3 * Copyright (c) 2009 Thilo Borgmann <thilo.borgmann _at_ mail.de>
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
25 * @author Thilo Borgmann <thilo.borgmann _at_ mail.de>
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 cs_switch; ///< if true, channel rearrangement is done
203 unsigned int num_blocks; ///< number of blocks used in the current frame
204 unsigned int s_max; ///< maximum Rice parameter allowed in entropy coding
205 uint8_t *bgmc_lut; ///< pointer at lookup tables used for BGMC
206 int *bgmc_lut_status; ///< pointer at lookup table status flags used for BGMC
207 int ltp_lag_length; ///< number of bits used for ltp lag value
208 int *const_block; ///< contains const_block flags for all channels
209 unsigned int *shift_lsbs; ///< contains shift_lsbs flags for all channels
210 unsigned int *opt_order; ///< contains opt_order flags for all channels
211 int *store_prev_samples; ///< contains store_prev_samples flags for all channels
212 int *use_ltp; ///< contains use_ltp flags for all channels
213 int *ltp_lag; ///< contains ltp lag values for all channels
214 int **ltp_gain; ///< gain values for ltp 5-tap filter for a channel
215 int *ltp_gain_buffer; ///< contains all gain values for ltp 5-tap filter
216 int32_t **quant_cof; ///< quantized parcor coefficients for a channel
217 int32_t *quant_cof_buffer; ///< contains all quantized parcor coefficients
218 int32_t **lpc_cof; ///< coefficients of the direct form prediction filter for a channel
219 int32_t *lpc_cof_buffer; ///< contains all coefficients of the direct form prediction filter
220 int32_t *lpc_cof_reversed_buffer; ///< temporary buffer to set up a reversed versio of lpc_cof_buffer
221 ALSChannelData **chan_data; ///< channel data for multi-channel correlation
222 ALSChannelData *chan_data_buffer; ///< contains channel data for all channels
223 int *reverted_channels; ///< stores a flag for each reverted channel
224 int32_t *prev_raw_samples; ///< contains unshifted raw samples from the previous block
225 int32_t **raw_samples; ///< decoded raw samples for each channel
226 int32_t *raw_buffer; ///< contains all decoded raw samples including carryover samples
227 uint8_t *crc_buffer; ///< buffer of byte order corrected samples used for CRC check
231 typedef struct ALSBlockData {
232 unsigned int block_length; ///< number of samples within the block
233 unsigned int ra_block; ///< if true, this is a random access block
234 int *const_block; ///< if true, this is a constant value block
235 int js_blocks; ///< true if this block contains a difference signal
236 unsigned int *shift_lsbs; ///< shift of values for this block
237 unsigned int *opt_order; ///< prediction order of this block
238 int *store_prev_samples;///< if true, carryover samples have to be stored
239 int *use_ltp; ///< if true, long-term prediction is used
240 int *ltp_lag; ///< lag value for long-term prediction
241 int *ltp_gain; ///< gain values for ltp 5-tap filter
242 int32_t *quant_cof; ///< quantized parcor coefficients
243 int32_t *lpc_cof; ///< coefficients of the direct form prediction
244 int32_t *raw_samples; ///< decoded raw samples / residuals for this block
245 int32_t *prev_raw_samples; ///< contains unshifted raw samples from the previous block
246 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)
253 AVCodecContext *avctx = ctx->avctx;
254 ALSSpecificConfig *sconf = &ctx->sconf;
256 ff_dlog(avctx, "resolution = %i\n", sconf->resolution);
257 ff_dlog(avctx, "floating = %i\n", sconf->floating);
258 ff_dlog(avctx, "frame_length = %i\n", sconf->frame_length);
259 ff_dlog(avctx, "ra_distance = %i\n", sconf->ra_distance);
260 ff_dlog(avctx, "ra_flag = %i\n", sconf->ra_flag);
261 ff_dlog(avctx, "adapt_order = %i\n", sconf->adapt_order);
262 ff_dlog(avctx, "coef_table = %i\n", sconf->coef_table);
263 ff_dlog(avctx, "long_term_prediction = %i\n", sconf->long_term_prediction);
264 ff_dlog(avctx, "max_order = %i\n", sconf->max_order);
265 ff_dlog(avctx, "block_switching = %i\n", sconf->block_switching);
266 ff_dlog(avctx, "bgmc = %i\n", sconf->bgmc);
267 ff_dlog(avctx, "sb_part = %i\n", sconf->sb_part);
268 ff_dlog(avctx, "joint_stereo = %i\n", sconf->joint_stereo);
269 ff_dlog(avctx, "mc_coding = %i\n", sconf->mc_coding);
270 ff_dlog(avctx, "chan_config = %i\n", sconf->chan_config);
271 ff_dlog(avctx, "chan_sort = %i\n", sconf->chan_sort);
272 ff_dlog(avctx, "RLSLMS = %i\n", sconf->rlslms);
273 ff_dlog(avctx, "chan_config_info = %i\n", sconf->chan_config_info);
278 /** Read an ALSSpecificConfig from a buffer into the output struct.
280 static av_cold int read_specific_config(ALSDecContext *ctx)
284 int i, config_offset;
285 MPEG4AudioConfig m4ac = {0};
286 ALSSpecificConfig *sconf = &ctx->sconf;
287 AVCodecContext *avctx = ctx->avctx;
288 uint32_t als_id, header_size, trailer_size;
291 if ((ret = init_get_bits8(&gb, avctx->extradata, avctx->extradata_size)) < 0)
294 config_offset = avpriv_mpeg4audio_get_config(&m4ac, avctx->extradata,
295 avctx->extradata_size * 8, 1);
297 if (config_offset < 0)
298 return AVERROR_INVALIDDATA;
300 skip_bits_long(&gb, config_offset);
302 if (get_bits_left(&gb) < (30 << 3))
303 return AVERROR_INVALIDDATA;
305 // read the fixed items
306 als_id = get_bits_long(&gb, 32);
307 avctx->sample_rate = m4ac.sample_rate;
308 skip_bits_long(&gb, 32); // sample rate already known
309 sconf->samples = get_bits_long(&gb, 32);
310 avctx->channels = m4ac.channels;
311 skip_bits(&gb, 16); // number of channels already known
312 skip_bits(&gb, 3); // skip file_type
313 sconf->resolution = get_bits(&gb, 3);
314 sconf->floating = get_bits1(&gb);
315 sconf->msb_first = get_bits1(&gb);
316 sconf->frame_length = get_bits(&gb, 16) + 1;
317 sconf->ra_distance = get_bits(&gb, 8);
318 sconf->ra_flag = get_bits(&gb, 2);
319 sconf->adapt_order = get_bits1(&gb);
320 sconf->coef_table = get_bits(&gb, 2);
321 sconf->long_term_prediction = get_bits1(&gb);
322 sconf->max_order = get_bits(&gb, 10);
323 sconf->block_switching = get_bits(&gb, 2);
324 sconf->bgmc = get_bits1(&gb);
325 sconf->sb_part = get_bits1(&gb);
326 sconf->joint_stereo = get_bits1(&gb);
327 sconf->mc_coding = get_bits1(&gb);
328 sconf->chan_config = get_bits1(&gb);
329 sconf->chan_sort = get_bits1(&gb);
330 sconf->crc_enabled = get_bits1(&gb);
331 sconf->rlslms = get_bits1(&gb);
332 skip_bits(&gb, 5); // skip 5 reserved bits
333 skip_bits1(&gb); // skip aux_data_enabled
336 // check for ALSSpecificConfig struct
337 if (als_id != MKBETAG('A','L','S','\0'))
338 return AVERROR_INVALIDDATA;
340 ctx->cur_frame_length = sconf->frame_length;
342 // read channel config
343 if (sconf->chan_config)
344 sconf->chan_config_info = get_bits(&gb, 16);
345 // TODO: use this to set avctx->channel_layout
348 // read channel sorting
349 if (sconf->chan_sort && avctx->channels > 1) {
350 int chan_pos_bits = av_ceil_log2(avctx->channels);
351 int bits_needed = avctx->channels * chan_pos_bits + 7;
352 if (get_bits_left(&gb) < bits_needed)
353 return AVERROR_INVALIDDATA;
355 if (!(sconf->chan_pos = av_malloc_array(avctx->channels, sizeof(*sconf->chan_pos))))
356 return AVERROR(ENOMEM);
360 for (i = 0; i < avctx->channels; i++) {
361 sconf->chan_pos[i] = -1;
364 for (i = 0; i < avctx->channels; i++) {
367 idx = get_bits(&gb, chan_pos_bits);
368 if (idx >= avctx->channels || sconf->chan_pos[idx] != -1) {
369 av_log(avctx, AV_LOG_WARNING, "Invalid channel reordering.\n");
373 sconf->chan_pos[idx] = i;
380 // read fixed header and trailer sizes,
381 // if size = 0xFFFFFFFF then there is no data field!
