar = filter ? 'I' : 'F';
539 // Filter is 0 for FIR, 1 for IIR.
542 if (m->filter_changed[channel][filter]++ > 1) {
543 av_log(m->avctx, AV_LOG_ERROR, "Filters may change only once per access unit.\n");
544 return AVERROR_INVALIDDATA;
547 order = get_bits(gbp, 4);
548 if (order > max_order) {
549 av_log(m->avctx, AV_LOG_ERROR,
550 "%cIR filter order %d is greater than maximum %d.\n",
551 fchar, order, max_order);
552 return AVERROR_INVALIDDATA;
557 int32_t *fcoeff = s->channel_params[channel].coeff[filter];
558 int coeff_bits, coeff_shift;
560 fp->shift = get_bits(gbp, 4);
562 coeff_bits = get_bits(gbp, 5);
563 coeff_shift = get_bits(gbp, 3);
564 if (coeff_bits < 1 || coeff_bits > 16) {
565 av_log(m->avctx, AV_LOG_ERROR,
566 "%cIR filter coeff_bits must be between 1 and 16.\n",
568 return AVERROR_INVALIDDATA;
570 if (coeff_bits + coeff_shift > 16) {
571 av_log(m->avctx, AV_LOG_ERROR,
572 "Sum of coeff_bits and coeff_shift for %cIR filter must be 16 or less.\n",
574 return AVERROR_INVALIDDATA;
577 for (i = 0; i < order; i++)
578 fcoeff[i] = get_sbits(gbp, coeff_bits) << coeff_shift;
580 if (get_bits1(gbp)) {
581 int state_bits, state_shift;
584 av_log(m->avctx, AV_LOG_ERROR,
585 "FIR filter has state data specified.\n");
586 return AVERROR_INVALIDDATA;
589 state_bits = get_bits(gbp, 4);
590 state_shift = get_bits(gbp, 4);
592 /* TODO: Check validity of state data. */
594 for (i = 0; i < order; i++)
595 fp->state[i] = get_sbits(gbp, state_bits) << state_shift;
602 /** Read parameters for primitive matrices. */
604 static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitContext *gbp)
606 SubStream *s = &m->substream[substr];
607 unsigned int mat, ch;
608 const int max_primitive_matrices = m->avctx->codec_id == CODEC_ID_MLP
610 : MAX_MATRICES_TRUEHD;
612 if (m->matrix_changed++ > 1) {
613 av_log(m->avctx, AV_LOG_ERROR, "Matrices may change only once per access unit.\n");
614 return AVERROR_INVALIDDATA;
617 s->num_primitive_matrices = get_bits(gbp, 4);
619 if (s->num_primitive_matrices > max_primitive_matrices) {
620 av_log(m->avctx, AV_LOG_ERROR,
621 "Number of primitive matrices cannot be greater than %d.\n",
622 max_primitive_matrices);
623 return AVERROR_INVALIDDATA;
626 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
627 int frac_bits, max_chan;
628 s->matrix_out_ch[mat] = get_bits(gbp, 4);
629 frac_bits = get_bits(gbp, 4);
630 s->lsb_bypass [mat] = get_bits1(gbp);
632 if (s->matrix_out_ch[mat] > s->max_matrix_channel) {
633 av_log(m->avctx, AV_LOG_ERROR,
634 "Invalid channel %d specified as output from matrix.\n",
635 s->matrix_out_ch[mat]);
636 return AVERROR_INVALIDDATA;
638 if (frac_bits > 14) {
639 av_log(m->avctx, AV_LOG_ERROR,
640 "Too many fractional bits specified.\n");
641 return AVERROR_INVALIDDATA;
644 max_chan = s->max_matrix_channel;
648 for (ch = 0; ch <= max_chan; ch++) {
651 coeff_val = get_sbits(gbp, frac_bits + 2);
653 s->matrix_coeff[mat][ch] = coeff_val << (14 - frac_bits);
657 s->matrix_noise_shift[mat] = get_bits(gbp, 4);
659 s->matrix_noise_shift[mat] = 0;
665 /** Read channel parameters. */
667 static int read_channel_params(MLPDecodeContext *m, unsigned int substr,
668 GetBitContext *gbp, unsigned int ch)
670 SubStream *s = &m->substream[substr];
671 ChannelParams *cp = &s->channel_params[ch];
672 FilterParams *fir = &cp->filter_params[FIR];
673 FilterParams *iir = &cp->filter_params[IIR];
676 if (s->param_presence_flags & PARAM_FIR)
678 if ((ret = read_filter_params(m, gbp, substr, ch, FIR)) < 0)
681 if (s->param_presence_flags & PARAM_IIR)
683 if ((ret = read_filter_params(m, gbp, substr, ch, IIR)) < 0)
686 if (fir->order + iir->order > 8) {
687 av_log(m->avctx, AV_LOG_ERROR, "Total filter orders too high.\n");
688 return AVERROR_INVALIDDATA;
691 if (fir->order && iir->order &&
692 fir->shift != iir->shift) {
693 av_log(m->avctx, AV_LOG_ERROR,
694 "FIR and IIR filters must use the same precision.\n");
695 return AVERROR_INVALIDDATA;
697 /* The FIR and IIR filters must have the same precision.
