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;
342 /** Read a major sync info header - contains high level information about
343 * the stream - sample rate, channel arrangement etc. Most of this
344 * information is not actually necessary for decoding, only for playback.
347 static int read_major_sync(MLPDecodeContext *m, GetBitContext *gb)
352 if (ff_mlp_read_major_sync(m->avctx, &mh, gb) != 0)
355 if (mh.group1_bits == 0) {
356 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown bits per sample\n");
359 if (mh.group2_bits > mh.group1_bits) {
360 av_log(m->avctx, AV_LOG_ERROR,
361 "Channel group 2 cannot have more bits per sample than group 1.\n");
365 if (mh.group2_samplerate && mh.group2_samplerate != mh.group1_samplerate) {
366 av_log(m->avctx, AV_LOG_ERROR,
367 "Channel groups with differing sample rates are not currently supported.\n");
371 if (mh.group1_samplerate == 0) {
372 av_log(m->avctx, AV_LOG_ERROR, "invalid/unknown sampling rate\n");
375 if (mh.group1_samplerate > MAX_SAMPLERATE) {
376 av_log(m->avctx, AV_LOG_ERROR,
377 "Sampling rate %d is greater than the supported maximum (%d).\n",
378 mh.group1_samplerate, MAX_SAMPLERATE);
381 if (mh.access_unit_size > MAX_BLOCKSIZE) {
382 av_log(m->avctx, AV_LOG_ERROR,
383 "Block size %d is greater than the supported maximum (%d).\n",
384 mh.access_unit_size, MAX_BLOCKSIZE);
387 if (mh.access_unit_size_pow2 > MAX_BLOCKSIZE_POW2) {
388 av_log(m->avctx, AV_LOG_ERROR,
389 "Block size pow2 %d is greater than the supported maximum (%d).\n",
390 mh.access_unit_size_pow2, MAX_BLOCKSIZE_POW2);
394 if (mh.num_substreams == 0)
396 if (mh.num_substreams > MAX_SUBSTREAMS) {
397 av_log(m->avctx, AV_LOG_ERROR,
398 "Number of substreams %d is larger than the maximum supported "
399 "by the decoder. %s\n", mh.num_substreams, sample_message);
403 m->access_unit_size = mh.access_unit_size;
404 m->access_unit_size_pow2 = mh.access_unit_size_pow2;
406 m->num_substreams = mh.num_substreams;
407 m->max_decoded_substream = m->num_substreams - 1;
409 m->avctx->sample_rate = mh.group1_samplerate;
410 m->avctx->frame_size = mh.access_unit_size;
412 #ifdef CONFIG_AUDIO_NONSHORT
413 m->avctx->bits_per_sample = mh.group1_bits;
414 if (mh.group1_bits > 16) {
415 m->avctx->sample_fmt = SAMPLE_FMT_S32;
420 for (substr = 0; substr < MAX_SUBSTREAMS; substr++)
421 m->substream[substr].restart_seen = 0;
426 /** Read a restart header from a block in a substream. This contains parameters
427 * required to decode the audio that do not change very often. Generally
428 * (always) present only in blocks following a major sync. */
430 static int read_restart_header(MLPDecodeContext *m, GetBitContext *gbp,
431 const uint8_t *buf, unsigned int substr)
433 SubStream *s = &m->substream[substr];
437 uint8_t lossless_check;
438 int start_count = get_bits_count(gbp);
