3 * Copyright (c) 2009 Thilo Borgmann <thilo.borgmann _at_ googlemail.com>
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
23 * @file libavcodec/alsdec.c
25 * @author Thilo Borgmann <thilo.borgmann _at_ googlemail.com>
35 #include "mpeg4audio.h"
36 #include "bytestream.h"
40 /** Rice parameters and corresponding index offsets for decoding the
41 * indices of scaled PARCOR values. The table choosen is set globally
42 * by the encoder and stored in ALSSpecificConfig.
44 static const int8_t parcor_rice_table[3][20][2] = {
45 { {-52, 4}, {-29, 5}, {-31, 4}, { 19, 4}, {-16, 4},
46 { 12, 3}, { -7, 3}, { 9, 3}, { -5, 3}, { 6, 3},
47 { -4, 3}, { 3, 3}, { -3, 2}, { 3, 2}, { -2, 2},
48 { 3, 2}, { -1, 2}, { 2, 2}, { -1, 2}, { 2, 2} },
49 { {-58, 3}, {-42, 4}, {-46, 4}, { 37, 5}, {-36, 4},
50 { 29, 4}, {-29, 4}, { 25, 4}, {-23, 4}, { 20, 4},
51 {-17, 4}, { 16, 4}, {-12, 4}, { 12, 3}, {-10, 4},
52 { 7, 3}, { -4, 4}, { 3, 3}, { -1, 3}, { 1, 3} },
53 { {-59, 3}, {-45, 5}, {-50, 4}, { 38, 4}, {-39, 4},
54 { 32, 4}, {-30, 4}, { 25, 3}, {-23, 3}, { 20, 3},
55 {-20, 3}, { 16, 3}, {-13, 3}, { 10, 3}, { -7, 3},
56 { 3, 3}, { 0, 3}, { -1, 3}, { 2, 3}, { -1, 2} }
60 /** Scaled PARCOR values used for the first two PARCOR coefficients.
61 * To be indexed by the Rice coded indices.
62 * Generated by: parcor_scaled_values[i] = 32 + ((i * (i+1)) << 7) - (1 << 20)
63 * Actual values are divided by 32 in order to be stored in 16 bits.
65 static const int16_t parcor_scaled_values[] = {
66 -1048544 / 32, -1048288 / 32, -1047776 / 32, -1047008 / 32,
67 -1045984 / 32, -1044704 / 32, -1043168 / 32, -1041376 / 32,
68 -1039328 / 32, -1037024 / 32, -1034464 / 32, -1031648 / 32,
69 -1028576 / 32, -1025248 / 32, -1021664 / 32, -1017824 / 32,
70 -1013728 / 32, -1009376 / 32, -1004768 / 32, -999904 / 32,
71 -994784 / 32, -989408 / 32, -983776 / 32, -977888 / 32,
72 -971744 / 32, -965344 / 32, -958688 / 32, -951776 / 32,
73 -944608 / 32, -937184 / 32, -929504 / 32, -921568 / 32,
74 -913376 / 32, -904928 / 32, -896224 / 32, -887264 / 32,
75 -878048 / 32, -868576 / 32, -858848 / 32, -848864 / 32,
76 -838624 / 32, -828128 / 32, -817376 / 32, -806368 / 32,
77 -795104 / 32, -783584 / 32, -771808 / 32, -759776 / 32,
78 -747488 / 32, -734944 / 32, -722144 / 32, -709088 / 32,
79 -695776 / 32, -682208 / 32, -668384 / 32, -654304 / 32,
80 -639968 / 32, -625376 / 32, -610528 / 32, -595424 / 32,
81 -580064 / 32, -564448 / 32, -548576 / 32, -532448 / 32,
82 -516064 / 32, -499424 / 32, -482528 / 32, -465376 / 32,
83 -447968 / 32, -430304 / 32, -412384 / 32, -394208 / 32,
84 -375776 / 32, -357088 / 32, -338144 / 32, -318944 / 32,
85 -299488 / 32, -279776 / 32, -259808 / 32, -239584 / 32,
86 -219104 / 32, -198368 / 32, -177376 / 32, -156128 / 32,
87 -134624 / 32, -112864 / 32, -90848 / 32, -68576 / 32,
88 -46048 / 32, -23264 / 32, -224 / 32, 23072 / 32,
89 46624 / 32, 70432 / 32, 94496 / 32, 118816 / 32,
90 143392 / 32, 168224 / 32, 193312 / 32, 218656 / 32,
91 244256 / 32, 270112 / 32, 296224 / 32, 322592 / 32,
92 349216 / 32, 376096 / 32, 403232 / 32, 430624 / 32,
93 458272 / 32, 486176 / 32, 514336 / 32, 542752 / 32,
94 571424 / 32, 600352 / 32, 629536 / 32, 658976 / 32,
95 688672 / 32, 718624 / 32, 748832 / 32, 779296 / 32,
96 810016 / 32, 840992 / 32, 872224 / 32, 903712 / 32,
97 935456 / 32, 967456 / 32, 999712 / 32, 1032224 / 32
101 /** Gain values of p(0) for long-term prediction.
102 * To be indexed by the Rice coded indices.