382 if (get_bits_left(&gb) < 64)
383 return AVERROR_INVALIDDATA;
385 header_size = get_bits_long(&gb, 32);
386 trailer_size = get_bits_long(&gb, 32);
387 if (header_size == 0xFFFFFFFF)
389 if (trailer_size == 0xFFFFFFFF)
392 ht_size = ((int64_t)(header_size) + (int64_t)(trailer_size)) << 3;
395 // skip the header and trailer data
396 if (get_bits_left(&gb) < ht_size)
397 return AVERROR_INVALIDDATA;
399 if (ht_size > INT32_MAX)
400 return AVERROR_PATCHWELCOME;
402 skip_bits_long(&gb, ht_size);
405 // initialize CRC calculation
406 if (sconf->crc_enabled) {
407 if (get_bits_left(&gb) < 32)
408 return AVERROR_INVALIDDATA;
410 if (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL)) {
411 ctx->crc_table = av_crc_get_table(AV_CRC_32_IEEE_LE);
412 ctx->crc = 0xFFFFFFFF;
413 ctx->crc_org = ~get_bits_long(&gb, 32);
415 skip_bits_long(&gb, 32);
419 // no need to read the rest of ALSSpecificConfig (ra_unit_size & aux data)
421 dprint_specific_config(ctx);
427 /** Check the ALSSpecificConfig for unsupported features.
429 static int check_specific_config(ALSDecContext *ctx)
431 ALSSpecificConfig *sconf = &ctx->sconf;
434 // report unsupported feature and set error value
435 #define MISSING_ERR(cond, str, errval) \
438 avpriv_report_missing_feature(ctx->avctx, \
444 MISSING_ERR(sconf->floating, "Floating point decoding", AVERROR_PATCHWELCOME);
445 MISSING_ERR(sconf->rlslms, "Adaptive RLS-LMS prediction", AVERROR_PATCHWELCOME);
451 /** Parse the bs_info field to extract the block partitioning used in
452 * block switching mode, refer to ISO/IEC 14496-3, section 11.6.2.
454 static void parse_bs_info(const uint32_t bs_info, unsigned int n,
455 unsigned int div, unsigned int **div_blocks,
456 unsigned int *num_blocks)
458 if (n < 31 && ((bs_info << n) & 0x40000000)) {
459 // if the level is valid and the investigated bit n is set
460 // then recursively check both children at bits (2n+1) and (2n+2)
463 parse_bs_info(bs_info, n + 1, div, div_blocks, num_blocks);
464 parse_bs_info(bs_info, n + 2, div, div_blocks, num_blocks);
466 // else the bit is not set or the last level has been reached
467 // (bit implicitly not set)
475 /** Read and decode a Rice codeword.
477 static int32_t decode_rice(GetBitContext *gb, unsigned int k)
479 int max = get_bits_left(gb) - k;
480 int q = get_unary(gb, 0, max);
481 int r = k ? get_bits1(gb) : !(q & 1);
485 q += get_bits_long(gb, k - 1);
493 /** Convert PARCOR coefficient k to direct filter coefficient.
495 static void parcor_to_lpc(unsigned int k, const int32_t *par, int32_t *cof)
499 for (i = 0, j = k - 1; i < j; i++, j--) {
500 int tmp1 = ((MUL64(par[k], cof[j]) + (1 << 19)) >> 20);
501 cof[j] += ((MUL64(par[k], cof[i]) + (1 << 19)) >> 20);
505 cof[i] += ((MUL64(par[k], cof[j]) + (1 << 19)) >> 20);
511 /** Read block switching field if necessary and set actual block sizes.
512 * Also assure that the block sizes of the last frame correspond to the
513 * actual number of samples.
515 static void get_block_sizes(ALSDecContext *ctx, unsigned int *div_blocks,
518 ALSSpecificConfig *sconf = &ctx->sconf;
519 GetBitContext *gb = &ctx->gb;
520 unsigned int *ptr_div_blocks = div_blocks;
523 if (sconf->block_switching) {
524 unsigned int bs_info_len = 1 << (sconf->block_switching + 2);
525 *bs_info = get_bits_long(gb, bs_info_len);
526 *bs_info <<= (32 - bs_info_len);
530 parse_bs_info(*bs_info, 0, 0, &ptr_div_blocks, &ctx->num_blocks);
532 // The last frame may have an overdetermined block structure given in
533 // the bitstream. In that case the defined block structure would need
534 // more samples than available to be consistent.
535 // The block structure is actually used but the block sizes are adapted
536 // to fit the actual number of available samples.
537 // Example: 5 samples, 2nd level block sizes: 2 2 2 2.
538 // This results in the actual block sizes: 2 2 1 0.
539 // This is not specified in 14496-3 but actually done by the reference
540 // codec RM22 revision 2.