698 * To simplify the filtering code, only the precision of the
699 * FIR filter is considered. If only the IIR filter is employed,
700 * the FIR filter precision is set to that of the IIR filter, so
701 * that the filtering code can use it. */
702 if (!fir->order && iir->order)
703 fir->shift = iir->shift;
705 if (s->param_presence_flags & PARAM_HUFFOFFSET)
707 cp->huff_offset = get_sbits(gbp, 15);
709 cp->codebook = get_bits(gbp, 2);
710 cp->huff_lsbs = get_bits(gbp, 5);
712 if (cp->huff_lsbs > 24) {
713 av_log(m->avctx, AV_LOG_ERROR, "Invalid huff_lsbs.\n");
714 return AVERROR_INVALIDDATA;
717 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
722 /** Read decoding parameters that change more often than those in the restart
725 static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
728 SubStream *s = &m->substream[substr];
732 if (s->param_presence_flags & PARAM_PRESENCE)
734 s->param_presence_flags = get_bits(gbp, 8);
736 if (s->param_presence_flags & PARAM_BLOCKSIZE)
737 if (get_bits1(gbp)) {
738 s->blocksize = get_bits(gbp, 9);
739 if (s->blocksize < 8 || s->blocksize > m->access_unit_size) {
740 av_log(m->avctx, AV_LOG_ERROR, "Invalid blocksize.");
742 return AVERROR_INVALIDDATA;
746 if (s->param_presence_flags & PARAM_MATRIX)
748 if ((ret = read_matrix_params(m, substr, gbp)) < 0)
751 if (s->param_presence_flags & PARAM_OUTSHIFT)
753 for (ch = 0; ch <= s->max_matrix_channel; ch++)
754 s->output_shift[ch] = get_sbits(gbp, 4);
756 if (s->param_presence_flags & PARAM_QUANTSTEP)
758 for (ch = 0; ch <= s->max_channel; ch++) {
759 ChannelParams *cp = &s->channel_params[ch];
761 s->quant_step_size[ch] = get_bits(gbp, 4);
763 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
766 for (ch = s->min_channel; ch <= s->max_channel; ch++)
768 if ((ret = read_channel_params(m, substr, gbp, ch)) < 0)
774 #define MSB_MASK(bits) (-1u << bits)
776 /** Generate PCM samples using the prediction filters and residual values
777 * read from the data stream, and update the filter state. */
779 static void filter_channel(MLPDecodeContext *m, unsigned int substr,
780 unsigned int channel)
782 SubStream *s = &m->substream[substr];
783 const int32_t *fircoeff = s->channel_params[channel].coeff[FIR];
784 int32_t state_buffer[NUM_FILTERS][MAX_BLOCKSIZE + MAX_FIR_ORDER];
785 int32_t *firbuf = state_buffer[FIR] + MAX_BLOCKSIZE;
786 int32_t *iirbuf = state_buffer[IIR] + MAX_BLOCKSIZE;
787 FilterParams *fir = &s->channel_params[channel].filter_params[FIR];
788 FilterParams *iir = &s->channel_params[channel].filter_params[IIR];
789 unsigned int filter_shift = fir->shift;
790 int32_t mask = MSB_MASK(s->quant_step_size[channel]);
792 memcpy(firbuf, fir->state, MAX_FIR_ORDER * sizeof(int32_t));
