3 * Copyright (c) 2009 Thilo Borgmann <thilo.borgmann _at_ googlemail.com>
5 * This file is part of Libav.
7 * Libav is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * Libav is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 * @author Thilo Borgmann <thilo.borgmann _at_ googlemail.com>
33 #include "mpeg4audio.h"
34 #include "bytestream.h"
38 #include "libavutil/samplefmt.h"
39 #include "libavutil/crc.h"
43 /** Rice parameters and corresponding index offsets for decoding the
44 * indices of scaled PARCOR values. The table chosen is set globally
45 * by the encoder and stored in ALSSpecificConfig.
47 static const int8_t parcor_rice_table[3][20][2] = {
48 { {-52, 4}, {-29, 5}, {-31, 4}, { 19, 4}, {-16, 4},
49 { 12, 3}, { -7, 3}, { 9, 3}, { -5, 3}, { 6, 3},
50 { -4, 3}, { 3, 3}, { -3, 2}, { 3, 2}, { -2, 2},
51 { 3, 2}, { -1, 2}, { 2, 2}, { -1, 2}, { 2, 2} },
52 { {-58, 3}, {-42, 4}, {-46, 4}, { 37, 5}, {-36, 4},
53 { 29, 4}, {-29, 4}, { 25, 4}, {-23, 4}, { 20, 4},
54 {-17, 4}, { 16, 4}, {-12, 4}, { 12, 3}, {-10, 4},
55 { 7, 3}, { -4, 4}, { 3, 3}, { -1, 3}, { 1, 3} },
56 { {-59, 3}, {-45, 5}, {-50, 4}, { 38, 4}, {-39, 4},
57 { 32, 4}, {-30, 4}, { 25, 3}, {-23, 3}, { 20, 3},
58 {-20, 3}, { 16, 3}, {-13, 3}, { 10, 3}, { -7, 3},
59 { 3, 3}, { 0, 3}, { -1, 3}, { 2, 3}, { -1, 2} }
63 /** Scaled PARCOR values used for the first two PARCOR coefficients.
64 * To be indexed by the Rice coded indices.
65 * Generated by: parcor_scaled_values[i] = 32 + ((i * (i+1)) << 7) - (1 << 20)
66 * Actual values are divided by 32 in order to be stored in 16 bits.
68 static const int16_t parcor_scaled_values[] = {
69 -1048544 / 32, -1048288 / 32, -1047776 / 32, -1047008 / 32,
70 -1045984 / 32, -1044704 / 32, -1043168 / 32, -1041376 / 32,
71 -1039328 / 32, -1037024 / 32, -1034464 / 32, -1031648 / 32,
72 -1028576 / 32, -1025248 / 32, -1021664 / 32, -1017824 / 32,
73 -1013728 / 32, -1009376 / 32, -1004768 / 32, -999904 / 32,
74 -994784 / 32, -989408 / 32, -983776 / 32, -977888 / 32,
75 -971744 / 32, -965344 / 32, -958688 / 32, -951776 / 32,
76 -944608 / 32, -937184 / 32, -929504 / 32, -921568 / 32,
77 -913376 / 32, -904928 / 32, -896224 / 32, -887264 / 32,
78 -878048 / 32, -868576 / 32, -858848 / 32, -848864 / 32,
79 -838624 / 32, -828128 / 32, -817376 / 32, -806368 / 32,
80 -795104 / 32, -783584 / 32, -771808 / 32, -759776 / 32,
81 -747488 / 32, -734944 / 32, -722144 / 32, -709088 / 32,
82 -695776 / 32, -682208 / 32, -668384 / 32, -654304 / 32,
83 -639968 / 32, -625376 / 32, -610528 / 32, -595424 / 32,
84 -580064 / 32, -564448 / 32, -548576 / 32, -532448 / 32,
85 -516064 / 32, -499424 / 32, -482528 / 32, -465376 / 32,
86 -447968 / 32, -430304 / 32, -412384 / 32, -394208 / 32,
87 -375776 / 32, -357088 / 32, -338144 / 32, -318944 / 32,
88 -299488 / 32, -279776 / 32, -259808 / 32, -239584 / 32,
89 -219104 / 32, -198368 / 32, -177376 / 32, -156128 / 32,
90 -134624 / 32, -112864 / 32, -90848 / 32, -68576 / 32,
91 -46048 / 32, -23264 / 32, -224 / 32, 23072 / 32,
92 46624 / 32, 70432 / 32, 94496 / 32, 118816 / 32,
93 143392 / 32, 168224 / 32, 193312 / 32, 218656 / 32,
94 244256 / 32, 270112 / 32, 296224 / 32, 322592 / 32,
95 349216 / 32, 376096 / 32, 403232 / 32, 430624 / 32,
96 458272 / 32, 486176 / 32, 514336 / 32, 542752 / 32,
97 571424 / 32, 600352 / 32, 629536 / 32, 658976 / 32,
98 688672 / 32, 718624 / 32, 748832 / 32, 779296 / 32,
99 810016 / 32, 840992 / 32, 872224 / 32, 903712 / 32,
100 935456 / 32, 967456 / 32, 999712 / 32, 1032224 / 32
104 /** Gain values of p(0) for long-term prediction.
105 * To be indexed by the Rice coded indices.
107 static const uint8_t ltp_gain_values [4][4] = {
115 /** Inter-channel weighting factors for multi-channel correlation.
116 * To be indexed by the Rice coded indices.
118 static const int16_t mcc_weightings[] = {
119 204, 192, 179, 166, 153, 140, 128, 115,
120 102, 89, 76, 64, 51, 38, 25, 12,
121 0, -12, -25, -38, -51, -64, -76, -89,
122 -102, -115, -128, -140, -153, -166, -179, -192
126 /** Tail codes used in arithmetic coding using block Gilbert-Moore codes.
128 static const uint8_t tail_code[16][6] = {
129 { 74, 44, 25, 13, 7, 3},
130 { 68, 42, 24, 13, 7, 3},
131 { 58, 39, 23, 13, 7, 3},
132 {126, 70, 37, 19, 10, 5},
133 {132, 70, 37, 20, 10, 5},
134 {124, 70, 38, 20, 10, 5},
135 {120, 69, 37, 20, 11, 5},
136 {116, 67, 37, 20, 11, 5},
137 {108, 66, 36, 20, 10, 5},
138 {102, 62, 36, 20, 10, 5},
139 { 88, 58, 34, 19, 10, 5},
140 {162, 89, 49, 25, 13, 7},
141 {156, 87, 49, 26, 14, 7},
142 {150, 86, 47, 26, 14, 7},
143 {142, 84, 47, 26, 14, 7},
144 {131, 79, 46, 26, 14, 7}
155 typedef struct ALSSpecificConfig {
156 uint32_t samples; ///< number of samples, 0xFFFFFFFF if unknown
157 int resolution; ///< 000 = 8-bit; 001 = 16-bit; 010 = 24-bit; 011 = 32-bit
158 int floating; ///< 1 = IEEE 32-bit floating-point, 0 = integer
159 int msb_first; ///< 1 = original CRC calculated on big-endian system, 0 = little-endian
160 int frame_length; ///< frame length for each frame (last frame may differ)
161 int ra_distance; ///< distance between RA frames (in frames, 0...255)
162 enum RA_Flag ra_flag; ///< indicates where the size of ra units is stored
163 int adapt_order; ///< adaptive order: 1 = on, 0 = off
164 int coef_table; ///< table index of Rice code parameters
165 int long_term_prediction; ///< long term prediction (LTP): 1 = on, 0 = off
166 int max_order; ///< maximum prediction order (0..1023)
167 int block_switching; ///< number of block switching levels
168 int bgmc; ///< "Block Gilbert-Moore Code": 1 = on, 0 = off (Rice coding only)
169 int sb_part; ///< sub-block partition
170 int joint_stereo; ///< joint stereo: 1 = on, 0 = off
171 int mc_coding; ///< extended inter-channel coding (multi channel coding): 1 = on, 0 = off
172 int chan_config; ///< indicates that a chan_config_info field is present
173 int chan_sort; ///< channel rearrangement: 1 = on, 0 = off
174 int rlslms; ///< use "Recursive Least Square-Least Mean Square" predictor: 1 = on, 0 = off
175 int chan_config_info; ///< mapping of channels to loudspeaker locations. Unused until setting channel configuration is implemented.
