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/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 /// Stream needs channel reordering to comply with FFmpeg's channel order
148 uint8_t needs_reordering;
150 /// number of PCM samples contained in each frame
151 int access_unit_size;
152 /// next power of two above the number of samples in each frame
153 int access_unit_size_pow2;
155 SubStream substream[MAX_SUBSTREAMS];
158 int filter_changed[MAX_CHANNELS][NUM_FILTERS];
160 int8_t noise_buffer[MAX_BLOCKSIZE_POW2];
161 int8_t bypassed_lsbs[MAX_BLOCKSIZE][MAX_CHANNELS];
162 int32_t sample_buffer[MAX_BLOCKSIZE][MAX_CHANNELS];
167 static const uint64_t thd_channel_order[] = {
168 AV_CH_FRONT_LEFT, AV_CH_FRONT_RIGHT, // LR
169 AV_CH_FRONT_CENTER, // C
170 AV_CH_LOW_FREQUENCY, // LFE
171 AV_CH_SIDE_LEFT, AV_CH_SIDE_RIGHT, // LRs
172 AV_CH_TOP_FRONT_LEFT, AV_CH_TOP_FRONT_RIGHT, // LRvh
173 AV_CH_FRONT_LEFT_OF_CENTER, AV_CH_FRONT_RIGHT_OF_CENTER, // LRc
174 AV_CH_BACK_LEFT, AV_CH_BACK_RIGHT, // LRrs
175 AV_CH_BACK_CENTER, // Cs
176 AV_CH_TOP_CENTER, // Ts
177 AV_CH_SURROUND_DIRECT_LEFT, AV_CH_SURROUND_DIRECT_RIGHT, // LRsd
178 AV_CH_WIDE_LEFT, AV_CH_WIDE_RIGHT, // LRw
179 AV_CH_TOP_FRONT_CENTER, // Cvh
180 AV_CH_LOW_FREQUENCY_2, // LFE2
183 static uint64_t thd_channel_layout_extract_channel(uint64_t channel_layout,
188 if (av_get_channel_layout_nb_channels(channel_layout) <= index)
191 for (i = 0; i < FF_ARRAY_ELEMS(thd_channel_order); i++)
192 if (channel_layout & thd_channel_order[i] && !index--)
193 return thd_channel_order[i];
197 static VLC huff_vlc[3];
199 /** Initialize static data, constant between all invocations of the codec. */
201 static av_cold void init_static(void)
203 if (!huff_vlc[0].bits) {
204 INIT_VLC_STATIC(&huff_vlc[0], VLC_BITS, 18,
205 &ff_mlp_huffman_tables[0][0][1], 2, 1,
206 &ff_mlp_huffman_tables[0][0][0], 2, 1, VLC_STATIC_SIZE);
207 INIT_VLC_STATIC(&huff_vlc[1], VLC_BITS, 16,
208 &ff_mlp_huffman_tables[1][0][1], 2, 1,
209 &ff_mlp_huffman_tables[1][0][0], 2, 1, VLC_STATIC_SIZE);
210 INIT_VLC_STATIC(&huff_vlc[2], VLC_BITS, 15,
211 &ff_mlp_huffman_tables[2][0][1], 2, 1,
212 &ff_mlp_huffman_tables[2][0][0], 2, 1, VLC_STATIC_SIZE);
218 static inline int32_t calculate_sign_huff(MLPDecodeContext *m,
219 unsigned int substr, unsigned int ch)
221 SubStream *s = &m->substream[substr];
222 ChannelParams *cp = &s->channel_params[ch];
223 int lsb_bits = cp->huff_lsbs - s->quant_step_size[ch];
224 int sign_shift = lsb_bits + (cp->codebook ? 2 - cp->codebook : -1);
225 int32_t sign_huff_offset = cp->huff_offset;
227 if (cp->codebook > 0)
228 sign_huff_offset -= 7 << lsb_bits;
231 sign_huff_offset -= 1 << sign_shift;
233 return sign_huff_offset;
236 /** Read a sample, consisting of either, both or neither of entropy-coded MSBs
239 static inline int read_huff_channels(MLPDecodeContext *m, GetBitContext *gbp,
240 unsigned int substr, unsigned int pos)
242 SubStream *s = &m->substream[substr];
243 unsigned int mat, channel;
245 for (mat = 0; mat < s->num_primitive_matrices; mat++)
246 if (s->lsb_bypass[mat])
247 m->bypassed_lsbs[pos + s->blockpos][mat] = get_bits1(gbp);
249 for (channel = s->min_channel; channel <= s->max_channel; channel++) {
250 ChannelParams *cp = &s->channel_params[channel];
251 int codebook = cp->codebook;
252 int quant_step_size = s->quant_step_size[channel];
253 int lsb_bits = cp->huff_lsbs - quant_step_size;
257 result = get_vlc2(gbp, huff_vlc[codebook-1].table,
258 VLC_BITS, (9 + VLC_BITS - 1) / VLC_BITS);
261 return AVERROR_INVALIDDATA;
264 result = (result << lsb_bits) + get_bits(gbp, lsb_bits);
266 result += cp->sign_huff_offset;
267 result <<= quant_step_size;
269 m->sample_buffer[pos + s->blockpos][channel] = result;
275 static av_cold int mlp_decode_init(AVCodecContext *avctx)
277 MLPDecodeContext *m = avctx->priv_data;
282 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
283 m->substream[substr].lossless_check_data = 0xffffffff;
284 ff_mlpdsp_init(&m->dsp);
289 /** Read a major sync info header - contains high level information about
290 * the stream - sample rate, channel arrangement etc. Most of this
291 * information is not actually necessary for decoding, only for playback.