793 memcpy(iirbuf, iir->state, MAX_IIR_ORDER * sizeof(int32_t));
795 m->dsp.mlp_filter_channel(firbuf, fircoeff,
796 fir->order, iir->order,
797 filter_shift, mask, s->blocksize,
798 &m->sample_buffer[s->blockpos][channel]);
800 memcpy(fir->state, firbuf - s->blocksize, MAX_FIR_ORDER * sizeof(int32_t));
801 memcpy(iir->state, iirbuf - s->blocksize, MAX_IIR_ORDER * sizeof(int32_t));
804 /** Read a block of PCM residual data (or actual if no filtering active). */
806 static int read_block_data(MLPDecodeContext *m, GetBitContext *gbp,
809 SubStream *s = &m->substream[substr];
810 unsigned int i, ch, expected_stream_pos = 0;
813 if (s->data_check_present) {
814 expected_stream_pos = get_bits_count(gbp);
815 expected_stream_pos += get_bits(gbp, 16);
816 av_log(m->avctx, AV_LOG_WARNING, "This file contains some features "
817 "we have not tested yet. %s\n", sample_message);
820 if (s->blockpos + s->blocksize > m->access_unit_size) {
821 av_log(m->avctx, AV_LOG_ERROR, "too many audio samples in frame\n");
822 return AVERROR_INVALIDDATA;
825 memset(&m->bypassed_lsbs[s->blockpos][0], 0,
826 s->blocksize * sizeof(m->bypassed_lsbs[0]));
828 for (i = 0; i < s->blocksize; i++)
829 if ((ret = read_huff_channels(m, gbp, substr, i)) < 0)
832 for (ch = s->min_channel; ch <= s->max_channel; ch++)
833 filter_channel(m, substr, ch);
835 s->blockpos += s->blocksize;
837 if (s->data_check_present) {
838 if (get_bits_count(gbp) != expected_stream_pos)
839 av_log(m->avctx, AV_LOG_ERROR, "block data length mismatch\n");
846 /** Data table used for TrueHD noise generation function. */
848 static const int8_t noise_table[256] = {
849 30, 51, 22, 54, 3, 7, -4, 38, 14, 55, 46, 81, 22, 58, -3, 2,
850 52, 31, -7, 51, 15, 44, 74, 30, 85, -17, 10, 33, 18, 80, 28, 62,
851 10, 32, 23, 69, 72, 26, 35, 17, 73, 60, 8, 56, 2, 6, -2, -5,
852 51, 4, 11, 50, 66, 76, 21, 44, 33, 47, 1, 26, 64, 48, 57, 40,
853 38, 16, -10, -28, 92, 22, -18, 29, -10, 5, -13, 49, 19, 24, 70, 34,
854 61, 48, 30, 14, -6, 25, 58, 33, 42, 60, 67, 17, 54, 17, 22, 30,
855 67, 44, -9, 50, -11, 43, 40, 32, 59, 82, 13, 49, -14, 55, 60, 36,
856 48, 49, 31, 47, 15, 12, 4, 65, 1, 23, 29, 39, 45, -2, 84, 69,
857 0, 72, 37, 57, 27, 41, -15, -16, 35, 31, 14, 61, 24, 0, 27, 24,
858 16, 41, 55, 34, 53, 9, 56, 12, 25, 29, 53, 5, 20, -20, -8, 20,
859 13, 28, -3, 78, 38, 16, 11, 62, 46, 29, 21, 24, 46, 65, 43, -23,
860 89, 18, 74, 21, 38, -12, 19, 12, -19, 8, 15, 33, 4, 57, 9, -8,
861 36, 35, 26, 28, 7, 83, 63, 79, 75, 11, 3, 87, 37, 47, 34, 40,
862 39, 19, 20, 42, 27, 34, 39, 77, 13, 42, 59, 64, 45, -1, 32, 37,
863 45, -5, 53, -6, 7, 36, 50, 23, 6, 32, 9, -21, 18, 71, 27, 52,
864 -25, 31, 35, 42, -1, 68, 63, 52, 26, 43, 66, 37, 41, 25, 40, 70,
867 /** Noise generation functions.
