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 "bitstream.h"
32 #include "libavutil/crc.h"
34 #include "mlp_parser.h"
36 /** Maximum number of channels that can be decoded. */
37 #define MAX_CHANNELS 16
39 /** Maximum number of matrices used in decoding; most streams have one matrix
40 * per output channel, but some rematrix a channel (usually 0) more than once.
43 #define MAX_MATRICES 15
45 /** Maximum number of substreams that can be decoded. This could also be set
46 * higher, but I haven't seen any examples with more than two. */
47 #define MAX_SUBSTREAMS 2
49 /** maximum sample frequency seen in files */
50 #define MAX_SAMPLERATE 192000
52 /** maximum number of audio samples within one access unit */
53 #define MAX_BLOCKSIZE (40 * (MAX_SAMPLERATE / 48000))
54 /** next power of two greater than MAX_BLOCKSIZE */
55 #define MAX_BLOCKSIZE_POW2 (64 * (MAX_SAMPLERATE / 48000))
57 /** number of allowed filters */
60 /** The maximum number of taps in either the IIR or FIR filter;
61 * I believe MLP actually specifies the maximum order for IIR filters as four,
62 * and that the sum of the orders of both filters must be <= 8. */
63 #define MAX_FILTER_ORDER 8
65 /** number of bits used for VLC lookup - longest Huffman code is 9 */
69 static const char* sample_message =
70 "Please file a bug report following the instructions at "
71 "http://ffmpeg.mplayerhq.hu/bugreports.html and include "
72 "a sample of this file.";
74 typedef struct SubStream {
75 //! Set if a valid restart header has been read. Otherwise the substream cannot be decoded.
79 /** restart header data */
80 //! The type of noise to be used in the rematrix stage.
83 //! The index of the first channel coded in this substream.
85 //! The index of the last channel coded in this substream.
87 //! The number of channels input into the rematrix stage.
88 uint8_t max_matrix_channel;
90 //! The left shift applied to random noise in 0x31ea substreams.
92 //! The current seed value for the pseudorandom noise generator(s).
93 uint32_t noisegen_seed;
95 //! Set if the substream contains extra info to check the size of VLC blocks.
96 uint8_t data_check_present;
98 //! Bitmask of which parameter sets are conveyed in a decoding parameter block.
99 uint8_t param_presence_flags;
100 #define PARAM_BLOCKSIZE (1 << 7)
101 #define PARAM_MATRIX (1 << 6)
102 #define PARAM_OUTSHIFT (1 << 5)
103 #define PARAM_QUANTSTEP (1 << 4)
104 #define PARAM_FIR (1 << 3)
105 #define PARAM_IIR (1 << 2)
106 #define PARAM_HUFFOFFSET (1 << 1)
112 //! Number of matrices to be applied.
113 uint8_t num_primitive_matrices;
115 //! matrix output channel
116 uint8_t matrix_out_ch[MAX_MATRICES];
118 //! Whether the LSBs of the matrix output are encoded in the bitstream.
119 uint8_t lsb_bypass[MAX_MATRICES];
120 //! Matrix coefficients, stored as 2.14 fixed point.
121 int32_t matrix_coeff[MAX_MATRICES][MAX_CHANNELS+2];
122 //! Left shift to apply to noise values in 0x31eb substreams.
123 uint8_t matrix_noise_shift[MAX_MATRICES];
126 //! Left shift to apply to Huffman-decoded residuals.
127 uint8_t quant_step_size[MAX_CHANNELS];
129 //! number of PCM samples in current audio block
131 //! Number of PCM samples decoded so far in this frame.
134 //! Left shift to apply to decoded PCM values to get final 24-bit output.
135 int8_t output_shift[MAX_CHANNELS];
137 //! Running XOR of all output samples.
138 int32_t lossless_check_data;
142 typedef struct MLPDecodeContext {
143 AVCodecContext *avctx;
145 //! Set if a valid major sync block has been read. Otherwise no decoding is possible.
146 uint8_t params_valid;
148 //! Number of substreams contained within this stream.
149 uint8_t num_substreams;
151 //! Index of the last substream to decode - further substreams are skipped.
152 uint8_t max_decoded_substream;
154 //! number of PCM samples contained in each frame
155 int access_unit_size;
156 //! next power of two above the number of samples in each frame
157 int access_unit_size_pow2;
159 SubStream substream[MAX_SUBSTREAMS];
165 //! number of taps in filter
166 uint8_t filter_order[MAX_CHANNELS][NUM_FILTERS];
167 //! Right shift to apply to output of filter.
168 uint8_t filter_shift[MAX_CHANNELS][NUM_FILTERS];
170 int32_t filter_coeff[MAX_CHANNELS][NUM_FILTERS][MAX_FILTER_ORDER];
171 int32_t filter_state[MAX_CHANNELS][NUM_FILTERS][MAX_FILTER_ORDER];
175 /** sample data coding information */
176 //! Offset to apply to residual values.