440 sync_word = get_bits(gbp, 13);
442 if (sync_word != 0x31ea >> 1) {
443 av_log(m->avctx, AV_LOG_ERROR,
444 "restart header sync incorrect (got 0x%04x)\n", sync_word);
447 s->noise_type = get_bits1(gbp);
449 skip_bits(gbp, 16); /* Output timestamp */
451 s->min_channel = get_bits(gbp, 4);
452 s->max_channel = get_bits(gbp, 4);
453 s->max_matrix_channel = get_bits(gbp, 4);
455 if (s->min_channel > s->max_channel) {
456 av_log(m->avctx, AV_LOG_ERROR,
457 "Substream min channel cannot be greater than max channel.\n");
461 if (m->avctx->request_channels > 0
462 && s->max_channel + 1 >= m->avctx->request_channels
463 && substr < m->max_decoded_substream) {
464 av_log(m->avctx, AV_LOG_INFO,
465 "Extracting %d channel downmix from substream %d. "
466 "Further substreams will be skipped.\n",
467 s->max_channel + 1, substr);
468 m->max_decoded_substream = substr;
471 s->noise_shift = get_bits(gbp, 4);
472 s->noisegen_seed = get_bits(gbp, 23);
476 s->data_check_present = get_bits1(gbp);
477 lossless_check = get_bits(gbp, 8);
478 if (substr == m->max_decoded_substream
479 && s->lossless_check_data != 0xffffffff) {
480 tmp = s->lossless_check_data;
484 if (tmp != lossless_check)
485 av_log(m->avctx, AV_LOG_WARNING,
486 "Lossless check failed - expected %02x, calculated %02x.\n",
487 lossless_check, tmp);
489 dprintf(m->avctx, "Lossless check passed for substream %d (%x).\n",
495 for (ch = 0; ch <= s->max_matrix_channel; ch++) {
496 int ch_assign = get_bits(gbp, 6);
497 dprintf(m->avctx, "ch_assign[%d][%d] = %d\n", substr, ch,
499 if (ch_assign != ch) {
500 av_log(m->avctx, AV_LOG_ERROR,
501 "Non-1:1 channel assignments are used in this stream. %s\n",
507 checksum = mlp_restart_checksum(buf, get_bits_count(gbp) - start_count);
509 if (checksum != get_bits(gbp, 8))
510 av_log(m->avctx, AV_LOG_ERROR, "restart header checksum error\n");
512 /* Set default decoding parameters. */
513 s->param_presence_flags = 0xff;
514 s->num_primitive_matrices = 0;
516 s->lossless_check_data = 0;
518 memset(s->output_shift , 0, sizeof(s->output_shift ));
519 memset(s->quant_step_size, 0, sizeof(s->quant_step_size));
521 for (ch = s->min_channel; ch <= s->max_channel; ch++) {
522 m->filter_order[ch][FIR] = 0;
523 m->filter_order[ch][IIR] = 0;
524 m->filter_shift[ch][FIR] = 0;
525 m->filter_shift[ch][IIR] = 0;
527 /* Default audio coding is 24-bit raw PCM. */
528 m->huff_offset [ch] = 0;
529 m->sign_huff_offset[ch] = (-1) << 23;
530 m->codebook [ch] = 0;
531 m->huff_lsbs [ch] = 24;
534 if (substr == m->max_decoded_substream) {
535 m->avctx->channels = s->max_channel + 1;
541 /** Read parameters for one of the prediction filters. */
543 static int read_filter_params(MLPDecodeContext *m, GetBitContext *gbp,
544 unsigned int channel, unsigned int filter)
546 const char fchar = filter ? 'I' : 'F';
549 // Filter is 0 for FIR, 1 for IIR.
