You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

2155 lines
77KB

  1. /*
  2. * MPEG-4 ALS decoder
  3. * Copyright (c) 2009 Thilo Borgmann <thilo.borgmann _at_ mail.de>
  4. *
  5. * This file is part of FFmpeg.
  6. *
  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.
  11. *
  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.
  16. *
  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
  20. */
  21. /**
  22. * @file
  23. * MPEG-4 ALS decoder
  24. * @author Thilo Borgmann <thilo.borgmann _at_ mail.de>
  25. */
  26. #include <inttypes.h>
  27. #include "avcodec.h"
  28. #include "get_bits.h"
  29. #include "unary.h"
  30. #include "mpeg4audio.h"
  31. #include "bytestream.h"
  32. #include "bgmc.h"
  33. #include "bswapdsp.h"
  34. #include "internal.h"
  35. #include "mlz.h"
  36. #include "libavutil/samplefmt.h"
  37. #include "libavutil/crc.h"
  38. #include "libavutil/softfloat_ieee754.h"
  39. #include "libavutil/intfloat.h"
  40. #include "libavutil/intreadwrite.h"
  41. #include <stdint.h>
  42. /** Rice parameters and corresponding index offsets for decoding the
  43. * indices of scaled PARCOR values. The table chosen is set globally
  44. * by the encoder and stored in ALSSpecificConfig.
  45. */
  46. static const int8_t parcor_rice_table[3][20][2] = {
  47. { {-52, 4}, {-29, 5}, {-31, 4}, { 19, 4}, {-16, 4},
  48. { 12, 3}, { -7, 3}, { 9, 3}, { -5, 3}, { 6, 3},
  49. { -4, 3}, { 3, 3}, { -3, 2}, { 3, 2}, { -2, 2},
  50. { 3, 2}, { -1, 2}, { 2, 2}, { -1, 2}, { 2, 2} },
  51. { {-58, 3}, {-42, 4}, {-46, 4}, { 37, 5}, {-36, 4},
  52. { 29, 4}, {-29, 4}, { 25, 4}, {-23, 4}, { 20, 4},
  53. {-17, 4}, { 16, 4}, {-12, 4}, { 12, 3}, {-10, 4},
  54. { 7, 3}, { -4, 4}, { 3, 3}, { -1, 3}, { 1, 3} },
  55. { {-59, 3}, {-45, 5}, {-50, 4}, { 38, 4}, {-39, 4},
  56. { 32, 4}, {-30, 4}, { 25, 3}, {-23, 3}, { 20, 3},
  57. {-20, 3}, { 16, 3}, {-13, 3}, { 10, 3}, { -7, 3},
  58. { 3, 3}, { 0, 3}, { -1, 3}, { 2, 3}, { -1, 2} }
  59. };
  60. /** Scaled PARCOR values used for the first two PARCOR coefficients.
  61. * To be indexed by the Rice coded indices.
  62. * Generated by: parcor_scaled_values[i] = 32 + ((i * (i+1)) << 7) - (1 << 20)
  63. * Actual values are divided by 32 in order to be stored in 16 bits.
  64. */
  65. static const int16_t parcor_scaled_values[] = {
  66. -1048544 / 32, -1048288 / 32, -1047776 / 32, -1047008 / 32,
  67. -1045984 / 32, -1044704 / 32, -1043168 / 32, -1041376 / 32,
  68. -1039328 / 32, -1037024 / 32, -1034464 / 32, -1031648 / 32,
  69. -1028576 / 32, -1025248 / 32, -1021664 / 32, -1017824 / 32,
  70. -1013728 / 32, -1009376 / 32, -1004768 / 32, -999904 / 32,
  71. -994784 / 32, -989408 / 32, -983776 / 32, -977888 / 32,
  72. -971744 / 32, -965344 / 32, -958688 / 32, -951776 / 32,
  73. -944608 / 32, -937184 / 32, -929504 / 32, -921568 / 32,
  74. -913376 / 32, -904928 / 32, -896224 / 32, -887264 / 32,
  75. -878048 / 32, -868576 / 32, -858848 / 32, -848864 / 32,
  76. -838624 / 32, -828128 / 32, -817376 / 32, -806368 / 32,
  77. -795104 / 32, -783584 / 32, -771808 / 32, -759776 / 32,
  78. -747488 / 32, -734944 / 32, -722144 / 32, -709088 / 32,
  79. -695776 / 32, -682208 / 32, -668384 / 32, -654304 / 32,
  80. -639968 / 32, -625376 / 32, -610528 / 32, -595424 / 32,
  81. -580064 / 32, -564448 / 32, -548576 / 32, -532448 / 32,
  82. -516064 / 32, -499424 / 32, -482528 / 32, -465376 / 32,
  83. -447968 / 32, -430304 / 32, -412384 / 32, -394208 / 32,
  84. -375776 / 32, -357088 / 32, -338144 / 32, -318944 / 32,
  85. -299488 / 32, -279776 / 32, -259808 / 32, -239584 / 32,
  86. -219104 / 32, -198368 / 32, -177376 / 32, -156128 / 32,
  87. -134624 / 32, -112864 / 32, -90848 / 32, -68576 / 32,
  88. -46048 / 32, -23264 / 32, -224 / 32, 23072 / 32,
  89. 46624 / 32, 70432 / 32, 94496 / 32, 118816 / 32,
  90. 143392 / 32, 168224 / 32, 193312 / 32, 218656 / 32,
  91. 244256 / 32, 270112 / 32, 296224 / 32, 322592 / 32,
  92. 349216 / 32, 376096 / 32, 403232 / 32, 430624 / 32,
  93. 458272 / 32, 486176 / 32, 514336 / 32, 542752 / 32,
  94. 571424 / 32, 600352 / 32, 629536 / 32, 658976 / 32,
  95. 688672 / 32, 718624 / 32, 748832 / 32, 779296 / 32,
  96. 810016 / 32, 840992 / 32, 872224 / 32, 903712 / 32,
  97. 935456 / 32, 967456 / 32, 999712 / 32, 1032224 / 32
  98. };
  99. /** Gain values of p(0) for long-term prediction.
  100. * To be indexed by the Rice coded indices.
  101. */
  102. static const uint8_t ltp_gain_values [4][4] = {
  103. { 0, 8, 16, 24},
  104. {32, 40, 48, 56},
  105. {64, 70, 76, 82},
  106. {88, 92, 96, 100}
  107. };
  108. /** Inter-channel weighting factors for multi-channel correlation.
  109. * To be indexed by the Rice coded indices.
  110. */
  111. static const int16_t mcc_weightings[] = {
  112. 204, 192, 179, 166, 153, 140, 128, 115,
  113. 102, 89, 76, 64, 51, 38, 25, 12,
  114. 0, -12, -25, -38, -51, -64, -76, -89,
  115. -102, -115, -128, -140, -153, -166, -179, -192
  116. };
  117. /** Tail codes used in arithmetic coding using block Gilbert-Moore codes.
  118. */
  119. static const uint8_t tail_code[16][6] = {
  120. { 74, 44, 25, 13, 7, 3},
  121. { 68, 42, 24, 13, 7, 3},
  122. { 58, 39, 23, 13, 7, 3},
  123. {126, 70, 37, 19, 10, 5},
  124. {132, 70, 37, 20, 10, 5},
  125. {124, 70, 38, 20, 10, 5},
  126. {120, 69, 37, 20, 11, 5},
  127. {116, 67, 37, 20, 11, 5},
  128. {108, 66, 36, 20, 10, 5},
  129. {102, 62, 36, 20, 10, 5},
  130. { 88, 58, 34, 19, 10, 5},
  131. {162, 89, 49, 25, 13, 7},
  132. {156, 87, 49, 26, 14, 7},
  133. {150, 86, 47, 26, 14, 7},
  134. {142, 84, 47, 26, 14, 7},
  135. {131, 79, 46, 26, 14, 7}
  136. };
  137. enum RA_Flag {
  138. RA_FLAG_NONE,
  139. RA_FLAG_FRAMES,
  140. RA_FLAG_HEADER
  141. };
  142. typedef struct ALSSpecificConfig {
  143. uint32_t samples; ///< number of samples, 0xFFFFFFFF if unknown
  144. int resolution; ///< 000 = 8-bit; 001 = 16-bit; 010 = 24-bit; 011 = 32-bit
  145. int floating; ///< 1 = IEEE 32-bit floating-point, 0 = integer
  146. int msb_first; ///< 1 = original CRC calculated on big-endian system, 0 = little-endian
  147. int frame_length; ///< frame length for each frame (last frame may differ)
  148. int ra_distance; ///< distance between RA frames (in frames, 0...255)
  149. enum RA_Flag ra_flag; ///< indicates where the size of ra units is stored
  150. int adapt_order; ///< adaptive order: 1 = on, 0 = off
  151. int coef_table; ///< table index of Rice code parameters
  152. int long_term_prediction; ///< long term prediction (LTP): 1 = on, 0 = off
  153. int max_order; ///< maximum prediction order (0..1023)
  154. int block_switching; ///< number of block switching levels
  155. int bgmc; ///< "Block Gilbert-Moore Code": 1 = on, 0 = off (Rice coding only)
  156. int sb_part; ///< sub-block partition
  157. int joint_stereo; ///< joint stereo: 1 = on, 0 = off
  158. int mc_coding; ///< extended inter-channel coding (multi channel coding): 1 = on, 0 = off
  159. int chan_config; ///< indicates that a chan_config_info field is present
  160. int chan_sort; ///< channel rearrangement: 1 = on, 0 = off
  161. int rlslms; ///< use "Recursive Least Square-Least Mean Square" predictor: 1 = on, 0 = off
  162. int chan_config_info; ///< mapping of channels to loudspeaker locations. Unused until setting channel configuration is implemented.