177 int16_t huff_offset[MAX_CHANNELS];
178 //! sign/rounding-corrected version of huff_offset
179 int32_t sign_huff_offset[MAX_CHANNELS];
180 //! Which VLC codebook to use to read residuals.
181 uint8_t codebook[MAX_CHANNELS];
182 //! Size of residual suffix not encoded using VLC.
183 uint8_t huff_lsbs[MAX_CHANNELS];
186 int8_t noise_buffer[MAX_BLOCKSIZE_POW2];
187 int8_t bypassed_lsbs[MAX_BLOCKSIZE][MAX_CHANNELS];
188 int32_t sample_buffer[MAX_BLOCKSIZE][MAX_CHANNELS+2];
191 /** Tables defining the Huffman codes.
192 * There are three entropy coding methods used in MLP (four if you count
193 * "none" as a method). These use the same sequences for codes starting with
194 * 00 or 01, but have different codes starting with 1. */
196 static const uint8_t huffman_tables[3][18][2] = {
197 { /* Huffman table 0, -7 - +10 */
198 {0x01, 9}, {0x01, 8}, {0x01, 7}, {0x01, 6}, {0x01, 5}, {0x01, 4}, {0x01, 3},
199 {0x04, 3}, {0x05, 3}, {0x06, 3}, {0x07, 3},
200 {0x03, 3}, {0x05, 4}, {0x09, 5}, {0x11, 6}, {0x21, 7}, {0x41, 8}, {0x81, 9},
201 }, { /* Huffman table 1, -7 - +8 */
202 {0x01, 9}, {0x01, 8}, {0x01, 7}, {0x01, 6}, {0x01, 5}, {0x01, 4}, {0x01, 3},
203 {0x02, 2}, {0x03, 2},
204 {0x03, 3}, {0x05, 4}, {0x09, 5}, {0x11, 6}, {0x21, 7}, {0x41, 8}, {0x81, 9},
205 }, { /* Huffman table 2, -7 - +7 */
206 {0x01, 9}, {0x01, 8}, {0x01, 7}, {0x01, 6}, {0x01, 5}, {0x01, 4}, {0x01, 3},
208 {0x03, 3}, {0x05, 4}, {0x09, 5}, {0x11, 6}, {0x21, 7}, {0x41, 8}, {0x81, 9},
212 static VLC huff_vlc[3];
214 static int crc_init = 0;
215 static AVCRC crc_63[1024];
216 static AVCRC crc_1D[1024];
219 /** Initialize static data, constant between all invocations of the codec. */
221 static av_cold void init_static()
223 INIT_VLC_STATIC(&huff_vlc[0], VLC_BITS, 18,
224 &huffman_tables[0][0][1], 2, 1,
225 &huffman_tables[0][0][0], 2, 1, 512);
226 INIT_VLC_STATIC(&huff_vlc[1], VLC_BITS, 16,
227 &huffman_tables[1][0][1], 2, 1,
228 &huffman_tables[1][0][0], 2, 1, 512);
229 INIT_VLC_STATIC(&huff_vlc[2], VLC_BITS, 15,
230 &huffman_tables[2][0][1], 2, 1,
231 &huffman_tables[2][0][0], 2, 1, 512);
234 av_crc_init(crc_63, 0, 8, 0x63, sizeof(crc_63));
235 av_crc_init(crc_1D, 0, 8, 0x1D, sizeof(crc_1D));
241 /** MLP uses checksums that seem to be based on the standard CRC algorithm, but
242 * are not (in implementation terms, the table lookup and XOR are reversed).