294 static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
299 if ((ret = ff_mlp_read_major_sync(m->avctx, &mh, gb)) != 0)
302 if (mh.group1_bits == 0) {
303 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown bits per sample\n");
304 return AVERROR_INVALIDDATA;
306 if (mh.group2_bits > mh.group1_bits) {
307 av_log(m->avctx, AV_LOG_ERROR,
308 "Channel group 2 cannot have more bits per sample than group 1.\n");
309 return AVERROR_INVALIDDATA;
312 if (mh.group2_samplerate && mh.group2_samplerate != mh.group1_samplerate) {
313 av_log(m->avctx, AV_LOG_ERROR,
314 "Channel groups with differing sample rates are not currently supported.\n");
315 return AVERROR_INVALIDDATA;
318 if (mh.group1_samplerate == 0) {
319 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown sampling rate\n");
320 return AVERROR_INVALIDDATA;
322 if (mh.group1_samplerate > MAX_SAMPLERATE) {
323 av_log(m->avctx, AV_LOG_ERROR,
324 "Sampling rate %d is greater than the supported maximum (%d).\n",
325 mh.group1_samplerate, MAX_SAMPLERATE);
326 return AVERROR_INVALIDDATA;
328 if (mh.access_unit_size > MAX_BLOCKSIZE) {
329 av_log(m->avctx, AV_LOG_ERROR,
330 "Block size %d is greater than the supported maximum (%d).\n",
331 mh.access_unit_size, MAX_BLOCKSIZE);
332 return AVERROR_INVALIDDATA;
334 if (mh.access_unit_size_pow2 > MAX_BLOCKSIZE_POW2) {
335 av_log(m->avctx, AV_LOG_ERROR,
336 "Block size pow2 %d is greater than the supported maximum (%d).\n",
337 mh.access_unit_size_pow2, MAX_BLOCKSIZE_POW2);
338 return AVERROR_INVALIDDATA;
341 if (mh.num_substreams == 0)
342 return AVERROR_INVALIDDATA;
343 if (m->avctx->codec_id == AV_CODEC_ID_MLP && mh.num_substreams > 2) {
344 av_log(m->avctx, AV_LOG_ERROR, "MLP only supports up to 2 substreams.\n");
345 return AVERROR_INVALIDDATA;
347 if (mh.num_substreams > MAX_SUBSTREAMS) {
348 avpriv_request_sample(m->avctx,
349 "%d substreams (more than the "
350 "maximum supported by the decoder)",
352 return AVERROR_PATCHWELCOME;
355 m->major_sync_header_size = mh.header_size;
357 m->access_unit_size = mh.access_unit_size;
358 m->access_unit_size_pow2 = mh.access_unit_size_pow2;
360 m->num_substreams = mh.num_substreams;
361 m->max_decoded_substream = m->num_substreams - 1;
363 m->avctx->sample_rate = mh.group1_samplerate;
364 m->avctx->frame_size = mh.access_unit_size;
366 m->avctx->bits_per_raw_sample = mh.group1_bits;
367 if (mh.group1_bits > 16)
368 m->avctx->sample_fmt = AV_SAMPLE_FMT_S32;
370 m->avctx->sample_fmt = AV_SAMPLE_FMT_S16;
371 m->dsp.mlp_pack_output = m->dsp.mlp_select_pack_output(m->substream[m->max_decoded_substream].ch_assign,
372 m->substream[m->max_decoded_substream].output_shift,
373 m->substream[m->max_decoded_substream].max_matrix_channel,
374 m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
377 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
378 m->substream[substr].restart_seen = 0;
380 /* Set the layout for each substream. When there's more than one, the first
381 * substream is Stereo. Subsequent substreams' layouts are indicated in the
383 if (m->avctx->codec_id == AV_CODEC_ID_MLP) {
384 if (mh.stream_type != 0xbb) {
385 avpriv_request_sample(m->avctx,
386 "unexpected stream_type %X in MLP",
388 return AVERROR_PATCHWELCOME;
390 if ((substr = (mh.num_substreams > 1)))
391 m->substream[0].ch_layout = AV_CH_LAYOUT_STEREO;
392 m->substream[substr].ch_layout = mh.channel_layout_mlp;
394 if (mh.stream_type != 0xba) {
395 avpriv_request_sample(m->avctx,
396 "unexpected stream_type %X in !MLP",
398 return AVERROR_PATCHWELCOME;
400 if ((substr = (mh.num_substreams > 1)))
401 m->substream[0].ch_layout = AV_CH_LAYOUT_STEREO;
402 if (mh.num_substreams > 2)
403 if (mh.channel_layout_thd_stream2)
404 m->substream[2].ch_layout = mh.channel_layout_thd_stream2;
406 m->substream[2].ch_layout = mh.channel_layout_thd_stream1;
407 m->substream[substr].ch_layout = mh.channel_layout_thd_stream1;
409 if (m->avctx->channels<=2 && m->substream[substr].ch_layout == AV_CH_LAYOUT_MONO && m->max_decoded_substream == 1) {
410 av_log(m->avctx, AV_LOG_DEBUG, "Mono stream with 2 substreams, ignoring 2nd\n");
411 m->max_decoded_substream = 0;
412 if (m->avctx->channels==2)
413 m->avctx->channel_layout = AV_CH_LAYOUT_STEREO;
417 m->needs_reordering = mh.channel_arrangement >= 18 && mh.channel_arrangement <= 20;
419 /* Parse the TrueHD decoder channel modifiers and set each substream's
420 * AVMatrixEncoding accordingly.