176 int *chan_pos; ///< original channel positions
177 int crc_enabled; ///< enable Cyclic Redundancy Checksum
181 typedef struct ALSChannelData {
191 typedef struct ALSDecContext {
192 AVCodecContext *avctx;
193 ALSSpecificConfig sconf;
195 BswapDSPContext bdsp;
196 const AVCRC *crc_table;
197 uint32_t crc_org; ///< CRC value of the original input data
198 uint32_t crc; ///< CRC value calculated from decoded data
199 unsigned int cur_frame_length; ///< length of the current frame to decode
200 unsigned int frame_id; ///< the frame ID / number of the current frame
201 unsigned int js_switch; ///< if true, joint-stereo decoding is enforced
202 unsigned int num_blocks; ///< number of blocks used in the current frame
203 unsigned int s_max; ///< maximum Rice parameter allowed in entropy coding
204 uint8_t *bgmc_lut; ///< pointer at lookup tables used for BGMC
205 int *bgmc_lut_status; ///< pointer at lookup table status flags used for BGMC
206 int ltp_lag_length; ///< number of bits used for ltp lag value
207 int *const_block; ///< contains const_block flags for all channels
208 unsigned int *shift_lsbs; ///< contains shift_lsbs flags for all channels
209 unsigned int *opt_order; ///< contains opt_order flags for all channels
210 int *store_prev_samples; ///< contains store_prev_samples flags for all channels
211 int *use_ltp; ///< contains use_ltp flags for all channels
212 int *ltp_lag; ///< contains ltp lag values for all channels
213 int **ltp_gain; ///< gain values for ltp 5-tap filter for a channel
214 int *ltp_gain_buffer; ///< contains all gain values for ltp 5-tap filter
215 int32_t **quant_cof; ///< quantized parcor coefficients for a channel
216 int32_t *quant_cof_buffer; ///< contains all quantized parcor coefficients
217 int32_t **lpc_cof; ///< coefficients of the direct form prediction filter for a channel
218 int32_t *lpc_cof_buffer; ///< contains all coefficients of the direct form prediction filter
219 int32_t *lpc_cof_reversed_buffer; ///< temporary buffer to set up a reversed versio of lpc_cof_buffer
220 ALSChannelData **chan_data; ///< channel data for multi-channel correlation
221 ALSChannelData *chan_data_buffer; ///< contains channel data for all channels
222 int *reverted_channels; ///< stores a flag for each reverted channel
223 int32_t *prev_raw_samples; ///< contains unshifted raw samples from the previous block
224 int32_t **raw_samples; ///< decoded raw samples for each channel
225 int32_t *raw_buffer; ///< contains all decoded raw samples including carryover samples
226 uint8_t *crc_buffer; ///< buffer of byte order corrected samples used for CRC check
230 typedef struct ALSBlockData {
231 unsigned int block_length; ///< number of samples within the block
232 unsigned int ra_block; ///< if true, this is a random access block
233 int *const_block; ///< if true, this is a constant value block
234 int js_blocks; ///< true if this block contains a difference signal
235 unsigned int *shift_lsbs; ///< shift of values for this block
236 unsigned int *opt_order; ///< prediction order of this block
237 int *store_prev_samples;///< if true, carryover samples have to be stored
238 int *use_ltp; ///< if true, long-term prediction is used
239 int *ltp_lag; ///< lag value for long-term prediction
240 int *ltp_gain; ///< gain values for ltp 5-tap filter
241 int32_t *quant_cof; ///< quantized parcor coefficients
242 int32_t *lpc_cof; ///< coefficients of the direct form prediction
243 int32_t *raw_samples; ///< decoded raw samples / residuals for this block
244 int32_t *prev_raw_samples; ///< contains unshifted raw samples from the previous block
245 int32_t *raw_other; ///< decoded raw samples of the other channel of a channel pair
250 static av_cold void dprint_specific_config(ALSDecContext *ctx)
252 AVCodecContext *avctx = ctx->avctx;
253 ALSSpecificConfig *sconf = &ctx->sconf;
255 ff_dlog(avctx, "resolution = %i\n", sconf->resolution);
256 ff_dlog(avctx, "floating = %i\n", sconf->floating);
257 ff_dlog(avctx, "frame_length = %i\n", sconf->frame_length);
258 ff_dlog(avctx, "ra_distance = %i\n", sconf->ra_distance);
259 ff_dlog(avctx, "ra_flag = %i\n", sconf->ra_flag);
260 ff_dlog(avctx, "adapt_order = %i\n", sconf->adapt_order);
261 ff_dlog(avctx, "coef_table = %i\n", sconf->coef_table);
262 ff_dlog(avctx, "long_term_prediction = %i\n", sconf->long_term_prediction);
263 ff_dlog(avctx, "max_order = %i\n", sconf->max_order);
264 ff_dlog(avctx, "block_switching = %i\n", sconf->block_switching);
265 ff_dlog(avctx, "bgmc = %i\n", sconf->bgmc);
266 ff_dlog(avctx, "sb_part = %i\n", sconf->sb_part);
267 ff_dlog(avctx, "joint_stereo = %i\n", sconf->joint_stereo);
268 ff_dlog(avctx, "mc_coding = %i\n", sconf->mc_coding);
269 ff_dlog(avctx, "chan_config = %i\n", sconf->chan_config);
270 ff_dlog(avctx, "chan_sort = %i\n", sconf->chan_sort);
271 ff_dlog(avctx, "RLSLMS = %i\n", sconf->rlslms);
272 ff_dlog(avctx, "chan_config_info = %i\n", sconf->chan_config_info);
275 #define dprint_specific_config(x) do {} while(0)
279 /** Read an ALSSpecificConfig from a buffer into the output struct.
281 static av_cold int read_specific_config(ALSDecContext *ctx)
285 int i, config_offset;
286 MPEG4AudioConfig m4ac;
287 ALSSpecificConfig *sconf = &ctx->sconf;
288 AVCodecContext *avctx = ctx->avctx;
289 uint32_t als_id, header_size, trailer_size;
291 init_get_bits(&gb, avctx->extradata, avctx->extradata_size * 8);
293 config_offset = avpriv_mpeg4audio_get_config(&m4ac, avctx->extradata,
294 avctx->extradata_size * 8, 1);
296 if (config_offset < 0)
297 return AVERROR_INVALIDDATA;
299 skip_bits_long(&gb, config_offset);
301 if (get_bits_left(&gb) < (30 << 3))
302 return AVERROR_INVALIDDATA;
304 // read the fixed items
305 als_id = get_bits_long(&gb, 32);
306 avctx->sample_rate = m4ac.sample_rate;
307 skip_bits_long(&gb, 32); // sample rate already known
308 sconf->samples = get_bits_long(&gb, 32);
309 avctx->channels = m4ac.channels;
310 skip_bits(&gb, 16); // number of channels already known
311 skip_bits(&gb, 3); // skip file_type
312 sconf->resolution = get_bits(&gb, 3);
313 sconf->floating = get_bits1(&gb);
314 sconf->msb_first = get_bits1(&gb);
315 sconf->frame_length = get_bits(&gb, 16) + 1;
316 sconf->ra_distance = get_bits(&gb, 8);
317 sconf->ra_flag = get_bits(&gb, 2);
318 sconf->adapt_order = get_bits1(&gb);
319 sconf->coef_table = get_bits(&gb, 2);
320 sconf->long_term_prediction = get_bits1(&gb);
321 sconf->max_order = get_bits(&gb, 10);
322 sconf->block_switching = get_bits(&gb, 2);
323 sconf->bgmc = get_bits1(&gb);
324 sconf->sb_part = get_bits1(&gb);
325 sconf->joint_stereo = get_bits1(&gb);
326 sconf->mc_coding = get_bits1(&gb);
327 sconf->chan_config = get_bits1(&gb);
328 sconf->chan_sort = get_bits1(&gb);
329 sconf->crc_enabled = get_bits1(&gb);
330 sconf->rlslms = get_bits1(&gb);
331 skip_bits(&gb, 5); // skip 5 reserved bits
332 skip_bits1(&gb); // skip aux_data_enabled
335 // check for ALSSpecificConfig struct
336 if (als_id != MKBETAG('A','L','S','\0'))
337 return AVERROR_INVALIDDATA;
339 ctx->cur_frame_length = sconf->frame_length;
341 // read channel config
342 if (sconf->chan_config)
343 sconf->chan_config_info = get_bits(&gb, 16);
344 // TODO: use this to set avctx->channel_layout
347 // read channel sorting
348 if (sconf->chan_sort && avctx->channels > 1) {
349 int chan_pos_bits = av_ceil_log2(avctx->channels);
350 int bits_needed = avctx->channels * chan_pos_bits + 7;
351 if (get_bits_left(&gb) < bits_needed)
352 return AVERROR_INVALIDDATA;
354 if (!(sconf->chan_pos = av_malloc(avctx->channels * sizeof(*sconf->chan_pos))))
355 return AVERROR(ENOMEM);
357 for (i = 0; i < avctx->channels; i++)
358 sconf->chan_pos[i] = get_bits(&gb, chan_pos_bits);
361 // TODO: use this to actually do channel sorting
363 sconf->chan_sort = 0;
367 // read fixed header and trailer sizes,
368 // if size = 0xFFFFFFFF then there is no data field!