541 // This appears to happen in case of an odd number of samples in the last
542 // frame which is actually not allowed by the block length switching part
544 // The ALS conformance files feature an odd number of samples in the last
547 for (b = 0; b < ctx->num_blocks; b++)
548 div_blocks[b] = ctx->sconf.frame_length >> div_blocks[b];
550 if (ctx->cur_frame_length != ctx->sconf.frame_length) {
551 unsigned int remaining = ctx->cur_frame_length;
553 for (b = 0; b < ctx->num_blocks; b++) {
554 if (remaining <= div_blocks[b]) {
555 div_blocks[b] = remaining;
556 ctx->num_blocks = b + 1;
560 remaining -= div_blocks[b];
566 /** Read the block data for a constant block
568 static int read_const_block_data(ALSDecContext *ctx, ALSBlockData *bd)
570 ALSSpecificConfig *sconf = &ctx->sconf;
571 AVCodecContext *avctx = ctx->avctx;
572 GetBitContext *gb = &ctx->gb;
574 if (bd->block_length <= 0)
575 return AVERROR_INVALIDDATA;
577 *bd->raw_samples = 0;
578 *bd->const_block = get_bits1(gb); // 1 = constant value, 0 = zero block (silence)
579 bd->js_blocks = get_bits1(gb);
581 // skip 5 reserved bits
584 if (*bd->const_block) {
585 unsigned int const_val_bits = sconf->floating ? 24 : avctx->bits_per_raw_sample;
586 *bd->raw_samples = get_sbits_long(gb, const_val_bits);
589 // ensure constant block decoding by reusing this field
590 *bd->const_block = 1;
596 /** Decode the block data for a constant block
598 static void decode_const_block_data(ALSDecContext *ctx, ALSBlockData *bd)
600 int smp = bd->block_length - 1;
601 int32_t val = *bd->raw_samples;
602 int32_t *dst = bd->raw_samples + 1;
604 // write raw samples into buffer
610 /** Read the block data for a non-constant block
612 static int read_var_block_data(ALSDecContext *ctx, ALSBlockData *bd)
614 ALSSpecificConfig *sconf = &ctx->sconf;
615 AVCodecContext *avctx = ctx->avctx;
616 GetBitContext *gb = &ctx->gb;
620 unsigned int sub_blocks, log2_sub_blocks, sb_length;
621 unsigned int start = 0;
622 unsigned int opt_order;
624 int32_t *quant_cof = bd->quant_cof;
625 int32_t *current_res;
628 // ensure variable block decoding by reusing this field
629 *bd->const_block = 0;
632 bd->js_blocks = get_bits1(gb);
634 opt_order = *bd->opt_order;
636 // determine the number of subblocks for entropy decoding
637 if (!sconf->bgmc && !sconf->sb_part) {
640 if (sconf->bgmc && sconf->sb_part)
641 log2_sub_blocks = get_bits(gb, 2);
643 log2_sub_blocks = 2 * get_bits1(gb);
646 sub_blocks = 1 << log2_sub_blocks;
648 // do not continue in case of a damaged stream since
649 // block_length must be evenly divisible by sub_blocks
650 if (bd->block_length & (sub_blocks - 1)) {
651 av_log(avctx, AV_LOG_WARNING,
652 "Block length is not evenly divisible by the number of subblocks.\n");
653 return AVERROR_INVALIDDATA;
656 sb_length = bd->block_length >> log2_sub_blocks;
659 s[0] = get_bits(gb, 8 + (sconf->resolution > 1));
660 for (k = 1; k < sub_blocks; k++)
661 s[k] = s[k - 1] + decode_rice(gb, 2);
663 for (k = 0; k < sub_blocks; k++) {
668 s[0] = get_bits(gb, 4 + (sconf->resolution > 1));
669 for (k = 1; k < sub_blocks; k++)
670 s[k] = s[k - 1] + decode_rice(gb, 0);
672 for (k = 1; k < sub_blocks; k++)
674 av_log(avctx, AV_LOG_ERROR, "k invalid for rice code.\n");
675 return AVERROR_INVALIDDATA;
679 *bd->shift_lsbs = get_bits(gb, 4) + 1;
681 *bd->store_prev_samples = (bd->js_blocks && bd->raw_other) || *bd->shift_lsbs;
684 if (!sconf->rlslms) {
685 if (sconf->adapt_order && sconf->max_order) {
686 int opt_order_length = av_ceil_log2(av_clip((bd->block_length >> 3) - 1,
687 2, sconf->max_order + 1));
688 *bd->opt_order = get_bits(gb, opt_order_length);
689 if (*bd->opt_order > sconf->max_order) {
690 *bd->opt_order = sconf->max_order;
691 av_log(avctx, AV_LOG_ERROR, "Predictor order too large.\n");
692 return AVERROR_INVALIDDATA;
695 *bd->opt_order = sconf->max_order;
697 if (*bd->opt_order > bd->block_length) {
698 *bd->opt_order = bd->block_length;
699 av_log(avctx, AV_LOG_ERROR, "Predictor order too large.\n");
700 return AVERROR_INVALIDDATA;
702 opt_order = *bd->opt_order;
707 if (sconf->coef_table == 3) {
710 // read coefficient 0
711 quant_cof[0] = 32 * parcor_scaled_values[get_bits(gb, 7)];
713 // read coefficient 1
715 quant_cof[1] = -32 * parcor_scaled_values[get_bits(gb, 7)];
717 // read coefficients 2 to opt_order
718 for (k = 2; k < opt_order; k++)
719 quant_cof[k] = get_bits(gb, 7);
724 // read coefficient 0 to 19
725 k_max = FFMIN(opt_order, 20);
726 for (k = 0; k < k_max; k++) {
727 int rice_param = parcor_rice_table[sconf->coef_table][k][1];
728 int offset = parcor_rice_table[sconf->coef_table][k][0];
729 quant_cof[k] = decode_rice(gb, rice_param) + offset;
730 if (quant_cof[k] < -64 || quant_cof[k] > 63) {
731 av_log(avctx, AV_LOG_ERROR,
732 "quant_cof %"PRIu32" is out of range.\n",
734 return AVERROR_INVALIDDATA;
738 // read coefficients 20 to 126
739 k_max = FFMIN(opt_order, 127);
740 for (; k < k_max; k++)
741 quant_cof[k] = decode_rice(gb, 2) + (k & 1);
743 // read coefficients 127 to opt_order
744 for (; k < opt_order; k++)
745 quant_cof[k] = decode_rice(gb, 1);
747 quant_cof[0] = 32 * parcor_scaled_values[quant_cof[0] + 64];
750 quant_cof[1] = -32 * parcor_scaled_values[quant_cof[1] + 64];
753 for (k = 2; k < opt_order; k++)
754 quant_cof[k] = (quant_cof[k] << 14) + (add_base << 13);
758 // read LTP gain and lag values
759 if (sconf->long_term_prediction) {
760 *bd->use_ltp = get_bits1(gb);
765 bd->ltp_gain[0] = decode_rice(gb, 1) << 3;
766 bd->ltp_gain[1] = decode_rice(gb, 2) << 3;
768 r = get_unary(gb, 0, 3);
770 bd->ltp_gain[2] = ltp_gain_values[r][c];
772 bd->ltp_gain[3] = decode_rice(gb, 2) << 3;
773 bd->ltp_gain[4] = decode_rice(gb, 1) << 3;
775 *bd->ltp_lag = get_bits(gb, ctx->ltp_lag_length);
776 *bd->ltp_lag += FFMAX(4, opt_order + 1);
780 // read first value and residuals in case of a random access block
783 bd->raw_samples[0] = decode_rice(gb, avctx->bits_per_raw_sample - 4);
785 bd->raw_samples[1] = decode_rice(gb, FFMIN(s[0] + 3, ctx->s_max));
787 bd->raw_samples[2] = decode_rice(gb, FFMIN(s[0] + 1, ctx->s_max));
789 start = FFMIN(opt_order, 3);
792 // read all residuals
796 unsigned int b = av_clip((av_ceil_log2(bd->block_length) - 3) >> 1, 0, 5);
798 // read most significant bits