422 * The meaning of the modifiers depends on the channel layout:
424 * - THD_CH_MODIFIER_LTRT, THD_CH_MODIFIER_LBINRBIN only apply to 2-channel
426 * - THD_CH_MODIFIER_MONO applies to 1-channel or 2-channel (dual mono)
428 * - THD_CH_MODIFIER_SURROUNDEX, THD_CH_MODIFIER_NOTSURROUNDEX only apply to
429 * layouts with an Ls/Rs channel pair
431 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
432 m->substream[substr].matrix_encoding = AV_MATRIX_ENCODING_NONE;
433 if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD) {
434 if (mh.num_substreams > 2 &&
435 mh.channel_layout_thd_stream2 & AV_CH_SIDE_LEFT &&
436 mh.channel_layout_thd_stream2 & AV_CH_SIDE_RIGHT &&
437 mh.channel_modifier_thd_stream2 == THD_CH_MODIFIER_SURROUNDEX)
438 m->substream[2].matrix_encoding = AV_MATRIX_ENCODING_DOLBYEX;
440 if (mh.num_substreams > 1 &&
441 mh.channel_layout_thd_stream1 & AV_CH_SIDE_LEFT &&
442 mh.channel_layout_thd_stream1 & AV_CH_SIDE_RIGHT &&
443 mh.channel_modifier_thd_stream1 == THD_CH_MODIFIER_SURROUNDEX)
444 m->substream[1].matrix_encoding = AV_MATRIX_ENCODING_DOLBYEX;
446 if (mh.num_substreams > 0)
447 switch (mh.channel_modifier_thd_stream0) {
448 case THD_CH_MODIFIER_LTRT:
449 m->substream[0].matrix_encoding = AV_MATRIX_ENCODING_DOLBY;
451 case THD_CH_MODIFIER_LBINRBIN:
452 m->substream[0].matrix_encoding = AV_MATRIX_ENCODING_DOLBYHEADPHONE;
462 /** Read a restart header from a block in a substream. This contains parameters
463 * required to decode the audio that do not change very often. Generally
464 * (always) present only in blocks following a major sync. */
466 static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
467 const uint8_t *buf, unsigned int substr)
469 SubStream *s = &m->substream[substr];
473 uint8_t lossless_check;
474 int start_count = get_bits_count(gbp);
475 int min_channel, max_channel, max_matrix_channel;
476 const int std_max_matrix_channel = m->avctx->codec_id == AV_CODEC_ID_MLP
477 ? MAX_MATRIX_CHANNEL_MLP
478 : MAX_MATRIX_CHANNEL_TRUEHD;
480 sync_word = get_bits(gbp, 13);
482 if (sync_word != 0x31ea >> 1) {
483 av_log(m->avctx, AV_LOG_ERROR,
484 "restart header sync incorrect (got 0x%04x)\n", sync_word);
485 return AVERROR_INVALIDDATA;
488 s->noise_type = get_bits1(gbp);
490 if (m->avctx->codec_id == AV_CODEC_ID_MLP && s->noise_type) {
491 av_log(m->avctx, AV_LOG_ERROR, "MLP must have 0x31ea sync word.\n");
492 return AVERROR_INVALIDDATA;
495 skip_bits(gbp, 16); /* Output timestamp */
497 min_channel = get_bits(gbp, 4);
498 max_channel = get_bits(gbp, 4);
499 max_matrix_channel = get_bits(gbp, 4);
501 if (max_matrix_channel > std_max_matrix_channel) {
502 av_log(m->avctx, AV_LOG_ERROR,
503 "Max matrix channel cannot be greater than %d.\n",
504 std_max_matrix_channel);
505 return AVERROR_INVALIDDATA;
508 if (max_channel != max_matrix_channel) {
509 av_log(m->avctx, AV_LOG_ERROR,
510 "Max channel must be equal max matrix channel.\n");
511 return AVERROR_INVALIDDATA;
514 /* This should happen for TrueHD streams with >6 channels and MLP's noise
515 * type. It is not yet known if this is allowed. */
516 if (max_channel > MAX_MATRIX_CHANNEL_MLP && !s->noise_type) {
517 avpriv_request_sample(m->avctx,
518 "%d channels (more than the "
519 "maximum supported by the decoder)",
521 return AVERROR_PATCHWELCOME;
524 if (min_channel > max_channel) {
525 av_log(m->avctx, AV_LOG_ERROR,
526 "Substream min channel cannot be greater than max channel.\n");
527 return AVERROR_INVALIDDATA;
530 s->min_channel = min_channel;
531 s->max_channel = max_channel;
532 s->max_matrix_channel = max_matrix_channel;
534 #if FF_API_REQUEST_CHANNELS
535 FF_DISABLE_DEPRECATION_WARNINGS
536 if (m->avctx->request_channels > 0 &&
537 m->avctx->request_channels <= s->max_channel + 1 &&
538 m->max_decoded_substream > substr) {
539 av_log(m->avctx, AV_LOG_DEBUG,
540 "Extracting %d-channel downmix from substream %d. "
541 "Further substreams will be skipped.\n",
542 s->max_channel + 1, substr);
543 m->max_decoded_substream = substr;
544 FF_ENABLE_DEPRECATION_WARNINGS
547 if (m->avctx->request_channel_layout && (s->ch_layout & m->avctx->request_channel_layout) ==
548 m->avctx->request_channel_layout && m->max_decoded_substream > substr) {
549 av_log(m->avctx, AV_LOG_DEBUG,
550 "Extracting %d-channel downmix (0x%"PRIx64") from substream %d. "