104 static const uint8_t ltp_gain_values [4][4] = {
120 uint32_t samples; ///< number of samples, 0xFFFFFFFF if unknown
121 int resolution; ///< 000 = 8-bit; 001 = 16-bit; 010 = 24-bit; 011 = 32-bit
122 int floating; ///< 1 = IEEE 32-bit floating-point, 0 = integer
123 int frame_length; ///< frame length for each frame (last frame may differ)
124 int ra_distance; ///< distance between RA frames (in frames, 0...255)
125 enum RA_Flag ra_flag; ///< indicates where the size of ra units is stored
126 int adapt_order; ///< adaptive order: 1 = on, 0 = off
127 int coef_table; ///< table index of Rice code parameters
128 int long_term_prediction; ///< long term prediction (LTP): 1 = on, 0 = off
129 int max_order; ///< maximum prediction order (0..1023)
130 int block_switching; ///< number of block switching levels
131 int bgmc; ///< "Block Gilbert-Moore Code": 1 = on, 0 = off (Rice coding only)
132 int sb_part; ///< sub-block partition
133 int joint_stereo; ///< joint stereo: 1 = on, 0 = off
134 int mc_coding; ///< extended inter-channel coding (multi channel coding): 1 = on, 0 = off
135 int chan_config; ///< indicates that a chan_config_info field is present
136 int chan_sort; ///< channel rearrangement: 1 = on, 0 = off
137 int rlslms; ///< use "Recursive Least Square-Least Mean Square" predictor: 1 = on, 0 = off
138 int chan_config_info; ///< mapping of channels to loudspeaker locations. Unused until setting channel configuration is implemented.
139 int *chan_pos; ///< original channel positions
140 uint32_t header_size; ///< header size of original audio file in bytes, provided for debugging
141 uint32_t trailer_size; ///< trailer size of original audio file in bytes, provided for debugging
146 AVCodecContext *avctx;
147 ALSSpecificConfig sconf;
149 unsigned int cur_frame_length; ///< length of the current frame to decode
150 unsigned int frame_id; ///< the frame ID / number of the current frame
151 unsigned int js_switch; ///< if true, joint-stereo decoding is enforced
152 unsigned int num_blocks; ///< number of blocks used in the current frame
153 int ltp_lag_length; ///< number of bits used for ltp lag value
154 int *use_ltp; ///< contains use_ltp flags for all channels
155 int *ltp_lag; ///< contains ltp lag values for all channels
156 int **ltp_gain; ///< gain values for ltp 5-tap filter for a channel
157 int *ltp_gain_buffer; ///< contains all gain values for ltp 5-tap filter
158 int32_t *quant_cof; ///< quantized parcor coefficients
159 int32_t *lpc_cof; ///< coefficients of the direct form prediction filter
160 int32_t *prev_raw_samples; ///< contains unshifted raw samples from the previous block
161 int32_t **raw_samples; ///< decoded raw samples for each channel
162 int32_t *raw_buffer; ///< contains all decoded raw samples including carryover samples
167 unsigned int block_length; ///< number of samples within the block
168 unsigned int ra_block; ///< if true, this is a random access block
169 int const_block; ///< if true, this is a constant value block
170 int32_t const_val; ///< the sample value of a constant block
171 int js_blocks; ///< true if this block contains a difference signal
172 unsigned int shift_lsbs; ///< shift of values for this block
173 unsigned int opt_order; ///< prediction order of this block
174 int store_prev_samples;///< if true, carryover samples have to be stored
175 int *use_ltp; ///< if true, long-term prediction is used
176 int *ltp_lag; ///< lag value for long-term prediction
177 int *ltp_gain; ///< gain values for ltp 5-tap filter
178 int32_t *quant_cof; ///< quantized parcor coefficients
179 int32_t *lpc_cof; ///< coefficients of the direct form prediction
180 int32_t *raw_samples; ///< decoded raw samples / residuals for this block
181 int32_t *prev_raw_samples; ///< contains unshifted raw samples from the previous block
182 int32_t *raw_other; ///< decoded raw samples of the other channel of a channel pair
186 static av_cold void dprint_specific_config(ALSDecContext *ctx)
189 AVCodecContext *avctx = ctx->avctx;
190 ALSSpecificConfig *sconf = &ctx->sconf;
192 dprintf(avctx, "resolution = %i\n", sconf->resolution);
193 dprintf(avctx, "floating = %i\n", sconf->floating);
194 dprintf(avctx, "frame_length = %i\n", sconf->frame_length);
195 dprintf(avctx, "ra_distance = %i\n", sconf->ra_distance);
196 dprintf(avctx, "ra_flag = %i\n", sconf->ra_flag);
197 dprintf(avctx, "adapt_order = %i\n", sconf->adapt_order);
198 dprintf(avctx, "coef_table = %i\n", sconf->coef_table);
199 dprintf(avctx, "long_term_prediction = %i\n", sconf->long_term_prediction);
200 dprintf(avctx, "max_order = %i\n", sconf->max_order);
201 dprintf(avctx, "block_switching = %i\n", sconf->block_switching);
202 dprintf(avctx, "bgmc = %i\n", sconf->bgmc);
203 dprintf(avctx, "sb_part = %i\n", sconf->sb_part);
204 dprintf(avctx, "joint_stereo = %i\n", sconf->joint_stereo);
205 dprintf(avctx, "mc_coding = %i\n", sconf->mc_coding);
206 dprintf(avctx, "chan_config = %i\n", sconf->chan_config);
207 dprintf(avctx, "chan_sort = %i\n", sconf->chan_sort);
208 dprintf(avctx, "RLSLMS = %i\n", sconf->rlslms);
209 dprintf(avctx, "chan_config_info = %i\n", sconf->chan_config_info);
210 dprintf(avctx, "header_size = %i\n", sconf->header_size);
211 dprintf(avctx, "trailer_size = %i\n", sconf->trailer_size);
216 /** Reads an ALSSpecificConfig from a buffer into the output struct.