  163. int *chan_pos; ///< original channel positions
  164. int crc_enabled; ///< enable Cyclic Redundancy Checksum
  165. } ALSSpecificConfig;
  166. typedef struct ALSChannelData {
  167. int stop_flag;
  168. int master_channel;
  169. int time_diff_flag;
  170. int time_diff_sign;
  171. int time_diff_index;
  172. int weighting[6];
  173. } ALSChannelData;
  174. typedef struct ALSDecContext {
  175. AVCodecContext *avctx;
  176. ALSSpecificConfig sconf;
  177. GetBitContext gb;
  178. BswapDSPContext bdsp;
  179. const AVCRC *crc_table;
  180. uint32_t crc_org; ///< CRC value of the original input data
  181. uint32_t crc; ///< CRC value calculated from decoded data
  182. unsigned int cur_frame_length; ///< length of the current frame to decode
  183. unsigned int frame_id; ///< the frame ID / number of the current frame
  184. unsigned int js_switch; ///< if true, joint-stereo decoding is enforced
  185. unsigned int cs_switch; ///< if true, channel rearrangement is done
  186. unsigned int num_blocks; ///< number of blocks used in the current frame
  187. unsigned int s_max; ///< maximum Rice parameter allowed in entropy coding
  188. uint8_t *bgmc_lut; ///< pointer at lookup tables used for BGMC
  189. int *bgmc_lut_status; ///< pointer at lookup table status flags used for BGMC
  190. int ltp_lag_length; ///< number of bits used for ltp lag value
  191. int *const_block; ///< contains const_block flags for all channels
  192. unsigned int *shift_lsbs; ///< contains shift_lsbs flags for all channels
  193. unsigned int *opt_order; ///< contains opt_order flags for all channels
  194. int *store_prev_samples; ///< contains store_prev_samples flags for all channels
  195. int *use_ltp; ///< contains use_ltp flags for all channels
  196. int *ltp_lag; ///< contains ltp lag values for all channels
  197. int **ltp_gain; ///< gain values for ltp 5-tap filter for a channel
  198. int *ltp_gain_buffer; ///< contains all gain values for ltp 5-tap filter
  199. int32_t **quant_cof; ///< quantized parcor coefficients for a channel
  200. int32_t *quant_cof_buffer; ///< contains all quantized parcor coefficients
  201. int32_t **lpc_cof; ///< coefficients of the direct form prediction filter for a channel
  202. int32_t *lpc_cof_buffer; ///< contains all coefficients of the direct form prediction filter
  203. int32_t *lpc_cof_reversed_buffer; ///< temporary buffer to set up a reversed versio of lpc_cof_buffer
  204. ALSChannelData **chan_data; ///< channel data for multi-channel correlation
  205. ALSChannelData *chan_data_buffer; ///< contains channel data for all channels
  206. int *reverted_channels; ///< stores a flag for each reverted channel
  207. int32_t *prev_raw_samples; ///< contains unshifted raw samples from the previous block
  208. int32_t **raw_samples; ///< decoded raw samples for each channel
  209. int32_t *raw_buffer; ///< contains all decoded raw samples including carryover samples
  210. uint8_t *crc_buffer; ///< buffer of byte order corrected samples used for CRC check
  211. MLZ* mlz; ///< masked lz decompression structure
  212. SoftFloat_IEEE754 *acf; ///< contains common multiplier for all channels
  213. int *last_acf_mantissa; ///< contains the last acf mantissa data of common multiplier for all channels
  214. int *shift_value; ///< value by which the binary point is to be shifted for all channels
  215. int *last_shift_value; ///< contains last shift value for all channels
  216. int **raw_mantissa; ///< decoded mantissa bits of the difference signal
  217. unsigned char *larray; ///< buffer to store the output of masked lz decompression
  218. int *nbits; ///< contains the number of bits to read for masked lz decompression for all samples
  219. } ALSDecContext;
  220. typedef struct ALSBlockData {
  221. unsigned int block_length; ///< number of samples within the block
  222. unsigned int ra_block; ///< if true, this is a random access block
  223. int *const_block; ///< if true, this is a constant value block
  224. int js_blocks; ///< true if this block contains a difference signal
  225. unsigned int *shift_lsbs; ///< shift of values for this block
  226. unsigned int *opt_order; ///< prediction order of this block
  227. int *store_prev_samples;///< if true, carryover samples have to be stored
  228. int *use_ltp; ///< if true, long-term prediction is used
  229. int *ltp_lag; ///< lag value for long-term prediction
  230. int *ltp_gain; ///< gain values for ltp 5-tap filter
  231. int32_t *quant_cof; ///< quantized parcor coefficients
  232. int32_t *lpc_cof; ///< coefficients of the direct form prediction
  233. int32_t *raw_samples; ///< decoded raw samples / residuals for this block
  234. int32_t *prev_raw_samples; ///< contains unshifted raw samples from the previous block
  235. int32_t *raw_other; ///< decoded raw samples of the other channel of a channel pair
  236. } ALSBlockData;
  237. static av_cold void dprint_specific_config(ALSDecContext *ctx)
  238. {
  239. #ifdef DEBUG
  240. AVCodecContext *avctx = ctx->avctx;
  241. ALSSpecificConfig *sconf = &ctx->sconf;
  242. ff_dlog(avctx, "resolution = %i\n", sconf->resolution);
  243. ff_dlog(avctx, "floating = %i\n", sconf->floating);
  244. ff_dlog(avctx, "frame_length = %i\n", sconf->frame_length);
  245. ff_dlog(avctx, "ra_distance = %i\n", sconf->ra_distance);
  246. ff_dlog(avctx, "ra_flag = %i\n", sconf->ra_flag);
  247. ff_dlog(avctx, "adapt_order = %i\n", sconf->adapt_order);
  248. ff_dlog(avctx, "coef_table = %i\n", sconf->coef_table);
  249. ff_dlog(avctx, "long_term_prediction = %i\n", sconf->long_term_prediction);
  250. ff_dlog(avctx, "max_order = %i\n", sconf->max_order);
  251. ff_dlog(avctx, "block_switching = %i\n", sconf->block_switching);
  252. ff_dlog(avctx, "bgmc = %i\n", sconf->bgmc);
  253. ff_dlog(avctx, "sb_part = %i\n", sconf->sb_part);
  254. ff_dlog(avctx, "joint_stereo = %i\n", sconf->joint_stereo);
  255. ff_dlog(avctx, "mc_coding = %i\n", sconf->mc_coding);
  256. ff_dlog(avctx, "chan_config = %i\n", sconf->chan_config);
  257. ff_dlog(avctx, "chan_sort = %i\n", sconf->chan_sort);
  258. ff_dlog(avctx, "RLSLMS = %i\n", sconf->rlslms);
  259. ff_dlog(avctx, "chan_config_info = %i\n", sconf->chan_config_info);
  260. #endif
  261. }
  262. /** Read an ALSSpecificConfig from a buffer into the output struct.
  263. */
  264. static av_cold int read_specific_config(ALSDecContext *ctx)
  265. {
  266. GetBitContext gb;
  267. uint64_t ht_size;
  268. int i, config_offset;
  269. MPEG4AudioConfig m4ac = {0};
  270. ALSSpecificConfig *sconf = &ctx->sconf;
  271. AVCodecContext *avctx = ctx->avctx;
  272. uint32_t als_id, header_size, trailer_size;
  273. int ret;
  274. if ((ret = init_get_bits8(&gb, avctx->extradata, avctx->extradata_size)) < 0)
  275. return ret;
  276. config_offset = avpriv_mpeg4audio_get_config(&m4ac, avctx->extradata,
  277. avctx->extradata_size * 8, 1);
  278. if (config_offset < 0)
  279. return AVERROR_INVALIDDATA;
  280. skip_bits_long(&gb, config_offset);
  281. if (get_bits_left(&gb) < (30 << 3))
  282. return AVERROR_INVALIDDATA;
  283. // read the fixed items
  284. als_id = get_bits_long(&gb, 32);
  285. avctx->sample_rate = m4ac.sample_rate;
  286. skip_bits_long(&gb, 32); // sample rate already known
  287. sconf->samples = get_bits_long(&gb, 32);
  288. avctx->channels = m4ac.channels;
  289. skip_bits(&gb, 16); // number of channels already known
  290. skip_bits(&gb, 3); // skip file_type
  291. sconf->resolution = get_bits(&gb, 3);
  292. sconf->floating = get_bits1(&gb);
  293. sconf->msb_first = get_bits1(&gb);
  294. sconf->frame_length = get_bits(&gb, 16) + 1;
  295. sconf->ra_distance = get_bits(&gb, 8);
  296. sconf->ra_flag = get_bits(&gb, 2);
  297. sconf->adapt_order = get_bits1(&gb);
  298. sconf->coef_table = get_bits(&gb, 2);
  299. sconf->long_term_prediction = get_bits1(&gb);
  300. sconf->max_order = get_bits(&gb, 10);
  301. sconf->block_switching = get_bits(&gb, 2);
  302. sconf->bgmc = get_bits1(&gb);
  303. sconf->sb_part = get_bits1(&gb);
  304. sconf->joint_stereo = get_bits1(&gb);
  305. sconf->mc_coding = get_bits1(&gb);
  306. sconf->chan_config = get_bits1(&gb);
  307. sconf->chan_sort = get_bits1(&gb);
  308. sconf->crc_enabled = get_bits1(&gb);
  309. sconf->rlslms = get_bits1(&gb);
  310. skip_bits(&gb, 5); // skip 5 reserved bits
  311. skip_bits1(&gb); // skip aux_data_enabled
  312. // check for ALSSpecificConfig struct
  313. if (als_id != MKBETAG('A','L','S','\0'))
  314. return AVERROR_INVALIDDATA;
  315. ctx->cur_frame_length = sconf->frame_length;
  316. // read channel config
  317. if (sconf->chan_config)
  318. sconf->chan_config_info = get_bits(&gb, 16);
  319. // TODO: use this to set avctx->channel_layout
  320. // read channel sorting
  321. if (sconf->chan_sort && avctx->channels > 1) {
  322. int chan_pos_bits = av_ceil_log2(avctx->channels);
  323. int bits_needed = avctx->channels * chan_pos_bits + 7;
  324. if (get_bits_left(&gb) < bits_needed)
  325. return AVERROR_INVALIDDATA;
  326. if (!(sconf->chan_pos = av_malloc_array(avctx->channels, sizeof(*sconf->chan_pos))))
  327. return AVERROR(ENOMEM);
  328. ctx->cs_switch = 1;
  329. for (i = 0; i < avctx->channels; i++) {
  330. sconf->chan_pos[i] = -1;
  331. }
  332. for (i = 0; i < avctx->channels; i++) {
  333. int idx;
  334. idx = get_bits(&gb, chan_pos_bits);
  335. if (idx >= avctx->channels || sconf->chan_pos[idx] != -1) {
  336. av_log(avctx, AV_LOG_WARNING, "Invalid channel reordering.\n");
  337. ctx->cs_switch = 0;
  338. break;
  339. }
  340. sconf->chan_pos[idx] = i;
  341. }
  342. align_get_bits(&gb);
  343. }
  344. // read fixed header and trailer sizes,
  345. // if size = 0xFFFFFFFF then there is no data field!
  346. if (get_bits_left(&gb) < 64)
  347. return AVERROR_INVALIDDATA;
  348. header_size = get_bits_long(&gb, 32);
  349. trailer_size = get_bits_long(&gb, 32);
  350. if (header_size == 0xFFFFFFFF)
  351. header_size = 0;
  352. if (trailer_size == 0xFFFFFFFF)
  353. trailer_size = 0;
  354. ht_size = ((int64_t)(header_size) + (int64_t)(trailer_size)) << 3;
  355. // skip the header and trailer data
  356. if (get_bits_left(&gb) < ht_size)
  357. return AVERROR_INVALIDDATA;
  358. if (ht_size > INT32_MAX)
  359. return AVERROR_PATCHWELCOME;
  360. skip_bits_long(&gb, ht_size);
  361. // initialize CRC calculation
  362. if (sconf->crc_enabled) {
  363. if (get_bits_left(&gb) < 32)
  364. return AVERROR_INVALIDDATA;
  365. if (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL)) {
  366. ctx->crc_table = av_crc_get_table(AV_CRC_32_IEEE_LE);
  367. ctx->crc = 0xFFFFFFFF;
  368. ctx->crc_org = ~get_bits_long(&gb, 32);
  369. } else
  370. skip_bits_long(&gb, 32);
  371. }
  372. // no need to read the rest of ALSSpecificConfig (ra_unit_size & aux data)
  373. dprint_specific_config(ctx);
  374. return 0;
  375. }
  376. /** Check the ALSSpecificConfig for unsupported features.
  377. */
  378. static int check_specific_config(ALSDecContext *ctx)
  379. {
  380. ALSSpecificConfig *sconf = &ctx->sconf;
  381. int error = 0;
  382. // report unsupported feature and set error value
  383. #define MISSING_ERR(cond, str, errval) \
  384. { \
  385. if (cond) { \
  386. avpriv_report_missing_feature(ctx->avctx, \
  387. str); \
  388. error = errval; \
  389. } \
  390. }
  391. MISSING_ERR(sconf->rlslms, "Adaptive RLS-LMS prediction", AVERROR_PATCHWELCOME);
  392. return error;
  393. }
  394. /** Parse the bs_info field to extract the block partitioning used in
  395. * block switching mode, refer to ISO/IEC 14496-3, section 11.6.2.
