3 * Copyright (c) 2007-2008 Ian Caulfield
5 * This file is part of Libav.
7 * Libav is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * Libav is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
30 #include "libavutil/internal.h"
31 #include "libavutil/intreadwrite.h"
32 #include "libavutil/channel_layout.h"
35 #include "libavutil/crc.h"
37 #include "mlp_parser.h"
42 /** number of bits used for VLC lookup - longest Huffman code is 9 */
45 #define VLC_STATIC_SIZE 64
48 #define VLC_STATIC_SIZE 512
51 typedef struct SubStream {
52 /// Set if a valid restart header has been read. Otherwise the substream cannot be decoded.
56 /** restart header data */
57 /// The type of noise to be used in the rematrix stage.
60 /// The index of the first channel coded in this substream.
62 /// The index of the last channel coded in this substream.
64 /// The number of channels input into the rematrix stage.
65 uint8_t max_matrix_channel;
66 /// For each channel output by the matrix, the output channel to map it to
67 uint8_t ch_assign[MAX_CHANNELS];
68 /// The channel layout for this substream
70 /// The matrix encoding mode for this substream
71 enum AVMatrixEncoding matrix_encoding;
73 /// Channel coding parameters for channels in the substream
74 ChannelParams channel_params[MAX_CHANNELS];
76 /// The left shift applied to random noise in 0x31ea substreams.
78 /// The current seed value for the pseudorandom noise generator(s).
79 uint32_t noisegen_seed;
81 /// Set if the substream contains extra info to check the size of VLC blocks.
82 uint8_t data_check_present;
84 /// Bitmask of which parameter sets are conveyed in a decoding parameter block.
85 uint8_t param_presence_flags;
86 #define PARAM_BLOCKSIZE (1 << 7)
87 #define PARAM_MATRIX (1 << 6)
88 #define PARAM_OUTSHIFT (1 << 5)
89 #define PARAM_QUANTSTEP (1 << 4)
90 #define PARAM_FIR (1 << 3)
91 #define PARAM_IIR (1 << 2)
92 #define PARAM_HUFFOFFSET (1 << 1)
93 #define PARAM_PRESENCE (1 << 0)
99 /// Number of matrices to be applied.
100 uint8_t num_primitive_matrices;
102 /// matrix output channel
103 uint8_t matrix_out_ch[MAX_MATRICES];
105 /// Whether the LSBs of the matrix output are encoded in the bitstream.
106 uint8_t lsb_bypass[MAX_MATRICES];
107 /// Matrix coefficients, stored as 2.14 fixed point.
108 int32_t matrix_coeff[MAX_MATRICES][MAX_CHANNELS];
109 /// Left shift to apply to noise values in 0x31eb substreams.
110 uint8_t matrix_noise_shift[MAX_MATRICES];
113 /// Left shift to apply to Huffman-decoded residuals.
114 uint8_t quant_step_size[MAX_CHANNELS];
116 /// number of PCM samples in current audio block
118 /// Number of PCM samples decoded so far in this frame.
121 /// Left shift to apply to decoded PCM values to get final 24-bit output.
122 int8_t output_shift[MAX_CHANNELS];
124 /// Running XOR of all output samples.
125 int32_t lossless_check_data;
129 typedef struct MLPDecodeContext {
130 AVCodecContext *avctx;
132 /// Current access unit being read has a major sync.
133 int is_major_sync_unit;
135 /// Size of the major sync unit, in bytes
136 int major_sync_header_size;
138 /// Set if a valid major sync block has been read. Otherwise no decoding is possible.
139 uint8_t params_valid;
141 /// Number of substreams contained within this stream.
142 uint8_t num_substreams;
144 /// Index of the last substream to decode - further substreams are skipped.
145 uint8_t max_decoded_substream;
147 /// number of PCM samples contained in each frame
148 int access_unit_size;
149 /// next power of two above the number of samples in each frame
150 int access_unit_size_pow2;
152 SubStream substream[MAX_SUBSTREAMS];
155 int filter_changed[MAX_CHANNELS][NUM_FILTERS];
157 int8_t noise_buffer[MAX_BLOCKSIZE_POW2];
158 int8_t bypassed_lsbs[MAX_BLOCKSIZE][MAX_CHANNELS];
159 int32_t sample_buffer[MAX_BLOCKSIZE][MAX_CHANNELS];
164 static const uint64_t thd_channel_order[] = {
165 AV_CH_FRONT_LEFT, AV_CH_FRONT_RIGHT, // LR
166 AV_CH_FRONT_CENTER, // C
167 AV_CH_LOW_FREQUENCY, // LFE
168 AV_CH_SIDE_LEFT, AV_CH_SIDE_RIGHT, // LRs
169 AV_CH_TOP_FRONT_LEFT, AV_CH_TOP_FRONT_RIGHT, // LRvh
170 AV_CH_FRONT_LEFT_OF_CENTER, AV_CH_FRONT_RIGHT_OF_CENTER, // LRc
171 AV_CH_BACK_LEFT, AV_CH_BACK_RIGHT, // LRrs
172 AV_CH_BACK_CENTER, // Cs
173 AV_CH_TOP_CENTER, // Ts
174 AV_CH_SURROUND_DIRECT_LEFT, AV_CH_SURROUND_DIRECT_RIGHT, // LRsd
175 AV_CH_WIDE_LEFT, AV_CH_WIDE_RIGHT, // LRw
176 AV_CH_TOP_FRONT_CENTER, // Cvh
177 AV_CH_LOW_FREQUENCY_2, // LFE2
180 static uint64_t thd_channel_layout_extract_channel(uint64_t channel_layout,
