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
30 #include "libavutil/intreadwrite.h"
31 #include "libavutil/channel_layout.h"
34 #include "libavutil/crc.h"
36 #include "mlp_parser.h"
40 /** number of bits used for VLC lookup - longest Huffman code is 9 */
43 typedef struct SubStream {
44 /// Set if a valid restart header has been read. Otherwise the substream cannot be decoded.
48 /** restart header data */
49 /// The type of noise to be used in the rematrix stage.
52 /// The index of the first channel coded in this substream.
54 /// The index of the last channel coded in this substream.
56 /// The number of channels input into the rematrix stage.
57 uint8_t max_matrix_channel;
58 /// For each channel output by the matrix, the output channel to map it to
59 uint8_t ch_assign[MAX_CHANNELS];
60 /// The channel layout for this substream
63 /// Channel coding parameters for channels in the substream
64 ChannelParams channel_params[MAX_CHANNELS];
66 /// The left shift applied to random noise in 0x31ea substreams.
68 /// The current seed value for the pseudorandom noise generator(s).
69 uint32_t noisegen_seed;
71 /// Set if the substream contains extra info to check the size of VLC blocks.
72 uint8_t data_check_present;
74 /// Bitmask of which parameter sets are conveyed in a decoding parameter block.
75 uint8_t param_presence_flags;
76 #define PARAM_BLOCKSIZE (1 << 7)
77 #define PARAM_MATRIX (1 << 6)
78 #define PARAM_OUTSHIFT (1 << 5)
79 #define PARAM_QUANTSTEP (1 << 4)
80 #define PARAM_FIR (1 << 3)
81 #define PARAM_IIR (1 << 2)
82 #define PARAM_HUFFOFFSET (1 << 1)
83 #define PARAM_PRESENCE (1 << 0)
89 /// Number of matrices to be applied.
90 uint8_t num_primitive_matrices;
92 /// matrix output channel
93 uint8_t matrix_out_ch[MAX_MATRICES];
95 /// Whether the LSBs of the matrix output are encoded in the bitstream.
96 uint8_t lsb_bypass[MAX_MATRICES];
97 /// Matrix coefficients, stored as 2.14 fixed point.
98 int32_t matrix_coeff[MAX_MATRICES][MAX_CHANNELS];
99 /// Left shift to apply to noise values in 0x31eb substreams.
100 uint8_t matrix_noise_shift[MAX_MATRICES];
103 /// Left shift to apply to Huffman-decoded residuals.
104 uint8_t quant_step_size[MAX_CHANNELS];
106 /// number of PCM samples in current audio block
108 /// Number of PCM samples decoded so far in this frame.
111 /// Left shift to apply to decoded PCM values to get final 24-bit output.
112 int8_t output_shift[MAX_CHANNELS];
114 /// Running XOR of all output samples.
115 int32_t lossless_check_data;
119 typedef struct MLPDecodeContext {
120 AVCodecContext *avctx;
122 /// Current access unit being read has a major sync.
123 int is_major_sync_unit;
125 /// Set if a valid major sync block has been read. Otherwise no decoding is possible.
126 uint8_t params_valid;
128 /// Number of substreams contained within this stream.
129 uint8_t num_substreams;
131 /// Index of the last substream to decode - further substreams are skipped.
132 uint8_t max_decoded_substream;
134 /// Stream needs channel reordering to comply with FFmpeg's channel order
135 uint8_t needs_reordering;
137 /// number of PCM samples contained in each frame
138 int access_unit_size;
139 /// next power of two above the number of samples in each frame
140 int access_unit_size_pow2;
142 SubStream substream[MAX_SUBSTREAMS];
145 int filter_changed[MAX_CHANNELS][NUM_FILTERS];
147 int8_t noise_buffer[MAX_BLOCKSIZE_POW2];
148 int8_t bypassed_lsbs[MAX_BLOCKSIZE][MAX_CHANNELS];
149 int32_t sample_buffer[MAX_BLOCKSIZE][MAX_CHANNELS];
154 static const uint64_t thd_channel_order[] = {
155 AV_CH_FRONT_LEFT, AV_CH_FRONT_RIGHT, // LR
156 AV_CH_FRONT_CENTER, // C
157 AV_CH_LOW_FREQUENCY, // LFE
158 AV_CH_SIDE_LEFT, AV_CH_SIDE_RIGHT, // LRs
159 AV_CH_TOP_FRONT_LEFT, AV_CH_TOP_FRONT_RIGHT, // LRvh
160 AV_CH_FRONT_LEFT_OF_CENTER, AV_CH_FRONT_RIGHT_OF_CENTER, // LRc
161 AV_CH_BACK_LEFT, AV_CH_BACK_RIGHT, // LRrs
162 AV_CH_BACK_CENTER, // Cs
163 AV_CH_TOP_CENTER, // Ts
164 AV_CH_SURROUND_DIRECT_LEFT, AV_CH_SURROUND_DIRECT_RIGHT, // LRsd
165 AV_CH_WIDE_LEFT, AV_CH_WIDE_RIGHT, // LRw
166 AV_CH_TOP_FRONT_CENTER, // Cvh
167 AV_CH_LOW_FREQUENCY_2, // LFE2
170 static uint64_t thd_channel_layout_extract_channel(uint64_t channel_layout,
175 if (av_get_channel_layout_nb_channels(channel_layout) <= index)
178 for (i = 0; i < FF_ARRAY_ELEMS(thd_channel_order); i++)