551 "Further substreams will be skipped.\n",
552 s->max_channel + 1, s->ch_layout, substr);
553 m->max_decoded_substream = substr;
556 s->noise_shift = get_bits(gbp, 4);
557 s->noisegen_seed = get_bits(gbp, 23);
561 s->data_check_present = get_bits1(gbp);
562 lossless_check = get_bits(gbp, 8);
563 if (substr == m->max_decoded_substream
564 && s->lossless_check_data != 0xffffffff) {
565 tmp = xor_32_to_8(s->lossless_check_data);
566 if (tmp != lossless_check)
567 av_log(m->avctx, AV_LOG_WARNING,
568 "Lossless check failed - expected %02x, calculated %02x.\n",
569 lossless_check, tmp);
574 memset(s->ch_assign, 0, sizeof(s->ch_assign));
576 for (ch = 0; ch <= s->max_matrix_channel; ch++) {
577 int ch_assign = get_bits(gbp, 6);
578 if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD) {
579 uint64_t channel = thd_channel_layout_extract_channel(s->ch_layout,
581 ch_assign = av_get_channel_layout_channel_index(s->ch_layout,
584 if ((unsigned)ch_assign > s->max_matrix_channel) {
585 avpriv_request_sample(m->avctx,
586 "Assignment of matrix channel %d to invalid output channel %d",
588 return AVERROR_PATCHWELCOME;
590 s->ch_assign[ch_assign] = ch;
593 checksum = ff_mlp_restart_checksum(buf, get_bits_count(gbp) - start_count);
595 if (checksum != get_bits(gbp, 8))
596 av_log(m->avctx, AV_LOG_ERROR, "restart header checksum error\n");
598 /* Set default decoding parameters. */
599 s->param_presence_flags = 0xff;
600 s->num_primitive_matrices = 0;
602 s->lossless_check_data = 0;
604 memset(s->output_shift , 0, sizeof(s->output_shift ));
605 memset(s->quant_step_size, 0, sizeof(s->quant_step_size));
607 for (ch = s->min_channel; ch <= s->max_channel; ch++) {
608 ChannelParams *cp = &s->channel_params[ch];
609 cp->filter_params[FIR].order = 0;
610 cp->filter_params[IIR].order = 0;
611 cp->filter_params[FIR].shift = 0;
612 cp->filter_params[IIR].shift = 0;
614 /* Default audio coding is 24-bit raw PCM. */
616 cp->sign_huff_offset = (-1) << 23;
621 if (substr == m->max_decoded_substream) {
622 m->avctx->channels = s->max_matrix_channel + 1;
623 m->avctx->channel_layout = s->ch_layout;
624 m->dsp.mlp_pack_output = m->dsp.mlp_select_pack_output(s->ch_assign,
626 s->max_matrix_channel,
627 m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
629 if (m->avctx->codec_id == AV_CODEC_ID_MLP && m->needs_reordering) {
630 if (m->avctx->channel_layout == (AV_CH_LAYOUT_QUAD|AV_CH_LOW_FREQUENCY) ||
631 m->avctx->channel_layout == AV_CH_LAYOUT_5POINT0_BACK) {
632 int i = s->ch_assign[4];
633 s->ch_assign[4] = s->ch_assign[3];
634 s->ch_assign[3] = s->ch_assign[2];
636 } else if (m->avctx->channel_layout == AV_CH_LAYOUT_5POINT1_BACK) {
637 FFSWAP(int, s->ch_assign[2], s->ch_assign[4]);
638 FFSWAP(int, s->ch_assign[3], s->ch_assign[5]);
647 /** Read parameters for one of the prediction filters. */
649 static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
650 unsigned int substr, unsigned int channel,
653 SubStream *s = &m->substream[substr];
654 FilterParams *fp = &s->channel_params[channel].filter_params[filter];
655 const int max_order = filter ? MAX_IIR_ORDER : MAX_FIR_ORDER;
656 const char fchar = filter ? 'I' : 'F';
659 // Filter is 0 for FIR, 1 for IIR.
660 av_assert0(filter < 2);
662 if (m->filter_changed[channel][filter]++ > 1) {
663 av_log(m->avctx, AV_LOG_ERROR, "Filters may change only once per access unit.\n");
664 return AVERROR_INVALIDDATA;
667 order = get_bits(gbp, 4);
668 if (order > max_order) {
669 av_log(m->avctx, AV_LOG_ERROR,
670 "%cIR filter order %d is greater than maximum %d.\n",
671 fchar, order, max_order);
672 return AVERROR_INVALIDDATA;
677 int32_t *fcoeff = s->channel_params[channel].coeff[filter];
678 int coeff_bits, coeff_shift;
680 fp->shift = get_bits(gbp, 4);
682 coeff_bits = get_bits(gbp, 5);
683 coeff_shift = get_bits(gbp, 3);
684 if (coeff_bits < 1 || coeff_bits > 16) {
685 av_log(m->avctx, AV_LOG_ERROR,
686 "%cIR filter coeff_bits must be between 1 and 16.\n",
688 return AVERROR_INVALIDDATA;
690 if (coeff_bits + coeff_shift > 16) {
691 av_log(m->avctx, AV_LOG_ERROR,
692 "Sum of coeff_bits and coeff_shift for %cIR filter must be 16 or less.\n",
694 return AVERROR_INVALIDDATA;
697 for (i = 0; i < order; i++)
698 fcoeff[i] = get_sbits(gbp, coeff_bits) << coeff_shift;
700 if (get_bits1(gbp)) {