185 if (av_get_channel_layout_nb_channels(channel_layout) <= index)
188 for (i = 0; i < FF_ARRAY_ELEMS(thd_channel_order); i++)
189 if (channel_layout & thd_channel_order[i] && !index--)
190 return thd_channel_order[i];
194 static VLC huff_vlc[3];
196 /** Initialize static data, constant between all invocations of the codec. */
198 static av_cold void init_static(void)
200 if (!huff_vlc[0].bits) {
201 INIT_VLC_STATIC(&huff_vlc[0], VLC_BITS, 18,
202 &ff_mlp_huffman_tables[0][0][1], 2, 1,
203 &ff_mlp_huffman_tables[0][0][0], 2, 1, VLC_STATIC_SIZE);
204 INIT_VLC_STATIC(&huff_vlc[1], VLC_BITS, 16,
205 &ff_mlp_huffman_tables[1][0][1], 2, 1,
206 &ff_mlp_huffman_tables[1][0][0], 2, 1, VLC_STATIC_SIZE);
207 INIT_VLC_STATIC(&huff_vlc[2], VLC_BITS, 15,
208 &ff_mlp_huffman_tables[2][0][1], 2, 1,
209 &ff_mlp_huffman_tables[2][0][0], 2, 1, VLC_STATIC_SIZE);
215 static inline int32_t calculate_sign_huff(MLPDecodeContext *m,
216 unsigned int substr, unsigned int ch)
218 SubStream *s = &m->substream[substr];
219 ChannelParams *cp = &s->channel_params[ch];
220 int lsb_bits = cp->huff_lsbs - s->quant_step_size[ch];
221 int sign_shift = lsb_bits + (cp->codebook ? 2 - cp->codebook : -1);
222 int32_t sign_huff_offset = cp->huff_offset;
224 if (cp->codebook > 0)
225 sign_huff_offset -= 7 << lsb_bits;
228 sign_huff_offset -= 1 << sign_shift;
230 return sign_huff_offset;
233 /** Read a sample, consisting of either, both or neither of entropy-coded MSBs
236 static inline int read_huff_channels(MLPDecodeContext *m, GetBitContext *gbp,
237 unsigned int substr, unsigned int pos)
239 SubStream *s = &m->substream[substr];
240 unsigned int mat, channel;
242 for (mat = 0; mat < s->num_primitive_matrices; mat++)
243 if (s->lsb_bypass[mat])
244 m->bypassed_lsbs[pos + s->blockpos][mat] = get_bits1(gbp);
246 for (channel = s->min_channel; channel <= s->max_channel; channel++) {
247 ChannelParams *cp = &s->channel_params[channel];
248 int codebook = cp->codebook;
249 int quant_step_size = s->quant_step_size[channel];
250 int lsb_bits = cp->huff_lsbs - quant_step_size;
254 result = get_vlc2(gbp, huff_vlc[codebook-1].table,
255 VLC_BITS, (9 + VLC_BITS - 1) / VLC_BITS);
258 return AVERROR_INVALIDDATA;
261 result = (result << lsb_bits) + get_bits(gbp, lsb_bits);
263 result += cp->sign_huff_offset;
264 result <<= quant_step_size;
266 m->sample_buffer[pos + s->blockpos][channel] = result;
272 static av_cold int mlp_decode_init(AVCodecContext *avctx)
274 MLPDecodeContext *m = avctx->priv_data;
279 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
280 m->substream[substr].lossless_check_data = 0xffffffff;
281 ff_mlpdsp_init(&m->dsp);
286 /** Read a major sync info header - contains high level information about
287 * the stream - sample rate, channel arrangement etc. Most of this
288 * information is not actually necessary for decoding, only for playback.
291 static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
296 if ((ret = ff_mlp_read_major_sync(m->avctx, &mh, gb)) != 0)
299 if (mh.group1_bits == 0) {
300 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown bits per sample\n");
301 return AVERROR_INVALIDDATA;
303 if (mh.group2_bits > mh.group1_bits) {
304 av_log(m->avctx, AV_LOG_ERROR,
305 "Channel group 2 cannot have more bits per sample than group 1.\n");
306 return AVERROR_INVALIDDATA;
309 if (mh.group2_samplerate && mh.group2_samplerate != mh.group1_samplerate) {
310 av_log(m->avctx, AV_LOG_ERROR,
311 "Channel groups with differing sample rates are not currently supported.\n");
312 return AVERROR_INVALIDDATA;
315 if (mh.group1_samplerate == 0) {
316 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown sampling rate\n");
317 return AVERROR_INVALIDDATA;
319 if (mh.group1_samplerate > MAX_SAMPLERATE) {
320 av_log(m->avctx, AV_LOG_ERROR,
321 "Sampling rate %d is greater than the supported maximum (%d).\n",
322 mh.group1_samplerate, MAX_SAMPLERATE);
323 return AVERROR_INVALIDDATA;
325 if (mh.access_unit_size > MAX_BLOCKSIZE) {
326 av_log(m->avctx, AV_LOG_ERROR,
327 "Block size %d is greater than the supported maximum (%d).\n",
328 mh.access_unit_size, MAX_BLOCKSIZE);
329 return AVERROR_INVALIDDATA;
331 if (mh.access_unit_size_pow2 > MAX_BLOCKSIZE_POW2) {
332 av_log(m->avctx, AV_LOG_ERROR,
333 "Block size pow2 %d is greater than the supported maximum (%d).\n",
334 mh.access_unit_size_pow2, MAX_BLOCKSIZE_POW2);
335 return AVERROR_INVALIDDATA;
338 if (mh.num_substreams == 0)
339 return AVERROR_INVALIDDATA;
340 if (m->avctx->codec_id == AV_CODEC_ID_MLP && mh.num_substreams > 2) {
341 av_log(m->avctx, AV_LOG_ERROR, "MLP only supports up to 2 substreams.\n");