803 ff_bgmc_decode_init(gb, &high, &low, &value);
805 current_res = bd->raw_samples + start;
807 for (sb = 0; sb < sub_blocks; sb++) {
808 unsigned int sb_len = sb_length - (sb ? 0 : start);
810 k [sb] = s[sb] > b ? s[sb] - b : 0;
811 delta[sb] = 5 - s[sb] + k[sb];
813 ff_bgmc_decode(gb, sb_len, current_res,
814 delta[sb], sx[sb], &high, &low, &value, ctx->bgmc_lut, ctx->bgmc_lut_status);
816 current_res += sb_len;
819 ff_bgmc_decode_end(gb);
822 // read least significant bits and tails
823 current_res = bd->raw_samples + start;
825 for (sb = 0; sb < sub_blocks; sb++, start = 0) {
826 unsigned int cur_tail_code = tail_code[sx[sb]][delta[sb]];
827 unsigned int cur_k = k[sb];
828 unsigned int cur_s = s[sb];
830 for (; start < sb_length; start++) {
831 int32_t res = *current_res;
833 if (res == cur_tail_code) {
834 unsigned int max_msb = (2 + (sx[sb] > 2) + (sx[sb] > 10))
837 res = decode_rice(gb, cur_s);
840 res += (max_msb ) << cur_k;
842 res -= (max_msb - 1) << cur_k;
845 if (res > cur_tail_code)
855 res |= get_bits_long(gb, cur_k);
859 *current_res++ = res;
863 current_res = bd->raw_samples + start;
865 for (sb = 0; sb < sub_blocks; sb++, start = 0)
866 for (; start < sb_length; start++)
867 *current_res++ = decode_rice(gb, s[sb]);
870 if (!sconf->mc_coding || ctx->js_switch)
877 /** Decode the block data for a non-constant block
879 static int decode_var_block_data(ALSDecContext *ctx, ALSBlockData *bd)
881 ALSSpecificConfig *sconf = &ctx->sconf;
882 unsigned int block_length = bd->block_length;
883 unsigned int smp = 0;
885 int opt_order = *bd->opt_order;
888 int32_t *quant_cof = bd->quant_cof;
889 int32_t *lpc_cof = bd->lpc_cof;
890 int32_t *raw_samples = bd->raw_samples;
891 int32_t *raw_samples_end = bd->raw_samples + bd->block_length;
892 int32_t *lpc_cof_reversed = ctx->lpc_cof_reversed_buffer;
894 // reverse long-term prediction
898 for (ltp_smp = FFMAX(*bd->ltp_lag - 2, 0); ltp_smp < block_length; ltp_smp++) {
899 int center = ltp_smp - *bd->ltp_lag;
900 int begin = FFMAX(0, center - 2);
901 int end = center + 3;
902 int tab = 5 - (end - begin);
907 for (base = begin; base < end; base++, tab++)
908 y += MUL64(bd->ltp_gain[tab], raw_samples[base]);
910 raw_samples[ltp_smp] += y >> 7;
914 // reconstruct all samples from residuals
916 for (smp = 0; smp < opt_order; smp++) {
919 for (sb = 0; sb < smp; sb++)
920 y += MUL64(lpc_cof[sb], raw_samples[-(sb + 1)]);
922 *raw_samples++ -= y >> 20;
923 parcor_to_lpc(smp, quant_cof, lpc_cof);
926 for (k = 0; k < opt_order; k++)
927 parcor_to_lpc(k, quant_cof, lpc_cof);
929 // store previous samples in case that they have to be altered
930 if (*bd->store_prev_samples)
931 memcpy(bd->prev_raw_samples, raw_samples - sconf->max_order,
932 sizeof(*bd->prev_raw_samples) * sconf->max_order);
934 // reconstruct difference signal for prediction (joint-stereo)
935 if (bd->js_blocks && bd->raw_other) {
936 int32_t *left, *right;
938 if (bd->raw_other > raw_samples) { // D = R - L
940 right = bd->raw_other;
941 } else { // D = R - L
942 left = bd->raw_other;
946 for (sb = -1; sb >= -sconf->max_order; sb--)
947 raw_samples[sb] = right[sb] - left[sb];
950 // reconstruct shifted signal
952 for (sb = -1; sb >= -sconf->max_order; sb--)
953 raw_samples[sb] >>= *bd->shift_lsbs;
956 // reverse linear prediction coefficients for efficiency
957 lpc_cof = lpc_cof + opt_order;
959 for (sb = 0; sb < opt_order; sb++)
960 lpc_cof_reversed[sb] = lpc_cof[-(sb + 1)];
962 // reconstruct raw samples
963 raw_samples = bd->raw_samples + smp;
964 lpc_cof = lpc_cof_reversed + opt_order;
966 for (; raw_samples < raw_samples_end; raw_samples++) {
969 for (sb = -opt_order; sb < 0; sb++)
970 y += MUL64(lpc_cof[sb], raw_samples[sb]);
972 *raw_samples -= y >> 20;
975 raw_samples = bd->raw_samples;
977 // restore previous samples in case that they have been altered
978 if (*bd->store_prev_samples)
979 memcpy(raw_samples - sconf->max_order, bd->prev_raw_samples,
980 sizeof(*raw_samples) * sconf->max_order);
986 /** Read the block data.
988 static int read_block(ALSDecContext *ctx, ALSBlockData *bd)
991 GetBitContext *gb = &ctx->gb;
994 // read block type flag and read the samples accordingly
996 ret = read_var_block_data(ctx, bd);
998 ret = read_const_block_data(ctx, bd);
1005 /** Decode the block data.
1007 static int decode_block(ALSDecContext *ctx, ALSBlockData *bd)
1012 // read block type flag and read the samples accordingly
1013 if (*bd->const_block)
1014 decode_const_block_data(ctx, bd);
1016 ret = decode_var_block_data(ctx, bd); // always return 0
1021 // TODO: read RLSLMS extension data
1023 if (*bd->shift_lsbs)
1024 for (smp = 0; smp < bd->block_length; smp++)
1025 bd->raw_samples[smp] <<= *bd->shift_lsbs;
1031 /** Read and decode block data successively.
1033 static int read_decode_block(ALSDecContext *ctx, ALSBlockData *bd)
1037 if ((ret = read_block(ctx, bd)) < 0)
1040 return decode_block(ctx, bd);
1044 /** Compute the number of samples left to decode for the current frame and
1045 * sets these samples to zero.
1047 static void zero_remaining(unsigned int b, unsigned int b_max,
1048 const unsigned int *div_blocks, int32_t *buf)
1050 unsigned int count = 0;
1053 count += div_blocks[b++];
1056 memset(buf, 0, sizeof(*buf) * count);
1060 /** Decode blocks independently.
1062 static int decode_blocks_ind(ALSDecContext *ctx, unsigned int ra_frame,
1063 unsigned int c, const unsigned int *div_blocks,
1064 unsigned int *js_blocks)
1068 ALSBlockData bd = { 0 };
1070 bd.ra_block = ra_frame;
1071 bd.const_block = ctx->const_block;
1072 bd.shift_lsbs = ctx->shift_lsbs;
1073 bd.opt_order = ctx->opt_order;
1074 bd.store_prev_samples = ctx->store_prev_samples;
1075 bd.use_ltp = ctx->use_ltp;
1076 bd.ltp_lag = ctx->ltp_lag;
1077 bd.ltp_gain = ctx->ltp_gain[0];
1078 bd.quant_cof = ctx->quant_cof[0];
1079 bd.lpc_cof = ctx->lpc_cof[0];
1080 bd.prev_raw_samples = ctx->prev_raw_samples;
1081 bd.raw_samples = ctx->raw_samples[c];
1084 for (b = 0; b < ctx->num_blocks; b++) {
1085 bd.block_length = div_blocks[b];
1087 if ((ret = read_decode_block(ctx, &bd)) < 0) {
1088 // damaged block, write zero for the rest of the frame
1089 zero_remaining(b, ctx->num_blocks, div_blocks, bd.raw_samples);
1092 bd.raw_samples += div_blocks[b];
1100 /** Decode blocks dependently.