701 int state_bits, state_shift;
704 av_log(m->avctx, AV_LOG_ERROR,
705 "FIR filter has state data specified.\n");
706 return AVERROR_INVALIDDATA;
709 state_bits = get_bits(gbp, 4);
710 state_shift = get_bits(gbp, 4);
712 /* TODO: Check validity of state data. */
714 for (i = 0; i < order; i++)
715 fp->state[i] = state_bits ? get_sbits(gbp, state_bits) << state_shift : 0;
722 /** Read parameters for primitive matrices. */
724 static int read_matrix_params(MLPDecodeContext *m, unsigned int substr, GetBitContext *gbp)
726 SubStream *s = &m->substream[substr];
727 unsigned int mat, ch;
728 const int max_primitive_matrices = m->avctx->codec_id == AV_CODEC_ID_MLP
730 : MAX_MATRICES_TRUEHD;
732 if (m->matrix_changed++ > 1) {
733 av_log(m->avctx, AV_LOG_ERROR, "Matrices may change only once per access unit.\n");
734 return AVERROR_INVALIDDATA;
737 s->num_primitive_matrices = get_bits(gbp, 4);
739 if (s->num_primitive_matrices > max_primitive_matrices) {
740 av_log(m->avctx, AV_LOG_ERROR,
741 "Number of primitive matrices cannot be greater than %d.\n",
742 max_primitive_matrices);
743 return AVERROR_INVALIDDATA;
746 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
747 int frac_bits, max_chan;
748 s->matrix_out_ch[mat] = get_bits(gbp, 4);
749 frac_bits = get_bits(gbp, 4);
750 s->lsb_bypass [mat] = get_bits1(gbp);
752 if (s->matrix_out_ch[mat] > s->max_matrix_channel) {
753 av_log(m->avctx, AV_LOG_ERROR,
754 "Invalid channel %d specified as output from matrix.\n",
755 s->matrix_out_ch[mat]);
756 return AVERROR_INVALIDDATA;
758 if (frac_bits > 14) {
759 av_log(m->avctx, AV_LOG_ERROR,
760 "Too many fractional bits specified.\n");
761 return AVERROR_INVALIDDATA;
764 max_chan = s->max_matrix_channel;
768 for (ch = 0; ch <= max_chan; ch++) {
771 coeff_val = get_sbits(gbp, frac_bits + 2);
773 s->matrix_coeff[mat][ch] = coeff_val << (14 - frac_bits);
777 s->matrix_noise_shift[mat] = get_bits(gbp, 4);
779 s->matrix_noise_shift[mat] = 0;
785 /** Read channel parameters. */
787 static int read_channel_params(MLPDecodeContext *m, unsigned int substr,
788 GetBitContext *gbp, unsigned int ch)
790 SubStream *s = &m->substream[substr];
791 ChannelParams *cp = &s->channel_params[ch];
792 FilterParams *fir = &cp->filter_params[FIR];
793 FilterParams *iir = &cp->filter_params[IIR];
796 if (s->param_presence_flags & PARAM_FIR)
798 if ((ret = read_filter_params(m, gbp, substr, ch, FIR)) < 0)
801 if (s->param_presence_flags & PARAM_IIR)
803 if ((ret = read_filter_params(m, gbp, substr, ch, IIR)) < 0)
806 if (fir->order + iir->order > 8) {
807 av_log(m->avctx, AV_LOG_ERROR, "Total filter orders too high.\n");
808 return AVERROR_INVALIDDATA;
811 if (fir->order && iir->order &&
812 fir->shift != iir->shift) {
813 av_log(m->avctx, AV_LOG_ERROR,
814 "FIR and IIR filters must use the same precision.\n");
815 return AVERROR_INVALIDDATA;
817 /* The FIR and IIR filters must have the same precision.
818 * To simplify the filtering code, only the precision of the
819 * FIR filter is considered. If only the IIR filter is employed,
820 * the FIR filter precision is set to that of the IIR filter, so
821 * that the filtering code can use it. */
822 if (!fir->order && iir->order)
823 fir->shift = iir->shift;
825 if (s->param_presence_flags & PARAM_HUFFOFFSET)
827 cp->huff_offset = get_sbits(gbp, 15);
829 cp->codebook = get_bits(gbp, 2);
830 cp->huff_lsbs = get_bits(gbp, 5);
832 if (cp->huff_lsbs > 24) {
833 av_log(m->avctx, AV_LOG_ERROR, "Invalid huff_lsbs.\n");
835 return AVERROR_INVALIDDATA;
838 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
843 /** Read decoding parameters that change more often than those in the restart
846 static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
849 SubStream *s = &m->substream[substr];
853 if (s->param_presence_flags & PARAM_PRESENCE)
855 s->param_presence_flags = get_bits(gbp, 8);
857 if (s->param_presence_flags & PARAM_BLOCKSIZE)
858 if (get_bits1(gbp)) {
859 s->blocksize = get_bits(gbp, 9);
860 if (s->blocksize < 8 || s->blocksize > m->access_unit_size) {
861 av_log(m->avctx, AV_LOG_ERROR, "Invalid blocksize.\n");
863 return AVERROR_INVALIDDATA;
867 if (s->param_presence_flags & PARAM_MATRIX)
869 if ((ret = read_matrix_params(m, substr, gbp)) < 0)
872 if (s->param_presence_flags & PARAM_OUTSHIFT)