243 * We can implement this behavior using a standard av_crc on all but the
244 * last element, then XOR that with the last element. */
246 static uint8_t mlp_checksum8(const uint8_t *buf, unsigned int buf_size)
248 uint8_t checksum = av_crc(crc_63, 0x3c, buf, buf_size - 1); // crc_63[0xa2] == 0x3c
249 checksum ^= buf[buf_size-1];
253 /** Calculate an 8-bit checksum over a restart header -- a non-multiple-of-8
254 * number of bits, starting two bits into the first byte of buf. */
256 static uint8_t mlp_restart_checksum(const uint8_t *buf, unsigned int bit_size)
259 int num_bytes = (bit_size + 2) / 8;
261 int crc = crc_1D[buf[0] & 0x3f];
262 crc = av_crc(crc_1D, crc, buf + 1, num_bytes - 2);
263 crc ^= buf[num_bytes - 1];
265 for (i = 0; i < ((bit_size + 2) & 7); i++) {
269 crc ^= (buf[num_bytes] >> (7 - i)) & 1;
275 static inline int32_t calculate_sign_huff(MLPDecodeContext *m,
276 unsigned int substr, unsigned int ch)
278 SubStream *s = &m->substream[substr];
279 int lsb_bits = m->huff_lsbs[ch] - s->quant_step_size[ch];
280 int sign_shift = lsb_bits + (m->codebook[ch] ? 2 - m->codebook[ch] : -1);
281 int32_t sign_huff_offset = m->huff_offset[ch];
283 if (m->codebook[ch] > 0)
284 sign_huff_offset -= 7 << lsb_bits;
287 sign_huff_offset -= 1 << sign_shift;
289 return sign_huff_offset;
292 /** Read a sample, consisting of either, both or neither of entropy-coded MSBs
295 static inline int read_huff_channels(MLPDecodeContext *m, GetBitContext *gbp,
296 unsigned int substr, unsigned int pos)
298 SubStream *s = &m->substream[substr];
299 unsigned int mat, channel;
301 for (mat = 0; mat < s->num_primitive_matrices; mat++)
302 if (s->lsb_bypass[mat])
303 m->bypassed_lsbs[pos + s->blockpos][mat] = get_bits1(gbp);
305 for (channel = s->min_channel; channel <= s->max_channel; channel++) {
306 int codebook = m->codebook[channel];
307 int quant_step_size = s->quant_step_size[channel];
308 int lsb_bits = m->huff_lsbs[channel] - quant_step_size;
312 result = get_vlc2(gbp, huff_vlc[codebook-1].table,
313 VLC_BITS, (9 + VLC_BITS - 1) / VLC_BITS);
319 result = (result << lsb_bits) + get_bits(gbp, lsb_bits);
321 result += m->sign_huff_offset[channel];
322 result <<= quant_step_size;
324 m->sample_buffer[pos + s->blockpos][channel] = result;
330 static av_cold int mlp_decode_init(AVCodecContext *avctx)
332 MLPDecodeContext *m = avctx->priv_data;
337 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
338 m->substream[substr].lossless_check_data = 0xffffffff;
339 avctx->sample_fmt = SAMPLE_FMT_S16;
343 /** Read a major sync info header - contains high level information about
344 * the stream - sample rate, channel arrangement etc. Most of this
345 * information is not actually necessary for decoding, only for playback.
348 static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
353 if (ff_mlp_read_major_sync(m->avctx, &mh, gb) != 0)
356 if (mh.group1_bits == 0) {
357 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown bits per sample\n");
360 if (mh.group2_bits > mh.group1_bits) {
361 av_log(m->avctx, AV_LOG_ERROR,
362 "Channel group 2 cannot have more bits per sample than group 1.\n");
366 if (mh.group2_samplerate && mh.group2_samplerate != mh.group1_samplerate) {
367 av_log(m->avctx, AV_LOG_ERROR,
368 "Channel groups with differing sample rates are not currently supported.\n");
372 if (mh.group1_samplerate == 0) {
373 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown sampling rate\n");
376 if (mh.group1_samplerate > MAX_SAMPLERATE) {
377 av_log(m->avctx, AV_LOG_ERROR,
378 "Sampling rate %d is greater than the supported maximum (%d).\n",
379 mh.group1_samplerate, MAX_SAMPLERATE);
382 if (mh.access_unit_size > MAX_BLOCKSIZE) {
383 av_log(m->avctx, AV_LOG_ERROR,
384 "Block size %d is greater than the supported maximum (%d).\n",
385 mh.access_unit_size, MAX_BLOCKSIZE);
388 if (mh.access_unit_size_pow2 > MAX_BLOCKSIZE_POW2) {
389 av_log(m->avctx, AV_LOG_ERROR,
390 "Block size pow2 %d is greater than the supported maximum (%d).\n",