552 order = get_bits(gbp, 4);
553 if (order > MAX_FILTER_ORDER) {
554 av_log(m->avctx, AV_LOG_ERROR,
555 "%cIR filter order %d is greater than maximum %d.\n",
556 fchar, order, MAX_FILTER_ORDER);
559 m->filter_order[channel][filter] = order;
562 int coeff_bits, coeff_shift;
564 m->filter_shift[channel][filter] = get_bits(gbp, 4);
566 coeff_bits = get_bits(gbp, 5);
567 coeff_shift = get_bits(gbp, 3);
568 if (coeff_bits < 1 || coeff_bits > 16) {
569 av_log(m->avctx, AV_LOG_ERROR,
570 "%cIR filter coeff_bits must be between 1 and 16.\n",
574 if (coeff_bits + coeff_shift > 16) {
575 av_log(m->avctx, AV_LOG_ERROR,
576 "Sum of coeff_bits and coeff_shift for %cIR filter must be 16 or less.\n",
581 for (i = 0; i < order; i++)
582 m->filter_coeff[channel][filter][i] =
583 get_sbits(gbp, coeff_bits) << coeff_shift;
585 if (get_bits1(gbp)) {
586 int state_bits, state_shift;
589 av_log(m->avctx, AV_LOG_ERROR,
590 "FIR filter has state data specified.\n");
594 state_bits = get_bits(gbp, 4);
595 state_shift = get_bits(gbp, 4);
597 /* TODO: Check validity of state data. */
599 for (i = 0; i < order; i++)
600 m->filter_state[channel][filter][i] =
601 get_sbits(gbp, state_bits) << state_shift;
608 /** Read decoding parameters that change more often than those in the restart
611 static int read_decoding_params(MLPDecodeContext *m, GetBitContext *gbp,
614 SubStream *s = &m->substream[substr];
615 unsigned int mat, ch;
618 s->param_presence_flags = get_bits(gbp, 8);
620 if (s->param_presence_flags & PARAM_BLOCKSIZE)
621 if (get_bits1(gbp)) {
622 s->blocksize = get_bits(gbp, 9);
623 if (s->blocksize > MAX_BLOCKSIZE) {
624 av_log(m->avctx, AV_LOG_ERROR, "block size too large\n");
630 if (s->param_presence_flags & PARAM_MATRIX)
631 if (get_bits1(gbp)) {
632 s->num_primitive_matrices = get_bits(gbp, 4);
634 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
635 int frac_bits, max_chan;
636 s->matrix_out_ch[mat] = get_bits(gbp, 4);
637 frac_bits = get_bits(gbp, 4);
638 s->lsb_bypass [mat] = get_bits1(gbp);
640 if (s->matrix_out_ch[mat] > s->max_channel) {
641 av_log(m->avctx, AV_LOG_ERROR,
642 "Invalid channel %d specified as output from matrix.\n",
643 s->matrix_out_ch[mat]);
646 if (frac_bits > 14) {
647 av_log(m->avctx, AV_LOG_ERROR,
648 "Too many fractional bits specified.\n");
652 max_chan = s->max_matrix_channel;
656 for (ch = 0; ch <= max_chan; ch++) {
659 coeff_val = get_sbits(gbp, frac_bits + 2);
661 s->matrix_coeff[mat][ch] = coeff_val << (14 - frac_bits);
665 s->matrix_noise_shift[mat] = get_bits(gbp, 4);
667 s->matrix_noise_shift[mat] = 0;
671 if (s->param_presence_flags & PARAM_OUTSHIFT)
673 for (ch = 0; ch <= s->max_matrix_channel; ch++) {
674 s->output_shift[ch] = get_bits(gbp, 4);
675 dprintf(m->avctx, "output shift[%d] = %d\n",
676 ch, s->output_shift[ch]);
680 if (s->param_presence_flags & PARAM_QUANTSTEP)
682 for (ch = 0; ch <= s->max_channel; ch++) {
683 s->quant_step_size[ch] = get_bits(gbp, 4);
686 m->sign_huff_offset[ch] = calculate_sign_huff(m, substr, ch);
689 for (ch = s->min_channel; ch <= s->max_channel; ch++)
690 if (get_bits1(gbp)) {
691 if (s->param_presence_flags & PARAM_FIR)
693 if (read_filter_params(m, gbp, ch, FIR) < 0)
696 if (s->param_presence_flags & PARAM_IIR)
698 if (read_filter_params(m, gbp, ch, IIR) < 0)
701 if (m->filter_order[ch][FIR] && m->filter_order[ch][IIR] &&
702 m->filter_shift[ch][FIR] != m->filter_shift[ch][IIR]) {
703 av_log(m->avctx, AV_LOG_ERROR,
704 "FIR and IIR filters must use the same precision.\n");
707 /* The FIR and IIR filters must have the same precision.