873 if (get_bits1(gbp)) {
874 for (ch = 0; ch <= s->max_matrix_channel; ch++)
875 s->output_shift[ch] = get_sbits(gbp, 4);
876 if (substr == m->max_decoded_substream)
877 m->dsp.mlp_pack_output = m->dsp.mlp_select_pack_output(s->ch_assign,
879 s->max_matrix_channel,
880 m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
883 if (s->param_presence_flags & PARAM_QUANTSTEP)
885 for (ch = 0; ch <= s->max_channel; ch++) {
886 ChannelParams *cp = &s->channel_params[ch];
888 s->quant_step_size[ch] = get_bits(gbp, 4);
890 cp->sign_huff_offset = calculate_sign_huff(m, substr, ch);
893 for (ch = s->min_channel; ch <= s->max_channel; ch++)
895 if ((ret = read_channel_params(m, substr, gbp, ch)) < 0)
901 #define MSB_MASK(bits) (-1u << (bits))
903 /** Generate PCM samples using the prediction filters and residual values
904 * read from the data stream, and update the filter state. */
906 static void filter_channel(MLPDecodeContext *m, unsigned int substr,
907 unsigned int channel)
909 SubStream *s = &m->substream[substr];
910 const int32_t *fircoeff = s->channel_params[channel].coeff[FIR];
911 int32_t state_buffer[NUM_FILTERS][MAX_BLOCKSIZE + MAX_FIR_ORDER];
912 int32_t *firbuf = state_buffer[FIR] + MAX_BLOCKSIZE;
913 int32_t *iirbuf = state_buffer[IIR] + MAX_BLOCKSIZE;
914 FilterParams *fir = &s->channel_params[channel].filter_params[FIR];
915 FilterParams *iir = &s->channel_params[channel].filter_params[IIR];
916 unsigned int filter_shift = fir->shift;
917 int32_t mask = MSB_MASK(s->quant_step_size[channel]);
919 memcpy(firbuf, fir->state, MAX_FIR_ORDER * sizeof(int32_t));
920 memcpy(iirbuf, iir->state, MAX_IIR_ORDER * sizeof(int32_t));
922 m->dsp.mlp_filter_channel(firbuf, fircoeff,
923 fir->order, iir->order,
924 filter_shift, mask, s->blocksize,
925 &m->sample_buffer[s->blockpos][channel]);
927 memcpy(fir->state, firbuf - s->blocksize, MAX_FIR_ORDER * sizeof(int32_t));
928 memcpy(iir->state, iirbuf - s->blocksize, MAX_IIR_ORDER * sizeof(int32_t));
931 /** Read a block of PCM residual data (or actual if no filtering active). */
933 static int read_block_data(MLPDecodeContext *m, GetBitContext *gbp,
936 SubStream *s = &m->substream[substr];
937 unsigned int i, ch, expected_stream_pos = 0;
940 if (s->data_check_present) {
941 expected_stream_pos = get_bits_count(gbp);
942 expected_stream_pos += get_bits(gbp, 16);
943 avpriv_request_sample(m->avctx,
944 "Substreams with VLC block size check info");
947 if (s->blockpos + s->blocksize > m->access_unit_size) {
948 av_log(m->avctx, AV_LOG_ERROR, "too many audio samples in frame\n");
949 return AVERROR_INVALIDDATA;
952 memset(&m->bypassed_lsbs[s->blockpos][0], 0,
953 s->blocksize * sizeof(m->bypassed_lsbs[0]));
955 for (i = 0; i < s->blocksize; i++)
956 if ((ret = read_huff_channels(m, gbp, substr, i)) < 0)
959 for (ch = s->min_channel; ch <= s->max_channel; ch++)
960 filter_channel(m, substr, ch);
962 s->blockpos += s->blocksize;
964 if (s->data_check_present) {
965 if (get_bits_count(gbp) != expected_stream_pos)
966 av_log(m->avctx, AV_LOG_ERROR, "block data length mismatch\n");
973 /** Data table used for TrueHD noise generation function. */
975 static const int8_t noise_table[256] = {
976 30, 51, 22, 54, 3, 7, -4, 38, 14, 55, 46, 81, 22, 58, -3, 2,
977 52, 31, -7, 51, 15, 44, 74, 30, 85, -17, 10, 33, 18, 80, 28, 62,
978 10, 32, 23, 69, 72, 26, 35, 17, 73, 60, 8, 56, 2, 6, -2, -5,
979 51, 4, 11, 50, 66, 76, 21, 44, 33, 47, 1, 26, 64, 48, 57, 40,
980 38, 16, -10, -28, 92, 22, -18, 29, -10, 5, -13, 49, 19, 24, 70, 34,
981 61, 48, 30, 14, -6, 25, 58, 33, 42, 60, 67, 17, 54, 17, 22, 30,
982 67, 44, -9, 50, -11, 43, 40, 32, 59, 82, 13, 49, -14, 55, 60, 36,
983 48, 49, 31, 47, 15, 12, 4, 65, 1, 23, 29, 39, 45, -2, 84, 69,
984 0, 72, 37, 57, 27, 41, -15, -16, 35, 31, 14, 61, 24, 0, 27, 24,
985 16, 41, 55, 34, 53, 9, 56, 12, 25, 29, 53, 5, 20, -20, -8, 20,
986 13, 28, -3, 78, 38, 16, 11, 62, 46, 29, 21, 24, 46, 65, 43, -23,
987 89, 18, 74, 21, 38, -12, 19, 12, -19, 8, 15, 33, 4, 57, 9, -8,
988 36, 35, 26, 28, 7, 83, 63, 79, 75, 11, 3, 87, 37, 47, 34, 40,
989 39, 19, 20, 42, 27, 34, 39, 77, 13, 42, 59, 64, 45, -1, 32, 37,
990 45, -5, 53, -6, 7, 36, 50, 23, 6, 32, 9, -21, 18, 71, 27, 52,
991 -25, 31, 35, 42, -1, 68, 63, 52, 26, 43, 66, 37, 41, 25, 40, 70,
994 /** Noise generation functions.