218 static av_cold int read_specific_config(ALSDecContext *ctx)
222 int i, config_offset, crc_enabled;
223 MPEG4AudioConfig m4ac;
224 ALSSpecificConfig *sconf = &ctx->sconf;
225 AVCodecContext *avctx = ctx->avctx;
228 init_get_bits(&gb, avctx->extradata, avctx->extradata_size * 8);
230 config_offset = ff_mpeg4audio_get_config(&m4ac, avctx->extradata,
231 avctx->extradata_size);
233 if (config_offset < 0)
236 skip_bits_long(&gb, config_offset);
238 if (get_bits_left(&gb) < (30 << 3))
241 // read the fixed items
242 als_id = get_bits_long(&gb, 32);
243 avctx->sample_rate = m4ac.sample_rate;
244 skip_bits_long(&gb, 32); // sample rate already known
245 sconf->samples = get_bits_long(&gb, 32);
246 avctx->channels = m4ac.channels;
247 skip_bits(&gb, 16); // number of channels already knwon
248 skip_bits(&gb, 3); // skip file_type
249 sconf->resolution = get_bits(&gb, 3);
250 sconf->floating = get_bits1(&gb);
251 skip_bits1(&gb); // skip msb_first
252 sconf->frame_length = get_bits(&gb, 16) + 1;
253 sconf->ra_distance = get_bits(&gb, 8);
254 sconf->ra_flag = get_bits(&gb, 2);
255 sconf->adapt_order = get_bits1(&gb);
256 sconf->coef_table = get_bits(&gb, 2);
257 sconf->long_term_prediction = get_bits1(&gb);
258 sconf->max_order = get_bits(&gb, 10);
259 sconf->block_switching = get_bits(&gb, 2);
260 sconf->bgmc = get_bits1(&gb);
261 sconf->sb_part = get_bits1(&gb);
262 sconf->joint_stereo = get_bits1(&gb);
263 sconf->mc_coding = get_bits1(&gb);
264 sconf->chan_config = get_bits1(&gb);
265 sconf->chan_sort = get_bits1(&gb);
266 crc_enabled = get_bits1(&gb);
267 sconf->rlslms = get_bits1(&gb);
268 skip_bits(&gb, 5); // skip 5 reserved bits
269 skip_bits1(&gb); // skip aux_data_enabled
272 // check for ALSSpecificConfig struct
273 if (als_id != MKBETAG('A','L','S','\0'))
276 ctx->cur_frame_length = sconf->frame_length;
278 // allocate quantized parcor coefficient buffer
279 if (!(ctx->quant_cof = av_malloc(sizeof(*ctx->quant_cof) * sconf->max_order)) ||
280 !(ctx->lpc_cof = av_malloc(sizeof(*ctx->lpc_cof) * sconf->max_order))) {
281 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
282 return AVERROR(ENOMEM);
285 // read channel config
286 if (sconf->chan_config)
287 sconf->chan_config_info = get_bits(&gb, 16);
288 // TODO: use this to set avctx->channel_layout
291 // read channel sorting
292 if (sconf->chan_sort && avctx->channels > 1) {
293 int chan_pos_bits = av_ceil_log2(avctx->channels);
294 int bits_needed = avctx->channels * chan_pos_bits + 7;
295 if (get_bits_left(&gb) < bits_needed)
298 if (!(sconf->chan_pos = av_malloc(avctx->channels * sizeof(*sconf->chan_pos))))
299 return AVERROR(ENOMEM);
301 for (i = 0; i < avctx->channels; i++)
302 sconf->chan_pos[i] = get_bits(&gb, chan_pos_bits);
305 // TODO: use this to actually do channel sorting
307 sconf->chan_sort = 0;
311 // read fixed header and trailer sizes,
312 // if size = 0xFFFFFFFF then there is no data field!
313 if (get_bits_left(&gb) < 64)
316 sconf->header_size = get_bits_long(&gb, 32);
317 sconf->trailer_size = get_bits_long(&gb, 32);
318 if (sconf->header_size == 0xFFFFFFFF)
319 sconf->header_size = 0;
320 if (sconf->trailer_size == 0xFFFFFFFF)
321 sconf->trailer_size = 0;
323 ht_size = ((int64_t)(sconf->header_size) + (int64_t)(sconf->trailer_size)) << 3;
326 // skip the header and trailer data
327 if (get_bits_left(&gb) < ht_size)
330 if (ht_size > INT32_MAX)
333 skip_bits_long(&gb, ht_size);
338 if (get_bits_left(&gb) < 32)
341 skip_bits_long(&gb, 32);
345 // no need to read the rest of ALSSpecificConfig (ra_unit_size & aux data)
347 dprint_specific_config(ctx);
353 /** Checks the ALSSpecificConfig for unsupported features.
355 static int check_specific_config(ALSDecContext *ctx)
357 ALSSpecificConfig *sconf = &ctx->sconf;
360 // report unsupported feature and set error value
361 #define MISSING_ERR(cond, str, errval) \
364 av_log_missing_feature(ctx->avctx, str, 0); \
369 MISSING_ERR(sconf->floating, "Floating point decoding", -1);
370 MISSING_ERR(sconf->bgmc, "BGMC entropy decoding", -1);
371 MISSING_ERR(sconf->mc_coding, "Multi-channel correlation", -1);
372 MISSING_ERR(sconf->rlslms, "Adaptive RLS-LMS prediction", -1);
373 MISSING_ERR(sconf->chan_sort, "Channel sorting", 0);
379 /** Parses the bs_info field to extract the block partitioning used in
380 * block switching mode, refer to ISO/IEC 14496-3, section 11.6.2.