708 * To simplify the filtering code, only the precision of the
709 * FIR filter is considered. If only the IIR filter is employed,
710 * the FIR filter precision is set to that of the IIR filter, so
711 * that the filtering code can use it. */
712 if (!m->filter_order[ch][FIR] && m->filter_order[ch][IIR])
713 m->filter_shift[ch][FIR] = m->filter_shift[ch][IIR];
715 if (s->param_presence_flags & PARAM_HUFFOFFSET)
717 m->huff_offset[ch] = get_sbits(gbp, 15);
719 m->codebook [ch] = get_bits(gbp, 2);
720 m->huff_lsbs[ch] = get_bits(gbp, 5);
722 m->sign_huff_offset[ch] = calculate_sign_huff(m, substr, ch);
730 #define MSB_MASK(bits) (-1u << bits)
732 /** Generate PCM samples using the prediction filters and residual values
733 * read from the data stream, and update the filter state. */
735 static void filter_channel(MLPDecodeContext *m, unsigned int substr,
736 unsigned int channel)
738 SubStream *s = &m->substream[substr];
739 int32_t filter_state_buffer[NUM_FILTERS][MAX_BLOCKSIZE + MAX_FILTER_ORDER];
740 unsigned int filter_shift = m->filter_shift[channel][FIR];
741 int32_t mask = MSB_MASK(s->quant_step_size[channel]);
742 int index = MAX_BLOCKSIZE;
745 for (j = 0; j < NUM_FILTERS; j++) {
746 memcpy(& filter_state_buffer [j][MAX_BLOCKSIZE],
747 &m->filter_state[channel][j][0],
748 MAX_FILTER_ORDER * sizeof(int32_t));
751 for (i = 0; i < s->blocksize; i++) {
752 int32_t residual = m->sample_buffer[i + s->blockpos][channel];
757 /* TODO: Move this code to DSPContext? */
759 for (j = 0; j < NUM_FILTERS; j++)
760 for (order = 0; order < m->filter_order[channel][j]; order++)
761 accum += (int64_t)filter_state_buffer[j][index + order] *
762 m->filter_coeff[channel][j][order];
764 accum = accum >> filter_shift;
765 result = (accum + residual) & mask;
769 filter_state_buffer[FIR][index] = result;
770 filter_state_buffer[IIR][index] = result - accum;
772 m->sample_buffer[i + s->blockpos][channel] = result;
775 for (j = 0; j < NUM_FILTERS; j++) {
776 memcpy(&m->filter_state[channel][j][0],
777 & filter_state_buffer [j][index],
778 MAX_FILTER_ORDER * sizeof(int32_t));
782 /** Read a block of PCM residual data (or actual if no filtering active). */
784 static int read_block_data(MLPDecodeContext *m, GetBitContext *gbp,
787 SubStream *s = &m->substream[substr];
788 unsigned int i, ch, expected_stream_pos = 0;
790 if (s->data_check_present) {
791 expected_stream_pos = get_bits_count(gbp);
792 expected_stream_pos += get_bits(gbp, 16);
793 av_log(m->avctx, AV_LOG_WARNING, "This file contains some features "
794 "we have not tested yet. %s\n", sample_message);
797 if (s->blockpos + s->blocksize > m->access_unit_size) {
798 av_log(m->avctx, AV_LOG_ERROR, "too many audio samples in frame\n");
802 memset(&m->bypassed_lsbs[s->blockpos][0], 0,
803 s->blocksize * sizeof(m->bypassed_lsbs[0]));
805 for (i = 0; i < s->blocksize; i++) {
806 if (read_huff_channels(m, gbp, substr, i) < 0)
810 for (ch = s->min_channel; ch <= s->max_channel; ch++) {
811 filter_channel(m, substr, ch);
814 s->blockpos += s->blocksize;
816 if (s->data_check_present) {
817 if (get_bits_count(gbp) != expected_stream_pos)
818 av_log(m->avctx, AV_LOG_ERROR, "block data length mismatch\n");