995 * I'm not sure what these are for - they seem to be some kind of pseudorandom
996 * sequence generators, used to generate noise data which is used when the
997 * channels are rematrixed. I'm not sure if they provide a practical benefit
998 * to compression, or just obfuscate the decoder. Are they for some kind of
1001 /** Generate two channels of noise, used in the matrix when
1002 * restart sync word == 0x31ea. */
1004 static void generate_2_noise_channels(MLPDecodeContext *m, unsigned int substr)
1006 SubStream *s = &m->substream[substr];
1008 uint32_t seed = s->noisegen_seed;
1009 unsigned int maxchan = s->max_matrix_channel;
1011 for (i = 0; i < s->blockpos; i++) {
1012 uint16_t seed_shr7 = seed >> 7;
1013 m->sample_buffer[i][maxchan+1] = ((int8_t)(seed >> 15)) << s->noise_shift;
1014 m->sample_buffer[i][maxchan+2] = ((int8_t) seed_shr7) << s->noise_shift;
1016 seed = (seed << 16) ^ seed_shr7 ^ (seed_shr7 << 5);
1019 s->noisegen_seed = seed;
1022 /** Generate a block of noise, used when restart sync word == 0x31eb. */
1024 static void fill_noise_buffer(MLPDecodeContext *m, unsigned int substr)
1026 SubStream *s = &m->substream[substr];
1028 uint32_t seed = s->noisegen_seed;
1030 for (i = 0; i < m->access_unit_size_pow2; i++) {
1031 uint8_t seed_shr15 = seed >> 15;
1032 m->noise_buffer[i] = noise_table[seed_shr15];
1033 seed = (seed << 8) ^ seed_shr15 ^ (seed_shr15 << 5);
1036 s->noisegen_seed = seed;
1040 /** Apply the channel matrices in turn to reconstruct the original audio
1043 static void rematrix_channels(MLPDecodeContext *m, unsigned int substr)
1045 SubStream *s = &m->substream[substr];
1047 unsigned int maxchan;
1049 maxchan = s->max_matrix_channel;
1050 if (!s->noise_type) {
1051 generate_2_noise_channels(m, substr);
1054 fill_noise_buffer(m, substr);
1057 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
1058 unsigned int dest_ch = s->matrix_out_ch[mat];
1059 m->dsp.mlp_rematrix_channel(&m->sample_buffer[0][0],
1060 s->matrix_coeff[mat],
1061 &m->bypassed_lsbs[0][mat],
1063 s->num_primitive_matrices - mat,
1067 s->matrix_noise_shift[mat],
1068 m->access_unit_size_pow2,
1069 MSB_MASK(s->quant_step_size[dest_ch]));
1073 /** Write the audio data into the output buffer. */
1075 static int output_data(MLPDecodeContext *m, unsigned int substr,
1076 AVFrame *frame, int *got_frame_ptr)
1078 AVCodecContext *avctx = m->avctx;
1079 SubStream *s = &m->substream[substr];
1081 int is32 = (m->avctx->sample_fmt == AV_SAMPLE_FMT_S32);
1083 if (m->avctx->channels != s->max_matrix_channel + 1) {
1084 av_log(m->avctx, AV_LOG_ERROR, "channel count mismatch\n");
1085 return AVERROR_INVALIDDATA;
1089 av_log(avctx, AV_LOG_ERROR, "No samples to output.\n");
1090 return AVERROR_INVALIDDATA;
1093 /* get output buffer */
1094 frame->nb_samples = s->blockpos;
1095 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1097 s->lossless_check_data = m->dsp.mlp_pack_output(s->lossless_check_data,
1103 s->max_matrix_channel,
1106 /* Update matrix encoding side data */
1107 if ((ret = ff_side_data_update_matrix_encoding(frame, s->matrix_encoding)) < 0)
1115 /** Read an access unit from the stream.
1116 * @return negative on error, 0 if not enough data is present in the input stream,
1117 * otherwise the number of bytes consumed. */
1119 static int read_access_unit(AVCodecContext *avctx, void* data,
1120 int *got_frame_ptr, AVPacket *avpkt)
1122 const uint8_t *buf = avpkt->data;
1123 int buf_size = avpkt->size;
1124 MLPDecodeContext *m = avctx->priv_data;
1126 unsigned int length, substr;
1127 unsigned int substream_start;
1128 unsigned int header_size = 4;
1129 unsigned int substr_header_size = 0;
1130 uint8_t substream_parity_present[MAX_SUBSTREAMS];
1131 uint16_t substream_data_len[MAX_SUBSTREAMS];
1132 uint8_t parity_bits;
1136 return AVERROR_INVALIDDATA;
1138 length = (AV_RB16(buf) & 0xfff) * 2;
1140 if (length < 4 || length > buf_size)
1141 return AVERROR_INVALIDDATA;
1143 init_get_bits(&gb, (buf + 4), (length - 4) * 8);
1145 m->is_major_sync_unit = 0;
1146 if (show_bits_long(&gb, 31) == (0xf8726fba >> 1)) {
1147 if (read_major_sync(m, &gb) < 0)
1149 m->is_major_sync_unit = 1;
1150 header_size += m->major_sync_header_size;
1153 if (!m->params_valid) {
1154 av_log(m->avctx, AV_LOG_WARNING,
1155 "Stream parameters not seen; skipping frame.\n");
1160 substream_start = 0;
1162 for (substr = 0; substr < m->num_substreams; substr++) {
1163 int extraword_present, checkdata_present, end, nonrestart_substr;
1165 extraword_present = get_bits1(&gb);
1166 nonrestart_substr = get_bits1(&gb);