369 if (get_bits_left(&gb) < 64)
370 return AVERROR_INVALIDDATA;
372 header_size = get_bits_long(&gb, 32);
373 trailer_size = get_bits_long(&gb, 32);
374 if (header_size == 0xFFFFFFFF)
376 if (trailer_size == 0xFFFFFFFF)
379 ht_size = ((int64_t)(header_size) + (int64_t)(trailer_size)) << 3;
382 // skip the header and trailer data
383 if (get_bits_left(&gb) < ht_size)
384 return AVERROR_INVALIDDATA;
386 if (ht_size > INT32_MAX)
387 return AVERROR_PATCHWELCOME;
389 skip_bits_long(&gb, ht_size);
392 // initialize CRC calculation
393 if (sconf->crc_enabled) {
394 if (get_bits_left(&gb) < 32)
395 return AVERROR_INVALIDDATA;
397 if (avctx->err_recognition & AV_EF_CRCCHECK) {
398 ctx->crc_table = av_crc_get_table(AV_CRC_32_IEEE_LE);
399 ctx->crc = 0xFFFFFFFF;
400 ctx->crc_org = ~get_bits_long(&gb, 32);
402 skip_bits_long(&gb, 32);
406 // no need to read the rest of ALSSpecificConfig (ra_unit_size & aux data)
408 dprint_specific_config(ctx);
414 /** Check the ALSSpecificConfig for unsupported features.
416 static int check_specific_config(ALSDecContext *ctx)
418 ALSSpecificConfig *sconf = &ctx->sconf;
421 // report unsupported feature and set error value
422 #define MISSING_ERR(cond, str, errval) \
425 avpriv_report_missing_feature(ctx->avctx, \
431 MISSING_ERR(sconf->floating, "Floating point decoding", AVERROR_PATCHWELCOME);
432 MISSING_ERR(sconf->rlslms, "Adaptive RLS-LMS prediction", AVERROR_PATCHWELCOME);
433 MISSING_ERR(sconf->chan_sort, "Channel sorting", 0);
439 /** Parse the bs_info field to extract the block partitioning used in
440 * block switching mode, refer to ISO/IEC 14496-3, section 11.6.2.
442 static void parse_bs_info(const uint32_t bs_info, unsigned int n,
443 unsigned int div, unsigned int **div_blocks,
444 unsigned int *num_blocks)
446 if (n < 31 && ((bs_info << n) & 0x40000000)) {
447 // if the level is valid and the investigated bit n is set
448 // then recursively check both children at bits (2n+1) and (2n+2)
451 parse_bs_info(bs_info, n + 1, div, div_blocks, num_blocks);
452 parse_bs_info(bs_info, n + 2, div, div_blocks, num_blocks);
454 // else the bit is not set or the last level has been reached
455 // (bit implicitly not set)
463 /** Read and decode a Rice codeword.
465 static int32_t decode_rice(GetBitContext *gb, unsigned int k)
467 int max = get_bits_left(gb) - k;
468 int q = get_unary(gb, 0, max);
469 int r = k ? get_bits1(gb) : !(q & 1);
473 q += get_bits_long(gb, k - 1);
481 /** Convert PARCOR coefficient k to direct filter coefficient.
483 static void parcor_to_lpc(unsigned int k, const int32_t *par, int32_t *cof)
487 for (i = 0, j = k - 1; i < j; i++, j--) {
488 int tmp1 = ((MUL64(par[k], cof[j]) + (1 << 19)) >> 20);
489 cof[j] += ((MUL64(par[k], cof[i]) + (1 << 19)) >> 20);
493 cof[i] += ((MUL64(par[k], cof[j]) + (1 << 19)) >> 20);
499 /** Read block switching field if necessary and set actual block sizes.
500 * Also assure that the block sizes of the last frame correspond to the
501 * actual number of samples.
503 static void get_block_sizes(ALSDecContext *ctx, unsigned int *div_blocks,
506 ALSSpecificConfig *sconf = &ctx->sconf;
507 GetBitContext *gb = &ctx->gb;
508 unsigned int *ptr_div_blocks = div_blocks;
511 if (sconf->block_switching) {
512 unsigned int bs_info_len = 1 << (sconf->block_switching + 2);
513 *bs_info = get_bits_long(gb, bs_info_len);
514 *bs_info <<= (32 - bs_info_len);
518 parse_bs_info(*bs_info, 0, 0, &ptr_div_blocks, &ctx->num_blocks);
520 // The last frame may have an overdetermined block structure given in
521 // the bitstream. In that case the defined block structure would need
522 // more samples than available to be consistent.
523 // The block structure is actually used but the block sizes are adapted
524 // to fit the actual number of available samples.
525 // Example: 5 samples, 2nd level block sizes: 2 2 2 2.
526 // This results in the actual block sizes: 2 2 1 0.
527 // This is not specified in 14496-3 but actually done by the reference
528 // codec RM22 revision 2.