179 if (channel_layout & thd_channel_order[i] && !index--)
180 return thd_channel_order[i];
184 static VLC huff_vlc[3];
186 /** Initialize static data, constant between all invocations of the codec. */
188 static av_cold void init_static(void)
190 if (!huff_vlc[0].bits) {
191 INIT_VLC_STATIC(&huff_vlc[0], VLC_BITS, 18,
192 &ff_mlp_huffman_tables[0][0][1], 2, 1,
193 &ff_mlp_huffman_tables[0][0][0], 2, 1, 512);
194 INIT_VLC_STATIC(&huff_vlc[1], VLC_BITS, 16,
195 &ff_mlp_huffman_tables[1][0][1], 2, 1,
196 &ff_mlp_huffman_tables[1][0][0], 2, 1, 512);
197 INIT_VLC_STATIC(&huff_vlc[2], VLC_BITS, 15,
198 &ff_mlp_huffman_tables[2][0][1], 2, 1,
199 &ff_mlp_huffman_tables[2][0][0], 2, 1, 512);
205 static inline int32_t calculate_sign_huff(MLPDecodeContext *m,
206 unsigned int substr, unsigned int ch)
208 SubStream *s = &m->substream[substr];
209 ChannelParams *cp = &s->channel_params[ch];
210 int lsb_bits = cp->huff_lsbs - s->quant_step_size[ch];
211 int sign_shift = lsb_bits + (cp->codebook ? 2 - cp->codebook : -1);
212 int32_t sign_huff_offset = cp->huff_offset;
214 if (cp->codebook > 0)
215 sign_huff_offset -= 7 << lsb_bits;
218 sign_huff_offset -= 1 << sign_shift;
220 return sign_huff_offset;
223 /** Read a sample, consisting of either, both or neither of entropy-coded MSBs
226 static inline int read_huff_channels(MLPDecodeContext *m, GetBitContext *gbp,
227 unsigned int substr, unsigned int pos)
229 SubStream *s = &m->substream[substr];
230 unsigned int mat, channel;
232 for (mat = 0; mat < s->num_primitive_matrices; mat++)
233 if (s->lsb_bypass[mat])
234 m->bypassed_lsbs[pos + s->blockpos][mat] = get_bits1(gbp);
236 for (channel = s->min_channel; channel <= s->max_channel; channel++) {
237 ChannelParams *cp = &s->channel_params[channel];
238 int codebook = cp->codebook;
239 int quant_step_size = s->quant_step_size[channel];
240 int lsb_bits = cp->huff_lsbs - quant_step_size;
244 result = get_vlc2(gbp, huff_vlc[codebook-1].table,
245 VLC_BITS, (9 + VLC_BITS - 1) / VLC_BITS);
248 return AVERROR_INVALIDDATA;
251 result = (result << lsb_bits) + get_bits(gbp, lsb_bits);
253 result += cp->sign_huff_offset;
254 result <<= quant_step_size;
256 m->sample_buffer[pos + s->blockpos][channel] = result;
262 static av_cold int mlp_decode_init(AVCodecContext *avctx)
264 MLPDecodeContext *m = avctx->priv_data;
269 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
270 m->substream[substr].lossless_check_data = 0xffffffff;
271 ff_mlpdsp_init(&m->dsp);
276 /** Read a major sync info header - contains high level information about
277 * the stream - sample rate, channel arrangement etc. Most of this
278 * information is not actually necessary for decoding, only for playback.
281 static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
286 if ((ret = ff_mlp_read_major_sync(m->avctx, &mh, gb)) != 0)
289 if (mh.group1_bits == 0) {
290 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown bits per sample\n");
291 return AVERROR_INVALIDDATA;
293 if (mh.group2_bits > mh.group1_bits) {
294 av_log(m->avctx, AV_LOG_ERROR,
295 "Channel group 2 cannot have more bits per sample than group 1.\n");
296 return AVERROR_INVALIDDATA;
299 if (mh.group2_samplerate && mh.group2_samplerate != mh.group1_samplerate) {
300 av_log(m->avctx, AV_LOG_ERROR,
301 "Channel groups with differing sample rates are not currently supported.\n");
302 return AVERROR_INVALIDDATA;
305 if (mh.group1_samplerate == 0) {
306 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown sampling rate\n");
307 return AVERROR_INVALIDDATA;
309 if (mh.group1_samplerate > MAX_SAMPLERATE) {
310 av_log(m->avctx, AV_LOG_ERROR,
311 "Sampling rate %d is greater than the supported maximum (%d).\n",
312 mh.group1_samplerate, MAX_SAMPLERATE);
313 return AVERROR_INVALIDDATA;
315 if (mh.access_unit_size > MAX_BLOCKSIZE) {
316 av_log(m->avctx, AV_LOG_ERROR,
317 "Block size %d is greater than the supported maximum (%d).\n",
318 mh.access_unit_size, MAX_BLOCKSIZE);
319 return AVERROR_INVALIDDATA;
321 if (mh.access_unit_size_pow2 > MAX_BLOCKSIZE_POW2) {
322 av_log(m->avctx, AV_LOG_ERROR,
323 "Block size pow2 %d is greater than the supported maximum (%d).\n",
324 mh.access_unit_size_pow2, MAX_BLOCKSIZE_POW2);
325 return AVERROR_INVALIDDATA;
328 if (mh.num_substreams == 0)
329 return AVERROR_INVALIDDATA;