1167 checkdata_present = get_bits1(&gb);
1170 end = get_bits(&gb, 12) * 2;
1172 substr_header_size += 2;
1174 if (extraword_present) {
1175 if (m->avctx->codec_id == AV_CODEC_ID_MLP) {
1176 av_log(m->avctx, AV_LOG_ERROR, "There must be no extraword for MLP.\n");
1180 substr_header_size += 2;
1183 if (!(nonrestart_substr ^ m->is_major_sync_unit)) {
1184 av_log(m->avctx, AV_LOG_ERROR, "Invalid nonrestart_substr.\n");
1188 if (end + header_size + substr_header_size > length) {
1189 av_log(m->avctx, AV_LOG_ERROR,
1190 "Indicated length of substream %d data goes off end of "
1191 "packet.\n", substr);
1192 end = length - header_size - substr_header_size;
1195 if (end < substream_start) {
1196 av_log(avctx, AV_LOG_ERROR,
1197 "Indicated end offset of substream %d data "
1198 "is smaller than calculated start offset.\n",
1203 if (substr > m->max_decoded_substream)
1206 substream_parity_present[substr] = checkdata_present;
1207 substream_data_len[substr] = end - substream_start;
1208 substream_start = end;
1211 parity_bits = ff_mlp_calculate_parity(buf, 4);
1212 parity_bits ^= ff_mlp_calculate_parity(buf + header_size, substr_header_size);
1214 if ((((parity_bits >> 4) ^ parity_bits) & 0xF) != 0xF) {
1215 av_log(avctx, AV_LOG_ERROR, "Parity check failed.\n");
1219 buf += header_size + substr_header_size;
1221 for (substr = 0; substr <= m->max_decoded_substream; substr++) {
1222 SubStream *s = &m->substream[substr];
1223 init_get_bits(&gb, buf, substream_data_len[substr] * 8);
1225 m->matrix_changed = 0;
1226 memset(m->filter_changed, 0, sizeof(m->filter_changed));
1230 if (get_bits1(&gb)) {
1231 if (get_bits1(&gb)) {
1232 /* A restart header should be present. */
1233 if (read_restart_header(m, &gb, buf, substr) < 0)
1235 s->restart_seen = 1;
1238 if (!s->restart_seen)
1240 if (read_decoding_params(m, &gb, substr) < 0)
1244 if (!s->restart_seen)
1247 if ((ret = read_block_data(m, &gb, substr)) < 0)
1250 if (get_bits_count(&gb) >= substream_data_len[substr] * 8)
1251 goto substream_length_mismatch;
1253 } while (!get_bits1(&gb));
1255 skip_bits(&gb, (-get_bits_count(&gb)) & 15);
1257 if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 32) {
1260 if (get_bits(&gb, 16) != 0xD234)
1261 return AVERROR_INVALIDDATA;
1263 shorten_by = get_bits(&gb, 16);
1264 if (m->avctx->codec_id == AV_CODEC_ID_TRUEHD && shorten_by & 0x2000)
1265 s->blockpos -= FFMIN(shorten_by & 0x1FFF, s->blockpos);
1266 else if (m->avctx->codec_id == AV_CODEC_ID_MLP && shorten_by != 0xD234)
1267 return AVERROR_INVALIDDATA;
1269 if (substr == m->max_decoded_substream)
1270 av_log(m->avctx, AV_LOG_INFO, "End of stream indicated.\n");
1273 if (substream_parity_present[substr]) {
1274 uint8_t parity, checksum;
1276 if (substream_data_len[substr] * 8 - get_bits_count(&gb) != 16)
1277 goto substream_length_mismatch;
1279 parity = ff_mlp_calculate_parity(buf, substream_data_len[substr] - 2);
1280 checksum = ff_mlp_checksum8 (buf, substream_data_len[substr] - 2);
1282 if ((get_bits(&gb, 8) ^ parity) != 0xa9 )
1283 av_log(m->avctx, AV_LOG_ERROR, "Substream %d parity check failed.\n", substr);
1284 if ( get_bits(&gb, 8) != checksum)
1285 av_log(m->avctx, AV_LOG_ERROR, "Substream %d checksum failed.\n" , substr);
1288 if (substream_data_len[substr] * 8 != get_bits_count(&gb))
1289 goto substream_length_mismatch;
1292 if (!s->restart_seen)
1293 av_log(m->avctx, AV_LOG_ERROR,
1294 "No restart header present in substream %d.\n", substr);
1296 buf += substream_data_len[substr];
1299 rematrix_channels(m, m->max_decoded_substream);
1301 if ((ret = output_data(m, m->max_decoded_substream, data, got_frame_ptr)) < 0)
1306 substream_length_mismatch:
1307 av_log(m->avctx, AV_LOG_ERROR, "substream %d length mismatch\n", substr);
1308 return AVERROR_INVALIDDATA;
1311 m->params_valid = 0;
1312 return AVERROR_INVALIDDATA;
1315 #if CONFIG_MLP_DECODER
1316 AVCodec ff_mlp_decoder = {
1318 .long_name = NULL_IF_CONFIG_SMALL("MLP (Meridian Lossless Packing)"),
1319 .type = AVMEDIA_TYPE_AUDIO,
1320 .id = AV_CODEC_ID_MLP,
1321 .priv_data_size = sizeof(MLPDecodeContext),
1322 .init = mlp_decode_init,
1323 .decode = read_access_unit,
1324 .capabilities = CODEC_CAP_DR1,
1327 #if CONFIG_TRUEHD_DECODER
1328 AVCodec ff_truehd_decoder = {
1330 .long_name = NULL_IF_CONFIG_SMALL("TrueHD"),
1331 .type = AVMEDIA_TYPE_AUDIO,
1332 .id = AV_CODEC_ID_TRUEHD,
1333 .priv_data_size = sizeof(MLPDecodeContext),
1334 .init = mlp_decode_init,
1335 .decode = read_access_unit,
1336 .capabilities = CODEC_CAP_DR1,
1338 #endif /* CONFIG_TRUEHD_DECODER */