529 // This appears to happen in case of an odd number of samples in the last
530 // frame which is actually not allowed by the block length switching part
532 // The ALS conformance files feature an odd number of samples in the last
535 for (b = 0; b < ctx->num_blocks; b++)
536 div_blocks[b] = ctx->sconf.frame_length >> div_blocks[b];
538 if (ctx->cur_frame_length != ctx->sconf.frame_length) {
539 unsigned int remaining = ctx->cur_frame_length;
541 for (b = 0; b < ctx->num_blocks; b++) {
542 if (remaining <= div_blocks[b]) {
543 div_blocks[b] = remaining;
544 ctx->num_blocks = b + 1;
548 remaining -= div_blocks[b];
554 /** Read the block data for a constant block
556 static void read_const_block_data(ALSDecContext *ctx, ALSBlockData *bd)
558 ALSSpecificConfig *sconf = &ctx->sconf;
559 AVCodecContext *avctx = ctx->avctx;
560 GetBitContext *gb = &ctx->gb;
562 *bd->raw_samples = 0;
563 *bd->const_block = get_bits1(gb); // 1 = constant value, 0 = zero block (silence)
564 bd->js_blocks = get_bits1(gb);
566 // skip 5 reserved bits
569 if (*bd->const_block) {
570 unsigned int const_val_bits = sconf->floating ? 24 : avctx->bits_per_raw_sample;
571 *bd->raw_samples = get_sbits_long(gb, const_val_bits);
574 // ensure constant block decoding by reusing this field
575 *bd->const_block = 1;
579 /** Decode the block data for a constant block
581 static void decode_const_block_data(ALSDecContext *ctx, ALSBlockData *bd)
583 int smp = bd->block_length - 1;
584 int32_t val = *bd->raw_samples;
585 int32_t *dst = bd->raw_samples + 1;
587 // write raw samples into buffer
593 /** Read the block data for a non-constant block
595 static int read_var_block_data(ALSDecContext *ctx, ALSBlockData *bd)
597 ALSSpecificConfig *sconf = &ctx->sconf;
598 AVCodecContext *avctx = ctx->avctx;
599 GetBitContext *gb = &ctx->gb;
603 unsigned int sub_blocks, log2_sub_blocks, sb_length;
604 unsigned int start = 0;
605 unsigned int opt_order;
607 int32_t *quant_cof = bd->quant_cof;
608 int32_t *current_res;
611 // ensure variable block decoding by reusing this field
612 *bd->const_block = 0;
615 bd->js_blocks = get_bits1(gb);
617 opt_order = *bd->opt_order;
619 // determine the number of subblocks for entropy decoding
620 if (!sconf->bgmc && !sconf->sb_part) {
623 if (sconf->bgmc && sconf->sb_part)
624 log2_sub_blocks = get_bits(gb, 2);
626 log2_sub_blocks = 2 * get_bits1(gb);
629 sub_blocks = 1 << log2_sub_blocks;
631 // do not continue in case of a damaged stream since
632 // block_length must be evenly divisible by sub_blocks
633 if (bd->block_length & (sub_blocks - 1)) {
634 av_log(avctx, AV_LOG_WARNING,
635 "Block length is not evenly divisible by the number of subblocks.\n");
636 return AVERROR_INVALIDDATA;
639 sb_length = bd->block_length >> log2_sub_blocks;
642 s[0] = get_bits(gb, 8 + (sconf->resolution > 1));
643 for (k = 1; k < sub_blocks; k++)
644 s[k] = s[k - 1] + decode_rice(gb, 2);
646 for (k = 0; k < sub_blocks; k++) {
651 s[0] = get_bits(gb, 4 + (sconf->resolution > 1));
652 for (k = 1; k < sub_blocks; k++)
653 s[k] = s[k - 1] + decode_rice(gb, 0);
655 for (k = 1; k < sub_blocks; k++)
657 av_log(avctx, AV_LOG_ERROR, "k invalid for rice code.\n");
658 return AVERROR_INVALIDDATA;
662 *bd->shift_lsbs = get_bits(gb, 4) + 1;
664 *bd->store_prev_samples = (bd->js_blocks && bd->raw_other) || *bd->shift_lsbs;
667 if (!sconf->rlslms) {
668 if (sconf->adapt_order && sconf->max_order) {
669 int opt_order_length = av_ceil_log2(av_clip((bd->block_length >> 3) - 1,
670 2, sconf->max_order + 1));
671 *bd->opt_order = get_bits(gb, opt_order_length);
672 if (*bd->opt_order > sconf->max_order) {
673 *bd->opt_order = sconf->max_order;
674 av_log(avctx, AV_LOG_ERROR, "Predictor order too large!\n");
675 return AVERROR_INVALIDDATA;
678 *bd->opt_order = sconf->max_order;
681 opt_order = *bd->opt_order;
686 if (sconf->coef_table == 3) {
689 // read coefficient 0
690 quant_cof[0] = 32 * parcor_scaled_values[get_bits(gb, 7)];
692 // read coefficient 1
694 quant_cof[1] = -32 * parcor_scaled_values[get_bits(gb, 7)];
696 // read coefficients 2 to opt_order
697 for (k = 2; k < opt_order; k++)
698 quant_cof[k] = get_bits(gb, 7);
703 // read coefficient 0 to 19
704 k_max = FFMIN(opt_order, 20);
705 for (k = 0; k < k_max; k++) {
706 int rice_param = parcor_rice_table[sconf->coef_table][k][1];
707 int offset = parcor_rice_table[sconf->coef_table][k][0];
708 quant_cof[k] = decode_rice(gb, rice_param) + offset;
709 if (quant_cof[k] < -64 || quant_cof[k] > 63) {
710 av_log(avctx, AV_LOG_ERROR,
711 "quant_cof %"PRIu32" is out of range\n",
713 return AVERROR_INVALIDDATA;
717 // read coefficients 20 to 126
718 k_max = FFMIN(opt_order, 127);
719 for (; k < k_max; k++)
720 quant_cof[k] = decode_rice(gb, 2) + (k & 1);
722 // read coefficients 127 to opt_order
723 for (; k < opt_order; k++)
724 quant_cof[k] = decode_rice(gb, 1);
726 quant_cof[0] = 32 * parcor_scaled_values[quant_cof[0] + 64];
729 quant_cof[1] = -32 * parcor_scaled_values[quant_cof[1] + 64];
732 for (k = 2; k < opt_order; k++)
733 quant_cof[k] = (quant_cof[k] << 14) + (add_base << 13);
737 // read LTP gain and lag values
738 if (sconf->long_term_prediction) {
739 *bd->use_ltp = get_bits1(gb);
744 bd->ltp_gain[0] = decode_rice(gb, 1) << 3;
745 bd->ltp_gain[1] = decode_rice(gb, 2) << 3;
747 r = get_unary(gb, 0, 3);
749 bd->ltp_gain[2] = ltp_gain_values[r][c];
751 bd->ltp_gain[3] = decode_rice(gb, 2) << 3;
752 bd->ltp_gain[4] = decode_rice(gb, 1) << 3;
754 *bd->ltp_lag = get_bits(gb, ctx->ltp_lag_length);
755 *bd->ltp_lag += FFMAX(4, opt_order + 1);
759 // read first value and residuals in case of a random access block
762 bd->raw_samples[0] = decode_rice(gb, avctx->bits_per_raw_sample - 4);
764 bd->raw_samples[1] = decode_rice(gb, FFMIN(s[0] + 3, ctx->s_max));
766 bd->raw_samples[2] = decode_rice(gb, FFMIN(s[0] + 1, ctx->s_max));
768 start = FFMIN(opt_order, 3);
771 // read all residuals
775 unsigned int b = av_clip((av_ceil_log2(bd->block_length) - 3) >> 1, 0, 5);
777 // read most significant bits
782 ff_bgmc_decode_init(gb, &high, &low, &value);
784 current_res = bd->raw_samples + start;
786 for (sb = 0; sb < sub_blocks; sb++) {
787 unsigned int sb_len = sb_length - (sb ? 0 : start);
789 k [sb] = s[sb] > b ? s[sb] - b : 0;
790 delta[sb] = 5 - s[sb] + k[sb];
792 ff_bgmc_decode(gb, sb_len, current_res,
793 delta[sb], sx[sb], &high, &low, &value, ctx->bgmc_lut, ctx->bgmc_lut_status);
795 current_res += sb_len;
798 ff_bgmc_decode_end(gb);
801 // read least significant bits and tails
802 current_res = bd->raw_samples + start;
804 for (sb = 0; sb < sub_blocks; sb++, start = 0) {
805 unsigned int cur_tail_code = tail_code[sx[sb]][delta[sb]];
806 unsigned int cur_k = k[sb];
807 unsigned int cur_s = s[sb];
809 for (; start < sb_length; start++) {
810 int32_t res = *current_res;
812 if (res == cur_tail_code) {
813 unsigned int max_msb = (2 + (sx[sb] > 2) + (sx[sb] > 10))
816 res = decode_rice(gb, cur_s);
819 res += (max_msb ) << cur_k;
821 res -= (max_msb - 1) << cur_k;
824 if (res > cur_tail_code)
834 res |= get_bits_long(gb, cur_k);
838 *current_res++ = res;
842 current_res = bd->raw_samples + start;
844 for (sb = 0; sb < sub_blocks; sb++, start = 0)
845 for (; start < sb_length; start++)
846 *current_res++ = decode_rice(gb, s[sb]);
849 if (!sconf->mc_coding || ctx->js_switch)
856 /** Decode the block data for a non-constant block
858 static int decode_var_block_data(ALSDecContext *ctx, ALSBlockData *bd)
860 ALSSpecificConfig *sconf = &ctx->sconf;
861 unsigned int block_length = bd->block_length;
862 unsigned int smp = 0;
864 int opt_order = *bd->opt_order;
867 int32_t *quant_cof = bd->quant_cof;
868 int32_t *lpc_cof = bd->lpc_cof;
869 int32_t *raw_samples = bd->raw_samples;
870 int32_t *raw_samples_end = bd->raw_samples + bd->block_length;
871 int32_t *lpc_cof_reversed = ctx->lpc_cof_reversed_buffer;
873 // reverse long-term prediction
877 for (ltp_smp = FFMAX(*bd->ltp_lag - 2, 0); ltp_smp < block_length; ltp_smp++) {
878 int center = ltp_smp - *bd->ltp_lag;
879 int begin = FFMAX(0, center - 2);
880 int end = center + 3;
881 int tab = 5 - (end - begin);
886 for (base = begin; base < end; base++, tab++)
887 y += MUL64(bd->ltp_gain[tab], raw_samples[base]);
889 raw_samples[ltp_smp] += y >> 7;
893 // reconstruct all samples from residuals
895 for (smp = 0; smp < opt_order; smp++) {
898 for (sb = 0; sb < smp; sb++)
899 y += MUL64(lpc_cof[sb], raw_samples[-(sb + 1)]);
901 *raw_samples++ -= y >> 20;
902 parcor_to_lpc(smp, quant_cof, lpc_cof);
905 for (k = 0; k < opt_order; k++)
906 parcor_to_lpc(k, quant_cof, lpc_cof);
908 // store previous samples in case that they have to be altered
909 if (*bd->store_prev_samples)
910 memcpy(bd->prev_raw_samples, raw_samples - sconf->max_order,
911 sizeof(*bd->prev_raw_samples) * sconf->max_order);
913 // reconstruct difference signal for prediction (joint-stereo)
914 if (bd->js_blocks && bd->raw_other) {
915 int32_t *left, *right;
917 if (bd->raw_other > raw_samples) { // D = R - L
919 right = bd->raw_other;
920 } else { // D = R - L
921 left = bd->raw_other;
925 for (sb = -1; sb >= -sconf->max_order; sb--)
926 raw_samples[sb] = right[sb] - left[sb];
929 // reconstruct shifted signal
931 for (sb = -1; sb >= -sconf->max_order; sb--)
932 raw_samples[sb] >>= *bd->shift_lsbs;
935 // reverse linear prediction coefficients for efficiency
936 lpc_cof = lpc_cof + opt_order;
938 for (sb = 0; sb < opt_order; sb++)
939 lpc_cof_reversed[sb] = lpc_cof[-(sb + 1)];
941 // reconstruct raw samples
942 raw_samples = bd->raw_samples + smp;
943 lpc_cof = lpc_cof_reversed + opt_order;
945 for (; raw_samples < raw_samples_end; raw_samples++) {
948 for (sb = -opt_order; sb < 0; sb++)
949 y += MUL64(lpc_cof[sb], raw_samples[sb]);
951 *raw_samples -= y >> 20;
954 raw_samples = bd->raw_samples;
956 // restore previous samples in case that they have been altered
957 if (*bd->store_prev_samples)
958 memcpy(raw_samples - sconf->max_order, bd->prev_raw_samples,
959 sizeof(*raw_samples) * sconf->max_order);
965 /** Read the block data.