330 if (m->avctx->codec_id == AV_CODEC_ID_MLP && mh.num_substreams > 2) {
331 av_log(m->avctx, AV_LOG_ERROR, "MLP only supports up to 2 substreams.\n");
332 return AVERROR_INVALIDDATA;
334 if (mh.num_substreams > MAX_SUBSTREAMS) {
335 av_log_ask_for_sample(m->avctx,
336 "Number of substreams %d is larger than the maximum supported "
337 "by the decoder.\n", mh.num_substreams);
338 return AVERROR_PATCHWELCOME;
341 m->access_unit_size = mh.access_unit_size;
342 m->access_unit_size_pow2 = mh.access_unit_size_pow2;
344 m->num_substreams = mh.num_substreams;
345 m->max_decoded_substream = m->num_substreams - 1;
347 m->avctx->sample_rate = mh.group1_samplerate;
348 m->avctx->frame_size = mh.access_unit_size;
350 m->avctx->bits_per_raw_sample = mh.group1_bits;
351 if (mh.group1_bits > 16)
352 m->avctx->sample_fmt = AV_SAMPLE_FMT_S32;
354 m->avctx->sample_fmt = AV_SAMPLE_FMT_S16;
357 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
358 m->substream[substr].restart_seen = 0;
360 /* Set the layout for each substream. When there's more than one, the first
361 * substream is Stereo. Subsequent substreams' layouts are indicated in the
363 if (m->avctx->codec_id == AV_CODEC_ID_MLP) {
364 if ((substr = (mh.num_substreams > 1)))
365 m->substream[0].ch_layout = AV_CH_LAYOUT_STEREO;
366 m->substream[substr].ch_layout = mh.channel_layout_mlp;
368 if ((substr = (mh.num_substreams > 1)))
369 m->substream[0].ch_layout = AV_CH_LAYOUT_STEREO;
370 if (mh.num_substreams > 2)
371 if (mh.channel_layout_thd_stream2)
372 m->substream[2].ch_layout = mh.channel_layout_thd_stream2;
374 m->substream[2].ch_layout = mh.channel_layout_thd_stream1;
375 m->substream[substr].ch_layout = mh.channel_layout_thd_stream1;
377 if (m->avctx->channels<=2 && m->substream[substr].ch_layout == AV_CH_LAYOUT_MONO && m->max_decoded_substream == 1) {
378 av_log(m->avctx, AV_LOG_DEBUG, "Mono stream with 2 substreams, ignoring 2nd\n");
379 m->max_decoded_substream = 0;
380 if (m->avctx->channels==2)
381 m->avctx->channel_layout = AV_CH_LAYOUT_STEREO;
385 m->needs_reordering = mh.channel_arrangement >= 18 && mh.channel_arrangement <= 20;
390 /** Read a restart header from a block in a substream. This contains parameters
391 * required to decode the audio that do not change very often. Generally
392 * (always) present only in blocks following a major sync. */
394 static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
395 const uint8_t *buf, unsigned int substr)
397 SubStream *s = &m->substream[substr];
401 uint8_t lossless_check;
402 int start_count = get_bits_count(gbp);
403 const int max_matrix_channel = m->avctx->codec_id == AV_CODEC_ID_MLP
404 ? MAX_MATRIX_CHANNEL_MLP
405 : MAX_MATRIX_CHANNEL_TRUEHD;
406 int max_channel, min_channel, matrix_channel;
408 sync_word = get_bits(gbp, 13);
410 if (sync_word != 0x31ea >> 1) {
411 av_log(m->avctx, AV_LOG_ERROR,
412 "restart header sync incorrect (got 0x%04x)\n", sync_word);
413 return AVERROR_INVALIDDATA;
416 s->noise_type = get_bits1(gbp);
418 if (m->avctx->codec_id == AV_CODEC_ID_MLP && s->noise_type) {
419 av_log(m->avctx, AV_LOG_ERROR, "MLP must have 0x31ea sync word.\n");
420 return AVERROR_INVALIDDATA;
423 skip_bits(gbp, 16); /* Output timestamp */
425 min_channel = get_bits(gbp, 4);
426 max_channel = get_bits(gbp, 4);
427 matrix_channel = get_bits(gbp, 4);
429 if (matrix_channel > max_matrix_channel) {
430 av_log(m->avctx, AV_LOG_ERROR,
431 "Max matrix channel cannot be greater than %d.\n",
433 return AVERROR_INVALIDDATA;
436 if (max_channel != matrix_channel) {
437 av_log(m->avctx, AV_LOG_ERROR,
438 "Max channel must be equal max matrix channel.\n");
439 return AVERROR_INVALIDDATA;
442 /* This should happen for TrueHD streams with >6 channels and MLP's noise
443 * type. It is not yet known if this is allowed. */
444 if (max_channel > MAX_MATRIX_CHANNEL_MLP && !s->noise_type) {
445 av_log_ask_for_sample(m->avctx,
446 "Number of channels %d is larger than the maximum supported "
447 "by the decoder.\n", max_channel + 2);
448 return AVERROR_PATCHWELCOME;
451 if (min_channel > max_channel) {
452 av_log(m->avctx, AV_LOG_ERROR,
453 "Substream min channel cannot be greater than max channel.\n");
454 return AVERROR_INVALIDDATA;
457 s->min_channel = min_channel;
458 s->max_channel = max_channel;
459 s->max_matrix_channel = matrix_channel;
461 #if FF_API_REQUEST_CHANNELS