1102 static int decode_blocks(ALSDecContext *ctx, unsigned int ra_frame,
1103 unsigned int c, const unsigned int *div_blocks,
1104 unsigned int *js_blocks)
1106 ALSSpecificConfig *sconf = &ctx->sconf;
1107 unsigned int offset = 0;
1110 ALSBlockData bd[2] = { { 0 } };
1112 bd[0].ra_block = ra_frame;
1113 bd[0].const_block = ctx->const_block;
1114 bd[0].shift_lsbs = ctx->shift_lsbs;
1115 bd[0].opt_order = ctx->opt_order;
1116 bd[0].store_prev_samples = ctx->store_prev_samples;
1117 bd[0].use_ltp = ctx->use_ltp;
1118 bd[0].ltp_lag = ctx->ltp_lag;
1119 bd[0].ltp_gain = ctx->ltp_gain[0];
1120 bd[0].quant_cof = ctx->quant_cof[0];
1121 bd[0].lpc_cof = ctx->lpc_cof[0];
1122 bd[0].prev_raw_samples = ctx->prev_raw_samples;
1123 bd[0].js_blocks = *js_blocks;
1125 bd[1].ra_block = ra_frame;
1126 bd[1].const_block = ctx->const_block;
1127 bd[1].shift_lsbs = ctx->shift_lsbs;
1128 bd[1].opt_order = ctx->opt_order;
1129 bd[1].store_prev_samples = ctx->store_prev_samples;
1130 bd[1].use_ltp = ctx->use_ltp;
1131 bd[1].ltp_lag = ctx->ltp_lag;
1132 bd[1].ltp_gain = ctx->ltp_gain[0];
1133 bd[1].quant_cof = ctx->quant_cof[0];
1134 bd[1].lpc_cof = ctx->lpc_cof[0];
1135 bd[1].prev_raw_samples = ctx->prev_raw_samples;
1136 bd[1].js_blocks = *(js_blocks + 1);
1138 // decode all blocks
1139 for (b = 0; b < ctx->num_blocks; b++) {
1142 bd[0].block_length = div_blocks[b];
1143 bd[1].block_length = div_blocks[b];
1145 bd[0].raw_samples = ctx->raw_samples[c ] + offset;
1146 bd[1].raw_samples = ctx->raw_samples[c + 1] + offset;
1148 bd[0].raw_other = bd[1].raw_samples;
1149 bd[1].raw_other = bd[0].raw_samples;
1151 if ((ret = read_decode_block(ctx, &bd[0])) < 0 ||
1152 (ret = read_decode_block(ctx, &bd[1])) < 0)
1155 // reconstruct joint-stereo blocks
1156 if (bd[0].js_blocks) {
1157 if (bd[1].js_blocks)
1158 av_log(ctx->avctx, AV_LOG_WARNING, "Invalid channel pair.\n");
1160 for (s = 0; s < div_blocks[b]; s++)
1161 bd[0].raw_samples[s] = bd[1].raw_samples[s] - bd[0].raw_samples[s];
1162 } else if (bd[1].js_blocks) {
1163 for (s = 0; s < div_blocks[b]; s++)
1164 bd[1].raw_samples[s] = bd[1].raw_samples[s] + bd[0].raw_samples[s];
1167 offset += div_blocks[b];
1172 // store carryover raw samples,
1173 // the others channel raw samples are stored by the calling function.
1174 memmove(ctx->raw_samples[c] - sconf->max_order,
1175 ctx->raw_samples[c] - sconf->max_order + sconf->frame_length,
1176 sizeof(*ctx->raw_samples[c]) * sconf->max_order);
1180 // damaged block, write zero for the rest of the frame
1181 zero_remaining(b, ctx->num_blocks, div_blocks, bd[0].raw_samples);
1182 zero_remaining(b, ctx->num_blocks, div_blocks, bd[1].raw_samples);
1186 static inline int als_weighting(GetBitContext *gb, int k, int off)
1188 int idx = av_clip(decode_rice(gb, k) + off,
1189 0, FF_ARRAY_ELEMS(mcc_weightings) - 1);
1190 return mcc_weightings[idx];
1193 /** Read the channel data.
1195 static int read_channel_data(ALSDecContext *ctx, ALSChannelData *cd, int c)
1197 GetBitContext *gb = &ctx->gb;
1198 ALSChannelData *current = cd;
1199 unsigned int channels = ctx->avctx->channels;
1202 while (entries < channels && !(current->stop_flag = get_bits1(gb))) {
1203 current->master_channel = get_bits_long(gb, av_ceil_log2(channels));
1205 if (current->master_channel >= channels) {
1206 av_log(ctx->avctx, AV_LOG_ERROR, "Invalid master channel.\n");
1207 return AVERROR_INVALIDDATA;
1210 if (current->master_channel != c) {
1211 current->time_diff_flag = get_bits1(gb);
1212 current->weighting[0] = als_weighting(gb, 1, 16);
1213 current->weighting[1] = als_weighting(gb, 2, 14);
1214 current->weighting[2] = als_weighting(gb, 1, 16);
1216 if (current->time_diff_flag) {
1217 current->weighting[3] = als_weighting(gb, 1, 16);
1218 current->weighting[4] = als_weighting(gb, 1, 16);
1219 current->weighting[5] = als_weighting(gb, 1, 16);
1221 current->time_diff_sign = get_bits1(gb);
1222 current->time_diff_index = get_bits(gb, ctx->ltp_lag_length - 3) + 3;
1230 if (entries == channels) {
1231 av_log(ctx->avctx, AV_LOG_ERROR, "Damaged channel data.\n");
1232 return AVERROR_INVALIDDATA;
1240 /** Recursively reverts the inter-channel correlation for a block.