  396. */
  397. static void parse_bs_info(const uint32_t bs_info, unsigned int n,
  398. unsigned int div, unsigned int **div_blocks,
  399. unsigned int *num_blocks)
  400. {
  401. if (n < 31 && ((bs_info << n) & 0x40000000)) {
  402. // if the level is valid and the investigated bit n is set
  403. // then recursively check both children at bits (2n+1) and (2n+2)
  404. n *= 2;
  405. div += 1;
  406. parse_bs_info(bs_info, n + 1, div, div_blocks, num_blocks);
  407. parse_bs_info(bs_info, n + 2, div, div_blocks, num_blocks);
  408. } else {
  409. // else the bit is not set or the last level has been reached
  410. // (bit implicitly not set)
  411. **div_blocks = div;
  412. (*div_blocks)++;
  413. (*num_blocks)++;
  414. }
  415. }
  416. /** Read and decode a Rice codeword.
  417. */
  418. static int32_t decode_rice(GetBitContext *gb, unsigned int k)
  419. {
  420. int max = get_bits_left(gb) - k;
  421. unsigned q = get_unary(gb, 0, max);
  422. int r = k ? get_bits1(gb) : !(q & 1);
  423. if (k > 1) {
  424. q <<= (k - 1);
  425. q += get_bits_long(gb, k - 1);
  426. } else if (!k) {
  427. q >>= 1;
  428. }
  429. return r ? q : ~q;
  430. }
  431. /** Convert PARCOR coefficient k to direct filter coefficient.
  432. */
  433. static void parcor_to_lpc(unsigned int k, const int32_t *par, int32_t *cof)
  434. {
  435. int i, j;
  436. for (i = 0, j = k - 1; i < j; i++, j--) {
  437. int tmp1 = ((MUL64(par[k], cof[j]) + (1 << 19)) >> 20);
  438. cof[j] += ((MUL64(par[k], cof[i]) + (1 << 19)) >> 20);
  439. cof[i] += tmp1;
  440. }
  441. if (i == j)
  442. cof[i] += ((MUL64(par[k], cof[j]) + (1 << 19)) >> 20);
  443. cof[k] = par[k];
  444. }
  445. /** Read block switching field if necessary and set actual block sizes.
  446. * Also assure that the block sizes of the last frame correspond to the
  447. * actual number of samples.
  448. */
  449. static void get_block_sizes(ALSDecContext *ctx, unsigned int *div_blocks,
  450. uint32_t *bs_info)
  451. {
  452. ALSSpecificConfig *sconf = &ctx->sconf;
  453. GetBitContext *gb = &ctx->gb;
  454. unsigned int *ptr_div_blocks = div_blocks;
  455. unsigned int b;
  456. if (sconf->block_switching) {
  457. unsigned int bs_info_len = 1 << (sconf->block_switching + 2);
  458. *bs_info = get_bits_long(gb, bs_info_len);
  459. *bs_info <<= (32 - bs_info_len);
  460. }
  461. ctx->num_blocks = 0;
  462. parse_bs_info(*bs_info, 0, 0, &ptr_div_blocks, &ctx->num_blocks);
  463. // The last frame may have an overdetermined block structure given in
  464. // the bitstream. In that case the defined block structure would need
  465. // more samples than available to be consistent.
  466. // The block structure is actually used but the block sizes are adapted
  467. // to fit the actual number of available samples.
  468. // Example: 5 samples, 2nd level block sizes: 2 2 2 2.
  469. // This results in the actual block sizes: 2 2 1 0.
  470. // This is not specified in 14496-3 but actually done by the reference
  471. // codec RM22 revision 2.
  472. // This appears to happen in case of an odd number of samples in the last
  473. // frame which is actually not allowed by the block length switching part
  474. // of 14496-3.
  475. // The ALS conformance files feature an odd number of samples in the last
  476. // frame.
  477. for (b = 0; b < ctx->num_blocks; b++)
  478. div_blocks[b] = ctx->sconf.frame_length >> div_blocks[b];
  479. if (ctx->cur_frame_length != ctx->sconf.frame_length) {
  480. unsigned int remaining = ctx->cur_frame_length;
  481. for (b = 0; b < ctx->num_blocks; b++) {
  482. if (remaining <= div_blocks[b]) {
  483. div_blocks[b] = remaining;
  484. ctx->num_blocks = b + 1;
  485. break;
  486. }
  487. remaining -= div_blocks[b];
  488. }
  489. }
  490. }
  491. /** Read the block data for a constant block
  492. */
  493. static int read_const_block_data(ALSDecContext *ctx, ALSBlockData *bd)
  494. {
  495. ALSSpecificConfig *sconf = &ctx->sconf;
  496. AVCodecContext *avctx = ctx->avctx;
  497. GetBitContext *gb = &ctx->gb;
  498. if (bd->block_length <= 0)
  499. return AVERROR_INVALIDDATA;
  500. *bd->raw_samples = 0;
  501. *bd->const_block = get_bits1(gb); // 1 = constant value, 0 = zero block (silence)
  502. bd->js_blocks = get_bits1(gb);
  503. // skip 5 reserved bits
  504. skip_bits(gb, 5);
  505. if (*bd->const_block) {
  506. unsigned int const_val_bits = sconf->floating ? 24 : avctx->bits_per_raw_sample;
  507. *bd->raw_samples = get_sbits_long(gb, const_val_bits);
  508. }
  509. // ensure constant block decoding by reusing this field
  510. *bd->const_block = 1;
  511. return 0;
  512. }
  513. /** Decode the block data for a constant block
  514. */
  515. static void decode_const_block_data(ALSDecContext *ctx, ALSBlockData *bd)
  516. {
  517. int smp = bd->block_length - 1;
  518. int32_t val = *bd->raw_samples;
  519. int32_t *dst = bd->raw_samples + 1;
  520. // write raw samples into buffer
  521. for (; smp; smp--)
  522. *dst++ = val;
  523. }
  524. /** Read the block data for a non-constant block
  525. */
  526. static int read_var_block_data(ALSDecContext *ctx, ALSBlockData *bd)
  527. {
  528. ALSSpecificConfig *sconf = &ctx->sconf;
  529. AVCodecContext *avctx = ctx->avctx;
  530. GetBitContext *gb = &ctx->gb;
  531. unsigned int k;
  532. unsigned int s[8];
  533. unsigned int sx[8];
  534. unsigned int sub_blocks, log2_sub_blocks, sb_length;
  535. unsigned int start = 0;
  536. unsigned int opt_order;
  537. int sb;
  538. int32_t *quant_cof = bd->quant_cof;
  539. int32_t *current_res;
  540. // ensure variable block decoding by reusing this field
  541. *bd->const_block = 0;
  542. *bd->opt_order = 1;
  543. bd->js_blocks = get_bits1(gb);
  544. opt_order = *bd->opt_order;
  545. // determine the number of subblocks for entropy decoding
  546. if (!sconf->bgmc && !sconf->sb_part) {
  547. log2_sub_blocks = 0;
  548. } else {
  549. if (sconf->bgmc && sconf->sb_part)
  550. log2_sub_blocks = get_bits(gb, 2);
  551. else
  552. log2_sub_blocks = 2 * get_bits1(gb);
  553. }
  554. sub_blocks = 1 << log2_sub_blocks;
  555. // do not continue in case of a damaged stream since
  556. // block_length must be evenly divisible by sub_blocks
  557. if (bd->block_length & (sub_blocks - 1)) {
  558. av_log(avctx, AV_LOG_WARNING,
  559. "Block length is not evenly divisible by the number of subblocks.\n");
  560. return AVERROR_INVALIDDATA;
  561. }
  562. sb_length = bd->block_length >> log2_sub_blocks;
  563. if (sconf->bgmc) {
  564. s[0] = get_bits(gb, 8 + (sconf->resolution > 1));
  565. for (k = 1; k < sub_blocks; k++)
  566. s[k] = s[k - 1] + decode_rice(gb, 2);
  567. for (k = 0; k < sub_blocks; k++) {
  568. sx[k] = s[k] & 0x0F;
  569. s [k] >>= 4;
  570. }
  571. } else {
  572. s[0] = get_bits(gb, 4 + (sconf->resolution > 1));
  573. for (k = 1; k < sub_blocks; k++)
  574. s[k] = s[k - 1] + decode_rice(gb, 0);
  575. }
  576. for (k = 1; k < sub_blocks; k++)
  577. if (s[k] > 32) {
  578. av_log(avctx, AV_LOG_ERROR, "k invalid for rice code.\n");
  579. return AVERROR_INVALIDDATA;
  580. }
  581. if (get_bits1(gb))
  582. *bd->shift_lsbs = get_bits(gb, 4) + 1;
  583. *bd->store_prev_samples = (bd->js_blocks && bd->raw_other) || *bd->shift_lsbs;
  584. if (!sconf->rlslms) {
  585. if (sconf->adapt_order && sconf->max_order) {
  586. int opt_order_length = av_ceil_log2(av_clip((bd->block_length >> 3) - 1,
  587. 2, sconf->max_order + 1));
  588. *bd->opt_order = get_bits(gb, opt_order_length);
  589. if (*bd->opt_order > sconf->max_order) {
  590. *bd->opt_order = sconf->max_order;
  591. av_log(avctx, AV_LOG_ERROR, "Predictor order too large.\n");
  592. return AVERROR_INVALIDDATA;
  593. }
  594. } else {
  595. *bd->opt_order = sconf->max_order;
  596. }
  597. if (*bd->opt_order > bd->block_length) {
  598. *bd->opt_order = bd->block_length;
  599. av_log(avctx, AV_LOG_ERROR, "Predictor order too large.\n");
  600. return AVERROR_INVALIDDATA;
  601. }
  602. opt_order = *bd->opt_order;
  603. if (opt_order) {
  604. int add_base;
  605. if (sconf->coef_table == 3) {
  606. add_base = 0x7F;
  607. // read coefficient 0
  608. quant_cof[0] = 32 * parcor_scaled_values[get_bits(gb, 7)];
  609. // read coefficient 1
  610. if (opt_order > 1)
  611. quant_cof[1] = -32 * parcor_scaled_values[get_bits(gb, 7)];
  612. // read coefficients 2 to opt_order
  613. for (k = 2; k < opt_order; k++)
  614. quant_cof[k] = get_bits(gb, 7);
  615. } else {
  616. int k_max;
  617. add_base = 1;
  618. // read coefficient 0 to 19
  619. k_max = FFMIN(opt_order, 20);
  620. for (k = 0; k < k_max; k++) {
  621. int rice_param = parcor_rice_table[sconf->coef_table][k][1];
  622. int offset = parcor_rice_table[sconf->coef_table][k][0];
  623. quant_cof[k] = decode_rice(gb, rice_param) + offset;
  624. if (quant_cof[k] < -64 || quant_cof[k] > 63) {
  625. av_log(avctx, AV_LOG_ERROR,
  626. "quant_cof %"PRId32" is out of range.\n",
  627. quant_cof[k]);
  628. return AVERROR_INVALIDDATA;
  629. }
  630. }
  631. // read coefficients 20 to 126
  632. k_max = FFMIN(opt_order, 127);
  633. for (; k < k_max; k++)
  634. quant_cof[k] = decode_rice(gb, 2) + (k & 1);
  635. // read coefficients 127 to opt_order
  636. for (; k < opt_order; k++)
  637. quant_cof[k] = decode_rice(gb, 1);
  638. quant_cof[0] = 32 * parcor_scaled_values[quant_cof[0] + 64];
  639. if (opt_order > 1)
  640. quant_cof[1] = -32 * parcor_scaled_values[quant_cof[1] + 64];
  641. }
  642. for (k = 2; k < opt_order; k++)