382 static void parse_bs_info(const uint32_t bs_info, unsigned int n,
383 unsigned int div, unsigned int **div_blocks,
384 unsigned int *num_blocks)
386 if (n < 31 && ((bs_info << n) & 0x40000000)) {
387 // if the level is valid and the investigated bit n is set
388 // then recursively check both children at bits (2n+1) and (2n+2)
391 parse_bs_info(bs_info, n + 1, div, div_blocks, num_blocks);
392 parse_bs_info(bs_info, n + 2, div, div_blocks, num_blocks);
394 // else the bit is not set or the last level has been reached
395 // (bit implicitly not set)
403 /** Reads and decodes a Rice codeword.
405 static int32_t decode_rice(GetBitContext *gb, unsigned int k)
407 int max = get_bits_left(gb) - k;
408 int q = get_unary(gb, 0, max);
409 int r = k ? get_bits1(gb) : !(q & 1);
413 q += get_bits_long(gb, k - 1);
421 /** Converts PARCOR coefficient k to direct filter coefficient.
423 static void parcor_to_lpc(unsigned int k, const int32_t *par, int32_t *cof)
427 for (i = 0, j = k - 1; i < j; i++, j--) {
428 int tmp1 = ((MUL64(par[k], cof[j]) + (1 << 19)) >> 20);
429 cof[j] += ((MUL64(par[k], cof[i]) + (1 << 19)) >> 20);
433 cof[i] += ((MUL64(par[k], cof[j]) + (1 << 19)) >> 20);
439 /** Reads block switching field if necessary and sets actual block sizes.
440 * Also assures that the block sizes of the last frame correspond to the
441 * actual number of samples.
443 static void get_block_sizes(ALSDecContext *ctx, unsigned int *div_blocks,
446 ALSSpecificConfig *sconf = &ctx->sconf;
447 GetBitContext *gb = &ctx->gb;
448 unsigned int *ptr_div_blocks = div_blocks;
451 if (sconf->block_switching) {
452 unsigned int bs_info_len = 1 << (sconf->block_switching + 2);
453 *bs_info = get_bits_long(gb, bs_info_len);
454 *bs_info <<= (32 - bs_info_len);
458 parse_bs_info(*bs_info, 0, 0, &ptr_div_blocks, &ctx->num_blocks);
460 // The last frame may have an overdetermined block structure given in
461 // the bitstream. In that case the defined block structure would need
462 // more samples than available to be consistent.
463 // The block structure is actually used but the block sizes are adapted
464 // to fit the actual number of available samples.
465 // Example: 5 samples, 2nd level block sizes: 2 2 2 2.
466 // This results in the actual block sizes: 2 2 1 0.
467 // This is not specified in 14496-3 but actually done by the reference
468 // codec RM22 revision 2.
469 // This appears to happen in case of an odd number of samples in the last
470 // frame which is actually not allowed by the block length switching part
472 // The ALS conformance files feature an odd number of samples in the last
475 for (b = 0; b < ctx->num_blocks; b++)
476 div_blocks[b] = ctx->sconf.frame_length >> div_blocks[b];
478 if (ctx->cur_frame_length != ctx->sconf.frame_length) {
479 unsigned int remaining = ctx->cur_frame_length;
481 for (b = 0; b < ctx->num_blocks; b++) {
482 if (remaining < div_blocks[b]) {
483 div_blocks[b] = remaining;
484 ctx->num_blocks = b + 1;
488 remaining -= div_blocks[b];
494 /** Reads the block data for a constant block
496 static void read_const_block_data(ALSDecContext *ctx, ALSBlockData *bd)
498 ALSSpecificConfig *sconf = &ctx->sconf;
499 AVCodecContext *avctx = ctx->avctx;
500 GetBitContext *gb = &ctx->gb;
503 bd->const_block = get_bits1(gb); // 1 = constant value, 0 = zero block (silence)
504 bd->js_blocks = get_bits1(gb);
506 // skip 5 reserved bits
509 if (bd->const_block) {
510 unsigned int const_val_bits = sconf->floating ? 24 : avctx->bits_per_raw_sample;
511 bd->const_val = get_sbits_long(gb, const_val_bits);
514 // ensure constant block decoding by reusing this field
519 /** Decodes the block data for a constant block
521 static void decode_const_block_data(ALSDecContext *ctx, ALSBlockData *bd)
523 int smp = bd->block_length;
524 int32_t val = bd->const_val;
525 int32_t *dst = bd->raw_samples;
527 // write raw samples into buffer
533 /** Reads the block data for a non-constant block