1242 static int revert_channel_correlation(ALSDecContext *ctx, ALSBlockData *bd,
1243 ALSChannelData **cd, int *reverted,
1244 unsigned int offset, int c)
1246 ALSChannelData *ch = cd[c];
1247 unsigned int dep = 0;
1248 unsigned int channels = ctx->avctx->channels;
1249 unsigned int channel_size = ctx->sconf.frame_length + ctx->sconf.max_order;
1256 while (dep < channels && !ch[dep].stop_flag) {
1257 revert_channel_correlation(ctx, bd, cd, reverted, offset,
1258 ch[dep].master_channel);
1263 if (dep == channels) {
1264 av_log(ctx->avctx, AV_LOG_WARNING, "Invalid channel correlation.\n");
1265 return AVERROR_INVALIDDATA;
1268 bd->const_block = ctx->const_block + c;
1269 bd->shift_lsbs = ctx->shift_lsbs + c;
1270 bd->opt_order = ctx->opt_order + c;
1271 bd->store_prev_samples = ctx->store_prev_samples + c;
1272 bd->use_ltp = ctx->use_ltp + c;
1273 bd->ltp_lag = ctx->ltp_lag + c;
1274 bd->ltp_gain = ctx->ltp_gain[c];
1275 bd->lpc_cof = ctx->lpc_cof[c];
1276 bd->quant_cof = ctx->quant_cof[c];
1277 bd->raw_samples = ctx->raw_samples[c] + offset;
1279 for (dep = 0; !ch[dep].stop_flag; dep++) {
1281 ptrdiff_t begin = 1;
1282 ptrdiff_t end = bd->block_length - 1;
1284 int32_t *master = ctx->raw_samples[ch[dep].master_channel] + offset;
1286 if (ch[dep].master_channel == c)
1289 if (ch[dep].time_diff_flag) {
1290 int t = ch[dep].time_diff_index;
1292 if (ch[dep].time_diff_sign) {
1295 av_log(ctx->avctx, AV_LOG_ERROR, "begin %td smaller than time diff index %d.\n", begin, t);
1296 return AVERROR_INVALIDDATA;
1301 av_log(ctx->avctx, AV_LOG_ERROR, "end %td smaller than time diff index %d.\n", end, t);
1302 return AVERROR_INVALIDDATA;
1307 if (FFMIN(begin - 1, begin - 1 + t) < ctx->raw_buffer - master ||
1308 FFMAX(end + 1, end + 1 + t) > ctx->raw_buffer + channels * channel_size - master) {
1309 av_log(ctx->avctx, AV_LOG_ERROR,
1310 "sample pointer range [%p, %p] not contained in raw_buffer [%p, %p].\n",
1311 master + FFMIN(begin - 1, begin - 1 + t), master + FFMAX(end + 1, end + 1 + t),
1312 ctx->raw_buffer, ctx->raw_buffer + channels * channel_size);
1313 return AVERROR_INVALIDDATA;
1316 for (smp = begin; smp < end; smp++) {
1318 MUL64(ch[dep].weighting[0], master[smp - 1 ]) +
1319 MUL64(ch[dep].weighting[1], master[smp ]) +
1320 MUL64(ch[dep].weighting[2], master[smp + 1 ]) +
1321 MUL64(ch[dep].weighting[3], master[smp - 1 + t]) +
1322 MUL64(ch[dep].weighting[4], master[smp + t]) +
1323 MUL64(ch[dep].weighting[5], master[smp + 1 + t]);
1325 bd->raw_samples[smp] += y >> 7;
1329 if (begin - 1 < ctx->raw_buffer - master ||
1330 end + 1 > ctx->raw_buffer + channels * channel_size - master) {
1331 av_log(ctx->avctx, AV_LOG_ERROR,
1332 "sample pointer range [%p, %p] not contained in raw_buffer [%p, %p].\n",
1333 master + begin - 1, master + end + 1,
1334 ctx->raw_buffer, ctx->raw_buffer + channels * channel_size);
1335 return AVERROR_INVALIDDATA;
1338 for (smp = begin; smp < end; smp++) {
1340 MUL64(ch[dep].weighting[0], master[smp - 1]) +
1341 MUL64(ch[dep].weighting[1], master[smp ]) +
1342 MUL64(ch[dep].weighting[2], master[smp + 1]);
1344 bd->raw_samples[smp] += y >> 7;
1353 /** Read the frame data.
1355 static int read_frame_data(ALSDecContext *ctx, unsigned int ra_frame)
1357 ALSSpecificConfig *sconf = &ctx->sconf;
1358 AVCodecContext *avctx = ctx->avctx;
1359 GetBitContext *gb = &ctx->gb;
1360 unsigned int div_blocks[32]; ///< block sizes.
1362 unsigned int js_blocks[2];
1363 uint32_t bs_info = 0;
1366 // skip the size of the ra unit if present in the frame
1367 if (sconf->ra_flag == RA_FLAG_FRAMES && ra_frame)
1368 skip_bits_long(gb, 32);
1370 if (sconf->mc_coding && sconf->joint_stereo) {
1371 ctx->js_switch = get_bits1(gb);
1375 if (!sconf->mc_coding || ctx->js_switch) {
1376 int independent_bs = !sconf->joint_stereo;
1378 for (c = 0; c < avctx->channels; c++) {
1382 get_block_sizes(ctx, div_blocks, &bs_info);
1384 // if joint_stereo and block_switching is set, independent decoding
1385 // is signaled via the first bit of bs_info
1386 if (sconf->joint_stereo && sconf->block_switching)
1390 // if this is the last channel, it has to be decoded independently
1391 if (c == avctx->channels - 1)
1394 if (independent_bs) {
1395 ret = decode_blocks_ind(ctx, ra_frame, c,
1396 div_blocks, js_blocks);
1401 ret = decode_blocks(ctx, ra_frame, c, div_blocks, js_blocks);
1408 // store carryover raw samples
1409 memmove(ctx->raw_samples[c] - sconf->max_order,
1410 ctx->raw_samples[c] - sconf->max_order + sconf->frame_length,
1411 sizeof(*ctx->raw_samples[c]) * sconf->max_order);
1413 } else { // multi-channel coding
1414 ALSBlockData bd = { 0 };
1416 int *reverted_channels = ctx->reverted_channels;
1417 unsigned int offset = 0;
1419 for (c = 0; c < avctx->channels; c++)
1420 if (ctx->chan_data[c] < ctx->chan_data_buffer) {
1421 av_log(ctx->avctx, AV_LOG_ERROR, "Invalid channel data.\n");
1422 return AVERROR_INVALIDDATA;
1425 memset(reverted_channels, 0, sizeof(*reverted_channels) * avctx->channels);
1427 bd.ra_block = ra_frame;
1428 bd.prev_raw_samples = ctx->prev_raw_samples;
1430 get_block_sizes(ctx, div_blocks, &bs_info);
1432 for (b = 0; b < ctx->num_blocks; b++) {
1433 bd.block_length = div_blocks[b];
1434 if (bd.block_length <= 0) {
1435 av_log(ctx->avctx, AV_LOG_WARNING,
1436 "Invalid block length %u in channel data!\n",
1441 for (c = 0; c < avctx->channels; c++) {
1442 bd.const_block = ctx->const_block + c;
1443 bd.shift_lsbs = ctx->shift_lsbs + c;
1444 bd.opt_order = ctx->opt_order + c;
1445 bd.store_prev_samples = ctx->store_prev_samples + c;
1446 bd.use_ltp = ctx->use_ltp + c;
1447 bd.ltp_lag = ctx->ltp_lag + c;
1448 bd.ltp_gain = ctx->ltp_gain[c];
1449 bd.lpc_cof = ctx->lpc_cof[c];
1450 bd.quant_cof = ctx->quant_cof[c];
1451 bd.raw_samples = ctx->raw_samples[c] + offset;
1452 bd.raw_other = NULL;
1454 if ((ret = read_block(ctx, &bd)) < 0)
1456 if ((ret = read_channel_data(ctx, ctx->chan_data[c], c)) < 0)
1460 for (c = 0; c < avctx->channels; c++) {
1461 ret = revert_channel_correlation(ctx, &bd, ctx->chan_data,
1462 reverted_channels, offset, c);
1466 for (c = 0; c < avctx->channels; c++) {
1467 bd.const_block = ctx->const_block + c;
1468 bd.shift_lsbs = ctx->shift_lsbs + c;
1469 bd.opt_order = ctx->opt_order + c;
1470 bd.store_prev_samples = ctx->store_prev_samples + c;
1471 bd.use_ltp = ctx->use_ltp + c;
1472 bd.ltp_lag = ctx->ltp_lag + c;
1473 bd.ltp_gain = ctx->ltp_gain[c];
1474 bd.lpc_cof = ctx->lpc_cof[c];
1475 bd.quant_cof = ctx->quant_cof[c];
1476 bd.raw_samples = ctx->raw_samples[c] + offset;
1478 if ((ret = decode_block(ctx, &bd)) < 0)
1482 memset(reverted_channels, 0, avctx->channels * sizeof(*reverted_channels));
1483 offset += div_blocks[b];
1487 // store carryover raw samples
1488 for (c = 0; c < avctx->channels; c++)
1489 memmove(ctx->raw_samples[c] - sconf->max_order,
1490 ctx->raw_samples[c] - sconf->max_order + sconf->frame_length,
1491 sizeof(*ctx->raw_samples[c]) * sconf->max_order);
1494 // TODO: read_diff_float_data
1496 if (get_bits_left(gb) < 0) {
1497 av_log(ctx->avctx, AV_LOG_ERROR, "Overread %d\n", -get_bits_left(gb));
1498 return AVERROR_INVALIDDATA;
1505 /** Decode an ALS frame.