462 if (m->avctx->request_channels > 0 &&
463 m->avctx->request_channels <= s->max_channel + 1 &&
464 m->max_decoded_substream > substr) {
465 av_log(m->avctx, AV_LOG_DEBUG,
466 "Extracting %d-channel downmix from substream %d. "
467 "Further substreams will be skipped.\n",
468 s->max_channel + 1, substr);
469 m->max_decoded_substream = substr;
472 if (m->avctx->request_channel_layout == s->ch_layout &&
473 m->max_decoded_substream > substr) {
474 av_log(m->avctx, AV_LOG_DEBUG,
475 "Extracting %d-channel downmix (0x%"PRIx64") from substream %d. "
476 "Further substreams will be skipped.\n",
477 s->max_channel + 1, s->ch_layout, substr);
478 m->max_decoded_substream = substr;
481 s->noise_shift = get_bits(gbp, 4);
482 s->noisegen_seed = get_bits(gbp, 23);
486 s->data_check_present = get_bits1(gbp);
487 lossless_check = get_bits(gbp, 8);
488 if (substr == m->max_decoded_substream
489 && s->lossless_check_data != 0xffffffff) {
490 tmp = xor_32_to_8(s->lossless_check_data);
491 if (tmp != lossless_check)
492 av_log(m->avctx, AV_LOG_WARNING,
493 "Lossless check failed - expected %02x, calculated %02x.\n",
494 lossless_check, tmp);
499 memset(s->ch_assign, 0, sizeof(s->ch_assign));
501 for (ch = 0; ch <= s->max_matrix_channel; ch++) {
502 int ch_assign = get_bits(gbp, 6);
503 if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD) {
504 uint64_t channel = thd_channel_layout_extract_channel(s->ch_layout,
506 ch_assign = av_get_channel_layout_channel_index(s->ch_layout,
509 if ((unsigned)ch_assign > s->max_matrix_channel) {
510 av_log_ask_for_sample(m->avctx,
511 "Assignment of matrix channel %d to invalid output channel %d.\n",
513 return AVERROR_PATCHWELCOME;
515 s->ch_assign[ch_assign] = ch;
518 checksum = ff_mlp_restart_checksum(buf, get_bits_count(gbp) - start_count);
520 if (checksum != get_bits(gbp, 8))
521 av_log(m->avctx, AV_LOG_ERROR, "restart header checksum error\n");
523 /* Set default decoding parameters. */
524 s->param_presence_flags = 0xff;
525 s->num_primitive_matrices = 0;
527 s->lossless_check_data = 0;
529 memset(s->output_shift , 0, sizeof(s->output_shift ));
530 memset(s->quant_step_size, 0, sizeof(s->quant_step_size));
532 for (ch = s->min_channel; ch <= s->max_channel; ch++) {
533 ChannelParams *cp = &s->channel_params[ch];
534 cp->filter_params[FIR].order = 0;
535 cp->filter_params[IIR].order = 0;
536 cp->filter_params[FIR].shift = 0;
537 cp->filter_params[IIR].shift = 0;
539 /* Default audio coding is 24-bit raw PCM. */
541 cp->sign_huff_offset = (-1) << 23;
546 if (substr == m->max_decoded_substream) {
547 m->avctx->channels = s->max_matrix_channel + 1;
548 m->avctx->channel_layout = s->ch_layout;
550 if (m->avctx->codec_id == AV_CODEC_ID_MLP && m->needs_reordering) {
551 if (m->avctx->channel_layout == (AV_CH_LAYOUT_QUAD|AV_CH_LOW_FREQUENCY) ||
552 m->avctx->channel_layout == AV_CH_LAYOUT_5POINT0_BACK) {
553 int i = s->ch_assign[4];
554 s->ch_assign[4] = s->ch_assign[3];
555 s->ch_assign[3] = s->ch_assign[2];
557 } else if (m->avctx->channel_layout == AV_CH_LAYOUT_5POINT1_BACK) {
558 FFSWAP(int, s->ch_assign[2], s->ch_assign[4]);
559 FFSWAP(int, s->ch_assign[3], s->ch_assign[5]);
568 /** Read parameters for one of the prediction filters. */
570 static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
571 unsigned int substr, unsigned int channel,
574 SubStream *s = &m->substream[substr];
575 FilterParams *fp = &s->channel_params[channel].filter_params[filter];
576 const int max_order = filter ? MAX_IIR_ORDER : MAX_FIR_ORDER;
577 const char fchar = filter ? 'I' : 'F';
580 // Filter is 0 for FIR, 1 for IIR.
581 av_assert0(filter < 2);
583 if (m->filter_changed[channel][filter]++ > 1) {
584 av_log(m->avctx, AV_LOG_ERROR, "Filters may change only once per access unit.\n");
585 return AVERROR_INVALIDDATA;
588 order = get_bits(gbp, 4);
589 if (order > max_order) {
590 av_log(m->avctx, AV_LOG_ERROR,
591 "%cIR filter order %d is greater than maximum %d.\n",
592 fchar, order, max_order);
593 return AVERROR_INVALIDDATA;
598 int32_t *fcoeff = s->channel_params[channel].coeff[filter];
599 int coeff_bits, coeff_shift;
601 fp->shift = get_bits(gbp, 4);
603 coeff_bits = get_bits(gbp, 5);
604 coeff_shift = get_bits(gbp, 3);
605 if (coeff_bits < 1 || coeff_bits > 16) {
606 av_log(m->avctx, AV_LOG_ERROR,
607 "%cIR filter coeff_bits must be between 1 and 16.\n",