391 mh.access_unit_size_pow2, MAX_BLOCKSIZE_POW2);
395 if (mh.num_substreams == 0)
397 if (mh.num_substreams > MAX_SUBSTREAMS) {
398 av_log(m->avctx, AV_LOG_ERROR,
399 "Number of substreams %d is larger than the maximum supported "
400 "by the decoder. %s\n", mh.num_substreams, sample_message);
404 m->access_unit_size = mh.access_unit_size;
405 m->access_unit_size_pow2 = mh.access_unit_size_pow2;
407 m->num_substreams = mh.num_substreams;
408 m->max_decoded_substream = m->num_substreams - 1;
410 m->avctx->sample_rate = mh.group1_samplerate;
411 m->avctx->frame_size = mh.access_unit_size;
413 #ifdef CONFIG_AUDIO_NONSHORT
414 m->avctx->bits_per_sample = mh.group1_bits;
415 if (mh.group1_bits > 16) {
416 m->avctx->sample_fmt = SAMPLE_FMT_S32;
421 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
422 m->substream[substr].restart_seen = 0;
427 /** Read a restart header from a block in a substream. This contains parameters
428 * required to decode the audio that do not change very often. Generally
429 * (always) present only in blocks following a major sync. */
431 static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
432 const uint8_t *buf, unsigned int substr)
434 SubStream *s = &m->substream[substr];
438 uint8_t lossless_check;
439 int start_count = get_bits_count(gbp);
441 sync_word = get_bits(gbp, 13);
443 if (sync_word != 0x31ea >> 1) {
444 av_log(m->avctx, AV_LOG_ERROR,
445 "restart header sync incorrect (got 0x%04x)\n", sync_word);
448 s->noise_type = get_bits1(gbp);
450 skip_bits(gbp, 16); /* Output timestamp */
452 s->min_channel = get_bits(gbp, 4);
453 s->max_channel = get_bits(gbp, 4);
454 s->max_matrix_channel = get_bits(gbp, 4);
456 if (s->min_channel > s->max_channel) {
457 av_log(m->avctx, AV_LOG_ERROR,
458 "Substream min channel cannot be greater than max channel.\n");
462 if (m->avctx->request_channels > 0
463 && s->max_channel + 1 >= m->avctx->request_channels
464 && substr < m->max_decoded_substream) {
465 av_log(m->avctx, AV_LOG_INFO,
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 s->noise_shift = get_bits(gbp, 4);
473 s->noisegen_seed = get_bits(gbp, 23);
477 s->data_check_present = get_bits1(gbp);
478 lossless_check = get_bits(gbp, 8);
479 if (substr == m->max_decoded_substream
480 && s->lossless_check_data != 0xffffffff) {
481 tmp = s->lossless_check_data;
485 if (tmp != lossless_check)
486 av_log(m->avctx, AV_LOG_WARNING,
487 "Lossless check failed - expected %02x, calculated %02x.\n",
488 lossless_check, tmp);
490 dprintf(m->avctx, "Lossless check passed for substream %d (%x).\n",
496 for (ch = 0; ch <= s->max_matrix_channel; ch++) {
497 int ch_assign = get_bits(gbp, 6);
498 dprintf(m->avctx, "ch_assign[%d][%d] = %d\n", substr, ch,
500 if (ch_assign != ch) {
501 av_log(m->avctx, AV_LOG_ERROR,
502 "Non-1:1 channel assignments are used in this stream. %s\n",
508 checksum = mlp_restart_checksum(buf, get_bits_count(gbp) - start_count);
510 if (checksum != get_bits(gbp, 8))
511 av_log(m->avctx, AV_LOG_ERROR, "restart header checksum error\n");
513 /* Set default decoding parameters. */
514 s->param_presence_flags = 0xff;
515 s->num_primitive_matrices = 0;
517 s->lossless_check_data = 0;
519 memset(s->output_shift , 0, sizeof(s->output_shift ));
520 memset(s->quant_step_size, 0, sizeof(s->quant_step_size));
522 for (ch = s->min_channel; ch <= s->max_channel; ch++) {
523 m->filter_order[ch][FIR] = 0;
524 m->filter_order[ch][IIR] = 0;
525 m->filter_shift[ch][FIR] = 0;
526 m->filter_shift[ch][IIR] = 0;
528 /* Default audio coding is 24-bit raw PCM. */
529 m->huff_offset [ch] = 0;
530 m->sign_huff_offset[ch] = (-1) << 23;
531 m->codebook [ch] = 0;
532 m->huff_lsbs [ch] = 24;
535 if (substr == m->max_decoded_substream) {
536 m->avctx->channels = s->max_channel + 1;
542 /** Read parameters for one of the prediction filters. */
544 static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
545 unsigned int channel, unsigned int filter)
547 const char fchar = filter ? 'I' : 'F';