868 * I'm not sure what these are for - they seem to be some kind of pseudorandom
869 * sequence generators, used to generate noise data which is used when the
870 * channels are rematrixed. I'm not sure if they provide a practical benefit
871 * to compression, or just obfuscate the decoder. Are they for some kind of
874 /** Generate two channels of noise, used in the matrix when
875 * restart sync word == 0x31ea. */
877 static void generate_2_noise_channels(MLPDecodeContext *m, unsigned int substr)
879 SubStream *s = &m->substream[substr];
881 uint32_t seed = s->noisegen_seed;
882 unsigned int maxchan = s->max_matrix_channel;
884 for (i = 0; i < s->blockpos; i++) {
885 uint16_t seed_shr7 = seed >> 7;
886 m->sample_buffer[i][maxchan+1] = ((int8_t)(seed >> 15)) << s->noise_shift;
887 m->sample_buffer[i][maxchan+2] = ((int8_t) seed_shr7) << s->noise_shift;
889 seed = (seed << 16) ^ seed_shr7 ^ (seed_shr7 << 5);
892 s->noisegen_seed = seed;
895 /** Generate a block of noise, used when restart sync word == 0x31eb. */
897 static void fill_noise_buffer(MLPDecodeContext *m, unsigned int substr)
899 SubStream *s = &m->substream[substr];
901 uint32_t seed = s->noisegen_seed;
903 for (i = 0; i < m->access_unit_size_pow2; i++) {
904 uint8_t seed_shr15 = seed >> 15;
905 m->noise_buffer[i] = noise_table[seed_shr15];
906 seed = (seed << 8) ^ seed_shr15 ^ (seed_shr15 << 5);
909 s->noisegen_seed = seed;
913 /** Apply the channel matrices in turn to reconstruct the original audio
916 static void rematrix_channels(MLPDecodeContext *m, unsigned int substr)
918 SubStream *s = &m->substream[substr];
919 unsigned int mat, src_ch, i;
920 unsigned int maxchan;
922 maxchan = s->max_matrix_channel;
923 if (!s->noise_type) {
924 generate_2_noise_channels(m, substr);
927 fill_noise_buffer(m, substr);
930 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
931 int matrix_noise_shift = s->matrix_noise_shift[mat];
932 unsigned int dest_ch = s->matrix_out_ch[mat];
933 int32_t mask = MSB_MASK(s->quant_step_size[dest_ch]);
934 int32_t *coeffs = s->matrix_coeff[mat];
935 int index = s->num_primitive_matrices - mat;
936 int index2 = 2 * index + 1;
938 /* TODO: DSPContext? */
940 for (i = 0; i < s->blockpos; i++) {
941 int32_t bypassed_lsb = m->bypassed_lsbs[i][mat];
942 int32_t *samples = m->sample_buffer[i];
945 for (src_ch = 0; src_ch <= maxchan; src_ch++)
946 accum += (int64_t) samples[src_ch] * coeffs[src_ch];
948 if (matrix_noise_shift) {
949 index &= m->access_unit_size_pow2 - 1;
950 accum += m->noise_buffer[index] << (matrix_noise_shift + 7);
954 samples[dest_ch] = ((accum >> 14) & mask) + bypassed_lsb;
959 /** Write the audio data into the output buffer. */
961 static int output_data(MLPDecodeContext *m, unsigned int substr,
962 void *data, int *got_frame_ptr)
964 AVCodecContext *avctx = m->avctx;
965 SubStream *s = &m->substream[substr];
966 unsigned int i, out_ch = 0;
970 int is32 = (m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
972 if (m->avctx->channels != s->max_matrix_channel + 1) {
973 av_log(m->avctx, AV_LOG_ERROR, "channel count mismatch\n");
974 return AVERROR_INVALIDDATA;
977 /* get output buffer */
978 m->frame.nb_samples = s->blockpos;
979 if ((ret = avctx->get_buffer(avctx, &m->frame)) < 0) {
980 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
983 data_32 = (int32_t *)m->frame.data[0];
984 data_16 = (int16_t *)m->frame.data[0];
986 for (i = 0; i < s->blockpos; i++) {
987 for (out_ch = 0; out_ch <= s->max_matrix_channel; out_ch++) {
988 int mat_ch = s->ch_assign[out_ch];
989 int32_t sample = m->sample_buffer[i][mat_ch]
990 << s->output_shift[mat_ch];
991 s->lossless_check_data ^= (sample & 0xffffff) << mat_ch;
992 if (is32) *data_32++ = sample << 8;
993 else *data_16++ = sample >> 8;
998 *(AVFrame *)data = m->frame;
1003 /** Read an access unit from the stream.