  643. quant_cof[k] = (quant_cof[k] << 14) + (add_base << 13);
  644. }
  645. }
  646. // read LTP gain and lag values
  647. if (sconf->long_term_prediction) {
  648. *bd->use_ltp = get_bits1(gb);
  649. if (*bd->use_ltp) {
  650. int r, c;
  651. bd->ltp_gain[0] = decode_rice(gb, 1) * 8;
  652. bd->ltp_gain[1] = decode_rice(gb, 2) * 8;
  653. r = get_unary(gb, 0, 4);
  654. c = get_bits(gb, 2);
  655. if (r >= 4) {
  656. av_log(avctx, AV_LOG_ERROR, "r overflow\n");
  657. return AVERROR_INVALIDDATA;
  658. }
  659. bd->ltp_gain[2] = ltp_gain_values[r][c];
  660. bd->ltp_gain[3] = decode_rice(gb, 2) * 8;
  661. bd->ltp_gain[4] = decode_rice(gb, 1) * 8;
  662. *bd->ltp_lag = get_bits(gb, ctx->ltp_lag_length);
  663. *bd->ltp_lag += FFMAX(4, opt_order + 1);
  664. }
  665. }
  666. // read first value and residuals in case of a random access block
  667. if (bd->ra_block) {
  668. if (opt_order)
  669. bd->raw_samples[0] = decode_rice(gb, avctx->bits_per_raw_sample - 4);
  670. if (opt_order > 1)
  671. bd->raw_samples[1] = decode_rice(gb, FFMIN(s[0] + 3, ctx->s_max));
  672. if (opt_order > 2)
  673. bd->raw_samples[2] = decode_rice(gb, FFMIN(s[0] + 1, ctx->s_max));
  674. start = FFMIN(opt_order, 3);
  675. }
  676. // read all residuals
  677. if (sconf->bgmc) {
  678. int delta[8];
  679. unsigned int k [8];
  680. unsigned int b = av_clip((av_ceil_log2(bd->block_length) - 3) >> 1, 0, 5);
  681. // read most significant bits
  682. unsigned int high;
  683. unsigned int low;
  684. unsigned int value;
  685. ff_bgmc_decode_init(gb, &high, &low, &value);
  686. current_res = bd->raw_samples + start;
  687. for (sb = 0; sb < sub_blocks; sb++) {
  688. unsigned int sb_len = sb_length - (sb ? 0 : start);
  689. k [sb] = s[sb] > b ? s[sb] - b : 0;
  690. delta[sb] = 5 - s[sb] + k[sb];
  691. ff_bgmc_decode(gb, sb_len, current_res,
  692. delta[sb], sx[sb], &high, &low, &value, ctx->bgmc_lut, ctx->bgmc_lut_status);
  693. current_res += sb_len;
  694. }
  695. ff_bgmc_decode_end(gb);
  696. // read least significant bits and tails
  697. current_res = bd->raw_samples + start;
  698. for (sb = 0; sb < sub_blocks; sb++, start = 0) {
  699. unsigned int cur_tail_code = tail_code[sx[sb]][delta[sb]];
  700. unsigned int cur_k = k[sb];
  701. unsigned int cur_s = s[sb];
  702. for (; start < sb_length; start++) {
  703. int32_t res = *current_res;
  704. if (res == cur_tail_code) {
  705. unsigned int max_msb = (2 + (sx[sb] > 2) + (sx[sb] > 10))
  706. << (5 - delta[sb]);
  707. res = decode_rice(gb, cur_s);
  708. if (res >= 0) {
  709. res += (max_msb ) << cur_k;
  710. } else {
  711. res -= (max_msb - 1) << cur_k;
  712. }
  713. } else {
  714. if (res > cur_tail_code)
  715. res--;
  716. if (res & 1)
  717. res = -res;
  718. res >>= 1;
  719. if (cur_k) {
  720. res <<= cur_k;
  721. res |= get_bits_long(gb, cur_k);
  722. }
  723. }
  724. *current_res++ = res;
  725. }
  726. }
  727. } else {
  728. current_res = bd->raw_samples + start;
  729. for (sb = 0; sb < sub_blocks; sb++, start = 0)
  730. for (; start < sb_length; start++)
  731. *current_res++ = decode_rice(gb, s[sb]);
  732. }
  733. return 0;
  734. }
  735. /** Decode the block data for a non-constant block
  736. */
  737. static int decode_var_block_data(ALSDecContext *ctx, ALSBlockData *bd)
  738. {
  739. ALSSpecificConfig *sconf = &ctx->sconf;
  740. unsigned int block_length = bd->block_length;
  741. unsigned int smp = 0;
  742. unsigned int k;
  743. int opt_order = *bd->opt_order;
  744. int sb;
  745. int64_t y;
  746. int32_t *quant_cof = bd->quant_cof;
  747. int32_t *lpc_cof = bd->lpc_cof;
  748. int32_t *raw_samples = bd->raw_samples;
  749. int32_t *raw_samples_end = bd->raw_samples + bd->block_length;
  750. int32_t *lpc_cof_reversed = ctx->lpc_cof_reversed_buffer;
  751. // reverse long-term prediction
  752. if (*bd->use_ltp) {
  753. int ltp_smp;
  754. for (ltp_smp = FFMAX(*bd->ltp_lag - 2, 0); ltp_smp < block_length; ltp_smp++) {
  755. int center = ltp_smp - *bd->ltp_lag;
  756. int begin = FFMAX(0, center - 2);
  757. int end = center + 3;
  758. int tab = 5 - (end - begin);
  759. int base;
  760. y = 1 << 6;
  761. for (base = begin; base < end; base++, tab++)
  762. y += MUL64(bd->ltp_gain[tab], raw_samples[base]);
  763. raw_samples[ltp_smp] += y >> 7;
  764. }
  765. }
  766. // reconstruct all samples from residuals
  767. if (bd->ra_block) {
  768. for (smp = 0; smp < opt_order; smp++) {
  769. y = 1 << 19;
  770. for (sb = 0; sb < smp; sb++)
  771. y += MUL64(lpc_cof[sb], raw_samples[-(sb + 1)]);
  772. *raw_samples++ -= y >> 20;
  773. parcor_to_lpc(smp, quant_cof, lpc_cof);
  774. }
  775. } else {
  776. for (k = 0; k < opt_order; k++)
  777. parcor_to_lpc(k, quant_cof, lpc_cof);
  778. // store previous samples in case that they have to be altered
  779. if (*bd->store_prev_samples)
  780. memcpy(bd->prev_raw_samples, raw_samples - sconf->max_order,
  781. sizeof(*bd->prev_raw_samples) * sconf->max_order);
  782. // reconstruct difference signal for prediction (joint-stereo)
  783. if (bd->js_blocks && bd->raw_other) {
  784. int32_t *left, *right;
  785. if (bd->raw_other > raw_samples) { // D = R - L
  786. left = raw_samples;
  787. right = bd->raw_other;
  788. } else { // D = R - L
  789. left = bd->raw_other;
  790. right = raw_samples;
  791. }
  792. for (sb = -1; sb >= -sconf->max_order; sb--)
  793. raw_samples[sb] = right[sb] - left[sb];
  794. }
  795. // reconstruct shifted signal
  796. if (*bd->shift_lsbs)
  797. for (sb = -1; sb >= -sconf->max_order; sb--)
  798. raw_samples[sb] >>= *bd->shift_lsbs;
  799. }
  800. // reverse linear prediction coefficients for efficiency
  801. lpc_cof = lpc_cof + opt_order;
  802. for (sb = 0; sb < opt_order; sb++)
  803. lpc_cof_reversed[sb] = lpc_cof[-(sb + 1)];
  804. // reconstruct raw samples
  805. raw_samples = bd->raw_samples + smp;
  806. lpc_cof = lpc_cof_reversed + opt_order;
  807. for (; raw_samples < raw_samples_end; raw_samples++) {
  808. y = 1 << 19;
  809. for (sb = -opt_order; sb < 0; sb++)
  810. y += MUL64(lpc_cof[sb], raw_samples[sb]);
  811. *raw_samples -= y >> 20;
  812. }
  813. raw_samples = bd->raw_samples;
  814. // restore previous samples in case that they have been altered
  815. if (*bd->store_prev_samples)
  816. memcpy(raw_samples - sconf->max_order, bd->prev_raw_samples,
  817. sizeof(*raw_samples) * sconf->max_order);
  818. return 0;
  819. }
  820. /** Read the block data.
  821. */
  822. static int read_block(ALSDecContext *ctx, ALSBlockData *bd)
  823. {
  824. int ret;
  825. GetBitContext *gb = &ctx->gb;
  826. ALSSpecificConfig *sconf = &ctx->sconf;
  827. *bd->shift_lsbs = 0;
  828. // read block type flag and read the samples accordingly
  829. if (get_bits1(gb)) {
  830. ret = read_var_block_data(ctx, bd);
  831. } else {
  832. ret = read_const_block_data(ctx, bd);
  833. }
  834. if (!sconf->mc_coding || ctx->js_switch)
  835. align_get_bits(gb);
  836. return ret;
  837. }
  838. /** Decode the block data.
  839. */
  840. static int decode_block(ALSDecContext *ctx, ALSBlockData *bd)
  841. {
  842. unsigned int smp;
  843. int ret = 0;
  844. // read block type flag and read the samples accordingly
  845. if (*bd->const_block)
  846. decode_const_block_data(ctx, bd);
  847. else
  848. ret = decode_var_block_data(ctx, bd); // always return 0
  849. if (ret < 0)
  850. return ret;
  851. // TODO: read RLSLMS extension data
  852. if (*bd->shift_lsbs)
  853. for (smp = 0; smp < bd->block_length; smp++)
  854. bd->raw_samples[smp] = (unsigned)bd->raw_samples[smp] << *bd->shift_lsbs;
  855. return 0;
  856. }
  857. /** Read and decode block data successively.
  858. */
  859. static int read_decode_block(ALSDecContext *ctx, ALSBlockData *bd)
  860. {
  861. int ret;
  862. if ((ret = read_block(ctx, bd)) < 0)
  863. return ret;
  864. return decode_block(ctx, bd);
  865. }
  866. /** Compute the number of samples left to decode for the current frame and
  867. * sets these samples to zero.
  868. */
  869. static void zero_remaining(unsigned int b, unsigned int b_max,
  870. const unsigned int *div_blocks, int32_t *buf)
  871. {
  872. unsigned int count = 0;
  873. while (b < b_max)
  874. count += div_blocks[b++];
  875. if (count)
  876. memset(buf, 0, sizeof(*buf) * count);
  877. }
  878. /** Decode blocks independently.
  879. */
  880. static int decode_blocks_ind(ALSDecContext *ctx, unsigned int ra_frame,
  881. unsigned int c, const unsigned int *div_blocks,
  882. unsigned int *js_blocks)
  883. {
  884. int ret;
  885. unsigned int b;
  886. ALSBlockData bd = { 0 };
  887. bd.ra_block = ra_frame;
  888. bd.const_block = ctx->const_block;
  889. bd.shift_lsbs = ctx->shift_lsbs;
  890. bd.opt_order = ctx->opt_order;
  891. bd.store_prev_samples = ctx->store_prev_samples;
  892. bd.use_ltp = ctx->use_ltp;
  893. bd.ltp_lag = ctx->ltp_lag;
  894. bd.ltp_gain = ctx->ltp_gain[0];
  895. bd.quant_cof = ctx->quant_cof[0];
  896. bd.lpc_cof = ctx->lpc_cof[0];
  897. bd.prev_raw_samples = ctx->prev_raw_samples;
  898. bd.raw_samples = ctx->raw_samples[c];
  899. for (b = 0; b < ctx->num_blocks; b++) {
  900. bd.block_length = div_blocks[b];
  901. if ((ret = read_decode_block(ctx, &bd)) < 0) {
  902. // damaged block, write zero for the rest of the frame
  903. zero_remaining(b, ctx->num_blocks, div_blocks, bd.raw_samples);
  904. return ret;
  905. }
  906. bd.raw_samples += div_blocks[b];
  907. bd.ra_block = 0;
  908. }
  909. return 0;
  910. }
  911. /** Decode blocks dependently.