609 return AVERROR_INVALIDDATA;
611 if (coeff_bits + coeff_shift > 16) {
612 av_log(m->avctx, AV_LOG_ERROR,
613 "Sum of coeff_bits and coeff_shift for %cIR filter must be 16 or less.\n",
615 return AVERROR_INVALIDDATA;
618 for (i = 0; i < order; i++)
619 fcoeff[i] = get_sbits(gbp, coeff_bits) << coeff_shift;
621 if (get_bits1(gbp)) {
622 int state_bits, state_shift;
625 av_log(m->avctx, AV_LOG_ERROR,
626 "FIR filter has state data specified.\n");
627 return AVERROR_INVALIDDATA;
630 state_bits = get_bits(gbp, 4);
631 state_shift = get_bits(gbp, 4);
633 /* TODO: Check validity of state data. */
635 for (i = 0; i < order; i++)
636 fp->state[i] = get_sbits(gbp, state_bits) << state_shift;
643 /** Read parameters for primitive matrices. */
645 static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitContext *gbp)
647 SubStream *s = &m->substream[substr];
648 unsigned int mat, ch;
649 const int max_primitive_matrices = m->avctx->codec_id == AV_CODEC_ID_MLP
651 : MAX_MATRICES_TRUEHD;
653 if (m->matrix_changed++ > 1) {
654 av_log(m->avctx, AV_LOG_ERROR, "Matrices may change only once per access unit.\n");
655 return AVERROR_INVALIDDATA;
658 s->num_primitive_matrices = get_bits(gbp, 4);
660 if (s->num_primitive_matrices > max_primitive_matrices) {
661 av_log(m->avctx, AV_LOG_ERROR,
662 "Number of primitive matrices cannot be greater than %d.\n",
663 max_primitive_matrices);
664 return AVERROR_INVALIDDATA;
667 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
668 int frac_bits, max_chan;
669 s->matrix_out_ch[mat] = get_bits(gbp, 4);
670 frac_bits = get_bits(gbp, 4);
671 s->lsb_bypass [mat] = get_bits1(gbp);
673 if (s->matrix_out_ch[mat] > s->max_matrix_channel) {
674 av_log(m->avctx, AV_LOG_ERROR,
675 "Invalid channel %d specified as output from matrix.\n",
676 s->matrix_out_ch[mat]);
677 return AVERROR_INVALIDDATA;
679 if (frac_bits > 14) {
680 av_log(m->avctx, AV_LOG_ERROR,
681 "Too many fractional bits specified.\n");
682 return AVERROR_INVALIDDATA;
685 max_chan = s->max_matrix_channel;
689 for (ch = 0; ch <= max_chan; ch++) {
692 coeff_val = get_sbits(gbp, frac_bits + 2);
694 s->matrix_coeff[mat][ch] = coeff_val << (14 - frac_bits);
698 s->matrix_noise_shift[mat] = get_bits(gbp, 4);
700 s->matrix_noise_shift[mat] = 0;
706 /** Read channel parameters. */
708 static int read_channel_params(MLPDecodeContext *m, unsigned int substr,
709 GetBitContext *gbp, unsigned int ch)
711 SubStream *s = &m->substream[substr];
712 ChannelParams *cp = &s->channel_params[ch];
713 FilterParams *fir = &cp->filter_params[FIR];
714 FilterParams *iir = &cp->filter_params[IIR];
717 if (s->param_presence_flags & PARAM_FIR)
719 if ((ret = read_filter_params(m, gbp, substr, ch, FIR)) < 0)
722 if (s->param_presence_flags & PARAM_IIR)
724 if ((ret = read_filter_params(m, gbp, substr, ch, IIR)) < 0)
727 if (fir->order + iir->order > 8) {
728 av_log(m->avctx, AV_LOG_ERROR, "Total filter orders too high.\n");
729 return AVERROR_INVALIDDATA;
732 if (fir->order && iir->order &&
733 fir->shift != iir->shift) {
734 av_log(m->avctx, AV_LOG_ERROR,
735 "FIR and IIR filters must use the same precision.\n");
736 return AVERROR_INVALIDDATA;
738 /* The FIR and IIR filters must have the same precision.
739 * To simplify the filtering code, only the precision of the
740 * FIR filter is considered. If only the IIR filter is employed,
741 * the FIR filter precision is set to that of the IIR filter, so
742 * that the filtering code can use it. */
743 if (!fir->order && iir->order)
744 fir->shift = iir->shift;
746 if (s->param_presence_flags & PARAM_HUFFOFFSET)
748 cp->huff_offset = get_sbits(gbp, 15);
750 cp->codebook = get_bits(gbp, 2);
751 cp->huff_lsbs = get_bits(gbp, 5);
753 if (cp->huff_lsbs > 24) {
754 av_log(m->avctx, AV_LOG_ERROR, "Invalid huff_lsbs.\n");
756 return AVERROR_INVALIDDATA;
759 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
764 /** Read decoding parameters that change more often than those in the restart
767 static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
770 SubStream *s = &m->substream[substr];
774 if (s->param_presence_flags & PARAM_PRESENCE)
776 s->param_presence_flags = get_bits(gbp, 8);
778 if (s->param_presence_flags & PARAM_BLOCKSIZE)
779 if (get_bits1(gbp)) {
780 s->blocksize = get_bits(gbp, 9);
781 if (s->blocksize < 8 || s->blocksize > m->access_unit_size) {