535 static int read_var_block_data(ALSDecContext *ctx, ALSBlockData *bd)
537 ALSSpecificConfig *sconf = &ctx->sconf;
538 AVCodecContext *avctx = ctx->avctx;
539 GetBitContext *gb = &ctx->gb;
542 unsigned int sub_blocks, log2_sub_blocks, sb_length;
543 unsigned int start = 0;
544 unsigned int opt_order;
546 int32_t *quant_cof = bd->quant_cof;
549 // ensure variable block decoding by reusing this field
553 bd->js_blocks = get_bits1(gb);
555 opt_order = bd->opt_order;
557 // determine the number of subblocks for entropy decoding
558 if (!sconf->bgmc && !sconf->sb_part) {
561 if (sconf->bgmc && sconf->sb_part)
562 log2_sub_blocks = get_bits(gb, 2);
564 log2_sub_blocks = 2 * get_bits1(gb);
567 sub_blocks = 1 << log2_sub_blocks;
569 // do not continue in case of a damaged stream since
570 // block_length must be evenly divisible by sub_blocks
571 if (bd->block_length & (sub_blocks - 1)) {
572 av_log(avctx, AV_LOG_WARNING,
573 "Block length is not evenly divisible by the number of subblocks.\n");
577 sb_length = bd->block_length >> log2_sub_blocks;
583 s[0] = get_bits(gb, 4 + (sconf->resolution > 1));
584 for (k = 1; k < sub_blocks; k++)
585 s[k] = s[k - 1] + decode_rice(gb, 0);
589 bd->shift_lsbs = get_bits(gb, 4) + 1;
591 bd->store_prev_samples = (bd->js_blocks && bd->raw_other) || bd->shift_lsbs;
594 if (!sconf->rlslms) {
595 if (sconf->adapt_order) {
596 int opt_order_length = av_ceil_log2(av_clip((bd->block_length >> 3) - 1,
597 2, sconf->max_order + 1));
598 bd->opt_order = get_bits(gb, opt_order_length);
600 bd->opt_order = sconf->max_order;
603 opt_order = bd->opt_order;
608 if (sconf->coef_table == 3) {
611 // read coefficient 0
612 quant_cof[0] = 32 * parcor_scaled_values[get_bits(gb, 7)];
614 // read coefficient 1
616 quant_cof[1] = -32 * parcor_scaled_values[get_bits(gb, 7)];
618 // read coefficients 2 to opt_order
619 for (k = 2; k < opt_order; k++)
620 quant_cof[k] = get_bits(gb, 7);
625 // read coefficient 0 to 19
626 k_max = FFMIN(opt_order, 20);
627 for (k = 0; k < k_max; k++) {
628 int rice_param = parcor_rice_table[sconf->coef_table][k][1];
629 int offset = parcor_rice_table[sconf->coef_table][k][0];
630 quant_cof[k] = decode_rice(gb, rice_param) + offset;
633 // read coefficients 20 to 126
634 k_max = FFMIN(opt_order, 127);
635 for (; k < k_max; k++)
636 quant_cof[k] = decode_rice(gb, 2) + (k & 1);
638 // read coefficients 127 to opt_order
639 for (; k < opt_order; k++)
640 quant_cof[k] = decode_rice(gb, 1);
642 quant_cof[0] = 32 * parcor_scaled_values[quant_cof[0] + 64];
645 quant_cof[1] = -32 * parcor_scaled_values[quant_cof[1] + 64];
648 for (k = 2; k < opt_order; k++)
649 quant_cof[k] = (quant_cof[k] << 14) + (add_base << 13);
653 // read LTP gain and lag values
654 if (sconf->long_term_prediction) {
655 *bd->use_ltp = get_bits1(gb);
658 bd->ltp_gain[0] = decode_rice(gb, 1) << 3;
659 bd->ltp_gain[1] = decode_rice(gb, 2) << 3;
661 bd->ltp_gain[2] = ltp_gain_values[get_unary(gb, 0, 4)][get_bits(gb, 2)];
663 bd->ltp_gain[3] = decode_rice(gb, 2) << 3;
664 bd->ltp_gain[4] = decode_rice(gb, 1) << 3;
666 *bd->ltp_lag = get_bits(gb, ctx->ltp_lag_length);
667 *bd->ltp_lag += FFMAX(4, opt_order + 1);
671 // read first value and residuals in case of a random access block
674 bd->raw_samples[0] = decode_rice(gb, avctx->bits_per_raw_sample - 4);
676 bd->raw_samples[1] = decode_rice(gb, s[0] + 3);
678 bd->raw_samples[2] = decode_rice(gb, s[0] + 1);
680 start = FFMIN(opt_order, 3);
683 // read all residuals
687 int32_t *current_res = bd->raw_samples + start;
689 for (sb = 0; sb < sub_blocks; sb++, start = 0)
690 for (; start < sb_length; start++)
691 *current_res++ = decode_rice(gb, s[sb]);
694 if (!sconf->mc_coding || ctx->js_switch)
701 /** Decodes the block data for a non-constant block
703 static int decode_var_block_data(ALSDecContext *ctx, ALSBlockData *bd)
705 ALSSpecificConfig *sconf = &ctx->sconf;
706 unsigned int block_length = bd->block_length;
707 unsigned int smp = 0;
709 unsigned int opt_order = bd->opt_order;