342 return AVERROR_INVALIDDATA;
344 if (mh.num_substreams > MAX_SUBSTREAMS) {
345 avpriv_request_sample(m->avctx,
346 "%d substreams (more than the "
347 "maximum supported by the decoder)",
349 return AVERROR_PATCHWELCOME;
352 m->major_sync_header_size = mh.header_size;
354 m->access_unit_size = mh.access_unit_size;
355 m->access_unit_size_pow2 = mh.access_unit_size_pow2;
357 m->num_substreams = mh.num_substreams;
359 /* limit to decoding 3 substreams, as the 4th is used by Dolby Atmos for non-audio data */
360 m->max_decoded_substream = FFMIN(m->num_substreams - 1, 2);
362 m->avctx->sample_rate = mh.group1_samplerate;
363 m->avctx->frame_size = mh.access_unit_size;
365 m->avctx->bits_per_raw_sample = mh.group1_bits;
366 if (mh.group1_bits > 16)
367 m->avctx->sample_fmt = AV_SAMPLE_FMT_S32;
369 m->avctx->sample_fmt = AV_SAMPLE_FMT_S16;
370 m->dsp.mlp_pack_output = m->dsp.mlp_select_pack_output(m->substream[m->max_decoded_substream].ch_assign,
371 m->substream[m->max_decoded_substream].output_shift,
372 m->substream[m->max_decoded_substream].max_matrix_channel,
373 m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
376 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
377 m->substream[substr].restart_seen = 0;
379 /* Set the layout for each substream. When there's more than one, the first
380 * substream is Stereo. Subsequent substreams' layouts are indicated in the
382 if (m->avctx->codec_id == AV_CODEC_ID_MLP) {
383 if ((substr = (mh.num_substreams > 1)))
384 m->substream[0].ch_layout = AV_CH_LAYOUT_STEREO;
385 m->substream[substr].ch_layout = mh.channel_layout_mlp;
387 if ((substr = (mh.num_substreams > 1)))
388 m->substream[0].ch_layout = AV_CH_LAYOUT_STEREO;
389 if (mh.num_substreams > 2)
390 if (mh.channel_layout_thd_stream2)
391 m->substream[2].ch_layout = mh.channel_layout_thd_stream2;
393 m->substream[2].ch_layout = mh.channel_layout_thd_stream1;
394 m->substream[substr].ch_layout = mh.channel_layout_thd_stream1;
397 /* Parse the TrueHD decoder channel modifiers and set each substream's
398 * AVMatrixEncoding accordingly.
400 * The meaning of the modifiers depends on the channel layout:
402 * - THD_CH_MODIFIER_LTRT, THD_CH_MODIFIER_LBINRBIN only apply to 2-channel
404 * - THD_CH_MODIFIER_MONO applies to 1-channel or 2-channel (dual mono)
406 * - THD_CH_MODIFIER_SURROUNDEX, THD_CH_MODIFIER_NOTSURROUNDEX only apply to
407 * layouts with an Ls/Rs channel pair
409 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
410 m->substream[substr].matrix_encoding = AV_MATRIX_ENCODING_NONE;
411 if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD) {
412 if (mh.num_substreams > 2 &&
413 mh.channel_layout_thd_stream2 & AV_CH_SIDE_LEFT &&
414 mh.channel_layout_thd_stream2 & AV_CH_SIDE_RIGHT &&
415 mh.channel_modifier_thd_stream2 == THD_CH_MODIFIER_SURROUNDEX)
416 m->substream[2].matrix_encoding = AV_MATRIX_ENCODING_DOLBYEX;
418 if (mh.num_substreams > 1 &&
419 mh.channel_layout_thd_stream1 & AV_CH_SIDE_LEFT &&
420 mh.channel_layout_thd_stream1 & AV_CH_SIDE_RIGHT &&
421 mh.channel_modifier_thd_stream1 == THD_CH_MODIFIER_SURROUNDEX)
422 m->substream[1].matrix_encoding = AV_MATRIX_ENCODING_DOLBYEX;
424 if (mh.num_substreams > 0)
425 switch (mh.channel_modifier_thd_stream0) {
426 case THD_CH_MODIFIER_LTRT:
427 m->substream[0].matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
429 case THD_CH_MODIFIER_LBINRBIN:
430 m->substream[0].matrix_encoding = AV_MATRIX_ENCODING_DOLBYHEADPHONE;
440 /** Read a restart header from a block in a substream. This contains parameters
441 * required to decode the audio that do not change very often. Generally
442 * (always) present only in blocks following a major sync. */
444 static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
445 const uint8_t *buf, unsigned int substr)
447 SubStream *s = &m->substream[substr];
451 uint8_t lossless_check;
452 int start_count = get_bits_count(gbp);
453 int min_channel, max_channel, max_matrix_channel;
454 const int std_max_matrix_channel = m->avctx->codec_id == AV_CODEC_ID_MLP
455 ? MAX_MATRIX_CHANNEL_MLP
456 : MAX_MATRIX_CHANNEL_TRUEHD;
458 sync_word = get_bits(gbp, 13);
460 if (sync_word != 0x31ea >> 1) {
461 av_log(m->avctx, AV_LOG_ERROR,
462 "restart header sync incorrect (got 0x%04x)\n", sync_word);
463 return AVERROR_INVALIDDATA;
466 s->noise_type = get_bits1(gbp);
468 if (m->avctx->codec_id == AV_CODEC_ID_MLP && s->noise_type) {
469 av_log(m->avctx, AV_LOG_ERROR, "MLP must have 0x31ea sync word.\n");
470 return AVERROR_INVALIDDATA;
473 skip_bits(gbp, 16); /* Output timestamp */