782 av_log(m->avctx, AV_LOG_ERROR, "Invalid blocksize.\n");
784 return AVERROR_INVALIDDATA;
788 if (s->param_presence_flags & PARAM_MATRIX)
790 if ((ret = read_matrix_params(m, substr, gbp)) < 0)
793 if (s->param_presence_flags & PARAM_OUTSHIFT)
795 for (ch = 0; ch <= s->max_matrix_channel; ch++)
796 s->output_shift[ch] = get_sbits(gbp, 4);
798 if (s->param_presence_flags & PARAM_QUANTSTEP)
800 for (ch = 0; ch <= s->max_channel; ch++) {
801 ChannelParams *cp = &s->channel_params[ch];
803 s->quant_step_size[ch] = get_bits(gbp, 4);
805 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
808 for (ch = s->min_channel; ch <= s->max_channel; ch++)
810 if ((ret = read_channel_params(m, substr, gbp, ch)) < 0)
816 #define MSB_MASK(bits) (-1u << bits)
818 /** Generate PCM samples using the prediction filters and residual values
819 * read from the data stream, and update the filter state. */
821 static void filter_channel(MLPDecodeContext *m, unsigned int substr,
822 unsigned int channel)
824 SubStream *s = &m->substream[substr];
825 const int32_t *fircoeff = s->channel_params[channel].coeff[FIR];
826 int32_t state_buffer[NUM_FILTERS][MAX_BLOCKSIZE + MAX_FIR_ORDER];
827 int32_t *firbuf = state_buffer[FIR] + MAX_BLOCKSIZE;
828 int32_t *iirbuf = state_buffer[IIR] + MAX_BLOCKSIZE;
829 FilterParams *fir = &s->channel_params[channel].filter_params[FIR];
830 FilterParams *iir = &s->channel_params[channel].filter_params[IIR];
831 unsigned int filter_shift = fir->shift;
832 int32_t mask = MSB_MASK(s->quant_step_size[channel]);
834 memcpy(firbuf, fir->state, MAX_FIR_ORDER * sizeof(int32_t));
835 memcpy(iirbuf, iir->state, MAX_IIR_ORDER * sizeof(int32_t));
837 m->dsp.mlp_filter_channel(firbuf, fircoeff,
838 fir->order, iir->order,
839 filter_shift, mask, s->blocksize,
840 &m->sample_buffer[s->blockpos][channel]);
842 memcpy(fir->state, firbuf - s->blocksize, MAX_FIR_ORDER * sizeof(int32_t));
843 memcpy(iir->state, iirbuf - s->blocksize, MAX_IIR_ORDER * sizeof(int32_t));
846 /** Read a block of PCM residual data (or actual if no filtering active). */
848 static int read_block_data(MLPDecodeContext *m, GetBitContext *gbp,
851 SubStream *s = &m->substream[substr];
852 unsigned int i, ch, expected_stream_pos = 0;
855 if (s->data_check_present) {
856 expected_stream_pos = get_bits_count(gbp);
857 expected_stream_pos += get_bits(gbp, 16);
858 av_log_ask_for_sample(m->avctx, "This file contains some features "
859 "we have not tested yet.\n");
862 if (s->blockpos + s->blocksize > m->access_unit_size) {
863 av_log(m->avctx, AV_LOG_ERROR, "too many audio samples in frame\n");
864 return AVERROR_INVALIDDATA;
867 memset(&m->bypassed_lsbs[s->blockpos][0], 0,
868 s->blocksize * sizeof(m->bypassed_lsbs[0]));
870 for (i = 0; i < s->blocksize; i++)
871 if ((ret = read_huff_channels(m, gbp, substr, i)) < 0)
874 for (ch = s->min_channel; ch <= s->max_channel; ch++)
875 filter_channel(m, substr, ch);
877 s->blockpos += s->blocksize;
879 if (s->data_check_present) {
880 if (get_bits_count(gbp) != expected_stream_pos)
881 av_log(m->avctx, AV_LOG_ERROR, "block data length mismatch\n");
888 /** Data table used for TrueHD noise generation function. */
890 static const int8_t noise_table[256] = {
891 30, 51, 22, 54, 3, 7, -4, 38, 14, 55, 46, 81, 22, 58, -3, 2,
892 52, 31, -7, 51, 15, 44, 74, 30, 85, -17, 10, 33, 18, 80, 28, 62,
893 10, 32, 23, 69, 72, 26, 35, 17, 73, 60, 8, 56, 2, 6, -2, -5,
894 51, 4, 11, 50, 66, 76, 21, 44, 33, 47, 1, 26, 64, 48, 57, 40,
895 38, 16, -10, -28, 92, 22, -18, 29, -10, 5, -13, 49, 19, 24, 70, 34,
896 61, 48, 30, 14, -6, 25, 58, 33, 42, 60, 67, 17, 54, 17, 22, 30,
897 67, 44, -9, 50, -11, 43, 40, 32, 59, 82, 13, 49, -14, 55, 60, 36,
898 48, 49, 31, 47, 15, 12, 4, 65, 1, 23, 29, 39, 45, -2, 84, 69,
899 0, 72, 37, 57, 27, 41, -15, -16, 35, 31, 14, 61, 24, 0, 27, 24,
900 16, 41, 55, 34, 53, 9, 56, 12, 25, 29, 53, 5, 20, -20, -8, 20,
901 13, 28, -3, 78, 38, 16, 11, 62, 46, 29, 21, 24, 46, 65, 43, -23,
902 89, 18, 74, 21, 38, -12, 19, 12, -19, 8, 15, 33, 4, 57, 9, -8,
903 36, 35, 26, 28, 7, 83, 63, 79, 75, 11, 3, 87, 37, 47, 34, 40,
904 39, 19, 20, 42, 27, 34, 39, 77, 13, 42, 59, 64, 45, -1, 32, 37,
905 45, -5, 53, -6, 7, 36, 50, 23, 6, 32, 9, -21, 18, 71, 27, 52,
906 -25, 31, 35, 42, -1, 68, 63, 52, 26, 43, 66, 37, 41, 25, 40, 70,
909 /** Noise generation functions.