1507 static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame_ptr,
1510 ALSDecContext *ctx = avctx->priv_data;
1511 AVFrame *frame = data;
1512 ALSSpecificConfig *sconf = &ctx->sconf;
1513 const uint8_t *buffer = avpkt->data;
1514 int buffer_size = avpkt->size;
1515 int invalid_frame, ret;
1516 unsigned int c, sample, ra_frame, bytes_read, shift;
1518 if ((ret = init_get_bits8(&ctx->gb, buffer, buffer_size)) < 0)
1521 // In the case that the distance between random access frames is set to zero
1522 // (sconf->ra_distance == 0) no frame is treated as a random access frame.
1523 // For the first frame, if prediction is used, all samples used from the
1524 // previous frame are assumed to be zero.
1525 ra_frame = sconf->ra_distance && !(ctx->frame_id % sconf->ra_distance);
1527 // the last frame to decode might have a different length
1528 if (sconf->samples != 0xFFFFFFFF)
1529 ctx->cur_frame_length = FFMIN(sconf->samples - ctx->frame_id * (uint64_t) sconf->frame_length,
1530 sconf->frame_length);
1532 ctx->cur_frame_length = sconf->frame_length;
1534 // decode the frame data
1535 if ((invalid_frame = read_frame_data(ctx, ra_frame)) < 0)
1536 av_log(ctx->avctx, AV_LOG_WARNING,
1537 "Reading frame data failed. Skipping RA unit.\n");
1541 /* get output buffer */
1542 frame->nb_samples = ctx->cur_frame_length;
1543 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1546 // transform decoded frame into output format
1547 #define INTERLEAVE_OUTPUT(bps) \
1549 int##bps##_t *dest = (int##bps##_t*)frame->data[0]; \
1550 shift = bps - ctx->avctx->bits_per_raw_sample; \
1551 if (!ctx->cs_switch) { \
1552 for (sample = 0; sample < ctx->cur_frame_length; sample++) \
1553 for (c = 0; c < avctx->channels; c++) \
1554 *dest++ = ctx->raw_samples[c][sample] << shift; \
1556 for (sample = 0; sample < ctx->cur_frame_length; sample++) \
1557 for (c = 0; c < avctx->channels; c++) \
1558 *dest++ = ctx->raw_samples[sconf->chan_pos[c]][sample] << shift; \
1562 if (ctx->avctx->bits_per_raw_sample <= 16) {
1563 INTERLEAVE_OUTPUT(16)
1565 INTERLEAVE_OUTPUT(32)
1569 if (sconf->crc_enabled && (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL))) {
1570 int swap = HAVE_BIGENDIAN != sconf->msb_first;
1572 if (ctx->avctx->bits_per_raw_sample == 24) {
1573 int32_t *src = (int32_t *)frame->data[0];
1576 sample < ctx->cur_frame_length * avctx->channels;
1581 v = av_bswap32(src[sample]);
1584 if (!HAVE_BIGENDIAN)
1587 ctx->crc = av_crc(ctx->crc_table, ctx->crc, (uint8_t*)(&v), 3);
1590 uint8_t *crc_source;
1593 if (ctx->avctx->bits_per_raw_sample <= 16) {
1594 int16_t *src = (int16_t*) frame->data[0];
1595 int16_t *dest = (int16_t*) ctx->crc_buffer;
1597 sample < ctx->cur_frame_length * avctx->channels;
1599 *dest++ = av_bswap16(src[sample]);
1601 ctx->bdsp.bswap_buf((uint32_t *) ctx->crc_buffer,
1602 (uint32_t *) frame->data[0],
1603 ctx->cur_frame_length * avctx->channels);
1605 crc_source = ctx->crc_buffer;
1607 crc_source = frame->data[0];
1610 ctx->crc = av_crc(ctx->crc_table, ctx->crc, crc_source,
1611 ctx->cur_frame_length * avctx->channels *
1612 av_get_bytes_per_sample(avctx->sample_fmt));
1616 // check CRC sums if this is the last frame
1617 if (ctx->cur_frame_length != sconf->frame_length &&
1618 ctx->crc_org != ctx->crc) {
1619 av_log(avctx, AV_LOG_ERROR, "CRC error.\n");
1620 if (avctx->err_recognition & AV_EF_EXPLODE)
1621 return AVERROR_INVALIDDATA;
1627 bytes_read = invalid_frame ? buffer_size :
1628 (get_bits_count(&ctx->gb) + 7) >> 3;
1634 /** Uninitialize the ALS decoder.
1636 static av_cold int decode_end(AVCodecContext *avctx)
1638 ALSDecContext *ctx = avctx->priv_data;
1640 av_freep(&ctx->sconf.chan_pos);
1642 ff_bgmc_end(&ctx->bgmc_lut, &ctx->bgmc_lut_status);
1644 av_freep(&ctx->const_block);
1645 av_freep(&ctx->shift_lsbs);
1646 av_freep(&ctx->opt_order);
1647 av_freep(&ctx->store_prev_samples);
1648 av_freep(&ctx->use_ltp);
1649 av_freep(&ctx->ltp_lag);
1650 av_freep(&ctx->ltp_gain);
1651 av_freep(&ctx->ltp_gain_buffer);
1652 av_freep(&ctx->quant_cof);
1653 av_freep(&ctx->lpc_cof);
1654 av_freep(&ctx->quant_cof_buffer);
1655 av_freep(&ctx->lpc_cof_buffer);
1656 av_freep(&ctx->lpc_cof_reversed_buffer);
1657 av_freep(&ctx->prev_raw_samples);
1658 av_freep(&ctx->raw_samples);
1659 av_freep(&ctx->raw_buffer);
1660 av_freep(&ctx->chan_data);
1661 av_freep(&ctx->chan_data_buffer);
1662 av_freep(&ctx->reverted_channels);
1663 av_freep(&ctx->crc_buffer);
1669 /** Initialize the ALS decoder.