475 min_channel = get_bits(gbp, 4);
476 max_channel = get_bits(gbp, 4);
477 max_matrix_channel = get_bits(gbp, 4);
479 if (max_matrix_channel > std_max_matrix_channel) {
480 av_log(m->avctx, AV_LOG_ERROR,
481 "Max matrix channel cannot be greater than %d.\n",
483 return AVERROR_INVALIDDATA;
486 if (max_channel != max_matrix_channel) {
487 av_log(m->avctx, AV_LOG_ERROR,
488 "Max channel must be equal max matrix channel.\n");
489 return AVERROR_INVALIDDATA;
492 /* This should happen for TrueHD streams with >6 channels and MLP's noise
493 * type. It is not yet known if this is allowed. */
494 if (s->max_channel > MAX_MATRIX_CHANNEL_MLP && !s->noise_type) {
495 avpriv_request_sample(m->avctx,
496 "%d channels (more than the "
497 "maximum supported by the decoder)",
499 return AVERROR_PATCHWELCOME;
502 if (min_channel > max_channel) {
503 av_log(m->avctx, AV_LOG_ERROR,
504 "Substream min channel cannot be greater than max channel.\n");
505 return AVERROR_INVALIDDATA;
508 s->min_channel = min_channel;
509 s->max_channel = max_channel;
510 s->max_matrix_channel = max_matrix_channel;
512 #if FF_API_REQUEST_CHANNELS
513 FF_DISABLE_DEPRECATION_WARNINGS
514 if (m->avctx->request_channels > 0 &&
515 m->avctx->request_channels <= s->max_channel + 1 &&
516 m->max_decoded_substream > substr) {
517 av_log(m->avctx, AV_LOG_DEBUG,
518 "Extracting %d-channel downmix from substream %d. "
519 "Further substreams will be skipped.\n",
520 s->max_channel + 1, substr);
521 m->max_decoded_substream = substr;
523 FF_ENABLE_DEPRECATION_WARNINGS
525 if (m->avctx->request_channel_layout && (s->ch_layout & m->avctx->request_channel_layout) ==
526 m->avctx->request_channel_layout && m->max_decoded_substream > substr) {
527 av_log(m->avctx, AV_LOG_DEBUG,
528 "Extracting %d-channel downmix (0x%"PRIx64") from substream %d. "
529 "Further substreams will be skipped.\n",
530 s->max_channel + 1, s->ch_layout, substr);
531 m->max_decoded_substream = substr;
534 s->noise_shift = get_bits(gbp, 4);
535 s->noisegen_seed = get_bits(gbp, 23);
539 s->data_check_present = get_bits1(gbp);
540 lossless_check = get_bits(gbp, 8);
541 if (substr == m->max_decoded_substream
542 && s->lossless_check_data != 0xffffffff) {
543 tmp = xor_32_to_8(s->lossless_check_data);
544 if (tmp != lossless_check)
545 av_log(m->avctx, AV_LOG_WARNING,
546 "Lossless check failed - expected %02x, calculated %02x.\n",
547 lossless_check, tmp);
552 memset(s->ch_assign, 0, sizeof(s->ch_assign));
554 for (ch = 0; ch <= s->max_matrix_channel; ch++) {
555 int ch_assign = get_bits(gbp, 6);
556 if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD) {
557 uint64_t channel = thd_channel_layout_extract_channel(s->ch_layout,
559 ch_assign = av_get_channel_layout_channel_index(s->ch_layout,
562 if (ch_assign < 0 || ch_assign > s->max_matrix_channel) {
563 avpriv_request_sample(m->avctx,
564 "Assignment of matrix channel %d to invalid output channel %d",
566 return AVERROR_PATCHWELCOME;
568 s->ch_assign[ch_assign] = ch;
571 checksum = ff_mlp_restart_checksum(buf, get_bits_count(gbp) - start_count);
573 if (checksum != get_bits(gbp, 8))
574 av_log(m->avctx, AV_LOG_ERROR, "restart header checksum error\n");
576 /* Set default decoding parameters. */
577 s->param_presence_flags = 0xff;
578 s->num_primitive_matrices = 0;
580 s->lossless_check_data = 0;
582 memset(s->output_shift , 0, sizeof(s->output_shift ));
583 memset(s->quant_step_size, 0, sizeof(s->quant_step_size));
585 for (ch = s->min_channel; ch <= s->max_channel; ch++) {
586 ChannelParams *cp = &s->channel_params[ch];
587 cp->filter_params[FIR].order = 0;
588 cp->filter_params[IIR].order = 0;
589 cp->filter_params[FIR].shift = 0;
590 cp->filter_params[IIR].shift = 0;
592 /* Default audio coding is 24-bit raw PCM. */
594 cp->sign_huff_offset = (-1) << 23;
599 if (substr == m->max_decoded_substream) {
600 m->avctx->channels = s->max_matrix_channel + 1;
601 m->avctx->channel_layout = s->ch_layout;
602 m->dsp.mlp_pack_output = m->dsp.mlp_select_pack_output(s->ch_assign,
604 s->max_matrix_channel,
605 m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
611 /** Read parameters for one of the prediction filters. */
613 static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
614 unsigned int substr, unsigned int channel,
617 SubStream *s = &m->substream[substr];
618 FilterParams *fp = &s->channel_params[channel].filter_params[filter];
619 const int max_order = filter ? MAX_IIR_ORDER : MAX_FIR_ORDER;
620 const char fchar = filter ? 'I' : 'F';
623 // Filter is 0 for FIR, 1 for IIR.