825 /** Data table used for TrueHD noise generation function. */
827 static const int8_t noise_table[256] = {
828 30, 51, 22, 54, 3, 7, -4, 38, 14, 55, 46, 81, 22, 58, -3, 2,
829 52, 31, -7, 51, 15, 44, 74, 30, 85, -17, 10, 33, 18, 80, 28, 62,
830 10, 32, 23, 69, 72, 26, 35, 17, 73, 60, 8, 56, 2, 6, -2, -5,
831 51, 4, 11, 50, 66, 76, 21, 44, 33, 47, 1, 26, 64, 48, 57, 40,
832 38, 16, -10, -28, 92, 22, -18, 29, -10, 5, -13, 49, 19, 24, 70, 34,
833 61, 48, 30, 14, -6, 25, 58, 33, 42, 60, 67, 17, 54, 17, 22, 30,
834 67, 44, -9, 50, -11, 43, 40, 32, 59, 82, 13, 49, -14, 55, 60, 36,
835 48, 49, 31, 47, 15, 12, 4, 65, 1, 23, 29, 39, 45, -2, 84, 69,
836 0, 72, 37, 57, 27, 41, -15, -16, 35, 31, 14, 61, 24, 0, 27, 24,
837 16, 41, 55, 34, 53, 9, 56, 12, 25, 29, 53, 5, 20, -20, -8, 20,
838 13, 28, -3, 78, 38, 16, 11, 62, 46, 29, 21, 24, 46, 65, 43, -23,
839 89, 18, 74, 21, 38, -12, 19, 12, -19, 8, 15, 33, 4, 57, 9, -8,
840 36, 35, 26, 28, 7, 83, 63, 79, 75, 11, 3, 87, 37, 47, 34, 40,
841 39, 19, 20, 42, 27, 34, 39, 77, 13, 42, 59, 64, 45, -1, 32, 37,
842 45, -5, 53, -6, 7, 36, 50, 23, 6, 32, 9, -21, 18, 71, 27, 52,
843 -25, 31, 35, 42, -1, 68, 63, 52, 26, 43, 66, 37, 41, 25, 40, 70,
846 /** Noise generation functions.
847 * I'm not sure what these are for - they seem to be some kind of pseudorandom
848 * sequence generators, used to generate noise data which is used when the
849 * channels are rematrixed. I'm not sure if they provide a practical benefit
850 * to compression, or just obfuscate the decoder. Are they for some kind of
853 /** Generate two channels of noise, used in the matrix when
854 * restart sync word == 0x31ea. */
856 static void generate_2_noise_channels(MLPDecodeContext *m, unsigned int substr)
858 SubStream *s = &m->substream[substr];
860 uint32_t seed = s->noisegen_seed;
861 unsigned int maxchan = s->max_matrix_channel;
863 for (i = 0; i < s->blockpos; i++) {
864 uint16_t seed_shr7 = seed >> 7;
865 m->sample_buffer[i][maxchan+1] = ((int8_t)(seed >> 15)) << s->noise_shift;
866 m->sample_buffer[i][maxchan+2] = ((int8_t) seed_shr7) << s->noise_shift;
868 seed = (seed << 16) ^ seed_shr7 ^ (seed_shr7 << 5);
871 s->noisegen_seed = seed;
874 /** Generate a block of noise, used when restart sync word == 0x31eb. */
876 static void fill_noise_buffer(MLPDecodeContext *m, unsigned int substr)
878 SubStream *s = &m->substream[substr];
880 uint32_t seed = s->noisegen_seed;
882 for (i = 0; i < m->access_unit_size_pow2; i++) {
883 uint8_t seed_shr15 = seed >> 15;
884 m->noise_buffer[i] = noise_table[seed_shr15];
885 seed = (seed << 8) ^ seed_shr15 ^ (seed_shr15 << 5);
888 s->noisegen_seed = seed;
892 /** Apply the channel matrices in turn to reconstruct the original audio
895 static void rematrix_channels(MLPDecodeContext *m, unsigned int substr)
897 SubStream *s = &m->substream[substr];
898 unsigned int mat, src_ch, i;
899 unsigned int maxchan;
901 maxchan = s->max_matrix_channel;
902 if (!s->noise_type) {
903 generate_2_noise_channels(m, substr);
906 fill_noise_buffer(m, substr);
909 for (mat = 0; mat < s->num_primitive_matrices; mat++) {
910 int matrix_noise_shift = s->matrix_noise_shift[mat];
911 unsigned int dest_ch = s->matrix_out_ch[mat];
912 int32_t mask = MSB_MASK(s->quant_step_size[dest_ch]);