910 * I'm not sure what these are for - they seem to be some kind of pseudorandom
911 * sequence generators, used to generate noise data which is used when the
912 * channels are rematrixed. I'm not sure if they provide a practical benefit
913 * to compression, or just obfuscate the decoder. Are they for some kind of
916 /** Generate two channels of noise, used in the matrix when
917 * restart sync word == 0x31ea. */
919 static void generate_2_noise_channels(MLPDecodeContext *m, unsigned int substr)
921 SubStream *s = &m->substream[substr];
923 uint32_t seed = s->noisegen_seed;
924 unsigned int maxchan = s->max_matrix_channel;
926 for (i = 0; i < s->blockpos; i++) {
927 uint16_t seed_shr7 = seed >> 7;
928 m->sample_buffer[i][maxchan+1] = ((int8_t)(seed >> 15)) << s->noise_shift;
929 m->sample_buffer[i][maxchan+2] = ((int8_t) seed_shr7) << s->noise_shift;
931 seed = (seed << 16) ^ seed_shr7 ^ (seed_shr7 << 5);
934 s->noisegen_seed = seed;
937 /** Generate a block of noise, used when restart sync word == 0x31eb. */
939 static void fill_noise_buffer(MLPDecodeContext *m, unsigned int substr)
941 SubStream *s = &m->substream[substr];
943 uint32_t seed = s->noisegen_seed;
945 for (i = 0; i < m->access_unit_size_pow2; i++) {
946 uint8_t seed_shr15 = seed >> 15;
947 m->noise_buffer[i] = noise_table[seed_shr15];
948 seed = (seed << 8) ^ seed_shr15 ^ (seed_shr15 << 5);
951 s->noisegen_seed = seed;
955 /** Apply the channel matrices in turn to reconstruct the original audio
958 static void rematrix_channels(MLPDecodeContext *m, unsigned int substr)
960 SubStream *s = &m->substream[substr];
961 unsigned int mat, src_ch, i;
962 unsigned int maxchan;
964 maxchan = s->max_matrix_channel;
965 if (!s->noise_type) {
966 generate_2_noise_channels(m, substr);
969 fill_noise_buffer(m, substr);
972 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
973 int matrix_noise_shift = s->matrix_noise_shift[mat];
974 unsigned int dest_ch = s->matrix_out_ch[mat];
975 int32_t mask = MSB_MASK(s->quant_step_size[dest_ch]);
976 int32_t *coeffs = s->matrix_coeff[mat];
977 int index = s->num_primitive_matrices - mat;
978 int index2 = 2 * index + 1;
980 /* TODO: DSPContext? */
982 for (i = 0; i < s->blockpos; i++) {
983 int32_t bypassed_lsb = m->bypassed_lsbs[i][mat];
984 int32_t *samples = m->sample_buffer[i];
987 for (src_ch = 0; src_ch <= maxchan; src_ch++)
988 accum += (int64_t) samples[src_ch] * coeffs[src_ch];
990 if (matrix_noise_shift) {
991 index &= m->access_unit_size_pow2 - 1;
992 accum += m->noise_buffer[index] << (matrix_noise_shift + 7);
996 samples[dest_ch] = ((accum >> 14) & mask) + bypassed_lsb;
1001 /** Write the audio data into the output buffer. */
1003 static int output_data(MLPDecodeContext *m, unsigned int substr,
1004 AVFrame *frame, int *got_frame_ptr)
1006 AVCodecContext *avctx = m->avctx;
1007 SubStream *s = &m->substream[substr];
1008 unsigned int i, out_ch = 0;
1012 int is32 = (m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
1014 if (m->avctx->channels != s->max_matrix_channel + 1) {
1015 av_log(m->avctx, AV_LOG_ERROR, "channel count mismatch\n");
1016 return AVERROR_INVALIDDATA;
1020 av_log(avctx, AV_LOG_ERROR, "No samples to output.\n");
1021 return AVERROR_INVALIDDATA;
1024 /* get output buffer */
1025 frame->nb_samples = s->blockpos;
1026 if ((ret = ff_get_buffer(avctx, frame)) < 0) {
1027 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1030 data_32 = (int32_t *)frame->data[0];
1031 data_16 = (int16_t *)frame->data[0];
1033 for (i = 0; i < s->blockpos; i++) {
1034 for (out_ch = 0; out_ch <= s->max_matrix_channel; out_ch++) {
1035 int mat_ch = s->ch_assign[out_ch];
1036 int32_t sample = m->sample_buffer[i][mat_ch]
1037 << s->output_shift[mat_ch];
1038 s->lossless_check_data ^= (sample & 0xffffff) << mat_ch;
1039 if (is32) *data_32++ = sample << 8;
1040 else *data_16++ = sample >> 8;
1049 /** Read an access unit from the stream.
1050 * @return negative on error, 0 if not enough data is present in the input stream,
1051 * otherwise the number of bytes consumed. */
1053 static int read_access_unit(AVCodecContext *avctx, void* data,
1054 int *got_frame_ptr, AVPacket *avpkt)
1056 const uint8_t *buf = avpkt->data;
1057 int buf_size = avpkt->size;
1058 MLPDecodeContext *m = avctx->priv_data;
1060 unsigned int length, substr;
1061 unsigned int substream_start;
1062 unsigned int header_size = 4;
1063 unsigned int substr_header_size = 0;
1064 uint8_t substream_parity_present[MAX_SUBSTREAMS];
1065 uint16_t substream_data_len[MAX_SUBSTREAMS];
1066 uint8_t parity_bits;
1072 length = (AV_RB16(buf) & 0xfff) * 2;
1074 if (length < 4 || length > buf_size)
1075 return AVERROR_INVALIDDATA;
1077 init_get_bits(&gb, (buf + 4), (length - 4) * 8);
1079 m->is_major_sync_unit = 0;
1080 if (show_bits_long(&gb, 31) == (0xf8726fba >> 1)) {
1081 if (read_major_sync(m, &gb) < 0)
1083 m->is_major_sync_unit = 1;
1087 if (!m->params_valid) {