967 static int read_block(ALSDecContext *ctx, ALSBlockData *bd)
970 GetBitContext *gb = &ctx->gb;
973 // read block type flag and read the samples accordingly
975 ret = read_var_block_data(ctx, bd);
977 read_const_block_data(ctx, bd);
984 /** Decode the block data.
986 static int decode_block(ALSDecContext *ctx, ALSBlockData *bd)
991 // read block type flag and read the samples accordingly
992 if (*bd->const_block)
993 decode_const_block_data(ctx, bd);
995 ret = decode_var_block_data(ctx, bd); // always return 0
1000 // TODO: read RLSLMS extension data
1002 if (*bd->shift_lsbs)
1003 for (smp = 0; smp < bd->block_length; smp++)
1004 bd->raw_samples[smp] <<= *bd->shift_lsbs;
1010 /** Read and decode block data successively.
1012 static int read_decode_block(ALSDecContext *ctx, ALSBlockData *bd)
1016 if ((ret = read_block(ctx, bd)) < 0)
1019 return decode_block(ctx, bd);
1023 /** Compute the number of samples left to decode for the current frame and
1024 * sets these samples to zero.
1026 static void zero_remaining(unsigned int b, unsigned int b_max,
1027 const unsigned int *div_blocks, int32_t *buf)
1029 unsigned int count = 0;
1031 for (; b < b_max; b++)
1032 count += div_blocks[b];
1035 memset(buf, 0, sizeof(*buf) * count);
1039 /** Decode blocks independently.
1041 static int decode_blocks_ind(ALSDecContext *ctx, unsigned int ra_frame,
1042 unsigned int c, const unsigned int *div_blocks,
1043 unsigned int *js_blocks)
1047 ALSBlockData bd = { 0 };
1049 bd.ra_block = ra_frame;
1050 bd.const_block = ctx->const_block;
1051 bd.shift_lsbs = ctx->shift_lsbs;
1052 bd.opt_order = ctx->opt_order;
1053 bd.store_prev_samples = ctx->store_prev_samples;
1054 bd.use_ltp = ctx->use_ltp;
1055 bd.ltp_lag = ctx->ltp_lag;
1056 bd.ltp_gain = ctx->ltp_gain[0];
1057 bd.quant_cof = ctx->quant_cof[0];
1058 bd.lpc_cof = ctx->lpc_cof[0];
1059 bd.prev_raw_samples = ctx->prev_raw_samples;
1060 bd.raw_samples = ctx->raw_samples[c];
1063 for (b = 0; b < ctx->num_blocks; b++) {
1064 bd.block_length = div_blocks[b];
1066 if ((ret = read_decode_block(ctx, &bd)) < 0) {
1067 // damaged block, write zero for the rest of the frame
1068 zero_remaining(b, ctx->num_blocks, div_blocks, bd.raw_samples);
1071 bd.raw_samples += div_blocks[b];
1079 /** Decode blocks dependently.
1081 static int decode_blocks(ALSDecContext *ctx, unsigned int ra_frame,
1082 unsigned int c, const unsigned int *div_blocks,
1083 unsigned int *js_blocks)
1085 ALSSpecificConfig *sconf = &ctx->sconf;
1086 unsigned int offset = 0;
1089 ALSBlockData bd[2] = { { 0 } };
1091 bd[0].ra_block = ra_frame;
1092 bd[0].const_block = ctx->const_block;
1093 bd[0].shift_lsbs = ctx->shift_lsbs;
1094 bd[0].opt_order = ctx->opt_order;
1095 bd[0].store_prev_samples = ctx->store_prev_samples;
1096 bd[0].use_ltp = ctx->use_ltp;
1097 bd[0].ltp_lag = ctx->ltp_lag;
1098 bd[0].ltp_gain = ctx->ltp_gain[0];
1099 bd[0].quant_cof = ctx->quant_cof[0];
1100 bd[0].lpc_cof = ctx->lpc_cof[0];
1101 bd[0].prev_raw_samples = ctx->prev_raw_samples;
1102 bd[0].js_blocks = *js_blocks;
1104 bd[1].ra_block = ra_frame;
1105 bd[1].const_block = ctx->const_block;
1106 bd[1].shift_lsbs = ctx->shift_lsbs;
1107 bd[1].opt_order = ctx->opt_order;
1108 bd[1].store_prev_samples = ctx->store_prev_samples;
1109 bd[1].use_ltp = ctx->use_ltp;
1110 bd[1].ltp_lag = ctx->ltp_lag;
1111 bd[1].ltp_gain = ctx->ltp_gain[0];
1112 bd[1].quant_cof = ctx->quant_cof[0];
1113 bd[1].lpc_cof = ctx->lpc_cof[0];
1114 bd[1].prev_raw_samples = ctx->prev_raw_samples;
1115 bd[1].js_blocks = *(js_blocks + 1);
1117 // decode all blocks
1118 for (b = 0; b < ctx->num_blocks; b++) {
1121 bd[0].block_length = div_blocks[b];
1122 bd[1].block_length = div_blocks[b];
1124 bd[0].raw_samples = ctx->raw_samples[c ] + offset;
1125 bd[1].raw_samples = ctx->raw_samples[c + 1] + offset;
1127 bd[0].raw_other = bd[1].raw_samples;
1128 bd[1].raw_other = bd[0].raw_samples;
1130 if ((ret = read_decode_block(ctx, &bd[0])) < 0 ||
1131 (ret = read_decode_block(ctx, &bd[1])) < 0)
1134 // reconstruct joint-stereo blocks
1135 if (bd[0].js_blocks) {
1136 if (bd[1].js_blocks)
1137 av_log(ctx->avctx, AV_LOG_WARNING, "Invalid channel pair!\n");
1139 for (s = 0; s < div_blocks[b]; s++)
1140 bd[0].raw_samples[s] = bd[1].raw_samples[s] - bd[0].raw_samples[s];
1141 } else if (bd[1].js_blocks) {
1142 for (s = 0; s < div_blocks[b]; s++)
1143 bd[1].raw_samples[s] = bd[1].raw_samples[s] + bd[0].raw_samples[s];
1146 offset += div_blocks[b];
1151 // store carryover raw samples,
1152 // the others channel raw samples are stored by the calling function.