626 if (m->filter_changed[channel][filter]++ > 1) {
627 av_log(m->avctx, AV_LOG_ERROR, "Filters may change only once per access unit.\n");
628 return AVERROR_INVALIDDATA;
631 order = get_bits(gbp, 4);
632 if (order > max_order) {
633 av_log(m->avctx, AV_LOG_ERROR,
634 "%cIR filter order %d is greater than maximum %d.\n",
635 fchar, order, max_order);
636 return AVERROR_INVALIDDATA;
641 int32_t *fcoeff = s->channel_params[channel].coeff[filter];
642 int coeff_bits, coeff_shift;
644 fp->shift = get_bits(gbp, 4);
646 coeff_bits = get_bits(gbp, 5);
647 coeff_shift = get_bits(gbp, 3);
648 if (coeff_bits < 1 || coeff_bits > 16) {
649 av_log(m->avctx, AV_LOG_ERROR,
650 "%cIR filter coeff_bits must be between 1 and 16.\n",
652 return AVERROR_INVALIDDATA;
654 if (coeff_bits + coeff_shift > 16) {
655 av_log(m->avctx, AV_LOG_ERROR,
656 "Sum of coeff_bits and coeff_shift for %cIR filter must be 16 or less.\n",
658 return AVERROR_INVALIDDATA;
661 for (i = 0; i < order; i++)
662 fcoeff[i] = get_sbits(gbp, coeff_bits) << coeff_shift;
664 if (get_bits1(gbp)) {
665 int state_bits, state_shift;
668 av_log(m->avctx, AV_LOG_ERROR,
669 "FIR filter has state data specified.\n");
670 return AVERROR_INVALIDDATA;
673 state_bits = get_bits(gbp, 4);
674 state_shift = get_bits(gbp, 4);
676 /* TODO: Check validity of state data. */
678 for (i = 0; i < order; i++)
679 fp->state[i] = get_sbits(gbp, state_bits) << state_shift;
686 /** Read parameters for primitive matrices. */
688 static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitContext *gbp)
690 SubStream *s = &m->substream[substr];
691 unsigned int mat, ch;
692 const int max_primitive_matrices = m->avctx->codec_id == AV_CODEC_ID_MLP
694 : MAX_MATRICES_TRUEHD;
696 if (m->matrix_changed++ > 1) {
697 av_log(m->avctx, AV_LOG_ERROR, "Matrices may change only once per access unit.\n");
698 return AVERROR_INVALIDDATA;
701 s->num_primitive_matrices = get_bits(gbp, 4);
703 if (s->num_primitive_matrices > max_primitive_matrices) {
704 av_log(m->avctx, AV_LOG_ERROR,
705 "Number of primitive matrices cannot be greater than %d.\n",
706 max_primitive_matrices);
707 return AVERROR_INVALIDDATA;
710 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
711 int frac_bits, max_chan;
712 s->matrix_out_ch[mat] = get_bits(gbp, 4);
713 frac_bits = get_bits(gbp, 4);
714 s->lsb_bypass [mat] = get_bits1(gbp);
716 if (s->matrix_out_ch[mat] > s->max_matrix_channel) {
717 av_log(m->avctx, AV_LOG_ERROR,
718 "Invalid channel %d specified as output from matrix.\n",
719 s->matrix_out_ch[mat]);
720 return AVERROR_INVALIDDATA;
722 if (frac_bits > 14) {
723 av_log(m->avctx, AV_LOG_ERROR,
724 "Too many fractional bits specified.\n");
725 return AVERROR_INVALIDDATA;
728 max_chan = s->max_matrix_channel;
732 for (ch = 0; ch <= max_chan; ch++) {
735 coeff_val = get_sbits(gbp, frac_bits + 2);
737 s->matrix_coeff[mat][ch] = coeff_val << (14 - frac_bits);
741 s->matrix_noise_shift[mat] = get_bits(gbp, 4);
743 s->matrix_noise_shift[mat] = 0;
749 /** Read channel parameters. */
751 static int read_channel_params(MLPDecodeContext *m, unsigned int substr,
752 GetBitContext *gbp, unsigned int ch)
754 SubStream *s = &m->substream[substr];
755 ChannelParams *cp = &s->channel_params[ch];
756 FilterParams *fir = &cp->filter_params[FIR];
757 FilterParams *iir = &cp->filter_params[IIR];
760 if (s->param_presence_flags & PARAM_FIR)
762 if ((ret = read_filter_params(m, gbp, substr, ch, FIR)) < 0)
765 if (s->param_presence_flags & PARAM_IIR)
767 if ((ret = read_filter_params(m, gbp, substr, ch, IIR)) < 0)
770 if (fir->order + iir->order > 8) {
771 av_log(m->avctx, AV_LOG_ERROR, "Total filter orders too high.\n");
772 return AVERROR_INVALIDDATA;
775 if (fir->order && iir->order &&
776 fir->shift != iir->shift) {
777 av_log(m->avctx, AV_LOG_ERROR,
778 "FIR and IIR filters must use the same precision.\n");
779 return AVERROR_INVALIDDATA;
781 /* The FIR and IIR filters must have the same precision.