  912. */
  913. static int decode_blocks(ALSDecContext *ctx, unsigned int ra_frame,
  914. unsigned int c, const unsigned int *div_blocks,
  915. unsigned int *js_blocks)
  916. {
  917. ALSSpecificConfig *sconf = &ctx->sconf;
  918. unsigned int offset = 0;
  919. unsigned int b;
  920. int ret;
  921. ALSBlockData bd[2] = { { 0 } };
  922. bd[0].ra_block = ra_frame;
  923. bd[0].const_block = ctx->const_block;
  924. bd[0].shift_lsbs = ctx->shift_lsbs;
  925. bd[0].opt_order = ctx->opt_order;
  926. bd[0].store_prev_samples = ctx->store_prev_samples;
  927. bd[0].use_ltp = ctx->use_ltp;
  928. bd[0].ltp_lag = ctx->ltp_lag;
  929. bd[0].ltp_gain = ctx->ltp_gain[0];
  930. bd[0].quant_cof = ctx->quant_cof[0];
  931. bd[0].lpc_cof = ctx->lpc_cof[0];
  932. bd[0].prev_raw_samples = ctx->prev_raw_samples;
  933. bd[0].js_blocks = *js_blocks;
  934. bd[1].ra_block = ra_frame;
  935. bd[1].const_block = ctx->const_block;
  936. bd[1].shift_lsbs = ctx->shift_lsbs;
  937. bd[1].opt_order = ctx->opt_order;
  938. bd[1].store_prev_samples = ctx->store_prev_samples;
  939. bd[1].use_ltp = ctx->use_ltp;
  940. bd[1].ltp_lag = ctx->ltp_lag;
  941. bd[1].ltp_gain = ctx->ltp_gain[0];
  942. bd[1].quant_cof = ctx->quant_cof[0];
  943. bd[1].lpc_cof = ctx->lpc_cof[0];
  944. bd[1].prev_raw_samples = ctx->prev_raw_samples;
  945. bd[1].js_blocks = *(js_blocks + 1);
  946. // decode all blocks
  947. for (b = 0; b < ctx->num_blocks; b++) {
  948. unsigned int s;
  949. bd[0].block_length = div_blocks[b];
  950. bd[1].block_length = div_blocks[b];
  951. bd[0].raw_samples = ctx->raw_samples[c ] + offset;
  952. bd[1].raw_samples = ctx->raw_samples[c + 1] + offset;
  953. bd[0].raw_other = bd[1].raw_samples;
  954. bd[1].raw_other = bd[0].raw_samples;
  955. if ((ret = read_decode_block(ctx, &bd[0])) < 0 ||
  956. (ret = read_decode_block(ctx, &bd[1])) < 0)
  957. goto fail;
  958. // reconstruct joint-stereo blocks
  959. if (bd[0].js_blocks) {
  960. if (bd[1].js_blocks)
  961. av_log(ctx->avctx, AV_LOG_WARNING, "Invalid channel pair.\n");
  962. for (s = 0; s < div_blocks[b]; s++)
  963. bd[0].raw_samples[s] = bd[1].raw_samples[s] - bd[0].raw_samples[s];
  964. } else if (bd[1].js_blocks) {
  965. for (s = 0; s < div_blocks[b]; s++)
  966. bd[1].raw_samples[s] = bd[1].raw_samples[s] + bd[0].raw_samples[s];
  967. }
  968. offset += div_blocks[b];
  969. bd[0].ra_block = 0;
  970. bd[1].ra_block = 0;
  971. }
  972. // store carryover raw samples,
  973. // the others channel raw samples are stored by the calling function.
  974. memmove(ctx->raw_samples[c] - sconf->max_order,
  975. ctx->raw_samples[c] - sconf->max_order + sconf->frame_length,
  976. sizeof(*ctx->raw_samples[c]) * sconf->max_order);
  977. return 0;
  978. fail:
  979. // damaged block, write zero for the rest of the frame
  980. zero_remaining(b, ctx->num_blocks, div_blocks, bd[0].raw_samples);
  981. zero_remaining(b, ctx->num_blocks, div_blocks, bd[1].raw_samples);
  982. return ret;
  983. }
  984. static inline int als_weighting(GetBitContext *gb, int k, int off)
  985. {
  986. int idx = av_clip(decode_rice(gb, k) + off,
  987. 0, FF_ARRAY_ELEMS(mcc_weightings) - 1);
  988. return mcc_weightings[idx];
  989. }
  990. /** Read the channel data.
  991. */
  992. static int read_channel_data(ALSDecContext *ctx, ALSChannelData *cd, int c)
  993. {
  994. GetBitContext *gb = &ctx->gb;
  995. ALSChannelData *current = cd;
  996. unsigned int channels = ctx->avctx->channels;
  997. int entries = 0;
  998. while (entries < channels && !(current->stop_flag = get_bits1(gb))) {
  999. current->master_channel = get_bits_long(gb, av_ceil_log2(channels));
  1000. if (current->master_channel >= channels) {
  1001. av_log(ctx->avctx, AV_LOG_ERROR, "Invalid master channel.\n");
  1002. return AVERROR_INVALIDDATA;
  1003. }
  1004. if (current->master_channel != c) {
  1005. current->time_diff_flag = get_bits1(gb);
  1006. current->weighting[0] = als_weighting(gb, 1, 16);
  1007. current->weighting[1] = als_weighting(gb, 2, 14);
  1008. current->weighting[2] = als_weighting(gb, 1, 16);
  1009. if (current->time_diff_flag) {
  1010. current->weighting[3] = als_weighting(gb, 1, 16);
  1011. current->weighting[4] = als_weighting(gb, 1, 16);
  1012. current->weighting[5] = als_weighting(gb, 1, 16);
  1013. current->time_diff_sign = get_bits1(gb);
  1014. current->time_diff_index = get_bits(gb, ctx->ltp_lag_length - 3) + 3;
  1015. }
  1016. }
  1017. current++;
  1018. entries++;
  1019. }
  1020. if (entries == channels) {
  1021. av_log(ctx->avctx, AV_LOG_ERROR, "Damaged channel data.\n");
  1022. return AVERROR_INVALIDDATA;
  1023. }
  1024. align_get_bits(gb);
  1025. return 0;
  1026. }
  1027. /** Recursively reverts the inter-channel correlation for a block.
  1028. */
  1029. static int revert_channel_correlation(ALSDecContext *ctx, ALSBlockData *bd,
  1030. ALSChannelData **cd, int *reverted,
  1031. unsigned int offset, int c)
  1032. {
  1033. ALSChannelData *ch = cd[c];
  1034. unsigned int dep = 0;
  1035. unsigned int channels = ctx->avctx->channels;
  1036. unsigned int channel_size = ctx->sconf.frame_length + ctx->sconf.max_order;
  1037. if (reverted[c])
  1038. return 0;
  1039. reverted[c] = 1;
  1040. while (dep < channels && !ch[dep].stop_flag) {
  1041. revert_channel_correlation(ctx, bd, cd, reverted, offset,
  1042. ch[dep].master_channel);
  1043. dep++;
  1044. }
  1045. if (dep == channels) {
  1046. av_log(ctx->avctx, AV_LOG_WARNING, "Invalid channel correlation.\n");
  1047. return AVERROR_INVALIDDATA;
  1048. }
  1049. bd->const_block = ctx->const_block + c;
  1050. bd->shift_lsbs = ctx->shift_lsbs + c;
  1051. bd->opt_order = ctx->opt_order + c;
  1052. bd->store_prev_samples = ctx->store_prev_samples + c;
  1053. bd->use_ltp = ctx->use_ltp + c;
  1054. bd->ltp_lag = ctx->ltp_lag + c;
  1055. bd->ltp_gain = ctx->ltp_gain[c];
  1056. bd->lpc_cof = ctx->lpc_cof[c];
  1057. bd->quant_cof = ctx->quant_cof[c];
  1058. bd->raw_samples = ctx->raw_samples[c] + offset;
  1059. for (dep = 0; !ch[dep].stop_flag; dep++) {
  1060. ptrdiff_t smp;
  1061. ptrdiff_t begin = 1;
  1062. ptrdiff_t end = bd->block_length - 1;
  1063. int64_t y;
  1064. int32_t *master = ctx->raw_samples[ch[dep].master_channel] + offset;
  1065. if (ch[dep].master_channel == c)
  1066. continue;
  1067. if (ch[dep].time_diff_flag) {
  1068. int t = ch[dep].time_diff_index;
  1069. if (ch[dep].time_diff_sign) {
  1070. t = -t;
  1071. if (begin < t) {
  1072. av_log(ctx->avctx, AV_LOG_ERROR, "begin %"PTRDIFF_SPECIFIER" smaller than time diff index %d.\n", begin, t);
  1073. return AVERROR_INVALIDDATA;
  1074. }
  1075. begin -= t;
  1076. } else {
  1077. if (end < t) {
  1078. av_log(ctx->avctx, AV_LOG_ERROR, "end %"PTRDIFF_SPECIFIER" smaller than time diff index %d.\n", end, t);
  1079. return AVERROR_INVALIDDATA;
  1080. }
  1081. end -= t;
  1082. }
  1083. if (FFMIN(begin - 1, begin - 1 + t) < ctx->raw_buffer - master ||
  1084. FFMAX(end + 1, end + 1 + t) > ctx->raw_buffer + channels * channel_size - master) {
  1085. av_log(ctx->avctx, AV_LOG_ERROR,
  1086. "sample pointer range [%p, %p] not contained in raw_buffer [%p, %p].\n",
  1087. master + FFMIN(begin - 1, begin - 1 + t), master + FFMAX(end + 1, end + 1 + t),
  1088. ctx->raw_buffer, ctx->raw_buffer + channels * channel_size);
  1089. return AVERROR_INVALIDDATA;
  1090. }
  1091. for (smp = begin; smp < end; smp++) {
  1092. y = (1 << 6) +
  1093. MUL64(ch[dep].weighting[0], master[smp - 1 ]) +
  1094. MUL64(ch[dep].weighting[1], master[smp ]) +
  1095. MUL64(ch[dep].weighting[2], master[smp + 1 ]) +
  1096. MUL64(ch[dep].weighting[3], master[smp - 1 + t]) +
  1097. MUL64(ch[dep].weighting[4], master[smp + t]) +
  1098. MUL64(ch[dep].weighting[5], master[smp + 1 + t]);
  1099. bd->raw_samples[smp] += y >> 7;
  1100. }
  1101. } else {
  1102. if (begin - 1 < ctx->raw_buffer - master ||
  1103. end + 1 > ctx->raw_buffer + channels * channel_size - master) {
  1104. av_log(ctx->avctx, AV_LOG_ERROR,
  1105. "sample pointer range [%p, %p] not contained in raw_buffer [%p, %p].\n",
  1106. master + begin - 1, master + end + 1,
  1107. ctx->raw_buffer, ctx->raw_buffer + channels * channel_size);
  1108. return AVERROR_INVALIDDATA;
  1109. }
  1110. for (smp = begin; smp < end; smp++) {
  1111. y = (1 << 6) +
  1112. MUL64(ch[dep].weighting[0], master[smp - 1]) +
  1113. MUL64(ch[dep].weighting[1], master[smp ]) +
  1114. MUL64(ch[dep].weighting[2], master[smp + 1]);
  1115. bd->raw_samples[smp] += y >> 7;
  1116. }
  1117. }
  1118. }
  1119. return 0;
  1120. }
  1121. /** multiply two softfloats and handle the rounding off
  1122. */
  1123. static SoftFloat_IEEE754 multiply(SoftFloat_IEEE754 a, SoftFloat_IEEE754 b) {
  1124. uint64_t mantissa_temp;
  1125. uint64_t mask_64;
  1126. int cutoff_bit_count;
  1127. unsigned char last_2_bits;
  1128. unsigned int mantissa;
  1129. int32_t sign;
  1130. uint32_t return_val = 0;
  1131. int bit_count = 48;
  1132. sign = a.sign ^ b.sign;
  1133. // Multiply mantissa bits in a 64-bit register
  1134. mantissa_temp = (uint64_t)a.mant * (uint64_t)b.mant;
  1135. mask_64 = (uint64_t)0x1 << 47;
  1136. if (!mantissa_temp)
  1137. return FLOAT_0;
  1138. // Count the valid bit count
  1139. while (!(mantissa_temp & mask_64) && mask_64) {
  1140. bit_count--;
  1141. mask_64 >>= 1;
  1142. }
  1143. // Round off
  1144. cutoff_bit_count = bit_count - 24;
  1145. if (cutoff_bit_count > 0) {
  1146. last_2_bits = (unsigned char)(((unsigned int)mantissa_temp >> (cutoff_bit_count - 1)) & 0x3 );
  1147. if ((last_2_bits == 0x3) || ((last_2_bits == 0x1) && ((unsigned int)mantissa_temp & ((0x1UL << (cutoff_bit_count - 1)) - 1)))) {
  1148. // Need to round up
  1149. mantissa_temp += (uint64_t)0x1 << cutoff_bit_count;
  1150. }
  1151. }
  1152. mantissa = (unsigned int)(mantissa_temp >> cutoff_bit_count);
  1153. // Need one more shift?