914 /* TODO: DSPContext? */
916 for (i = 0; i < s->blockpos; i++) {
918 for (src_ch = 0; src_ch <= maxchan; src_ch++) {
919 accum += (int64_t)m->sample_buffer[i][src_ch]
920 * s->matrix_coeff[mat][src_ch];
922 if (matrix_noise_shift) {
923 uint32_t index = s->num_primitive_matrices - mat;
924 index = (i * (index * 2 + 1) + index) & (m->access_unit_size_pow2 - 1);
925 accum += m->noise_buffer[index] << (matrix_noise_shift + 7);
927 m->sample_buffer[i][dest_ch] = ((accum >> 14) & mask)
928 + m->bypassed_lsbs[i][mat];
933 /** Write the audio data into the output buffer. */
935 static int output_data_internal(MLPDecodeContext *m, unsigned int substr,
936 uint8_t *data, unsigned int *data_size, int is32)
938 SubStream *s = &m->substream[substr];
939 unsigned int i, ch = 0;
940 int32_t *data_32 = (int32_t*) data;
941 int16_t *data_16 = (int16_t*) data;
943 if (*data_size < (s->max_channel + 1) * s->blockpos * (is32 ? 4 : 2))
946 for (i = 0; i < s->blockpos; i++) {
947 for (ch = 0; ch <= s->max_channel; ch++) {
948 int32_t sample = m->sample_buffer[i][ch] << s->output_shift[ch];
949 s->lossless_check_data ^= (sample & 0xffffff) << ch;
950 if (is32) *data_32++ = sample << 8;
951 else *data_16++ = sample >> 8;
955 *data_size = i * ch * (is32 ? 4 : 2);
960 static int output_data(MLPDecodeContext *m, unsigned int substr,
961 uint8_t *data, unsigned int *data_size)
963 if (m->avctx->sample_fmt == SAMPLE_FMT_S32)
964 return output_data_internal(m, substr, data, data_size, 1);
966 return output_data_internal(m, substr, data, data_size, 0);
970 /** XOR together all the bytes of a buffer.
971 * Does this belong in dspcontext? */
973 static uint8_t calculate_parity(const uint8_t *buf, unsigned int buf_size)
975 uint32_t scratch = 0;
976 const uint8_t *buf_end = buf + buf_size;
978 for (; buf < buf_end - 3; buf += 4)
979 scratch ^= *((const uint32_t*)buf);
981 scratch ^= scratch >> 16;
982 scratch ^= scratch >> 8;
984 for (; buf < buf_end; buf++)
990 /** Read an access unit from the stream.
991 * Returns < 0 on error, 0 if not enough data is present in the input stream
992 * otherwise returns the number of bytes consumed. */
994 static int read_access_unit(AVCodecContext *avctx, void* data, int *data_size,
995 const uint8_t *buf, int buf_size)
997 MLPDecodeContext *m = avctx->priv_data;
999 unsigned int length, substr;
1000 unsigned int substream_start;
1001 unsigned int header_size = 4;
1002 unsigned int substr_header_size = 0;
1003 uint8_t substream_parity_present[MAX_SUBSTREAMS];
1004 uint16_t substream_data_len[MAX_SUBSTREAMS];
1005 uint8_t parity_bits;
1010 length = (AV_RB16(buf) & 0xfff) * 2;
1012 if (length > buf_size)
1015 init_get_bits(&gb, (buf + 4), (length - 4) * 8);
1017 if (show_bits_long(&gb, 31) == (0xf8726fba >> 1)) {
1018 dprintf(m->avctx, "Found major sync.\n");
1019 if (read_major_sync(m, &gb) < 0)
1024 if (!m->params_valid) {
1025 av_log(m->avctx, AV_LOG_WARNING,
1026 "Stream parameters not seen; skipping frame.\n");
1031 substream_start = 0;
1033 for (substr = 0; substr < m->num_substreams; substr++) {
1034 int extraword_present, checkdata_present, end;
1036 extraword_present = get_bits1(&gb);
1038 checkdata_present = get_bits1(&gb);
1041 end = get_bits(&gb, 12) * 2;