1004 * @return negative on error, 0 if not enough data is present in the input stream,
1005 * otherwise the number of bytes consumed. */
1007 static int read_access_unit(AVCodecContext *avctx, void* data,
1008 int *got_frame_ptr, AVPacket *avpkt)
1010 const uint8_t *buf = avpkt->data;
1011 int buf_size = avpkt->size;
1012 MLPDecodeContext *m = avctx->priv_data;
1014 unsigned int length, substr;
1015 unsigned int substream_start;
1016 unsigned int header_size = 4;
1017 unsigned int substr_header_size = 0;
1018 uint8_t substream_parity_present[MAX_SUBSTREAMS];
1019 uint16_t substream_data_len[MAX_SUBSTREAMS];
1020 uint8_t parity_bits;
1026 length = (AV_RB16(buf) & 0xfff) * 2;
1028 if (length < 4 || length > buf_size)
1029 return AVERROR_INVALIDDATA;
1031 init_get_bits(&gb, (buf + 4), (length - 4) * 8);
1033 m->is_major_sync_unit = 0;
1034 if (show_bits_long(&gb, 31) == (0xf8726fba >> 1)) {
1035 if (read_major_sync(m, &gb) < 0)
1037 m->is_major_sync_unit = 1;
1041 if (!m->params_valid) {
1042 av_log(m->avctx, AV_LOG_WARNING,
1043 "Stream parameters not seen; skipping frame.\n");
1048 substream_start = 0;
1050 for (substr = 0; substr < m->num_substreams; substr++) {
1051 int extraword_present, checkdata_present, end, nonrestart_substr;
1053 extraword_present = get_bits1(&gb);
1054 nonrestart_substr = get_bits1(&gb);
1055 checkdata_present = get_bits1(&gb);
1058 end = get_bits(&gb, 12) * 2;
1060 substr_header_size += 2;
1062 if (extraword_present) {
1063 if (m->avctx->codec_id == CODEC_ID_MLP) {
1064 av_log(m->avctx, AV_LOG_ERROR, "There must be no extraword for MLP.\n");
1068 substr_header_size += 2;
1071 if (!(nonrestart_substr ^ m->is_major_sync_unit)) {
1072 av_log(m->avctx, AV_LOG_ERROR, "Invalid nonrestart_substr.\n");
1076 if (end + header_size + substr_header_size > length) {
1077 av_log(m->avctx, AV_LOG_ERROR,
1078 "Indicated length of substream %d data goes off end of "
1079 "packet.\n", substr);
1080 end = length - header_size - substr_header_size;
1083 if (end < substream_start) {
1084 av_log(avctx, AV_LOG_ERROR,
1085 "Indicated end offset of substream %d data "
1086 "is smaller than calculated start offset.\n",
1091 if (substr > m->max_decoded_substream)
1094 substream_parity_present[substr] = checkdata_present;
1095 substream_data_len[substr] = end - substream_start;
1096 substream_start = end;
1099 parity_bits = ff_mlp_calculate_parity(buf, 4);
1100 parity_bits ^= ff_mlp_calculate_parity(buf + header_size, substr_header_size);
1102 if ((((parity_bits >> 4) ^ parity_bits) & 0xF) != 0xF) {
1103 av_log(avctx, AV_LOG_ERROR, "Parity check failed.\n");
1107 buf += header_size + substr_header_size;
1109 for (substr = 0; substr <= m->max_decoded_substream; substr++) {
1110 SubStream *s = &m->substream[substr];
1111 init_get_bits(&gb, buf, substream_data_len[substr] * 8);
1113 m->matrix_changed = 0;
1114 memset(m->filter_changed, 0, sizeof(m->filter_changed));