550 // Filter is 0 for FIR, 1 for IIR.
553 order = get_bits(gbp, 4);
554 if (order > MAX_FILTER_ORDER) {
555 av_log(m->avctx, AV_LOG_ERROR,
556 "%cIR filter order %d is greater than maximum %d.\n",
557 fchar, order, MAX_FILTER_ORDER);
560 m->filter_order[channel][filter] = order;
563 int coeff_bits, coeff_shift;
565 m->filter_shift[channel][filter] = get_bits(gbp, 4);
567 coeff_bits = get_bits(gbp, 5);
568 coeff_shift = get_bits(gbp, 3);
569 if (coeff_bits < 1 || coeff_bits > 16) {
570 av_log(m->avctx, AV_LOG_ERROR,
571 "%cIR filter coeff_bits must be between 1 and 16.\n",
575 if (coeff_bits + coeff_shift > 16) {
576 av_log(m->avctx, AV_LOG_ERROR,
577 "Sum of coeff_bits and coeff_shift for %cIR filter must be 16 or less.\n",
582 for (i = 0; i < order; i++)
583 m->filter_coeff[channel][filter][i] =
584 get_sbits(gbp, coeff_bits) << coeff_shift;
586 if (get_bits1(gbp)) {
587 int state_bits, state_shift;
590 av_log(m->avctx, AV_LOG_ERROR,
591 "FIR filter has state data specified.\n");
595 state_bits = get_bits(gbp, 4);
596 state_shift = get_bits(gbp, 4);
598 /* TODO: Check validity of state data. */
600 for (i = 0; i < order; i++)
601 m->filter_state[channel][filter][i] =
602 get_sbits(gbp, state_bits) << state_shift;
609 /** Read decoding parameters that change more often than those in the restart
612 static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
615 SubStream *s = &m->substream[substr];
616 unsigned int mat, ch;
619 s->param_presence_flags = get_bits(gbp, 8);
621 if (s->param_presence_flags & PARAM_BLOCKSIZE)
622 if (get_bits1(gbp)) {
623 s->blocksize = get_bits(gbp, 9);
624 if (s->blocksize > MAX_BLOCKSIZE) {
625 av_log(m->avctx, AV_LOG_ERROR, "block size too large\n");
631 if (s->param_presence_flags & PARAM_MATRIX)
632 if (get_bits1(gbp)) {
633 s->num_primitive_matrices = get_bits(gbp, 4);
635 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
636 int frac_bits, max_chan;
637 s->matrix_out_ch[mat] = get_bits(gbp, 4);
638 frac_bits = get_bits(gbp, 4);
639 s->lsb_bypass [mat] = get_bits1(gbp);
641 if (s->matrix_out_ch[mat] > s->max_channel) {
642 av_log(m->avctx, AV_LOG_ERROR,
643 "Invalid channel %d specified as output from matrix.\n",
644 s->matrix_out_ch[mat]);
647 if (frac_bits > 14) {
648 av_log(m->avctx, AV_LOG_ERROR,
649 "Too many fractional bits specified.\n");
653 max_chan = s->max_matrix_channel;
657 for (ch = 0; ch <= max_chan; ch++) {
660 coeff_val = get_sbits(gbp, frac_bits + 2);
662 s->matrix_coeff[mat][ch] = coeff_val << (14 - frac_bits);
666 s->matrix_noise_shift[mat] = get_bits(gbp, 4);
668 s->matrix_noise_shift[mat] = 0;
672 if (s->param_presence_flags & PARAM_OUTSHIFT)
674 for (ch = 0; ch <= s->max_matrix_channel; ch++) {
675 s->output_shift[ch] = get_bits(gbp, 4);
676 dprintf(m->avctx, "output shift[%d] = %d\n",
677 ch, s->output_shift[ch]);
681 if (s->param_presence_flags & PARAM_QUANTSTEP)
683 for (ch = 0; ch <= s->max_channel; ch++) {
684 s->quant_step_size[ch] = get_bits(gbp, 4);
687 m->sign_huff_offset[ch] = calculate_sign_huff(m, substr, ch);
690 for (ch = s->min_channel; ch <= s->max_channel; ch++)
691 if (get_bits1(gbp)) {
692 if (s->param_presence_flags & PARAM_FIR)
694 if (read_filter_params(m, gbp, ch, FIR) < 0)
697 if (s->param_presence_flags & PARAM_IIR)
699 if (read_filter_params(m, gbp, ch, IIR) < 0)
702 if (m->filter_order[ch][FIR] && m->filter_order[ch][IIR] &&
703 m->filter_shift[ch][FIR] != m->filter_shift[ch][IIR]) {
704 av_log(m->avctx, AV_LOG_ERROR,
705 "FIR and IIR filters must use the same precision.\n");
708 /* The FIR and IIR filters must have the same precision.
709 * To simplify the filtering code, only the precision of the
710 * FIR filter is considered. If only the IIR filter is employed,
711 * the FIR filter precision is set to that of the IIR filter, so
712 * that the filtering code can use it. */
713 if (!m->filter_order[ch][FIR] && m->filter_order[ch][IIR])
714 m->filter_shift[ch][FIR] = m->filter_shift[ch][IIR];
716 if (s->param_presence_flags & PARAM_HUFFOFFSET)