1043 substr_header_size += 2;
1045 if (extraword_present) {
1047 substr_header_size += 2;
1050 if (end + header_size + substr_header_size > length) {
1051 av_log(m->avctx, AV_LOG_ERROR,
1052 "Indicated length of substream %d data goes off end of "
1053 "packet.\n", substr);
1054 end = length - header_size - substr_header_size;
1057 if (end < substream_start) {
1058 av_log(avctx, AV_LOG_ERROR,
1059 "Indicated end offset of substream %d data "
1060 "is smaller than calculated start offset.\n",
1065 if (substr > m->max_decoded_substream)
1068 substream_parity_present[substr] = checkdata_present;
1069 substream_data_len[substr] = end - substream_start;
1070 substream_start = end;
1073 parity_bits = calculate_parity(buf, 4);
1074 parity_bits ^= calculate_parity(buf + header_size, substr_header_size);
1076 if ((((parity_bits >> 4) ^ parity_bits) & 0xF) != 0xF) {
1077 av_log(avctx, AV_LOG_ERROR, "Parity check failed.\n");
1081 buf += header_size + substr_header_size;
1083 for (substr = 0; substr <= m->max_decoded_substream; substr++) {
1084 SubStream *s = &m->substream[substr];
1085 init_get_bits(&gb, buf, substream_data_len[substr] * 8);
1089 if (get_bits1(&gb)) {
1090 if (get_bits1(&gb)) {
1091 /* A restart header should be present. */
1092 if (read_restart_header(m, &gb, buf, substr) < 0)
1094 s->restart_seen = 1;
1097 if (!s->restart_seen) {
1098 av_log(m->avctx, AV_LOG_ERROR,
1099 "No restart header present in substream %d.\n",
1104 if (read_decoding_params(m, &gb, substr) < 0)
1108 if (!s->restart_seen) {
1109 av_log(m->avctx, AV_LOG_ERROR,
1110 "No restart header present in substream %d.\n",
1115 if (read_block_data(m, &gb, substr) < 0)
1118 } while ((get_bits_count(&gb) < substream_data_len[substr] * 8)
1119 && get_bits1(&gb) == 0);
1121 skip_bits(&gb, (-get_bits_count(&gb)) & 15);
1122 if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 32 &&
1123 (show_bits_long(&gb, 32) == 0xd234d234 ||
1124 show_bits_long(&gb, 20) == 0xd234e)) {
1126 if (substr == m->max_decoded_substream)
1127 av_log(m->avctx, AV_LOG_INFO, "End of stream indicated.\n");
1129 if (get_bits1(&gb)) {
1130 int shorten_by = get_bits(&gb, 13);
1131 shorten_by = FFMIN(shorten_by, s->blockpos);
1132 s->blockpos -= shorten_by;
1136 if (substream_data_len[substr] * 8 - get_bits_count(&gb) >= 16 &&
1137 substream_parity_present[substr]) {
1138 uint8_t parity, checksum;
1140 parity = calculate_parity(buf, substream_data_len[substr] - 2);
1141 if ((parity ^ get_bits(&gb, 8)) != 0xa9)
1142 av_log(m->avctx, AV_LOG_ERROR,
1143 "Substream %d parity check failed.\n", substr);
1145 checksum = mlp_checksum8(buf, substream_data_len[substr] - 2);
1146 if (checksum != get_bits(&gb, 8))
1147 av_log(m->avctx, AV_LOG_ERROR, "Substream %d checksum failed.\n",
1150 if (substream_data_len[substr] * 8 != get_bits_count(&gb)) {
1151 av_log(m->avctx, AV_LOG_ERROR, "substream %d length mismatch\n",
1157 buf += substream_data_len[substr];
1160 rematrix_channels(m, m->max_decoded_substream);
1162 if (output_data(m, m->max_decoded_substream, data, data_size) < 0)
1168 m->params_valid = 0;
1172 AVCodec mlp_decoder = {
1176 sizeof(MLPDecodeContext),
1181 .long_name = NULL_IF_CONFIG_SMALL("Meridian Lossless Packing"),