1118 if (get_bits1(&gb)) {
1119 if (get_bits1(&gb)) {
1120 /* A restart header should be present. */
1121 if (read_restart_header(m, &gb, buf, substr) < 0)
1123 s->restart_seen = 1;
1126 if (!s->restart_seen)
1128 if (read_decoding_params(m, &gb, substr) < 0)
1132 if (!s->restart_seen)
1135 if ((ret = read_block_data(m, &gb, substr)) < 0)
1138 if (get_bits_count(&gb) >= substream_data_len[substr] * 8)
1139 goto substream_length_mismatch;
1141 } while (!get_bits1(&gb));
1143 skip_bits(&gb, (-get_bits_count(&gb)) & 15);
1145 if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 32) {
1148 if (get_bits(&gb, 16) != 0xD234)
1149 return AVERROR_INVALIDDATA;
1151 shorten_by = get_bits(&gb, 16);
1152 if (m->avctx->codec_id == CODEC_ID_TRUEHD && shorten_by & 0x2000)
1153 s->blockpos -= FFMIN(shorten_by & 0x1FFF, s->blockpos);
1154 else if (m->avctx->codec_id == CODEC_ID_MLP && shorten_by != 0xD234)
1155 return AVERROR_INVALIDDATA;
1157 if (substr == m->max_decoded_substream)
1158 av_log(m->avctx, AV_LOG_INFO, "End of stream indicated.\n");
1161 if (substream_parity_present[substr]) {
1162 uint8_t parity, checksum;
1164 if (substream_data_len[substr] * 8 - get_bits_count(&gb) != 16)
1165 goto substream_length_mismatch;
1167 parity = ff_mlp_calculate_parity(buf, substream_data_len[substr] - 2);
1168 checksum = ff_mlp_checksum8 (buf, substream_data_len[substr] - 2);
1170 if ((get_bits(&gb, 8) ^ parity) != 0xa9 )
1171 av_log(m->avctx, AV_LOG_ERROR, "Substream %d parity check failed.\n", substr);
1172 if ( get_bits(&gb, 8) != checksum)
1173 av_log(m->avctx, AV_LOG_ERROR, "Substream %d checksum failed.\n" , substr);
1176 if (substream_data_len[substr] * 8 != get_bits_count(&gb))
1177 goto substream_length_mismatch;
1180 if (!s->restart_seen)
1181 av_log(m->avctx, AV_LOG_ERROR,
1182 "No restart header present in substream %d.\n", substr);
1184 buf += substream_data_len[substr];
1187 rematrix_channels(m, m->max_decoded_substream);
1189 if ((ret = output_data(m, m->max_decoded_substream, data, got_frame_ptr)) < 0)
1194 substream_length_mismatch:
1195 av_log(m->avctx, AV_LOG_ERROR, "substream %d length mismatch\n", substr);
1196 return AVERROR_INVALIDDATA;
1199 m->params_valid = 0;
1200 return AVERROR_INVALIDDATA;
1203 AVCodec ff_mlp_decoder = {
1205 .type = AVMEDIA_TYPE_AUDIO,
1207 .priv_data_size = sizeof(MLPDecodeContext),
1208 .init = mlp_decode_init,
1209 .decode = read_access_unit,
1210 .capabilities = CODEC_CAP_DR1,
1211 .long_name = NULL_IF_CONFIG_SMALL("MLP (Meridian Lossless Packing)"),
1214 #if CONFIG_TRUEHD_DECODER
1215 AVCodec ff_truehd_decoder = {
1217 .type = AVMEDIA_TYPE_AUDIO,
1218 .id = CODEC_ID_TRUEHD,
1219 .priv_data_size = sizeof(MLPDecodeContext),
1220 .init = mlp_decode_init,
1221 .decode = read_access_unit,
1222 .capabilities = CODEC_CAP_DR1,
1223 .long_name = NULL_IF_CONFIG_SMALL("TrueHD"),
1225 #endif /* CONFIG_TRUEHD_DECODER */