718 m->huff_offset[ch] = get_sbits(gbp, 15);
720 m->codebook [ch] = get_bits(gbp, 2);
721 m->huff_lsbs[ch] = get_bits(gbp, 5);
723 m->sign_huff_offset[ch] = calculate_sign_huff(m, substr, ch);
731 #define MSB_MASK(bits) (-1u << bits)
733 /** Generate PCM samples using the prediction filters and residual values
734 * read from the data stream, and update the filter state. */
736 static void filter_channel(MLPDecodeContext *m, unsigned int substr,
737 unsigned int channel)
739 SubStream *s = &m->substream[substr];
740 int32_t filter_state_buffer[NUM_FILTERS][MAX_BLOCKSIZE + MAX_FILTER_ORDER];
741 unsigned int filter_shift = m->filter_shift[channel][FIR];
742 int32_t mask = MSB_MASK(s->quant_step_size[channel]);
743 int index = MAX_BLOCKSIZE;
746 for (j = 0; j < NUM_FILTERS; j++) {
747 memcpy(& filter_state_buffer [j][MAX_BLOCKSIZE],
748 &m->filter_state[channel][j][0],
749 MAX_FILTER_ORDER * sizeof(int32_t));
752 for (i = 0; i < s->blocksize; i++) {
753 int32_t residual = m->sample_buffer[i + s->blockpos][channel];
758 /* TODO: Move this code to DSPContext? */
760 for (j = 0; j < NUM_FILTERS; j++)
761 for (order = 0; order < m->filter_order[channel][j]; order++)
762 accum += (int64_t)filter_state_buffer[j][index + order] *
763 m->filter_coeff[channel][j][order];
765 accum = accum >> filter_shift;
766 result = (accum + residual) & mask;
770 filter_state_buffer[FIR][index] = result;
771 filter_state_buffer[IIR][index] = result - accum;
773 m->sample_buffer[i + s->blockpos][channel] = result;
776 for (j = 0; j < NUM_FILTERS; j++) {
777 memcpy(&m->filter_state[channel][j][0],
778 & filter_state_buffer [j][index],
779 MAX_FILTER_ORDER * sizeof(int32_t));
783 /** Read a block of PCM residual data (or actual if no filtering active). */
785 static int read_block_data(MLPDecodeContext *m, GetBitContext *gbp,
788 SubStream *s = &m->substream[substr];
789 unsigned int i, ch, expected_stream_pos = 0;
791 if (s->data_check_present) {
792 expected_stream_pos = get_bits_count(gbp);
793 expected_stream_pos += get_bits(gbp, 16);
794 av_log(m->avctx, AV_LOG_WARNING, "This file contains some features "
795 "we have not tested yet. %s\n", sample_message);
798 if (s->blockpos + s->blocksize > m->access_unit_size) {
799 av_log(m->avctx, AV_LOG_ERROR, "too many audio samples in frame\n");
803 memset(&m->bypassed_lsbs[s->blockpos][0], 0,
804 s->blocksize * sizeof(m->bypassed_lsbs[0]));
806 for (i = 0; i < s->blocksize; i++) {
807 if (read_huff_channels(m, gbp, substr, i) < 0)
811 for (ch = s->min_channel; ch <= s->max_channel; ch++) {
812 filter_channel(m, substr, ch);
815 s->blockpos += s->blocksize;
817 if (s->data_check_present) {
818 if (get_bits_count(gbp) != expected_stream_pos)
819 av_log(m->avctx, AV_LOG_ERROR, "block data length mismatch\n");
826 /** Data table used for TrueHD noise generation function. */
828 static const int8_t noise_table[256] = {
829 30, 51, 22, 54, 3, 7, -4, 38, 14, 55, 46, 81, 22, 58, -3, 2,
830 52, 31, -7, 51, 15, 44, 74, 30, 85, -17, 10, 33, 18, 80, 28, 62,
831 10, 32, 23, 69, 72, 26, 35, 17, 73, 60, 8, 56, 2, 6, -2, -5,
832 51, 4, 11, 50, 66, 76, 21, 44, 33, 47, 1, 26, 64, 48, 57, 40,
833 38, 16, -10, -28, 92, 22, -18, 29, -10, 5, -13, 49, 19, 24, 70, 34,
834 61, 48, 30, 14, -6, 25, 58, 33, 42, 60, 67, 17, 54, 17, 22, 30,
835 67, 44, -9, 50, -11, 43, 40, 32, 59, 82, 13, 49, -14, 55, 60, 36,
836 48, 49, 31, 47, 15, 12, 4, 65, 1, 23, 29, 39, 45, -2, 84, 69,
837 0, 72, 37, 57, 27, 41, -15, -16, 35, 31, 14, 61, 24, 0, 27, 24,
838 16, 41, 55, 34, 53, 9, 56, 12, 25, 29, 53, 5, 20, -20, -8, 20,
839 13, 28, -3, 78, 38, 16, 11, 62, 46, 29, 21, 24, 46, 65, 43, -23,
840 89, 18, 74, 21, 38, -12, 19, 12, -19, 8, 15, 33, 4, 57, 9, -8,
841 36, 35, 26, 28, 7, 83, 63, 79, 75, 11, 3, 87, 37, 47, 34, 40,
842 39, 19, 20, 42, 27, 34, 39, 77, 13, 42, 59, 64, 45, -1, 32, 37,
843 45, -5, 53, -6, 7, 36, 50, 23, 6, 32, 9, -21, 18, 71, 27, 52,
844 -25, 31, 35, 42, -1, 68, 63, 52, 26, 43, 66, 37, 41, 25, 40, 70,
847 /** Noise generation functions.