712 int32_t *quant_cof = bd->quant_cof;
713 int32_t *lpc_cof = bd->lpc_cof;
714 int32_t *raw_samples = bd->raw_samples;
716 // reverse long-term prediction
720 for (ltp_smp = FFMAX(*bd->ltp_lag - 2, 0); ltp_smp < block_length; ltp_smp++) {
721 int center = ltp_smp - *bd->ltp_lag;
722 int begin = FFMAX(0, center - 2);
723 int end = center + 3;
724 int tab = 5 - (end - begin);
729 for (base = begin; base < end; base++, tab++)
730 y += MUL64(bd->ltp_gain[tab], raw_samples[base]);
732 raw_samples[ltp_smp] += y >> 7;
736 // reconstruct all samples from residuals
738 for (smp = 0; smp < opt_order; smp++) {
741 for (sb = 0; sb < smp; sb++)
742 y += MUL64(lpc_cof[sb],raw_samples[smp - (sb + 1)]);
744 raw_samples[smp] -= y >> 20;
745 parcor_to_lpc(smp, quant_cof, lpc_cof);
748 for (k = 0; k < opt_order; k++)
749 parcor_to_lpc(k, quant_cof, lpc_cof);
751 // store previous samples in case that they have to be altered
752 if (bd->store_prev_samples)
753 memcpy(bd->prev_raw_samples, raw_samples - sconf->max_order,
754 sizeof(*bd->prev_raw_samples) * sconf->max_order);
756 // reconstruct difference signal for prediction (joint-stereo)
757 if (bd->js_blocks && bd->raw_other) {
758 int32_t *left, *right;
760 if (bd->raw_other > raw_samples) { // D = R - L
762 right = bd->raw_other;
763 } else { // D = R - L
764 left = bd->raw_other;
768 for (sb = -1; sb >= -sconf->max_order; sb--)
769 raw_samples[sb] = right[sb] - left[sb];
772 // reconstruct shifted signal
774 for (sb = -1; sb >= -sconf->max_order; sb--)
775 raw_samples[sb] >>= bd->shift_lsbs;
778 // reconstruct raw samples
779 for (; smp < bd->block_length; smp++) {
782 for (sb = 0; sb < opt_order; sb++)
783 y += MUL64(bd->lpc_cof[sb],raw_samples[smp - (sb + 1)]);
785 raw_samples[smp] -= y >> 20;
788 // restore previous samples in case that they have been altered
789 if (bd->store_prev_samples)
790 memcpy(raw_samples - sconf->max_order, bd->prev_raw_samples,
791 sizeof(*raw_samples) * sconf->max_order);
797 /** Reads the block data.
799 static int read_block(ALSDecContext *ctx, ALSBlockData *bd)
801 GetBitContext *gb = &ctx->gb;
803 // read block type flag and read the samples accordingly
805 if (read_var_block_data(ctx, bd))
808 read_const_block_data(ctx, bd);
815 /** Decodes the block data.
817 static int decode_block(ALSDecContext *ctx, ALSBlockData *bd)
821 // read block type flag and read the samples accordingly
823 decode_const_block_data(ctx, bd);
824 else if (decode_var_block_data(ctx, bd))
827 // TODO: read RLSLMS extension data
830 for (smp = 0; smp < bd->block_length; smp++)
831 bd->raw_samples[smp] <<= bd->shift_lsbs;
837 /** Reads and decodes block data successively.
839 static int read_decode_block(ALSDecContext *ctx, ALSBlockData *bd)
843 ret = read_block(ctx, bd);
848 ret = decode_block(ctx, bd);
854 /** Computes the number of samples left to decode for the current frame and
855 * sets these samples to zero.
857 static void zero_remaining(unsigned int b, unsigned int b_max,
858 const unsigned int *div_blocks, int32_t *buf)
860 unsigned int count = 0;
863 count += div_blocks[b];
866 memset(buf, 0, sizeof(*buf) * count);
870 /** Decodes blocks independently.
872 static int decode_blocks_ind(ALSDecContext *ctx, unsigned int ra_frame,
873 unsigned int c, const unsigned int *div_blocks,
874 unsigned int *js_blocks)
879 memset(&bd, 0, sizeof(ALSBlockData));
881 bd.ra_block = ra_frame;
882 bd.use_ltp = ctx->use_ltp;
883 bd.ltp_lag = ctx->ltp_lag;
884 bd.ltp_gain = ctx->ltp_gain[0];
885 bd.quant_cof = ctx->quant_cof;
886 bd.lpc_cof = ctx->lpc_cof;
887 bd.prev_raw_samples = ctx->prev_raw_samples;
888 bd.raw_samples = ctx->raw_samples[c];
891 for (b = 0; b < ctx->num_blocks; b++) {
893 bd.block_length = div_blocks[b];
895 if (read_decode_block(ctx, &bd)) {
896 // damaged block, write zero for the rest of the frame
897 zero_remaining(b, ctx->num_blocks, div_blocks, bd.raw_samples);
900 bd.raw_samples += div_blocks[b];
908 /** Decodes blocks dependently.