1671 static av_cold int decode_init(AVCodecContext *avctx)
1674 unsigned int channel_size;
1675 int num_buffers, ret;
1676 ALSDecContext *ctx = avctx->priv_data;
1677 ALSSpecificConfig *sconf = &ctx->sconf;
1680 if (!avctx->extradata) {
1681 av_log(avctx, AV_LOG_ERROR, "Missing required ALS extradata.\n");
1682 return AVERROR_INVALIDDATA;
1685 if ((ret = read_specific_config(ctx)) < 0) {
1686 av_log(avctx, AV_LOG_ERROR, "Reading ALSSpecificConfig failed.\n");
1690 if ((ret = check_specific_config(ctx)) < 0) {
1695 ret = ff_bgmc_init(avctx, &ctx->bgmc_lut, &ctx->bgmc_lut_status);
1699 if (sconf->floating) {
1700 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
1701 avctx->bits_per_raw_sample = 32;
1703 avctx->sample_fmt = sconf->resolution > 1
1704 ? AV_SAMPLE_FMT_S32 : AV_SAMPLE_FMT_S16;
1705 avctx->bits_per_raw_sample = (sconf->resolution + 1) * 8;
1706 if (avctx->bits_per_raw_sample > 32) {
1707 av_log(avctx, AV_LOG_ERROR, "Bits per raw sample %d larger than 32.\n",
1708 avctx->bits_per_raw_sample);
1709 ret = AVERROR_INVALIDDATA;
1714 // set maximum Rice parameter for progressive decoding based on resolution
1715 // This is not specified in 14496-3 but actually done by the reference
1716 // codec RM22 revision 2.
1717 ctx->s_max = sconf->resolution > 1 ? 31 : 15;
1719 // set lag value for long-term prediction
1720 ctx->ltp_lag_length = 8 + (avctx->sample_rate >= 96000) +
1721 (avctx->sample_rate >= 192000);
1723 // allocate quantized parcor coefficient buffer
1724 num_buffers = sconf->mc_coding ? avctx->channels : 1;
1726 ctx->quant_cof = av_malloc_array(num_buffers, sizeof(*ctx->quant_cof));
1727 ctx->lpc_cof = av_malloc_array(num_buffers, sizeof(*ctx->lpc_cof));
1728 ctx->quant_cof_buffer = av_malloc_array(num_buffers * sconf->max_order,
1729 sizeof(*ctx->quant_cof_buffer));
1730 ctx->lpc_cof_buffer = av_malloc_array(num_buffers * sconf->max_order,
1731 sizeof(*ctx->lpc_cof_buffer));
1732 ctx->lpc_cof_reversed_buffer = av_malloc_array(sconf->max_order,
1733 sizeof(*ctx->lpc_cof_buffer));
1735 if (!ctx->quant_cof || !ctx->lpc_cof ||
1736 !ctx->quant_cof_buffer || !ctx->lpc_cof_buffer ||
1737 !ctx->lpc_cof_reversed_buffer) {
1738 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
1739 ret = AVERROR(ENOMEM);
1743 // assign quantized parcor coefficient buffers
1744 for (c = 0; c < num_buffers; c++) {
1745 ctx->quant_cof[c] = ctx->quant_cof_buffer + c * sconf->max_order;
1746 ctx->lpc_cof[c] = ctx->lpc_cof_buffer + c * sconf->max_order;
1749 // allocate and assign lag and gain data buffer for ltp mode
1750 ctx->const_block = av_malloc_array(num_buffers, sizeof(*ctx->const_block));
1751 ctx->shift_lsbs = av_malloc_array(num_buffers, sizeof(*ctx->shift_lsbs));
1752 ctx->opt_order = av_malloc_array(num_buffers, sizeof(*ctx->opt_order));
1753 ctx->store_prev_samples = av_malloc_array(num_buffers, sizeof(*ctx->store_prev_samples));
1754 ctx->use_ltp = av_mallocz_array(num_buffers, sizeof(*ctx->use_ltp));
1755 ctx->ltp_lag = av_malloc_array(num_buffers, sizeof(*ctx->ltp_lag));
1756 ctx->ltp_gain = av_malloc_array(num_buffers, sizeof(*ctx->ltp_gain));
1757 ctx->ltp_gain_buffer = av_malloc_array(num_buffers * 5, sizeof(*ctx->ltp_gain_buffer));
1759 if (!ctx->const_block || !ctx->shift_lsbs ||
1760 !ctx->opt_order || !ctx->store_prev_samples ||
1761 !ctx->use_ltp || !ctx->ltp_lag ||
1762 !ctx->ltp_gain || !ctx->ltp_gain_buffer) {
1763 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
1764 ret = AVERROR(ENOMEM);
1768 for (c = 0; c < num_buffers; c++)
1769 ctx->ltp_gain[c] = ctx->ltp_gain_buffer + c * 5;
1771 // allocate and assign channel data buffer for mcc mode
1772 if (sconf->mc_coding) {
1773 ctx->chan_data_buffer = av_mallocz_array(num_buffers * num_buffers,
1774 sizeof(*ctx->chan_data_buffer));
1775 ctx->chan_data = av_mallocz_array(num_buffers,
1776 sizeof(*ctx->chan_data));
1777 ctx->reverted_channels = av_malloc_array(num_buffers,
1778 sizeof(*ctx->reverted_channels));
1780 if (!ctx->chan_data_buffer || !ctx->chan_data || !ctx->reverted_channels) {
1781 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
1782 ret = AVERROR(ENOMEM);
1786 for (c = 0; c < num_buffers; c++)
1787 ctx->chan_data[c] = ctx->chan_data_buffer + c * num_buffers;
1789 ctx->chan_data = NULL;
1790 ctx->chan_data_buffer = NULL;
1791 ctx->reverted_channels = NULL;
1794 channel_size = sconf->frame_length + sconf->max_order;
1796 ctx->prev_raw_samples = av_malloc_array(sconf->max_order, sizeof(*ctx->prev_raw_samples));
1797 ctx->raw_buffer = av_mallocz_array(avctx->channels * channel_size, sizeof(*ctx->raw_buffer));
1798 ctx->raw_samples = av_malloc_array(avctx->channels, sizeof(*ctx->raw_samples));
1800 // allocate previous raw sample buffer
1801 if (!ctx->prev_raw_samples || !ctx->raw_buffer|| !ctx->raw_samples) {
1802 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
1803 ret = AVERROR(ENOMEM);
1807 // assign raw samples buffers
1808 ctx->raw_samples[0] = ctx->raw_buffer + sconf->max_order;
1809 for (c = 1; c < avctx->channels; c++)
1810 ctx->raw_samples[c] = ctx->raw_samples[c - 1] + channel_size;
1812 // allocate crc buffer
1813 if (HAVE_BIGENDIAN != sconf->msb_first && sconf->crc_enabled &&
1814 (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL))) {
1815 ctx->crc_buffer = av_malloc_array(ctx->cur_frame_length *
1817 av_get_bytes_per_sample(avctx->sample_fmt),
1818 sizeof(*ctx->crc_buffer));
1819 if (!ctx->crc_buffer) {
1820 av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
1821 ret = AVERROR(ENOMEM);
1826 ff_bswapdsp_init(&ctx->bdsp);
1836 /** Flush (reset) the frame ID after seeking.
1838 static av_cold void flush(AVCodecContext *avctx)
1840 ALSDecContext *ctx = avctx->priv_data;
1846 AVCodec ff_als_decoder = {
1848 .long_name = NULL_IF_CONFIG_SMALL("MPEG-4 Audio Lossless Coding (ALS)"),
1849 .type = AVMEDIA_TYPE_AUDIO,
1850 .id = AV_CODEC_ID_MP4ALS,
1851 .priv_data_size = sizeof(ALSDecContext),
1852 .init = decode_init,
1853 .close = decode_end,
1854 .decode = decode_frame,
1856 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,