782 * To simplify the filtering code, only the precision of the
783 * FIR filter is considered. If only the IIR filter is employed,
784 * the FIR filter precision is set to that of the IIR filter, so
785 * that the filtering code can use it. */
786 if (!fir->order && iir->order)
787 fir->shift = iir->shift;
789 if (s->param_presence_flags & PARAM_HUFFOFFSET)
791 cp->huff_offset = get_sbits(gbp, 15);
793 cp->codebook = get_bits(gbp, 2);
794 cp->huff_lsbs = get_bits(gbp, 5);
796 if (cp->huff_lsbs > 24) {
797 av_log(m->avctx, AV_LOG_ERROR, "Invalid huff_lsbs.\n");
798 return AVERROR_INVALIDDATA;
801 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
806 /** Read decoding parameters that change more often than those in the restart
809 static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
812 SubStream *s = &m->substream[substr];
816 if (s->param_presence_flags & PARAM_PRESENCE)
818 s->param_presence_flags = get_bits(gbp, 8);
820 if (s->param_presence_flags & PARAM_BLOCKSIZE)
821 if (get_bits1(gbp)) {
822 s->blocksize = get_bits(gbp, 9);
823 if (s->blocksize < 8 || s->blocksize > m->access_unit_size) {
824 av_log(m->avctx, AV_LOG_ERROR, "Invalid blocksize.");
826 return AVERROR_INVALIDDATA;
830 if (s->param_presence_flags & PARAM_MATRIX)
832 if ((ret = read_matrix_params(m, substr, gbp)) < 0)
835 if (s->param_presence_flags & PARAM_OUTSHIFT)
836 if (get_bits1(gbp)) {
837 for (ch = 0; ch <= s->max_matrix_channel; ch++)
838 s->output_shift[ch] = get_sbits(gbp, 4);
839 if (substr == m->max_decoded_substream)
840 m->dsp.mlp_pack_output = m->dsp.mlp_select_pack_output(s->ch_assign,
842 s->max_matrix_channel,
843 m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
846 if (s->param_presence_flags & PARAM_QUANTSTEP)
848 for (ch = 0; ch <= s->max_channel; ch++) {
849 ChannelParams *cp = &s->channel_params[ch];
851 s->quant_step_size[ch] = get_bits(gbp, 4);
853 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
856 for (ch = s->min_channel; ch <= s->max_channel; ch++)
858 if ((ret = read_channel_params(m, substr, gbp, ch)) < 0)
864 #define MSB_MASK(bits) (-1u << bits)
866 /** Generate PCM samples using the prediction filters and residual values
867 * read from the data stream, and update the filter state. */
869 static void filter_channel(MLPDecodeContext *m, unsigned int substr,
870 unsigned int channel)
872 SubStream *s = &m->substream[substr];
873 const int32_t *fircoeff = s->channel_params[channel].coeff[FIR];
874 int32_t state_buffer[NUM_FILTERS][MAX_BLOCKSIZE + MAX_FIR_ORDER];
875 int32_t *firbuf = state_buffer[FIR] + MAX_BLOCKSIZE;
876 int32_t *iirbuf = state_buffer[IIR] + MAX_BLOCKSIZE;
877 FilterParams *fir = &s->channel_params[channel].filter_params[FIR];
878 FilterParams *iir = &s->channel_params[channel].filter_params[IIR];
879 unsigned int filter_shift = fir->shift;
880 int32_t mask = MSB_MASK(s->quant_step_size[channel]);
882 memcpy(firbuf, fir->state, MAX_FIR_ORDER * sizeof(int32_t));
883 memcpy(iirbuf, iir->state, MAX_IIR_ORDER * sizeof(int32_t));
885 m->dsp.mlp_filter_channel(firbuf, fircoeff,
886 fir->order, iir->order,
887 filter_shift, mask, s->blocksize,
888 &m->sample_buffer[s->blockpos][channel]);
890 memcpy(fir->state, firbuf - s->blocksize, MAX_FIR_ORDER * sizeof(int32_t));
891 memcpy(iir->state, iirbuf - s->blocksize, MAX_IIR_ORDER * sizeof(int32_t));
894 /** Read a block of PCM residual data (or actual if no filtering active). */
896 static int read_block_data(MLPDecodeContext *m, GetBitContext *gbp,
899 SubStream *s = &m->substream[substr];
900 unsigned int i, ch, expected_stream_pos = 0;
903 if (s->data_check_present) {
904 expected_stream_pos = get_bits_count(gbp);
905 expected_stream_pos += get_bits(gbp, 16);
906 avpriv_request_sample(m->avctx,
907 "Substreams with VLC block size check info");
910 if (s->blockpos + s->blocksize > m->access_unit_size) {
911 av_log(m->avctx, AV_LOG_ERROR, "too many audio samples in frame\n");
912 return AVERROR_INVALIDDATA;
915 memset(&m->bypassed_lsbs[s->blockpos][0], 0,
916 s->blocksize * sizeof(m->bypassed_lsbs[0]));
918 for (i = 0; i < s->blocksize; i++)
919 if ((ret = read_huff_channels(m, gbp, substr, i)) < 0)
922 for (ch = s->min_channel; ch <= s->max_channel; ch++)
923 filter_channel(m, substr, ch);
925 s->blockpos += s->blocksize;
927 if (s->data_check_present) {
928 if (get_bits_count(gbp) != expected_stream_pos)
929 av_log(m->avctx, AV_LOG_ERROR, "block data length mismatch\n");
936 /** Data table used for TrueHD noise generation function. */
938 static const int8_t noise_table[256] = {
939 30, 51, 22, 54, 3, 7, -4, 38, 14, 55, 46, 81, 22, 58, -3, 2,
940 52, 31, -7, 51, 15, 44, 74, 30, 85, -17, 10, 33, 18, 80, 28, 62,
941 10, 32, 23, 69, 72, 26, 35, 17, 73, 60, 8, 56, 2, 6, -2, -5,
942 51, 4, 11, 50, 66, 76, 21, 44, 33, 47, 1, 26, 64, 48, 57, 40,
943 38, 16, -10, -28, 92, 22, -18, 29, -10, 5, -13, 49, 19, 24, 70, 34,
944 61, 48, 30, 14, -6, 25, 58, 33, 42, 60, 67, 17, 54, 17, 22, 30,
945 67, 44, -9, 50, -11, 43, 40, 32, 59, 82, 13, 49, -14, 55, 60, 36,
946 48, 49, 31, 47, 15, 12, 4, 65, 1, 23, 29, 39, 45, -2, 84, 69,
947 0, 72, 37, 57, 27, 41, -15, -16, 35, 31, 14, 61, 24, 0, 27, 24,
948 16, 41, 55, 34, 53, 9, 56, 12, 25, 29, 53, 5, 20, -20, -8, 20,
949 13, 28, -3, 78, 38, 16, 11, 62, 46, 29, 21, 24, 46, 65, 43, -23,
950 89, 18, 74, 21, 38, -12, 19, 12, -19, 8, 15, 33, 4, 57, 9, -8,
951 36, 35, 26, 28, 7, 83, 63, 79, 75, 11, 3, 87, 37, 47, 34, 40,
952 39, 19, 20, 42, 27, 34, 39, 77, 13, 42, 59, 64, 45, -1, 32, 37,
953 45, -5, 53, -6, 7, 36, 50, 23, 6, 32, 9, -21, 18, 71, 27, 52,
954 -25, 31, 35, 42, -1, 68, 63, 52, 26, 43, 66, 37, 41, 25, 40, 70,
957 /** Noise generation functions.