910 static int decode_blocks(ALSDecContext *ctx, unsigned int ra_frame,
911 unsigned int c, const unsigned int *div_blocks,
912 unsigned int *js_blocks)
914 ALSSpecificConfig *sconf = &ctx->sconf;
915 unsigned int offset = 0;
919 memset(bd, 0, 2 * sizeof(ALSBlockData));
921 bd[0].ra_block = ra_frame;
922 bd[0].use_ltp = ctx->use_ltp;
923 bd[0].ltp_lag = ctx->ltp_lag;
924 bd[0].ltp_gain = ctx->ltp_gain[0];
925 bd[0].quant_cof = ctx->quant_cof;
926 bd[0].lpc_cof = ctx->lpc_cof;
927 bd[0].prev_raw_samples = ctx->prev_raw_samples;
928 bd[0].js_blocks = *js_blocks;
930 bd[1].ra_block = ra_frame;
931 bd[1].use_ltp = ctx->use_ltp;
932 bd[1].ltp_lag = ctx->ltp_lag;
933 bd[1].ltp_gain = ctx->ltp_gain[0];
934 bd[1].quant_cof = ctx->quant_cof;
935 bd[1].lpc_cof = ctx->lpc_cof;
936 bd[1].prev_raw_samples = ctx->prev_raw_samples;
937 bd[1].js_blocks = *(js_blocks + 1);
940 for (b = 0; b < ctx->num_blocks; b++) {
943 bd[0].shift_lsbs = 0;
944 bd[1].shift_lsbs = 0;
946 bd[0].block_length = div_blocks[b];
947 bd[1].block_length = div_blocks[b];
949 bd[0].raw_samples = ctx->raw_samples[c ] + offset;
950 bd[1].raw_samples = ctx->raw_samples[c + 1] + offset;
952 bd[0].raw_other = bd[1].raw_samples;
953 bd[1].raw_other = bd[0].raw_samples;
955 if(read_decode_block(ctx, &bd[0]) || read_decode_block(ctx, &bd[1])) {
956 // damaged block, write zero for the rest of the frame
957 zero_remaining(b, ctx->num_blocks, div_blocks, bd[0].raw_samples);
958 zero_remaining(b, ctx->num_blocks, div_blocks, bd[1].raw_samples);
962 // reconstruct joint-stereo blocks
963 if (bd[0].js_blocks) {
965 av_log(ctx->avctx, AV_LOG_WARNING, "Invalid channel pair!\n");
967 for (s = 0; s < div_blocks[b]; s++)
968 bd[0].raw_samples[s] = bd[1].raw_samples[s] - bd[0].raw_samples[s];
969 } else if (bd[1].js_blocks) {
970 for (s = 0; s < div_blocks[b]; s++)
971 bd[1].raw_samples[s] = bd[1].raw_samples[s] + bd[0].raw_samples[s];
974 offset += div_blocks[b];
979 // store carryover raw samples,
980 // the others channel raw samples are stored by the calling function.
981 memmove(ctx->raw_samples[c] - sconf->max_order,
982 ctx->raw_samples[c] - sconf->max_order + sconf->frame_length,
983 sizeof(*ctx->raw_samples[c]) * sconf->max_order);
989 /** Reads the frame data.
991 static int read_frame_data(ALSDecContext *ctx, unsigned int ra_frame)
993 ALSSpecificConfig *sconf = &ctx->sconf;
994 AVCodecContext *avctx = ctx->avctx;
995 GetBitContext *gb = &ctx->gb;
996 unsigned int div_blocks[32]; ///< block sizes.
998 unsigned int js_blocks[2];
1000 uint32_t bs_info = 0;
1002 // skip the size of the ra unit if present in the frame
1003 if (sconf->ra_flag == RA_FLAG_FRAMES && ra_frame)
1004 skip_bits_long(gb, 32);
1006 if (sconf->mc_coding && sconf->joint_stereo) {
1007 ctx->js_switch = get_bits1(gb);
1011 if (!sconf->mc_coding || ctx->js_switch) {
1012 int independent_bs = !sconf->joint_stereo;
1014 for (c = 0; c < avctx->channels; c++) {
1018 get_block_sizes(ctx, div_blocks, &bs_info);
1020 // if joint_stereo and block_switching is set, independent decoding
1021 // is signaled via the first bit of bs_info
1022 if (sconf->joint_stereo && sconf->block_switching)
1026 // if this is the last channel, it has to be decoded independently
1027 if (c == avctx->channels - 1)
1030 if (independent_bs) {
1031 if (decode_blocks_ind(ctx, ra_frame, c, div_blocks, js_blocks))
1036 if (decode_blocks(ctx, ra_frame, c, div_blocks, js_blocks))
1042 // store carryover raw samples
1043 memmove(ctx->raw_samples[c] - sconf->max_order,
1044 ctx->raw_samples[c] - sconf->max_order + sconf->frame_length,
1045 sizeof(*ctx->raw_samples[c]) * sconf->max_order);
1047 } else { // multi-channel coding
1048 get_block_sizes(ctx, div_blocks, &bs_info);
1050 // TODO: multi channel coding might use a temporary buffer instead as
1051 // the actual channel is not known when read_block-data is called
1052 if (decode_blocks_ind(ctx, ra_frame, 0, div_blocks, js_blocks))
1054 // TODO: read_channel_data
1057 // TODO: read_diff_float_data
1063 /** Decodes an ALS frame.
1065 static int decode_frame(AVCodecContext *avctx,
1066 void *data, int *data_size,
1069 ALSDecContext *ctx = avctx->priv_data;
1070 ALSSpecificConfig *sconf = &ctx->sconf;
1071 const uint8_t *buffer = avpkt->data;
1072 int buffer_size = avpkt->size;
1073 int invalid_frame, size;
1074 unsigned int c, sample, ra_frame, bytes_read, shift;
1076 init_get_bits(&ctx->gb, buffer, buffer_size * 8);
1078 // In the case that the distance between random access frames is set to zero
1079 // (sconf->ra_distance == 0) no frame is treated as a random access frame.
1080 // For the first frame, if prediction is used, all samples used from the
1081 // previous frame are assumed to be zero.