1153 memmove(ctx->raw_samples[c] - sconf->max_order,
1154 ctx->raw_samples[c] - sconf->max_order + sconf->frame_length,
1155 sizeof(*ctx->raw_samples[c]) * sconf->max_order);
1159 // damaged block, write zero for the rest of the frame
1160 zero_remaining(b, ctx->num_blocks, div_blocks, bd[0].raw_samples);
1161 zero_remaining(b, ctx->num_blocks, div_blocks, bd[1].raw_samples);
1165 static inline int als_weighting(GetBitContext *gb, int k, int off)
1167 int idx = av_clip(decode_rice(gb, k) + off,
1168 0, FF_ARRAY_ELEMS(mcc_weightings) - 1);
1169 return mcc_weightings[idx];
1172 /** Read the channel data.
1174 static int read_channel_data(ALSDecContext *ctx, ALSChannelData *cd, int c)
1176 GetBitContext *gb = &ctx->gb;
1177 ALSChannelData *current = cd;
1178 unsigned int channels = ctx->avctx->channels;
1181 while (entries < channels && !(current->stop_flag = get_bits1(gb))) {
1182 current->master_channel = get_bits_long(gb, av_ceil_log2(channels));
1184 if (current->master_channel >= channels) {
1185 av_log(ctx->avctx, AV_LOG_ERROR, "Invalid master channel!\n");
1186 return AVERROR_INVALIDDATA;
1189 if (current->master_channel != c) {
1190 current->time_diff_flag = get_bits1(gb);
1191 current->weighting[0] = als_weighting(gb, 1, 16);
1192 current->weighting[1] = als_weighting(gb, 2, 14);
1193 current->weighting[2] = als_weighting(gb, 1, 16);
1195 if (current->time_diff_flag) {
1196 current->weighting[3] = als_weighting(gb, 1, 16);
1197 current->weighting[4] = als_weighting(gb, 1, 16);
1198 current->weighting[5] = als_weighting(gb, 1, 16);
1200 current->time_diff_sign = get_bits1(gb);
1201 current->time_diff_index = get_bits(gb, ctx->ltp_lag_length - 3) + 3;
1209 if (entries == channels) {
1210 av_log(ctx->avctx, AV_LOG_ERROR, "Damaged channel data!\n");
1211 return AVERROR_INVALIDDATA;
1219 /** Recursively reverts the inter-channel correlation for a block.
1221 static int revert_channel_correlation(ALSDecContext *ctx, ALSBlockData *bd,
1222 ALSChannelData **cd, int *reverted,
1223 unsigned int offset, int c)
1225 ALSChannelData *ch = cd[c];
1226 unsigned int dep = 0;
1227 unsigned int channels = ctx->avctx->channels;
1228 unsigned int channel_size = ctx->sconf.frame_length + ctx->sconf.max_order;
1235 while (dep < channels && !ch[dep].stop_flag) {
1236 revert_channel_correlation(ctx, bd, cd, reverted, offset,
1237 ch[dep].master_channel);
1242 if (dep == channels) {
1243 av_log(ctx->avctx, AV_LOG_WARNING, "Invalid channel correlation!\n");
1244 return AVERROR_INVALIDDATA;
1247 bd->const_block = ctx->const_block + c;
1248 bd->shift_lsbs = ctx->shift_lsbs + c;
1249 bd->opt_order = ctx->opt_order + c;
1250 bd->store_prev_samples = ctx->store_prev_samples + c;
1251 bd->use_ltp = ctx->use_ltp + c;
1252 bd->ltp_lag = ctx->ltp_lag + c;
1253 bd->ltp_gain = ctx->ltp_gain[c];
1254 bd->lpc_cof = ctx->lpc_cof[c];
1255 bd->quant_cof = ctx->quant_cof[c];
1256 bd->raw_samples = ctx->raw_samples[c] + offset;
1259 while (!ch[dep].stop_flag) {
1261 ptrdiff_t begin = 1;
1262 ptrdiff_t end = bd->block_length - 1;
1264 int32_t *master = ctx->raw_samples[ch[dep].master_channel] + offset;
1266 if (ch[dep].time_diff_flag) {
1267 int t = ch[dep].time_diff_index;
1269 if (ch[dep].time_diff_sign) {
1276 if (FFMIN(begin - 1, begin - 1 + t) < ctx->raw_buffer - master ||
1277 FFMAX(end + 1, end + 1 + t) > ctx->raw_buffer + channels * channel_size - master) {
1278 av_log(ctx->avctx, AV_LOG_ERROR,
1279 "sample pointer range [%p, %p] not contained in raw_buffer [%p, %p].\n",
1280 master + FFMIN(begin - 1, begin - 1 + t), master + FFMAX(end + 1, end + 1 + t),
1281 ctx->raw_buffer, ctx->raw_buffer + channels * channel_size);
1282 return AVERROR_INVALIDDATA;
1285 for (smp = begin; smp < end; smp++) {
1287 MUL64(ch[dep].weighting[0], master[smp - 1 ]) +
1288 MUL64(ch[dep].weighting[1], master[smp ]) +
1289 MUL64(ch[dep].weighting[2], master[smp + 1 ]) +
1290 MUL64(ch[dep].weighting[3], master[smp - 1 + t]) +
1291 MUL64(ch[dep].weighting[4], master[smp + t]) +
1292 MUL64(ch[dep].weighting[5], master[smp + 1 + t]);
1294 bd->raw_samples[smp] += y >> 7;
1298 if (begin - 1 < ctx->raw_buffer - master ||
1299 end + 1 > ctx->raw_buffer + channels * channel_size - master) {
1300 av_log(ctx->avctx, AV_LOG_ERROR,
1301 "sample pointer range [%p, %p] not contained in raw_buffer [%p, %p].\n",
1302 master + begin - 1, master + end + 1,
1303 ctx->raw_buffer, ctx->raw_buffer + channels * channel_size);
1304 return AVERROR_INVALIDDATA;
1307 for (smp = begin; smp < end; smp++) {
1309 MUL64(ch[dep].weighting[0], master[smp - 1]) +
1310 MUL64(ch[dep].weighting[1], master[smp ]) +
1311 MUL64(ch[dep].weighting[2], master[smp + 1]);
1313 bd->raw_samples[smp] += y >> 7;
1324 /** Read the frame data.
1326 static int read_frame_data(ALSDecContext *ctx, unsigned int ra_frame)
1328 ALSSpecificConfig *sconf = &ctx->sconf;
1329 AVCodecContext *avctx = ctx->avctx;
1330 GetBitContext *gb = &ctx->gb;
1331 unsigned int div_blocks[32]; ///< block sizes.