958 * I'm not sure what these are for - they seem to be some kind of pseudorandom
959 * sequence generators, used to generate noise data which is used when the
960 * channels are rematrixed. I'm not sure if they provide a practical benefit
961 * to compression, or just obfuscate the decoder. Are they for some kind of
964 /** Generate two channels of noise, used in the matrix when
965 * restart sync word == 0x31ea. */
967 static void generate_2_noise_channels(MLPDecodeContext *m, unsigned int substr)
969 SubStream *s = &m->substream[substr];
971 uint32_t seed = s->noisegen_seed;
972 unsigned int maxchan = s->max_matrix_channel;
974 for (i = 0; i < s->blockpos; i++) {
975 uint16_t seed_shr7 = seed >> 7;
976 m->sample_buffer[i][maxchan+1] = ((int8_t)(seed >> 15)) << s->noise_shift;
977 m->sample_buffer[i][maxchan+2] = ((int8_t) seed_shr7) << s->noise_shift;
979 seed = (seed << 16) ^ seed_shr7 ^ (seed_shr7 << 5);
982 s->noisegen_seed = seed;
985 /** Generate a block of noise, used when restart sync word == 0x31eb. */
987 static void fill_noise_buffer(MLPDecodeContext *m, unsigned int substr)
989 SubStream *s = &m->substream[substr];
991 uint32_t seed = s->noisegen_seed;
993 for (i = 0; i < m->access_unit_size_pow2; i++) {
994 uint8_t seed_shr15 = seed >> 15;
995 m->noise_buffer[i] = noise_table[seed_shr15];
996 seed = (seed << 8) ^ seed_shr15 ^ (seed_shr15 << 5);
999 s->noisegen_seed = seed;
1003 /** Apply the channel matrices in turn to reconstruct the original audio
1006 static void rematrix_channels(MLPDecodeContext *m, unsigned int substr)
1008 SubStream *s = &m->substream[substr];
1010 unsigned int maxchan;
1012 maxchan = s->max_matrix_channel;
1013 if (!s->noise_type) {
1014 generate_2_noise_channels(m, substr);
1017 fill_noise_buffer(m, substr);
1020 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
1021 unsigned int dest_ch = s->matrix_out_ch[mat];
1022 m->dsp.mlp_rematrix_channel(&m->sample_buffer[0][0],
1023 s->matrix_coeff[mat],
1024 &m->bypassed_lsbs[0][mat],
1026 s->num_primitive_matrices - mat,
1030 s->matrix_noise_shift[mat],
1031 m->access_unit_size_pow2,
1032 MSB_MASK(s->quant_step_size[dest_ch]));
1036 /** Write the audio data into the output buffer. */
1038 static int output_data(MLPDecodeContext *m, unsigned int substr,
1039 AVFrame *frame, int *got_frame_ptr)
1041 AVCodecContext *avctx = m->avctx;
1042 SubStream *s = &m->substream[substr];
1044 int is32 = (m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
1046 if (m->avctx->channels != s->max_matrix_channel + 1) {
1047 av_log(m->avctx, AV_LOG_ERROR, "channel count mismatch\n");
1048 return AVERROR_INVALIDDATA;
1052 av_log(avctx, AV_LOG_ERROR, "No samples to output.\n");
1053 return AVERROR_INVALIDDATA;
1056 /* get output buffer */
1057 frame->nb_samples = s->blockpos;
1058 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1059 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1062 s->lossless_check_data = m->dsp.mlp_pack_output(s->lossless_check_data,
1068 s->max_matrix_channel,
1071 /* Update matrix encoding side data */
1072 if ((ret = ff_side_data_update_matrix_encoding(frame, s->matrix_encoding)) < 0)
1080 /** Read an access unit from the stream.
1081 * @return negative on error, 0 if not enough data is present in the input stream,
1082 * otherwise the number of bytes consumed. */
1084 static int read_access_unit(AVCodecContext *avctx, void* data,
1085 int *got_frame_ptr, AVPacket *avpkt)
1087 const uint8_t *buf = avpkt->data;
1088 int buf_size = avpkt->size;
1089 MLPDecodeContext *m = avctx->priv_data;
1091 unsigned int length, substr;
1092 unsigned int substream_start;
1093 unsigned int header_size = 4;
1094 unsigned int substr_header_size = 0;
1095 uint8_t substream_parity_present[MAX_SUBSTREAMS];
1096 uint16_t substream_data_len[MAX_SUBSTREAMS];
1097 uint8_t parity_bits;
1103 length = (AV_RB16(buf) & 0xfff) * 2;
1105 if (length < 4 || length > buf_size)
1106 return AVERROR_INVALIDDATA;
1108 init_get_bits(&gb, (buf + 4), (length - 4) * 8);
1110 m->is_major_sync_unit = 0;
1111 if (show_bits_long(&gb, 31) == (0xf8726fba >> 1)) {
1112 if (read_major_sync(m, &gb) < 0)
1114 m->is_major_sync_unit = 1;
1115 header_size += m->major_sync_header_size;
1118 if (!m->params_valid) {
1119 av_log(m->avctx, AV_LOG_WARNING,
1120 "Stream parameters not seen; skipping frame.\n");
1125 substream_start = 0;
1127 for (substr = 0; substr < m->num_substreams; substr++) {
1128 int extraword_present, checkdata_present, end, nonrestart_substr;
1130 extraword_present = get_bits1(&gb);