1082 ra_frame = sconf->ra_distance && !(ctx->frame_id % sconf->ra_distance);
1084 // the last frame to decode might have a different length
1085 if (sconf->samples != 0xFFFFFFFF)
1086 ctx->cur_frame_length = FFMIN(sconf->samples - ctx->frame_id * (uint64_t) sconf->frame_length,
1087 sconf->frame_length);
1089 ctx->cur_frame_length = sconf->frame_length;
1091 // decode the frame data
1092 if ((invalid_frame = read_frame_data(ctx, ra_frame) < 0))
1093 av_log(ctx->avctx, AV_LOG_WARNING,
1094 "Reading frame data failed. Skipping RA unit.\n");
1098 // check for size of decoded data
1099 size = ctx->cur_frame_length * avctx->channels *
1100 (av_get_bits_per_sample_format(avctx->sample_fmt) >> 3);
1102 if (size > *data_size) {
1103 av_log(avctx, AV_LOG_ERROR, "Decoded data exceeds buffer size.\n");
1109 // transform decoded frame into output format
1110 #define INTERLEAVE_OUTPUT(bps) \
1112 int##bps##_t *dest = (int##bps##_t*) data; \
1113 shift = bps - ctx->avctx->bits_per_raw_sample; \
1114 for (sample = 0; sample < ctx->cur_frame_length; sample++) \
1115 for (c = 0; c < avctx->channels; c++) \
1116 *dest++ = ctx->raw_samples[c][sample] << shift; \
1119 if (ctx->avctx->bits_per_raw_sample <= 16) {
1120 INTERLEAVE_OUTPUT(16)
1122 INTERLEAVE_OUTPUT(32)
1125 bytes_read = invalid_frame ? buffer_size :
1126 (get_bits_count(&ctx->gb) + 7) >> 3;
1132 /** Uninitializes the ALS decoder.
1134 static av_cold int decode_end(AVCodecContext *avctx)
1136 ALSDecContext *ctx = avctx->priv_data;
1138 av_freep(&ctx->sconf.chan_pos);
1140 av_freep(&ctx->use_ltp);
1141 av_freep(&ctx->ltp_lag);
1142 av_freep(&ctx->ltp_gain);
1143 av_freep(&ctx->ltp_gain_buffer);
1144 av_freep(&ctx->quant_cof);
1145 av_freep(&ctx->lpc_cof);
1146 av_freep(&ctx->prev_raw_samples);
1147 av_freep(&ctx->raw_samples);
1148 av_freep(&ctx->raw_buffer);
1154 /** Initializes the ALS decoder.
1156 static av_cold int decode_init(AVCodecContext *avctx)
1159 unsigned int channel_size;
1161 ALSDecContext *ctx = avctx->priv_data;
1162 ALSSpecificConfig *sconf = &ctx->sconf;
1165 if (!avctx->extradata) {
1166 av_log(avctx, AV_LOG_ERROR, "Missing required ALS extradata.\n");
1170 if (read_specific_config(ctx)) {
1171 av_log(avctx, AV_LOG_ERROR, "Reading ALSSpecificConfig failed.\n");
1176 if (check_specific_config(ctx)) {
1181 if (sconf->floating) {
1182 avctx->sample_fmt = SAMPLE_FMT_FLT;
1183 avctx->bits_per_raw_sample = 32;
1185 avctx->sample_fmt = sconf->resolution > 1
1186 ? SAMPLE_FMT_S32 : SAMPLE_FMT_S16;
1187 avctx->bits_per_raw_sample = (sconf->resolution + 1) * 8;
1190 // set lag value for long-term prediction
1191 ctx->ltp_lag_length = 8 + (avctx->sample_rate >= 96000) +
1192 (avctx->sample_rate >= 192000);
1194 // allocate quantized parcor coefficient buffer
1195 num_buffers = sconf->mc_coding ? avctx->channels : 1;
1197 // allocate and assign lag and gain data buffer for ltp mode
1198 ctx->use_ltp = av_mallocz(sizeof(*ctx->use_ltp) * num_buffers);
1199 ctx->ltp_lag = av_malloc (sizeof(*ctx->ltp_lag) * num_buffers);
1200 ctx->ltp_gain = av_malloc (sizeof(*ctx->ltp_gain) * num_buffers);
1201 ctx->ltp_gain_buffer = av_malloc (sizeof(*ctx->ltp_gain_buffer) *
1204 if (!ctx->use_ltp || !ctx->ltp_lag ||
1205 !ctx->ltp_gain || !ctx->ltp_gain_buffer) {
1206 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
1208 return AVERROR(ENOMEM);
1211 for (c = 0; c < num_buffers; c++)
1212 ctx->ltp_gain[c] = ctx->ltp_gain_buffer + c * 5;
1214 avctx->frame_size = sconf->frame_length;
1215 channel_size = sconf->frame_length + sconf->max_order;
1217 ctx->prev_raw_samples = av_malloc (sizeof(*ctx->prev_raw_samples) * sconf->max_order);
1218 ctx->raw_buffer = av_mallocz(sizeof(*ctx-> raw_buffer) * avctx->channels * channel_size);
1219 ctx->raw_samples = av_malloc (sizeof(*ctx-> raw_samples) * avctx->channels);
1221 // allocate previous raw sample buffer
1222 if (!ctx->prev_raw_samples || !ctx->raw_buffer|| !ctx->raw_samples) {
1223 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
1225 return AVERROR(ENOMEM);
1228 // assign raw samples buffers
1229 ctx->raw_samples[0] = ctx->raw_buffer + sconf->max_order;
1230 for (c = 1; c < avctx->channels; c++)
1231 ctx->raw_samples[c] = ctx->raw_samples[c - 1] + channel_size;
1237 /** Flushes (resets) the frame ID after seeking.
1239 static av_cold void flush(AVCodecContext *avctx)
1241 ALSDecContext *ctx = avctx->priv_data;
1247 AVCodec als_decoder = {
1251 sizeof(ALSDecContext),
1257 .capabilities = CODEC_CAP_SUBFRAMES,
1258 .long_name = NULL_IF_CONFIG_SMALL("MPEG-4 Audio Lossless Coding (ALS)"),