1333 unsigned int js_blocks[2];
1334 uint32_t bs_info = 0;
1337 // skip the size of the ra unit if present in the frame
1338 if (sconf->ra_flag == RA_FLAG_FRAMES && ra_frame)
1339 skip_bits_long(gb, 32);
1341 if (sconf->mc_coding && sconf->joint_stereo) {
1342 ctx->js_switch = get_bits1(gb);
1346 if (!sconf->mc_coding || ctx->js_switch) {
1347 int independent_bs = !sconf->joint_stereo;
1349 for (c = 0; c < avctx->channels; c++) {
1353 get_block_sizes(ctx, div_blocks, &bs_info);
1355 // if joint_stereo and block_switching is set, independent decoding
1356 // is signaled via the first bit of bs_info
1357 if (sconf->joint_stereo && sconf->block_switching)
1361 // if this is the last channel, it has to be decoded independently
1362 if (c == avctx->channels - 1)
1365 if (independent_bs) {
1366 ret = decode_blocks_ind(ctx, ra_frame, c,
1367 div_blocks, js_blocks);
1372 ret = decode_blocks(ctx, ra_frame, c, div_blocks, js_blocks);
1379 // store carryover raw samples
1380 memmove(ctx->raw_samples[c] - sconf->max_order,
1381 ctx->raw_samples[c] - sconf->max_order + sconf->frame_length,
1382 sizeof(*ctx->raw_samples[c]) * sconf->max_order);
1384 } else { // multi-channel coding
1385 ALSBlockData bd = { 0 };
1387 int *reverted_channels = ctx->reverted_channels;
1388 unsigned int offset = 0;
1390 for (c = 0; c < avctx->channels; c++)
1391 if (ctx->chan_data[c] < ctx->chan_data_buffer) {
1392 av_log(ctx->avctx, AV_LOG_ERROR, "Invalid channel data!\n");
1393 return AVERROR_INVALIDDATA;
1396 memset(reverted_channels, 0, sizeof(*reverted_channels) * avctx->channels);
1398 bd.ra_block = ra_frame;
1399 bd.prev_raw_samples = ctx->prev_raw_samples;
1401 get_block_sizes(ctx, div_blocks, &bs_info);
1403 for (b = 0; b < ctx->num_blocks; b++) {
1404 bd.block_length = div_blocks[b];
1405 if (bd.block_length <= 0) {
1406 av_log(ctx->avctx, AV_LOG_WARNING,
1407 "Invalid block length %u in channel data!\n",
1412 for (c = 0; c < avctx->channels; c++) {
1413 bd.const_block = ctx->const_block + c;
1414 bd.shift_lsbs = ctx->shift_lsbs + c;
1415 bd.opt_order = ctx->opt_order + c;
1416 bd.store_prev_samples = ctx->store_prev_samples + c;
1417 bd.use_ltp = ctx->use_ltp + c;
1418 bd.ltp_lag = ctx->ltp_lag + c;
1419 bd.ltp_gain = ctx->ltp_gain[c];
1420 bd.lpc_cof = ctx->lpc_cof[c];
1421 bd.quant_cof = ctx->quant_cof[c];
1422 bd.raw_samples = ctx->raw_samples[c] + offset;
1423 bd.raw_other = NULL;
1425 if ((ret = read_block(ctx, &bd)) < 0)
1427 if ((ret = read_channel_data(ctx, ctx->chan_data[c], c)) < 0)
1431 for (c = 0; c < avctx->channels; c++) {
1432 ret = revert_channel_correlation(ctx, &bd, ctx->chan_data,
1433 reverted_channels, offset, c);
1437 for (c = 0; c < avctx->channels; c++) {
1438 bd.const_block = ctx->const_block + c;
1439 bd.shift_lsbs = ctx->shift_lsbs + c;
1440 bd.opt_order = ctx->opt_order + c;
1441 bd.store_prev_samples = ctx->store_prev_samples + c;
1442 bd.use_ltp = ctx->use_ltp + c;
1443 bd.ltp_lag = ctx->ltp_lag + c;
1444 bd.ltp_gain = ctx->ltp_gain[c];
1445 bd.lpc_cof = ctx->lpc_cof[c];
1446 bd.quant_cof = ctx->quant_cof[c];
1447 bd.raw_samples = ctx->raw_samples[c] + offset;
1448 if ((ret = decode_block(ctx, &bd)) < 0)
1452 memset(reverted_channels, 0, avctx->channels * sizeof(*reverted_channels));
1453 offset += div_blocks[b];
1457 // store carryover raw samples
1458 for (c = 0; c < avctx->channels; c++)
1459 memmove(ctx->raw_samples[c] - sconf->max_order,
1460 ctx->raw_samples[c] - sconf->max_order + sconf->frame_length,
1461 sizeof(*ctx->raw_samples[c]) * sconf->max_order);
1464 // TODO: read_diff_float_data
1470 /** Decode an ALS frame.
1472 static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame_ptr,
1475 ALSDecContext *ctx = avctx->priv_data;
1476 AVFrame *frame = data;
1477 ALSSpecificConfig *sconf = &ctx->sconf;
1478 const uint8_t *buffer = avpkt->data;
1479 int buffer_size = avpkt->size;
1480 int invalid_frame, ret;
1481 unsigned int c, sample, ra_frame, bytes_read, shift;
1483 init_get_bits(&ctx->gb, buffer, buffer_size * 8);
1485 // In the case that the distance between random access frames is set to zero
1486 // (sconf->ra_distance == 0) no frame is treated as a random access frame.
1487 // For the first frame, if prediction is used, all samples used from the
1488 // previous frame are assumed to be zero.
1489 ra_frame = sconf->ra_distance && !(ctx->frame_id % sconf->ra_distance);
1491 // the last frame to decode might have a different length
1492 if (sconf->samples != 0xFFFFFFFF)
1493 ctx->cur_frame_length = FFMIN(sconf->samples - ctx->frame_id * (uint64_t) sconf->frame_length,
1494 sconf->frame_length);
1496 ctx->cur_frame_length = sconf->frame_length;
1498 // decode the frame data
1499 if ((invalid_frame = read_frame_data(ctx, ra_frame)) < 0)
1500 av_log(ctx->avctx, AV_LOG_WARNING,
1501 "Reading frame data failed. Skipping RA unit.\n");
1505 /* get output buffer */
1506 frame->nb_samples = ctx->cur_frame_length;
1507 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1508 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1512 // transform decoded frame into output format
1513 #define INTERLEAVE_OUTPUT(bps) \
1515 int##bps##_t *dest = (int##bps##_t*)frame->data[0]; \
1516 shift = bps - ctx->avctx->bits_per_raw_sample; \
1517 for (sample = 0; sample < ctx->cur_frame_length; sample++) \
1518 for (c = 0; c < avctx->channels; c++) \
1519 *dest++ = ctx->raw_samples[c][sample] << shift; \
1522 if (ctx->avctx->bits_per_raw_sample <= 16) {
1523 INTERLEAVE_OUTPUT(16)
1525 INTERLEAVE_OUTPUT(32)
1529 if (sconf->crc_enabled && (avctx->err_recognition & AV_EF_CRCCHECK)) {
1530 int swap = HAVE_BIGENDIAN != sconf->msb_first;
1532 if (ctx->avctx->bits_per_raw_sample == 24) {
1533 int32_t *src = (int32_t *)frame->data[0];
1536 sample < ctx->cur_frame_length * avctx->channels;
1541 v = av_bswap32(src[sample]);
1544 if (!HAVE_BIGENDIAN)
1547 ctx->crc = av_crc(ctx->crc_table, ctx->crc, (uint8_t*)(&v), 3);
1550 uint8_t *crc_source;
1553 if (ctx->avctx->bits_per_raw_sample <= 16) {
1554 int16_t *src = (int16_t*) frame->data[0];
1555 int16_t *dest = (int16_t*) ctx->crc_buffer;
1557 sample < ctx->cur_frame_length * avctx->channels;
1559 *dest++ = av_bswap16(src[sample]);
1561 ctx->bdsp.bswap_buf((uint32_t *) ctx->crc_buffer,
1562 (uint32_t *) frame->data[0],
1563 ctx->cur_frame_length * avctx->channels);
1565 crc_source = ctx->crc_buffer;
1567 crc_source = frame->data[0];
1570 ctx->crc = av_crc(ctx->crc_table, ctx->crc, crc_source,
1571 ctx->cur_frame_length * avctx->channels *
1572 av_get_bytes_per_sample(avctx->sample_fmt));
1576 // check CRC sums if this is the last frame
1577 if (ctx->cur_frame_length != sconf->frame_length &&
1578 ctx->crc_org != ctx->crc) {
1579 av_log(avctx, AV_LOG_ERROR, "CRC error.\n");
1580 if (avctx->err_recognition & AV_EF_EXPLODE)
1581 return AVERROR_INVALIDDATA;
1587 bytes_read = invalid_frame ? buffer_size :
1588 (get_bits_count(&ctx->gb) + 7) >> 3;
1594 /** Uninitialize the ALS decoder.
1596 static av_cold int decode_end(AVCodecContext *avctx)
1598 ALSDecContext *ctx = avctx->priv_data;
1600 av_freep(&ctx->sconf.chan_pos);
1602 ff_bgmc_end(&ctx->bgmc_lut, &ctx->bgmc_lut_status);
1604 av_freep(&ctx->const_block);
1605 av_freep(&ctx->shift_lsbs);
1606 av_freep(&ctx->opt_order);
1607 av_freep(&ctx->store_prev_samples);
1608 av_freep(&ctx->use_ltp);
1609 av_freep(&ctx->ltp_lag);
1610 av_freep(&ctx->ltp_gain);
1611 av_freep(&ctx->ltp_gain_buffer);
1612 av_freep(&ctx->quant_cof);
1613 av_freep(&ctx->lpc_cof);
1614 av_freep(&ctx->quant_cof_buffer);
1615 av_freep(&ctx->lpc_cof_buffer);
1616 av_freep(&ctx->lpc_cof_reversed_buffer);
1617 av_freep(&ctx->prev_raw_samples);
1618 av_freep(&ctx->raw_samples);
1619 av_freep(&ctx->raw_buffer);
1620 av_freep(&ctx->chan_data);
1621 av_freep(&ctx->chan_data_buffer);
1622 av_freep(&ctx->reverted_channels);
1623 av_freep(&ctx->crc_buffer);
1629 /** Initialize the ALS decoder.