1131 nonrestart_substr = get_bits1(&gb);
1132 checkdata_present = get_bits1(&gb);
1135 end = get_bits(&gb, 12) * 2;
1137 substr_header_size += 2;
1139 if (extraword_present) {
1140 if (m->avctx->codec_id == AV_CODEC_ID_MLP) {
1141 av_log(m->avctx, AV_LOG_ERROR, "There must be no extraword for MLP.\n");
1145 substr_header_size += 2;
1148 if (!(nonrestart_substr ^ m->is_major_sync_unit)) {
1149 av_log(m->avctx, AV_LOG_ERROR, "Invalid nonrestart_substr.\n");
1153 if (end + header_size + substr_header_size > length) {
1154 av_log(m->avctx, AV_LOG_ERROR,
1155 "Indicated length of substream %d data goes off end of "
1156 "packet.\n", substr);
1157 end = length - header_size - substr_header_size;
1160 if (end < substream_start) {
1161 av_log(avctx, AV_LOG_ERROR,
1162 "Indicated end offset of substream %d data "
1163 "is smaller than calculated start offset.\n",
1168 if (substr > m->max_decoded_substream)
1171 substream_parity_present[substr] = checkdata_present;
1172 substream_data_len[substr] = end - substream_start;
1173 substream_start = end;
1176 parity_bits = ff_mlp_calculate_parity(buf, 4);
1177 parity_bits ^= ff_mlp_calculate_parity(buf + header_size, substr_header_size);
1179 if ((((parity_bits >> 4) ^ parity_bits) & 0xF) != 0xF) {
1180 av_log(avctx, AV_LOG_ERROR, "Parity check failed.\n");
1184 buf += header_size + substr_header_size;
1186 for (substr = 0; substr <= m->max_decoded_substream; substr++) {
1187 SubStream *s = &m->substream[substr];
1188 init_get_bits(&gb, buf, substream_data_len[substr] * 8);
1190 m->matrix_changed = 0;
1191 memset(m->filter_changed, 0, sizeof(m->filter_changed));
1195 if (get_bits1(&gb)) {
1196 if (get_bits1(&gb)) {
1197 /* A restart header should be present. */
1198 if (read_restart_header(m, &gb, buf, substr) < 0)
1200 s->restart_seen = 1;
1203 if (!s->restart_seen)
1205 if (read_decoding_params(m, &gb, substr) < 0)
1209 if (!s->restart_seen)
1212 if ((ret = read_block_data(m, &gb, substr)) < 0)
1215 if (get_bits_count(&gb) >= substream_data_len[substr] * 8)
1216 goto substream_length_mismatch;
1218 } while (!get_bits1(&gb));
1220 skip_bits(&gb, (-get_bits_count(&gb)) & 15);
1222 if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 32) {
1225 if (get_bits(&gb, 16) != 0xD234)
1226 return AVERROR_INVALIDDATA;
1228 shorten_by = get_bits(&gb, 16);
1229 if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD && shorten_by & 0x2000)
1230 s->blockpos -= FFMIN(shorten_by & 0x1FFF, s->blockpos);
1231 else if (m->avctx->codec_id == AV_CODEC_ID_MLP && shorten_by != 0xD234)
1232 return AVERROR_INVALIDDATA;
1234 if (substr == m->max_decoded_substream)
1235 av_log(m->avctx, AV_LOG_INFO, "End of stream indicated.\n");
1238 if (substream_parity_present[substr]) {
1239 uint8_t parity, checksum;
1241 if (substream_data_len[substr] * 8 - get_bits_count(&gb) != 16)
1242 goto substream_length_mismatch;
1244 parity = ff_mlp_calculate_parity(buf, substream_data_len[substr] - 2);
1245 checksum = ff_mlp_checksum8 (buf, substream_data_len[substr] - 2);
1247 if ((get_bits(&gb, 8) ^ parity) != 0xa9 )
1248 av_log(m->avctx, AV_LOG_ERROR, "Substream %d parity check failed.\n", substr);
1249 if ( get_bits(&gb, 8) != checksum)
1250 av_log(m->avctx, AV_LOG_ERROR, "Substream %d checksum failed.\n" , substr);
1253 if (substream_data_len[substr] * 8 != get_bits_count(&gb))
1254 goto substream_length_mismatch;
1257 if (!s->restart_seen)
1258 av_log(m->avctx, AV_LOG_ERROR,
1259 "No restart header present in substream %d.\n", substr);
1261 buf += substream_data_len[substr];
1264 rematrix_channels(m, m->max_decoded_substream);
1266 if ((ret = output_data(m, m->max_decoded_substream, data, got_frame_ptr)) < 0)
1271 substream_length_mismatch:
1272 av_log(m->avctx, AV_LOG_ERROR, "substream %d length mismatch\n", substr);
1273 return AVERROR_INVALIDDATA;
1276 m->params_valid = 0;
1277 return AVERROR_INVALIDDATA;
1280 AVCodec ff_mlp_decoder = {
1282 .long_name = NULL_IF_CONFIG_SMALL("MLP (Meridian Lossless Packing)"),
1283 .type = AVMEDIA_TYPE_AUDIO,
1284 .id = AV_CODEC_ID_MLP,
1285 .priv_data_size = sizeof(MLPDecodeContext),
1286 .init = mlp_decode_init,
1287 .decode = read_access_unit,
1288 .capabilities = CODEC_CAP_DR1,
1291 #if CONFIG_TRUEHD_DECODER
1292 AVCodec ff_truehd_decoder = {
1294 .long_name = NULL_IF_CONFIG_SMALL("TrueHD"),
1295 .type = AVMEDIA_TYPE_AUDIO,
1296 .id = AV_CODEC_ID_TRUEHD,
1297 .priv_data_size = sizeof(MLPDecodeContext),
1298 .init = mlp_decode_init,
1299 .decode = read_access_unit,
1300 .capabilities = CODEC_CAP_DR1,
1302 #endif /* CONFIG_TRUEHD_DECODER */