1088 av_log(m->avctx, AV_LOG_WARNING,
1089 "Stream parameters not seen; skipping frame.\n");
1094 substream_start = 0;
1096 for (substr = 0; substr < m->num_substreams; substr++) {
1097 int extraword_present, checkdata_present, end, nonrestart_substr;
1099 extraword_present = get_bits1(&gb);
1100 nonrestart_substr = get_bits1(&gb);
1101 checkdata_present = get_bits1(&gb);
1104 end = get_bits(&gb, 12) * 2;
1106 substr_header_size += 2;
1108 if (extraword_present) {
1109 if (m->avctx->codec_id == AV_CODEC_ID_MLP) {
1110 av_log(m->avctx, AV_LOG_ERROR, "There must be no extraword for MLP.\n");
1114 substr_header_size += 2;
1117 if (!(nonrestart_substr ^ m->is_major_sync_unit)) {
1118 av_log(m->avctx, AV_LOG_ERROR, "Invalid nonrestart_substr.\n");
1122 if (end + header_size + substr_header_size > length) {
1123 av_log(m->avctx, AV_LOG_ERROR,
1124 "Indicated length of substream %d data goes off end of "
1125 "packet.\n", substr);
1126 end = length - header_size - substr_header_size;
1129 if (end < substream_start) {
1130 av_log(avctx, AV_LOG_ERROR,
1131 "Indicated end offset of substream %d data "
1132 "is smaller than calculated start offset.\n",
1137 if (substr > m->max_decoded_substream)
1140 substream_parity_present[substr] = checkdata_present;
1141 substream_data_len[substr] = end - substream_start;
1142 substream_start = end;
1145 parity_bits = ff_mlp_calculate_parity(buf, 4);
1146 parity_bits ^= ff_mlp_calculate_parity(buf + header_size, substr_header_size);
1148 if ((((parity_bits >> 4) ^ parity_bits) & 0xF) != 0xF) {
1149 av_log(avctx, AV_LOG_ERROR, "Parity check failed.\n");
1153 buf += header_size + substr_header_size;
1155 for (substr = 0; substr <= m->max_decoded_substream; substr++) {
1156 SubStream *s = &m->substream[substr];
1157 init_get_bits(&gb, buf, substream_data_len[substr] * 8);
1159 m->matrix_changed = 0;
1160 memset(m->filter_changed, 0, sizeof(m->filter_changed));
1164 if (get_bits1(&gb)) {
1165 if (get_bits1(&gb)) {
1166 /* A restart header should be present. */
1167 if (read_restart_header(m, &gb, buf, substr) < 0)
1169 s->restart_seen = 1;
1172 if (!s->restart_seen)
1174 if (read_decoding_params(m, &gb, substr) < 0)
1178 if (!s->restart_seen)
1181 if ((ret = read_block_data(m, &gb, substr)) < 0)
1184 if (get_bits_count(&gb) >= substream_data_len[substr] * 8)
1185 goto substream_length_mismatch;
1187 } while (!get_bits1(&gb));
1189 skip_bits(&gb, (-get_bits_count(&gb)) & 15);
1191 if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 32) {
1194 if (get_bits(&gb, 16) != 0xD234)
1195 return AVERROR_INVALIDDATA;
1197 shorten_by = get_bits(&gb, 16);
1198 if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD && shorten_by & 0x2000)
1199 s->blockpos -= FFMIN(shorten_by & 0x1FFF, s->blockpos);
1200 else if (m->avctx->codec_id == AV_CODEC_ID_MLP && shorten_by != 0xD234)
1201 return AVERROR_INVALIDDATA;
1203 if (substr == m->max_decoded_substream)
1204 av_log(m->avctx, AV_LOG_INFO, "End of stream indicated.\n");
1207 if (substream_parity_present[substr]) {
1208 uint8_t parity, checksum;
1210 if (substream_data_len[substr] * 8 - get_bits_count(&gb) != 16)
1211 goto substream_length_mismatch;
1213 parity = ff_mlp_calculate_parity(buf, substream_data_len[substr] - 2);
1214 checksum = ff_mlp_checksum8 (buf, substream_data_len[substr] - 2);
1216 if ((get_bits(&gb, 8) ^ parity) != 0xa9 )
1217 av_log(m->avctx, AV_LOG_ERROR, "Substream %d parity check failed.\n", substr);
1218 if ( get_bits(&gb, 8) != checksum)
1219 av_log(m->avctx, AV_LOG_ERROR, "Substream %d checksum failed.\n" , substr);
1222 if (substream_data_len[substr] * 8 != get_bits_count(&gb))
1223 goto substream_length_mismatch;
1226 if (!s->restart_seen)
1227 av_log(m->avctx, AV_LOG_ERROR,
1228 "No restart header present in substream %d.\n", substr);
1230 buf += substream_data_len[substr];
1233 rematrix_channels(m, m->max_decoded_substream);
1235 if ((ret = output_data(m, m->max_decoded_substream, data, got_frame_ptr)) < 0)
1240 substream_length_mismatch:
1241 av_log(m->avctx, AV_LOG_ERROR, "substream %d length mismatch\n", substr);
1242 return AVERROR_INVALIDDATA;
1245 m->params_valid = 0;
1246 return AVERROR_INVALIDDATA;
1249 #if CONFIG_MLP_DECODER
1250 AVCodec ff_mlp_decoder = {
1252 .type = AVMEDIA_TYPE_AUDIO,
1253 .id = AV_CODEC_ID_MLP,
1254 .priv_data_size = sizeof(MLPDecodeContext),
1255 .init = mlp_decode_init,
1256 .decode = read_access_unit,
1257 .capabilities = CODEC_CAP_DR1,
1258 .long_name = NULL_IF_CONFIG_SMALL("MLP (Meridian Lossless Packing)"),
1261 #if CONFIG_TRUEHD_DECODER
1262 AVCodec ff_truehd_decoder = {
1264 .type = AVMEDIA_TYPE_AUDIO,
1265 .id = AV_CODEC_ID_TRUEHD,
1266 .priv_data_size = sizeof(MLPDecodeContext),
1267 .init = mlp_decode_init,
1268 .decode = read_access_unit,
1269 .capabilities = CODEC_CAP_DR1,
1270 .long_name = NULL_IF_CONFIG_SMALL("TrueHD"),
1272 #endif /* CONFIG_TRUEHD_DECODER */