  1154. if (mantissa & 0x01000000ul) {
  1155. bit_count++;
  1156. mantissa >>= 1;
  1157. }
  1158. if (!sign) {
  1159. return_val = 0x80000000U;
  1160. }
  1161. return_val |= (a.exp + b.exp + bit_count - 47) << 23;
  1162. return_val |= mantissa;
  1163. return av_bits2sf_ieee754(return_val);
  1164. }
  1165. /** Read and decode the floating point sample data
  1166. */
  1167. static int read_diff_float_data(ALSDecContext *ctx, unsigned int ra_frame) {
  1168. AVCodecContext *avctx = ctx->avctx;
  1169. GetBitContext *gb = &ctx->gb;
  1170. SoftFloat_IEEE754 *acf = ctx->acf;
  1171. int *shift_value = ctx->shift_value;
  1172. int *last_shift_value = ctx->last_shift_value;
  1173. int *last_acf_mantissa = ctx->last_acf_mantissa;
  1174. int **raw_mantissa = ctx->raw_mantissa;
  1175. int *nbits = ctx->nbits;
  1176. unsigned char *larray = ctx->larray;
  1177. int frame_length = ctx->cur_frame_length;
  1178. SoftFloat_IEEE754 scale = av_int2sf_ieee754(0x1u, 23);
  1179. unsigned int partA_flag;
  1180. unsigned int highest_byte;
  1181. unsigned int shift_amp;
  1182. uint32_t tmp_32;
  1183. int use_acf;
  1184. int nchars;
  1185. int i;
  1186. int c;
  1187. long k;
  1188. long nbits_aligned;
  1189. unsigned long acc;
  1190. unsigned long j;
  1191. uint32_t sign;
  1192. uint32_t e;
  1193. uint32_t mantissa;
  1194. skip_bits_long(gb, 32); //num_bytes_diff_float
  1195. use_acf = get_bits1(gb);
  1196. if (ra_frame) {
  1197. memset(last_acf_mantissa, 0, avctx->channels * sizeof(*last_acf_mantissa));
  1198. memset(last_shift_value, 0, avctx->channels * sizeof(*last_shift_value) );
  1199. ff_mlz_flush_dict(ctx->mlz);
  1200. }
  1201. for (c = 0; c < avctx->channels; ++c) {
  1202. if (use_acf) {
  1203. //acf_flag
  1204. if (get_bits1(gb)) {
  1205. tmp_32 = get_bits(gb, 23);
  1206. last_acf_mantissa[c] = tmp_32;
  1207. } else {
  1208. tmp_32 = last_acf_mantissa[c];
  1209. }
  1210. acf[c] = av_bits2sf_ieee754(tmp_32);
  1211. } else {
  1212. acf[c] = FLOAT_1;
  1213. }
  1214. highest_byte = get_bits(gb, 2);
  1215. partA_flag = get_bits1(gb);
  1216. shift_amp = get_bits1(gb);
  1217. if (shift_amp) {
  1218. shift_value[c] = get_bits(gb, 8);
  1219. last_shift_value[c] = shift_value[c];
  1220. } else {
  1221. shift_value[c] = last_shift_value[c];
  1222. }
  1223. if (partA_flag) {
  1224. if (!get_bits1(gb)) { //uncompressed
  1225. for (i = 0; i < frame_length; ++i) {
  1226. if (ctx->raw_samples[c][i] == 0) {
  1227. ctx->raw_mantissa[c][i] = get_bits_long(gb, 32);
  1228. }
  1229. }
  1230. } else { //compressed
  1231. nchars = 0;
  1232. for (i = 0; i < frame_length; ++i) {
  1233. if (ctx->raw_samples[c][i] == 0) {
  1234. nchars += 4;
  1235. }
  1236. }
  1237. tmp_32 = ff_mlz_decompression(ctx->mlz, gb, nchars, larray);
  1238. if(tmp_32 != nchars) {
  1239. av_log(ctx->avctx, AV_LOG_ERROR, "Error in MLZ decompression (%d, %d).\n", tmp_32, nchars);
  1240. return AVERROR_INVALIDDATA;
  1241. }
  1242. for (i = 0; i < frame_length; ++i) {
  1243. ctx->raw_mantissa[c][i] = AV_RB32(larray);
  1244. }
  1245. }
  1246. }
  1247. //decode part B
  1248. if (highest_byte) {
  1249. for (i = 0; i < frame_length; ++i) {
  1250. if (ctx->raw_samples[c][i] != 0) {
  1251. //The following logic is taken from Tabel 14.45 and 14.46 from the ISO spec
  1252. if (av_cmp_sf_ieee754(acf[c], FLOAT_1)) {
  1253. nbits[i] = 23 - av_log2(abs(ctx->raw_samples[c][i]));
  1254. } else {
  1255. nbits[i] = 23;
  1256. }
  1257. nbits[i] = FFMIN(nbits[i], highest_byte*8);
  1258. }
  1259. }
  1260. if (!get_bits1(gb)) { //uncompressed
  1261. for (i = 0; i < frame_length; ++i) {
  1262. if (ctx->raw_samples[c][i] != 0) {
  1263. raw_mantissa[c][i] = get_bitsz(gb, nbits[i]);
  1264. }
  1265. }
  1266. } else { //compressed
  1267. nchars = 0;
  1268. for (i = 0; i < frame_length; ++i) {
  1269. if (ctx->raw_samples[c][i]) {
  1270. nchars += (int) nbits[i] / 8;
  1271. if (nbits[i] & 7) {
  1272. ++nchars;
  1273. }
  1274. }
  1275. }
  1276. tmp_32 = ff_mlz_decompression(ctx->mlz, gb, nchars, larray);
  1277. if(tmp_32 != nchars) {
  1278. av_log(ctx->avctx, AV_LOG_ERROR, "Error in MLZ decompression (%d, %d).\n", tmp_32, nchars);
  1279. return AVERROR_INVALIDDATA;
  1280. }
  1281. j = 0;
  1282. for (i = 0; i < frame_length; ++i) {
  1283. if (ctx->raw_samples[c][i]) {
  1284. if (nbits[i] & 7) {
  1285. nbits_aligned = 8 * ((unsigned int)(nbits[i] / 8) + 1);
  1286. } else {
  1287. nbits_aligned = nbits[i];
  1288. }
  1289. acc = 0;
  1290. for (k = 0; k < nbits_aligned/8; ++k) {
  1291. acc = (acc << 8) + larray[j++];
  1292. }
  1293. acc >>= (nbits_aligned - nbits[i]);
  1294. raw_mantissa[c][i] = acc;
  1295. }
  1296. }
  1297. }
  1298. }
  1299. for (i = 0; i < frame_length; ++i) {
  1300. SoftFloat_IEEE754 pcm_sf = av_int2sf_ieee754(ctx->raw_samples[c][i], 0);
  1301. pcm_sf = av_div_sf_ieee754(pcm_sf, scale);
  1302. if (ctx->raw_samples[c][i] != 0) {
  1303. if (!av_cmp_sf_ieee754(acf[c], FLOAT_1)) {
  1304. pcm_sf = multiply(acf[c], pcm_sf);
  1305. }
  1306. sign = pcm_sf.sign;
  1307. e = pcm_sf.exp;
  1308. mantissa = (pcm_sf.mant | 0x800000) + raw_mantissa[c][i];
  1309. while(mantissa >= 0x1000000) {
  1310. e++;
  1311. mantissa >>= 1;
  1312. }
  1313. if (mantissa) e += (shift_value[c] - 127);
  1314. mantissa &= 0x007fffffUL;
  1315. tmp_32 = (sign << 31) | ((e + EXP_BIAS) << 23) | (mantissa);
  1316. ctx->raw_samples[c][i] = tmp_32;
  1317. } else {
  1318. ctx->raw_samples[c][i] = raw_mantissa[c][i] & 0x007fffffUL;
  1319. }
  1320. }
  1321. align_get_bits(gb);
  1322. }
  1323. return 0;
  1324. }
  1325. /** Read the frame data.
  1326. */
  1327. static int read_frame_data(ALSDecContext *ctx, unsigned int ra_frame)
  1328. {
  1329. ALSSpecificConfig *sconf = &ctx->sconf;
  1330. AVCodecContext *avctx = ctx->avctx;
  1331. GetBitContext *gb = &ctx->gb;
  1332. unsigned int div_blocks[32]; ///< block sizes.