848 * I'm not sure what these are for - they seem to be some kind of pseudorandom
849 * sequence generators, used to generate noise data which is used when the
850 * channels are rematrixed. I'm not sure if they provide a practical benefit
851 * to compression, or just obfuscate the decoder. Are they for some kind of
854 /** Generate two channels of noise, used in the matrix when
855 * restart sync word == 0x31ea. */
857 static void generate_2_noise_channels(MLPDecodeContext *m, unsigned int substr)
859 SubStream *s = &m->substream[substr];
861 uint32_t seed = s->noisegen_seed;
862 unsigned int maxchan = s->max_matrix_channel;
864 for (i = 0; i < s->blockpos; i++) {
865 uint16_t seed_shr7 = seed >> 7;
866 m->sample_buffer[i][maxchan+1] = ((int8_t)(seed >> 15)) << s->noise_shift;
867 m->sample_buffer[i][maxchan+2] = ((int8_t) seed_shr7) << s->noise_shift;
869 seed = (seed << 16) ^ seed_shr7 ^ (seed_shr7 << 5);
872 s->noisegen_seed = seed;
875 /** Generate a block of noise, used when restart sync word == 0x31eb. */
877 static void fill_noise_buffer(MLPDecodeContext *m, unsigned int substr)
879 SubStream *s = &m->substream[substr];
881 uint32_t seed = s->noisegen_seed;
883 for (i = 0; i < m->access_unit_size_pow2; i++) {
884 uint8_t seed_shr15 = seed >> 15;
885 m->noise_buffer[i] = noise_table[seed_shr15];
886 seed = (seed << 8) ^ seed_shr15 ^ (seed_shr15 << 5);
889 s->noisegen_seed = seed;
893 /** Apply the channel matrices in turn to reconstruct the original audio
896 static void rematrix_channels(MLPDecodeContext *m, unsigned int substr)
898 SubStream *s = &m->substream[substr];
899 unsigned int mat, src_ch, i;
900 unsigned int maxchan;
902 maxchan = s->max_matrix_channel;
903 if (!s->noise_type) {
904 generate_2_noise_channels(m, substr);
907 fill_noise_buffer(m, substr);
910 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
911 int matrix_noise_shift = s->matrix_noise_shift[mat];
912 unsigned int dest_ch = s->matrix_out_ch[mat];
913 int32_t mask = MSB_MASK(s->quant_step_size[dest_ch]);
915 /* TODO: DSPContext? */
917 for (i = 0; i < s->blockpos; i++) {
919 for (src_ch = 0; src_ch <= maxchan; src_ch++) {
920 accum += (int64_t)m->sample_buffer[i][src_ch]
921 * s->matrix_coeff[mat][src_ch];
923 if (matrix_noise_shift) {
924 uint32_t index = s->num_primitive_matrices - mat;
925 index = (i * (index * 2 + 1) + index) & (m->access_unit_size_pow2 - 1);
926 accum += m->noise_buffer[index] << (matrix_noise_shift + 7);
928 m->sample_buffer[i][dest_ch] = ((accum >> 14) & mask)
929 + m->bypassed_lsbs[i][mat];
934 /** Write the audio data into the output buffer. */
936 static int output_data_internal(MLPDecodeContext *m, unsigned int substr,
937 uint8_t *data, unsigned int *data_size, int is32)
939 SubStream *s = &m->substream[substr];
940 unsigned int i, ch = 0;
941 int32_t *data_32 = (int32_t*) data;
942 int16_t *data_16 = (int16_t*) data;
944 if (*data_size < (s->max_channel + 1) * s->blockpos * (is32 ? 4 : 2))
947 for (i = 0; i < s->blockpos; i++) {
948 for (ch = 0; ch <= s->max_channel; ch++) {
949 int32_t sample = m->sample_buffer[i][ch] << s->output_shift[ch];
950 s->lossless_check_data ^= (sample & 0xffffff) << ch;
951 if (is32) *data_32++ = sample << 8;
952 else *data_16++ = sample >> 8;
956 *data_size = i * ch * (is32 ? 4 : 2);
961 static int output_data(MLPDecodeContext *m, unsigned int substr,
962 uint8_t *data, unsigned int *data_size)
964 if (m->avctx->sample_fmt == SAMPLE_FMT_S32)
965 return output_data_internal(m, substr, data, data_size, 1);
967 return output_data_internal(m, substr, data, data_size, 0);
971 /** XOR together all the bytes of a buffer.
972 * Does this belong in dspcontext? */
974 static uint8_t calculate_parity(const uint8_t *buf, unsigned int buf_size)
976 uint32_t scratch = 0;
977 const uint8_t *buf_end = buf + buf_size;
979 for (; buf < buf_end - 3; buf += 4)
980 scratch ^= *((const uint32_t*)buf);
982 scratch ^= scratch >> 16;
983 scratch ^= scratch >> 8;
985 for (; buf < buf_end; buf++)
991 /** Read an access unit from the stream.