1631 static av_cold int decode_init(AVCodecContext *avctx)
1634 unsigned int channel_size;
1635 int num_buffers, ret;
1636 ALSDecContext *ctx = avctx->priv_data;
1637 ALSSpecificConfig *sconf = &ctx->sconf;
1640 if (!avctx->extradata) {
1641 av_log(avctx, AV_LOG_ERROR, "Missing required ALS extradata.\n");
1642 return AVERROR_INVALIDDATA;
1645 if ((ret = read_specific_config(ctx)) < 0) {
1646 av_log(avctx, AV_LOG_ERROR, "Reading ALSSpecificConfig failed.\n");
1650 if ((ret = check_specific_config(ctx)) < 0) {
1655 ret = ff_bgmc_init(avctx, &ctx->bgmc_lut, &ctx->bgmc_lut_status);
1659 if (sconf->floating) {
1660 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
1661 avctx->bits_per_raw_sample = 32;
1663 avctx->sample_fmt = sconf->resolution > 1
1664 ? AV_SAMPLE_FMT_S32 : AV_SAMPLE_FMT_S16;
1665 avctx->bits_per_raw_sample = (sconf->resolution + 1) * 8;
1666 if (avctx->bits_per_raw_sample > 32) {
1667 av_log(avctx, AV_LOG_ERROR, "Bits per raw sample %d larger than 32.\n",
1668 avctx->bits_per_raw_sample);
1669 ret = AVERROR_INVALIDDATA;
1674 // set maximum Rice parameter for progressive decoding based on resolution
1675 // This is not specified in 14496-3 but actually done by the reference
1676 // codec RM22 revision 2.
1677 ctx->s_max = sconf->resolution > 1 ? 31 : 15;
1679 // set lag value for long-term prediction
1680 ctx->ltp_lag_length = 8 + (avctx->sample_rate >= 96000) +
1681 (avctx->sample_rate >= 192000);
1683 // allocate quantized parcor coefficient buffer
1684 num_buffers = sconf->mc_coding ? avctx->channels : 1;
1686 ctx->quant_cof = av_malloc(sizeof(*ctx->quant_cof) * num_buffers);
1687 ctx->lpc_cof = av_malloc(sizeof(*ctx->lpc_cof) * num_buffers);
1688 ctx->quant_cof_buffer = av_malloc(sizeof(*ctx->quant_cof_buffer) *
1689 num_buffers * sconf->max_order);
1690 ctx->lpc_cof_buffer = av_malloc(sizeof(*ctx->lpc_cof_buffer) *
1691 num_buffers * sconf->max_order);
1692 ctx->lpc_cof_reversed_buffer = av_malloc(sizeof(*ctx->lpc_cof_buffer) *
1695 if (!ctx->quant_cof || !ctx->lpc_cof ||
1696 !ctx->quant_cof_buffer || !ctx->lpc_cof_buffer ||
1697 !ctx->lpc_cof_reversed_buffer) {
1698 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
1699 ret = AVERROR(ENOMEM);
1703 // assign quantized parcor coefficient buffers
1704 for (c = 0; c < num_buffers; c++) {
1705 ctx->quant_cof[c] = ctx->quant_cof_buffer + c * sconf->max_order;
1706 ctx->lpc_cof[c] = ctx->lpc_cof_buffer + c * sconf->max_order;
1709 // allocate and assign lag and gain data buffer for ltp mode
1710 ctx->const_block = av_malloc (sizeof(*ctx->const_block) * num_buffers);
1711 ctx->shift_lsbs = av_malloc (sizeof(*ctx->shift_lsbs) * num_buffers);
1712 ctx->opt_order = av_malloc (sizeof(*ctx->opt_order) * num_buffers);
1713 ctx->store_prev_samples = av_malloc(sizeof(*ctx->store_prev_samples) * num_buffers);
1714 ctx->use_ltp = av_mallocz(sizeof(*ctx->use_ltp) * num_buffers);
1715 ctx->ltp_lag = av_malloc (sizeof(*ctx->ltp_lag) * num_buffers);
1716 ctx->ltp_gain = av_malloc (sizeof(*ctx->ltp_gain) * num_buffers);
1717 ctx->ltp_gain_buffer = av_malloc (sizeof(*ctx->ltp_gain_buffer) *
1720 if (!ctx->const_block || !ctx->shift_lsbs ||
1721 !ctx->opt_order || !ctx->store_prev_samples ||
1722 !ctx->use_ltp || !ctx->ltp_lag ||
1723 !ctx->ltp_gain || !ctx->ltp_gain_buffer) {
1724 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
1725 ret = AVERROR(ENOMEM);
1729 for (c = 0; c < num_buffers; c++)
1730 ctx->ltp_gain[c] = ctx->ltp_gain_buffer + c * 5;
1732 // allocate and assign channel data buffer for mcc mode
1733 if (sconf->mc_coding) {
1734 ctx->chan_data_buffer = av_malloc(sizeof(*ctx->chan_data_buffer) *
1735 num_buffers * num_buffers);
1736 ctx->chan_data = av_malloc(sizeof(*ctx->chan_data) *
1738 ctx->reverted_channels = av_malloc(sizeof(*ctx->reverted_channels) *
1741 if (!ctx->chan_data_buffer || !ctx->chan_data || !ctx->reverted_channels) {
1742 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
1743 ret = AVERROR(ENOMEM);
1747 for (c = 0; c < num_buffers; c++)
1748 ctx->chan_data[c] = ctx->chan_data_buffer + c * num_buffers;
1750 ctx->chan_data = NULL;
1751 ctx->chan_data_buffer = NULL;
1752 ctx->reverted_channels = NULL;
1755 channel_size = sconf->frame_length + sconf->max_order;
1757 ctx->prev_raw_samples = av_malloc (sizeof(*ctx->prev_raw_samples) * sconf->max_order);
1758 ctx->raw_buffer = av_mallocz(sizeof(*ctx-> raw_buffer) * avctx->channels * channel_size);
1759 ctx->raw_samples = av_malloc (sizeof(*ctx-> raw_samples) * avctx->channels);
1761 // allocate previous raw sample buffer
1762 if (!ctx->prev_raw_samples || !ctx->raw_buffer|| !ctx->raw_samples) {
1763 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
1764 ret = AVERROR(ENOMEM);
1768 // assign raw samples buffers
1769 ctx->raw_samples[0] = ctx->raw_buffer + sconf->max_order;
1770 for (c = 1; c < avctx->channels; c++)
1771 ctx->raw_samples[c] = ctx->raw_samples[c - 1] + channel_size;
1773 // allocate crc buffer
1774 if (HAVE_BIGENDIAN != sconf->msb_first && sconf->crc_enabled &&
1775 (avctx->err_recognition & AV_EF_CRCCHECK)) {
1776 ctx->crc_buffer = av_malloc(sizeof(*ctx->crc_buffer) *
1777 ctx->cur_frame_length *
1779 av_get_bytes_per_sample(avctx->sample_fmt));
1780 if (!ctx->crc_buffer) {
1781 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
1782 ret = AVERROR(ENOMEM);
1787 ff_bswapdsp_init(&ctx->bdsp);
1797 /** Flush (reset) the frame ID after seeking.
1799 static av_cold void flush(AVCodecContext *avctx)
1801 ALSDecContext *ctx = avctx->priv_data;
1807 AVCodec ff_als_decoder = {
1809 .long_name = NULL_IF_CONFIG_SMALL("MPEG-4 Audio Lossless Coding (ALS)"),
1810 .type = AVMEDIA_TYPE_AUDIO,
1811 .id = AV_CODEC_ID_MP4ALS,
1812 .priv_data_size = sizeof(ALSDecContext),
1813 .init = decode_init,
1814 .close = decode_end,
1815 .decode = decode_frame,
1817 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,