  1333. unsigned int c;
  1334. unsigned int js_blocks[2];
  1335. uint32_t bs_info = 0;
  1336. int ret;
  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);
  1340. if (sconf->mc_coding && sconf->joint_stereo) {
  1341. ctx->js_switch = get_bits1(gb);
  1342. align_get_bits(gb);
  1343. }
  1344. if (!sconf->mc_coding || ctx->js_switch) {
  1345. int independent_bs = !sconf->joint_stereo;
  1346. for (c = 0; c < avctx->channels; c++) {
  1347. js_blocks[0] = 0;
  1348. js_blocks[1] = 0;
  1349. get_block_sizes(ctx, div_blocks, &bs_info);
  1350. // if joint_stereo and block_switching is set, independent decoding
  1351. // is signaled via the first bit of bs_info
  1352. if (sconf->joint_stereo && sconf->block_switching)
  1353. if (bs_info >> 31)
  1354. independent_bs = 2;
  1355. // if this is the last channel, it has to be decoded independently
  1356. if (c == avctx->channels - 1)
  1357. independent_bs = 1;
  1358. if (independent_bs) {
  1359. ret = decode_blocks_ind(ctx, ra_frame, c,
  1360. div_blocks, js_blocks);
  1361. if (ret < 0)
  1362. return ret;
  1363. independent_bs--;
  1364. } else {
  1365. ret = decode_blocks(ctx, ra_frame, c, div_blocks, js_blocks);
  1366. if (ret < 0)
  1367. return ret;
  1368. c++;
  1369. }
  1370. // store carryover raw samples
  1371. memmove(ctx->raw_samples[c] - sconf->max_order,
  1372. ctx->raw_samples[c] - sconf->max_order + sconf->frame_length,
  1373. sizeof(*ctx->raw_samples[c]) * sconf->max_order);
  1374. }
  1375. } else { // multi-channel coding
  1376. ALSBlockData bd = { 0 };
  1377. int b, ret;
  1378. int *reverted_channels = ctx->reverted_channels;
  1379. unsigned int offset = 0;
  1380. for (c = 0; c < avctx->channels; c++)
  1381. if (ctx->chan_data[c] < ctx->chan_data_buffer) {
  1382. av_log(ctx->avctx, AV_LOG_ERROR, "Invalid channel data.\n");
  1383. return AVERROR_INVALIDDATA;
  1384. }
  1385. memset(reverted_channels, 0, sizeof(*reverted_channels) * avctx->channels);
  1386. bd.ra_block = ra_frame;
  1387. bd.prev_raw_samples = ctx->prev_raw_samples;
  1388. get_block_sizes(ctx, div_blocks, &bs_info);
  1389. for (b = 0; b < ctx->num_blocks; b++) {
  1390. bd.block_length = div_blocks[b];
  1391. if (bd.block_length <= 0) {
  1392. av_log(ctx->avctx, AV_LOG_WARNING,
  1393. "Invalid block length %u in channel data!\n",
  1394. bd.block_length);
  1395. continue;
  1396. }
  1397. for (c = 0; c < avctx->channels; c++) {
  1398. bd.const_block = ctx->const_block + c;
  1399. bd.shift_lsbs = ctx->shift_lsbs + c;
  1400. bd.opt_order = ctx->opt_order + c;
  1401. bd.store_prev_samples = ctx->store_prev_samples + c;
  1402. bd.use_ltp = ctx->use_ltp + c;
  1403. bd.ltp_lag = ctx->ltp_lag + c;
  1404. bd.ltp_gain = ctx->ltp_gain[c];
  1405. bd.lpc_cof = ctx->lpc_cof[c];
  1406. bd.quant_cof = ctx->quant_cof[c];
  1407. bd.raw_samples = ctx->raw_samples[c] + offset;
  1408. bd.raw_other = NULL;
  1409. if ((ret = read_block(ctx, &bd)) < 0)
  1410. return ret;
  1411. if ((ret = read_channel_data(ctx, ctx->chan_data[c], c)) < 0)
  1412. return ret;
  1413. }
  1414. for (c = 0; c < avctx->channels; c++) {
  1415. ret = revert_channel_correlation(ctx, &bd, ctx->chan_data,
  1416. reverted_channels, offset, c);
  1417. if (ret < 0)
  1418. return ret;
  1419. }
  1420. for (c = 0; c < avctx->channels; c++) {
  1421. bd.const_block = ctx->const_block + c;
  1422. bd.shift_lsbs = ctx->shift_lsbs + c;
  1423. bd.opt_order = ctx->opt_order + c;
  1424. bd.store_prev_samples = ctx->store_prev_samples + c;
  1425. bd.use_ltp = ctx->use_ltp + c;
  1426. bd.ltp_lag = ctx->ltp_lag + c;
  1427. bd.ltp_gain = ctx->ltp_gain[c];
  1428. bd.lpc_cof = ctx->lpc_cof[c];
  1429. bd.quant_cof = ctx->quant_cof[c];
  1430. bd.raw_samples = ctx->raw_samples[c] + offset;
  1431. if ((ret = decode_block(ctx, &bd)) < 0)
  1432. return ret;
  1433. }
  1434. memset(reverted_channels, 0, avctx->channels * sizeof(*reverted_channels));
  1435. offset += div_blocks[b];
  1436. bd.ra_block = 0;
  1437. }
  1438. // store carryover raw samples
  1439. for (c = 0; c < avctx->channels; c++)
  1440. memmove(ctx->raw_samples[c] - sconf->max_order,
  1441. ctx->raw_samples[c] - sconf->max_order + sconf->frame_length,
  1442. sizeof(*ctx->raw_samples[c]) * sconf->max_order);
  1443. }
  1444. if (sconf->floating) {
  1445. read_diff_float_data(ctx, ra_frame);
  1446. }
  1447. if (get_bits_left(gb) < 0) {
  1448. av_log(ctx->avctx, AV_LOG_ERROR, "Overread %d\n", -get_bits_left(gb));
  1449. return AVERROR_INVALIDDATA;
  1450. }
  1451. return 0;
  1452. }
  1453. /** Decode an ALS frame.
  1454. */
  1455. static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame_ptr,
  1456. AVPacket *avpkt)
  1457. {
  1458. ALSDecContext *ctx = avctx->priv_data;
  1459. AVFrame *frame = data;
  1460. ALSSpecificConfig *sconf = &ctx->sconf;
  1461. const uint8_t *buffer = avpkt->data;
  1462. int buffer_size = avpkt->size;
  1463. int invalid_frame, ret;
  1464. unsigned int c, sample, ra_frame, bytes_read, shift;
  1465. if ((ret = init_get_bits8(&ctx->gb, buffer, buffer_size)) < 0)
  1466. return ret;
  1467. // In the case that the distance between random access frames is set to zero
  1468. // (sconf->ra_distance == 0) no frame is treated as a random access frame.
  1469. // For the first frame, if prediction is used, all samples used from the
  1470. // previous frame are assumed to be zero.
  1471. ra_frame = sconf->ra_distance && !(ctx->frame_id % sconf->ra_distance);
  1472. // the last frame to decode might have a different length
  1473. if (sconf->samples != 0xFFFFFFFF)
  1474. ctx->cur_frame_length = FFMIN(sconf->samples - ctx->frame_id * (uint64_t) sconf->frame_length,
  1475. sconf->frame_length);
  1476. else
  1477. ctx->cur_frame_length = sconf->frame_length;
  1478. // decode the frame data
  1479. if ((invalid_frame = read_frame_data(ctx, ra_frame)) < 0)
  1480. av_log(ctx->avctx, AV_LOG_WARNING,
  1481. "Reading frame data failed. Skipping RA unit.\n");
  1482. ctx->frame_id++;
  1483. /* get output buffer */
  1484. frame->nb_samples = ctx->cur_frame_length;
  1485. if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
  1486. return ret;
  1487. // transform decoded frame into output format
  1488. #define INTERLEAVE_OUTPUT(bps) \
  1489. { \
  1490. int##bps##_t *dest = (int##bps##_t*)frame->data[0]; \
  1491. shift = bps - ctx->avctx->bits_per_raw_sample; \
  1492. if (!ctx->cs_switch) { \
  1493. for (sample = 0; sample < ctx->cur_frame_length; sample++) \
  1494. for (c = 0; c < avctx->channels; c++) \
  1495. *dest++ = ctx->raw_samples[c][sample] * (1U << shift); \
  1496. } else { \
  1497. for (sample = 0; sample < ctx->cur_frame_length; sample++) \
  1498. for (c = 0; c < avctx->channels; c++) \
  1499. *dest++ = ctx->raw_samples[sconf->chan_pos[c]][sample] * (1U << shift); \
  1500. } \
  1501. }
  1502. if (ctx->avctx->bits_per_raw_sample <= 16) {
  1503. INTERLEAVE_OUTPUT(16)
  1504. } else {
  1505. INTERLEAVE_OUTPUT(32)
  1506. }
  1507. // update CRC
  1508. if (sconf->crc_enabled && (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL))) {
  1509. int swap = HAVE_BIGENDIAN != sconf->msb_first;
  1510. if (ctx->avctx->bits_per_raw_sample == 24) {
  1511. int32_t *src = (int32_t *)frame->data[0];
  1512. for (sample = 0;
  1513. sample < ctx->cur_frame_length * avctx->channels;
  1514. sample++) {
  1515. int32_t v;
  1516. if (swap)
  1517. v = av_bswap32(src[sample]);
  1518. else
  1519. v = src[sample];
  1520. if (!HAVE_BIGENDIAN)
  1521. v >>= 8;
  1522. ctx->crc = av_crc(ctx->crc_table, ctx->crc, (uint8_t*)(&v), 3);
  1523. }
  1524. } else {
  1525. uint8_t *crc_source;
  1526. if (swap) {
  1527. if (ctx->avctx->bits_per_raw_sample <= 16) {
  1528. int16_t *src = (int16_t*) frame->data[0];
  1529. int16_t *dest = (int16_t*) ctx->crc_buffer;
  1530. for (sample = 0;
  1531. sample < ctx->cur_frame_length * avctx->channels;
  1532. sample++)
  1533. *dest++ = av_bswap16(src[sample]);
  1534. } else {
  1535. ctx->bdsp.bswap_buf((uint32_t *) ctx->crc_buffer,
  1536. (uint32_t *) frame->data[0],
  1537. ctx->cur_frame_length * avctx->channels);
  1538. }
  1539. crc_source = ctx->crc_buffer;
  1540. } else {
  1541. crc_source = frame->data[0];
  1542. }
  1543. ctx->crc = av_crc(ctx->crc_table, ctx->crc, crc_source,
  1544. ctx->cur_frame_length * avctx->channels *
  1545. av_get_bytes_per_sample(avctx->sample_fmt));
  1546. }
  1547. // check CRC sums if this is the last frame
  1548. if (ctx->cur_frame_length != sconf->frame_length &&
  1549. ctx->crc_org != ctx->crc) {
  1550. av_log(avctx, AV_LOG_ERROR, "CRC error.\n");
  1551. if (avctx->err_recognition & AV_EF_EXPLODE)
  1552. return AVERROR_INVALIDDATA;
  1553. }
  1554. }
  1555. *got_frame_ptr = 1;
  1556. bytes_read = invalid_frame ? buffer_size :
  1557. (get_bits_count(&ctx->gb) + 7) >> 3;
  1558. return bytes_read;
  1559. }
  1560. /** Uninitialize the ALS decoder.
  1561. */
  1562. static av_cold int decode_end(AVCodecContext *avctx)
  1563. {
  1564. ALSDecContext *ctx = avctx->priv_data;
  1565. int i;
  1566. av_freep(&ctx->sconf.chan_pos);
  1567. ff_bgmc_end(&ctx->bgmc_lut, &ctx->bgmc_lut_status);
  1568. av_freep(&ctx->const_block);
  1569. av_freep(&ctx->shift_lsbs);
  1570. av_freep(&ctx->opt_order);
  1571. av_freep(&ctx->store_prev_samples);
  1572. av_freep(&ctx->use_ltp);
  1573. av_freep(&ctx->ltp_lag);
  1574. av_freep(&ctx->ltp_gain);
  1575. av_freep(&ctx->ltp_gain_buffer);
  1576. av_freep(&ctx->quant_cof);
  1577. av_freep(&ctx->lpc_cof);
  1578. av_freep(&ctx->quant_cof_buffer);
  1579. av_freep(&ctx->lpc_cof_buffer);
  1580. av_freep(&ctx->lpc_cof_reversed_buffer);
  1581. av_freep(&ctx->prev_raw_samples);
  1582. av_freep(&ctx->raw_samples);
  1583. av_freep(&ctx->raw_buffer);
  1584. av_freep(&ctx->chan_data);
  1585. av_freep(&ctx->chan_data_buffer);
  1586. av_freep(&ctx->reverted_channels);
  1587. av_freep(&ctx->crc_buffer);
  1588. if (ctx->mlz) {
  1589. av_freep(&ctx->mlz->dict);
  1590. av_freep(&ctx->mlz);
  1591. }
  1592. av_freep(&ctx->acf);
  1593. av_freep(&ctx->last_acf_mantissa);
  1594. av_freep(&ctx->shift_value);
  1595. av_freep(&ctx->last_shift_value);
  1596. if (ctx->raw_mantissa) {
  1597. for (i = 0; i < avctx->channels; i++) {
  1598. av_freep(&ctx->raw_mantissa[i]);
  1599. }
  1600. av_freep(&ctx->raw_mantissa);
  1601. }
  1602. av_freep(&ctx->larray);
  1603. av_freep(&ctx->nbits);
  1604. return 0;
  1605. }
  1606. /** Initialize the ALS decoder.