992 * Returns < 0 on error, 0 if not enough data is present in the input stream
993 * otherwise returns the number of bytes consumed. */
995 static int read_access_unit(AVCodecContext *avctx, void* data, int *data_size,
996 const uint8_t *buf, int buf_size)
998 MLPDecodeContext *m = avctx->priv_data;
1000 unsigned int length, substr;
1001 unsigned int substream_start;
1002 unsigned int header_size = 4;
1003 unsigned int substr_header_size = 0;
1004 uint8_t substream_parity_present[MAX_SUBSTREAMS];
1005 uint16_t substream_data_len[MAX_SUBSTREAMS];
1006 uint8_t parity_bits;
1011 length = (AV_RB16(buf) & 0xfff) * 2;
1013 if (length > buf_size)
1016 init_get_bits(&gb, (buf + 4), (length - 4) * 8);
1018 if (show_bits_long(&gb, 31) == (0xf8726fba >> 1)) {
1019 dprintf(m->avctx, "Found major sync.\n");
1020 if (read_major_sync(m, &gb) < 0)
1025 if (!m->params_valid) {
1026 av_log(m->avctx, AV_LOG_WARNING,
1027 "Stream parameters not seen; skipping frame.\n");
1032 substream_start = 0;
1034 for (substr = 0; substr < m->num_substreams; substr++) {
1035 int extraword_present, checkdata_present, end;
1037 extraword_present = get_bits1(&gb);
1039 checkdata_present = get_bits1(&gb);
1042 end = get_bits(&gb, 12) * 2;
1044 substr_header_size += 2;
1046 if (extraword_present) {
1048 substr_header_size += 2;
1051 if (end + header_size + substr_header_size > length) {
1052 av_log(m->avctx, AV_LOG_ERROR,
1053 "Indicated length of substream %d data goes off end of "
1054 "packet.\n", substr);
1055 end = length - header_size - substr_header_size;
1058 if (end < substream_start) {
1059 av_log(avctx, AV_LOG_ERROR,
1060 "Indicated end offset of substream %d data "
1061 "is smaller than calculated start offset.\n",
1066 if (substr > m->max_decoded_substream)
1069 substream_parity_present[substr] = checkdata_present;
1070 substream_data_len[substr] = end - substream_start;
1071 substream_start = end;
1074 parity_bits = calculate_parity(buf, 4);
1075 parity_bits ^= calculate_parity(buf + header_size, substr_header_size);
1077 if ((((parity_bits >> 4) ^ parity_bits) & 0xF) != 0xF) {
1078 av_log(avctx, AV_LOG_ERROR, "Parity check failed.\n");
1082 buf += header_size + substr_header_size;
1084 for (substr = 0; substr <= m->max_decoded_substream; substr++) {
1085 SubStream *s = &m->substream[substr];
1086 init_get_bits(&gb, buf, substream_data_len[substr] * 8);
1090 if (get_bits1(&gb)) {
1091 if (get_bits1(&gb)) {
1092 /* A restart header should be present. */
1093 if (read_restart_header(m, &gb, buf, substr) < 0)
1095 s->restart_seen = 1;
1098 if (!s->restart_seen) {
1099 av_log(m->avctx, AV_LOG_ERROR,
1100 "No restart header present in substream %d.\n",
1105 if (read_decoding_params(m, &gb, substr) < 0)
1109 if (!s->restart_seen) {
1110 av_log(m->avctx, AV_LOG_ERROR,
1111 "No restart header present in substream %d.\n",
1116 if (read_block_data(m, &gb, substr) < 0)
1119 } while ((get_bits_count(&gb) < substream_data_len[substr] * 8)
1120 && get_bits1(&gb) == 0);
1122 skip_bits(&gb, (-get_bits_count(&gb)) & 15);
1123 if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 32 &&
1124 (show_bits_long(&gb, 32) == 0xd234d234 ||
1125 show_bits_long(&gb, 20) == 0xd234e)) {
1127 if (substr == m->max_decoded_substream)
1128 av_log(m->avctx, AV_LOG_INFO, "End of stream indicated.\n");
1130 if (get_bits1(&gb)) {
1131 int shorten_by = get_bits(&gb, 13);
1132 shorten_by = FFMIN(shorten_by, s->blockpos);
1133 s->blockpos -= shorten_by;
1137 if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 16 &&
1138 substream_parity_present[substr]) {
1139 uint8_t parity, checksum;
1141 parity = calculate_parity(buf, substream_data_len[substr] - 2);
1142 if ((parity ^ get_bits(&gb, 8)) != 0xa9)
1143 av_log(m->avctx, AV_LOG_ERROR,
1144 "Substream %d parity check failed.\n", substr);
1146 checksum = mlp_checksum8(buf, substream_data_len[substr] - 2);
1147 if (checksum != get_bits(&gb, 8))
1148 av_log(m->avctx, AV_LOG_ERROR, "Substream %d checksum failed.\n",
1151 if (substream_data_len[substr] * 8 != get_bits_count(&gb)) {
1152 av_log(m->avctx, AV_LOG_ERROR, "substream %d length mismatch\n",
1158 buf += substream_data_len[substr];
1161 rematrix_channels(m, m->max_decoded_substream);
1163 if (output_data(m, m->max_decoded_substream, data, data_size) < 0)
1169 m->params_valid = 0;
1173 AVCodec mlp_decoder = {
1177 sizeof(MLPDecodeContext),
1182 .long_name = NULL_IF_CONFIG_SMALL("Meridian Lossless Packing"),