  1607. */
  1608. static av_cold int decode_init(AVCodecContext *avctx)
  1609. {
  1610. unsigned int c;
  1611. unsigned int channel_size;
  1612. int num_buffers, ret;
  1613. ALSDecContext *ctx = avctx->priv_data;
  1614. ALSSpecificConfig *sconf = &ctx->sconf;
  1615. ctx->avctx = avctx;
  1616. if (!avctx->extradata) {
  1617. av_log(avctx, AV_LOG_ERROR, "Missing required ALS extradata.\n");
  1618. return AVERROR_INVALIDDATA;
  1619. }
  1620. if ((ret = read_specific_config(ctx)) < 0) {
  1621. av_log(avctx, AV_LOG_ERROR, "Reading ALSSpecificConfig failed.\n");
  1622. goto fail;
  1623. }
  1624. if ((ret = check_specific_config(ctx)) < 0) {
  1625. goto fail;
  1626. }
  1627. if (sconf->bgmc) {
  1628. ret = ff_bgmc_init(avctx, &ctx->bgmc_lut, &ctx->bgmc_lut_status);
  1629. if (ret < 0)
  1630. goto fail;
  1631. }
  1632. if (sconf->floating) {
  1633. avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
  1634. avctx->bits_per_raw_sample = 32;
  1635. } else {
  1636. avctx->sample_fmt = sconf->resolution > 1
  1637. ? AV_SAMPLE_FMT_S32 : AV_SAMPLE_FMT_S16;
  1638. avctx->bits_per_raw_sample = (sconf->resolution + 1) * 8;
  1639. if (avctx->bits_per_raw_sample > 32) {
  1640. av_log(avctx, AV_LOG_ERROR, "Bits per raw sample %d larger than 32.\n",
  1641. avctx->bits_per_raw_sample);
  1642. ret = AVERROR_INVALIDDATA;
  1643. goto fail;
  1644. }
  1645. }
  1646. // set maximum Rice parameter for progressive decoding based on resolution
  1647. // This is not specified in 14496-3 but actually done by the reference
  1648. // codec RM22 revision 2.
  1649. ctx->s_max = sconf->resolution > 1 ? 31 : 15;
  1650. // set lag value for long-term prediction
  1651. ctx->ltp_lag_length = 8 + (avctx->sample_rate >= 96000) +
  1652. (avctx->sample_rate >= 192000);
  1653. // allocate quantized parcor coefficient buffer
  1654. num_buffers = sconf->mc_coding ? avctx->channels : 1;
  1655. ctx->quant_cof = av_malloc_array(num_buffers, sizeof(*ctx->quant_cof));
  1656. ctx->lpc_cof = av_malloc_array(num_buffers, sizeof(*ctx->lpc_cof));
  1657. ctx->quant_cof_buffer = av_malloc_array(num_buffers * sconf->max_order,
  1658. sizeof(*ctx->quant_cof_buffer));
  1659. ctx->lpc_cof_buffer = av_malloc_array(num_buffers * sconf->max_order,
  1660. sizeof(*ctx->lpc_cof_buffer));
  1661. ctx->lpc_cof_reversed_buffer = av_malloc_array(sconf->max_order,
  1662. sizeof(*ctx->lpc_cof_buffer));
  1663. if (!ctx->quant_cof || !ctx->lpc_cof ||
  1664. !ctx->quant_cof_buffer || !ctx->lpc_cof_buffer ||
  1665. !ctx->lpc_cof_reversed_buffer) {
  1666. av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
  1667. ret = AVERROR(ENOMEM);
  1668. goto fail;
  1669. }
  1670. // assign quantized parcor coefficient buffers
  1671. for (c = 0; c < num_buffers; c++) {
  1672. ctx->quant_cof[c] = ctx->quant_cof_buffer + c * sconf->max_order;
  1673. ctx->lpc_cof[c] = ctx->lpc_cof_buffer + c * sconf->max_order;
  1674. }
  1675. // allocate and assign lag and gain data buffer for ltp mode
  1676. ctx->const_block = av_malloc_array(num_buffers, sizeof(*ctx->const_block));
  1677. ctx->shift_lsbs = av_malloc_array(num_buffers, sizeof(*ctx->shift_lsbs));
  1678. ctx->opt_order = av_malloc_array(num_buffers, sizeof(*ctx->opt_order));
  1679. ctx->store_prev_samples = av_malloc_array(num_buffers, sizeof(*ctx->store_prev_samples));
  1680. ctx->use_ltp = av_mallocz_array(num_buffers, sizeof(*ctx->use_ltp));
  1681. ctx->ltp_lag = av_malloc_array(num_buffers, sizeof(*ctx->ltp_lag));
  1682. ctx->ltp_gain = av_malloc_array(num_buffers, sizeof(*ctx->ltp_gain));
  1683. ctx->ltp_gain_buffer = av_malloc_array(num_buffers * 5, sizeof(*ctx->ltp_gain_buffer));
  1684. if (!ctx->const_block || !ctx->shift_lsbs ||
  1685. !ctx->opt_order || !ctx->store_prev_samples ||
  1686. !ctx->use_ltp || !ctx->ltp_lag ||
  1687. !ctx->ltp_gain || !ctx->ltp_gain_buffer) {
  1688. av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
  1689. ret = AVERROR(ENOMEM);
  1690. goto fail;
  1691. }
  1692. for (c = 0; c < num_buffers; c++)
  1693. ctx->ltp_gain[c] = ctx->ltp_gain_buffer + c * 5;
  1694. // allocate and assign channel data buffer for mcc mode
  1695. if (sconf->mc_coding) {
  1696. ctx->chan_data_buffer = av_mallocz_array(num_buffers * num_buffers,
  1697. sizeof(*ctx->chan_data_buffer));
  1698. ctx->chan_data = av_mallocz_array(num_buffers,
  1699. sizeof(*ctx->chan_data));
  1700. ctx->reverted_channels = av_malloc_array(num_buffers,
  1701. sizeof(*ctx->reverted_channels));
  1702. if (!ctx->chan_data_buffer || !ctx->chan_data || !ctx->reverted_channels) {
  1703. av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
  1704. ret = AVERROR(ENOMEM);
  1705. goto fail;
  1706. }
  1707. for (c = 0; c < num_buffers; c++)
  1708. ctx->chan_data[c] = ctx->chan_data_buffer + c * num_buffers;
  1709. } else {
  1710. ctx->chan_data = NULL;
  1711. ctx->chan_data_buffer = NULL;
  1712. ctx->reverted_channels = NULL;
  1713. }
  1714. channel_size = sconf->frame_length + sconf->max_order;
  1715. ctx->prev_raw_samples = av_malloc_array(sconf->max_order, sizeof(*ctx->prev_raw_samples));
  1716. ctx->raw_buffer = av_mallocz_array(avctx->channels * channel_size, sizeof(*ctx->raw_buffer));
  1717. ctx->raw_samples = av_malloc_array(avctx->channels, sizeof(*ctx->raw_samples));
  1718. if (sconf->floating) {
  1719. ctx->acf = av_malloc_array(avctx->channels, sizeof(*ctx->acf));
  1720. ctx->shift_value = av_malloc_array(avctx->channels, sizeof(*ctx->shift_value));
  1721. ctx->last_shift_value = av_malloc_array(avctx->channels, sizeof(*ctx->last_shift_value));
  1722. ctx->last_acf_mantissa = av_malloc_array(avctx->channels, sizeof(*ctx->last_acf_mantissa));
  1723. ctx->raw_mantissa = av_mallocz_array(avctx->channels, sizeof(*ctx->raw_mantissa));
  1724. ctx->larray = av_malloc_array(ctx->cur_frame_length * 4, sizeof(*ctx->larray));
  1725. ctx->nbits = av_malloc_array(ctx->cur_frame_length, sizeof(*ctx->nbits));
  1726. ctx->mlz = av_mallocz(sizeof(*ctx->mlz));
  1727. if (!ctx->mlz || !ctx->acf || !ctx->shift_value || !ctx->last_shift_value
  1728. || !ctx->last_acf_mantissa || !ctx->raw_mantissa) {
  1729. av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
  1730. ret = AVERROR(ENOMEM);
  1731. goto fail;
  1732. }
  1733. ff_mlz_init_dict(avctx, ctx->mlz);
  1734. ff_mlz_flush_dict(ctx->mlz);
  1735. for (c = 0; c < avctx->channels; ++c) {
  1736. ctx->raw_mantissa[c] = av_mallocz_array(ctx->cur_frame_length, sizeof(**ctx->raw_mantissa));
  1737. }
  1738. }
  1739. // allocate previous raw sample buffer
  1740. if (!ctx->prev_raw_samples || !ctx->raw_buffer|| !ctx->raw_samples) {
  1741. av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
  1742. ret = AVERROR(ENOMEM);
  1743. goto fail;
  1744. }
  1745. // assign raw samples buffers
  1746. ctx->raw_samples[0] = ctx->raw_buffer + sconf->max_order;
  1747. for (c = 1; c < avctx->channels; c++)
  1748. ctx->raw_samples[c] = ctx->raw_samples[c - 1] + channel_size;
  1749. // allocate crc buffer
  1750. if (HAVE_BIGENDIAN != sconf->msb_first && sconf->crc_enabled &&
  1751. (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL))) {
  1752. ctx->crc_buffer = av_malloc_array(ctx->cur_frame_length *
  1753. avctx->channels *
  1754. av_get_bytes_per_sample(avctx->sample_fmt),
  1755. sizeof(*ctx->crc_buffer));
  1756. if (!ctx->crc_buffer) {
  1757. av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
  1758. ret = AVERROR(ENOMEM);
  1759. goto fail;
  1760. }
  1761. }
  1762. ff_bswapdsp_init(&ctx->bdsp);
  1763. return 0;
  1764. fail:
  1765. decode_end(avctx);
  1766. return ret;
  1767. }
  1768. /** Flush (reset) the frame ID after seeking.
  1769. */
  1770. static av_cold void flush(AVCodecContext *avctx)
  1771. {
  1772. ALSDecContext *ctx = avctx->priv_data;
  1773. ctx->frame_id = 0;
  1774. }
  1775. AVCodec ff_als_decoder = {
  1776. .name = "als",
  1777. .long_name = NULL_IF_CONFIG_SMALL("MPEG-4 Audio Lossless Coding (ALS)"),
  1778. .type = AVMEDIA_TYPE_AUDIO,
  1779. .id = AV_CODEC_ID_MP4ALS,
  1780. .priv_data_size = sizeof(ALSDecContext),
  1781. .init = decode_init,
  1782. .close = decode_end,
  1783. .decode = decode_frame,
  1784. .flush = flush,
  1785. .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
  1786. };