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.

2146 lines
76KB

  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 "bgmc.h"
  32. #include "bswapdsp.h"
  33. #include "internal.h"
  34. #include "mlz.h"
  35. #include "libavutil/samplefmt.h"
  36. #include "libavutil/crc.h"
  37. #include "libavutil/softfloat_ieee754.h"
  38. #include "libavutil/intfloat.h"
  39. #include "libavutil/intreadwrite.h"
  40. #include <stdint.h>
  41. /** Rice parameters and corresponding index offsets for decoding the
  42. * indices of scaled PARCOR values. The table chosen is set globally
  43. * by the encoder and stored in ALSSpecificConfig.
  44. */
  45. static const int8_t parcor_rice_table[3][20][2] = {
  46. { {-52, 4}, {-29, 5}, {-31, 4}, { 19, 4}, {-16, 4},
  47. { 12, 3}, { -7, 3}, { 9, 3}, { -5, 3}, { 6, 3},
  48. { -4, 3}, { 3, 3}, { -3, 2}, { 3, 2}, { -2, 2},
  49. { 3, 2}, { -1, 2}, { 2, 2}, { -1, 2}, { 2, 2} },
  50. { {-58, 3}, {-42, 4}, {-46, 4}, { 37, 5}, {-36, 4},
  51. { 29, 4}, {-29, 4}, { 25, 4}, {-23, 4}, { 20, 4},
  52. {-17, 4}, { 16, 4}, {-12, 4}, { 12, 3}, {-10, 4},
  53. { 7, 3}, { -4, 4}, { 3, 3}, { -1, 3}, { 1, 3} },
  54. { {-59, 3}, {-45, 5}, {-50, 4}, { 38, 4}, {-39, 4},
  55. { 32, 4}, {-30, 4}, { 25, 3}, {-23, 3}, { 20, 3},
  56. {-20, 3}, { 16, 3}, {-13, 3}, { 10, 3}, { -7, 3},
  57. { 3, 3}, { 0, 3}, { -1, 3}, { 2, 3}, { -1, 2} }
  58. };
  59. /** Scaled PARCOR values used for the first two PARCOR coefficients.
  60. * To be indexed by the Rice coded indices.
  61. * Generated by: parcor_scaled_values[i] = 32 + ((i * (i+1)) << 7) - (1 << 20)
  62. * Actual values are divided by 32 in order to be stored in 16 bits.
  63. */
  64. static const int16_t parcor_scaled_values[] = {
  65. -1048544 / 32, -1048288 / 32, -1047776 / 32, -1047008 / 32,
  66. -1045984 / 32, -1044704 / 32, -1043168 / 32, -1041376 / 32,
  67. -1039328 / 32, -1037024 / 32, -1034464 / 32, -1031648 / 32,
  68. -1028576 / 32, -1025248 / 32, -1021664 / 32, -1017824 / 32,
  69. -1013728 / 32, -1009376 / 32, -1004768 / 32, -999904 / 32,
  70. -994784 / 32, -989408 / 32, -983776 / 32, -977888 / 32,
  71. -971744 / 32, -965344 / 32, -958688 / 32, -951776 / 32,
  72. -944608 / 32, -937184 / 32, -929504 / 32, -921568 / 32,
  73. -913376 / 32, -904928 / 32, -896224 / 32, -887264 / 32,
  74. -878048 / 32, -868576 / 32, -858848 / 32, -848864 / 32,
  75. -838624 / 32, -828128 / 32, -817376 / 32, -806368 / 32,
  76. -795104 / 32, -783584 / 32, -771808 / 32, -759776 / 32,
  77. -747488 / 32, -734944 / 32, -722144 / 32, -709088 / 32,
  78. -695776 / 32, -682208 / 32, -668384 / 32, -654304 / 32,
  79. -639968 / 32, -625376 / 32, -610528 / 32, -595424 / 32,
  80. -580064 / 32, -564448 / 32, -548576 / 32, -532448 / 32,
  81. -516064 / 32, -499424 / 32, -482528 / 32, -465376 / 32,
  82. -447968 / 32, -430304 / 32, -412384 / 32, -394208 / 32,
  83. -375776 / 32, -357088 / 32, -338144 / 32, -318944 / 32,
  84. -299488 / 32, -279776 / 32, -259808 / 32, -239584 / 32,
  85. -219104 / 32, -198368 / 32, -177376 / 32, -156128 / 32,
  86. -134624 / 32, -112864 / 32, -90848 / 32, -68576 / 32,
  87. -46048 / 32, -23264 / 32, -224 / 32, 23072 / 32,
  88. 46624 / 32, 70432 / 32, 94496 / 32, 118816 / 32,
  89. 143392 / 32, 168224 / 32, 193312 / 32, 218656 / 32,
  90. 244256 / 32, 270112 / 32, 296224 / 32, 322592 / 32,
  91. 349216 / 32, 376096 / 32, 403232 / 32, 430624 / 32,
  92. 458272 / 32, 486176 / 32, 514336 / 32, 542752 / 32,
  93. 571424 / 32, 600352 / 32, 629536 / 32, 658976 / 32,
  94. 688672 / 32, 718624 / 32, 748832 / 32, 779296 / 32,
  95. 810016 / 32, 840992 / 32, 872224 / 32, 903712 / 32,
  96. 935456 / 32, 967456 / 32, 999712 / 32, 1032224 / 32
  97. };
  98. /** Gain values of p(0) for long-term prediction.
  99. * To be indexed by the Rice coded indices.
  100. */
  101. static const uint8_t ltp_gain_values [4][4] = {
  102. { 0, 8, 16, 24},
  103. {32, 40, 48, 56},
  104. {64, 70, 76, 82},
  105. {88, 92, 96, 100}
  106. };
  107. /** Inter-channel weighting factors for multi-channel correlation.
  108. * To be indexed by the Rice coded indices.
  109. */
  110. static const int16_t mcc_weightings[] = {
  111. 204, 192, 179, 166, 153, 140, 128, 115,
  112. 102, 89, 76, 64, 51, 38, 25, 12,
  113. 0, -12, -25, -38, -51, -64, -76, -89,
  114. -102, -115, -128, -140, -153, -166, -179, -192
  115. };
  116. /** Tail codes used in arithmetic coding using block Gilbert-Moore codes.
  117. */
  118. static const uint8_t tail_code[16][6] = {
  119. { 74, 44, 25, 13, 7, 3},
  120. { 68, 42, 24, 13, 7, 3},
  121. { 58, 39, 23, 13, 7, 3},
  122. {126, 70, 37, 19, 10, 5},
  123. {132, 70, 37, 20, 10, 5},
  124. {124, 70, 38, 20, 10, 5},
  125. {120, 69, 37, 20, 11, 5},
  126. {116, 67, 37, 20, 11, 5},
  127. {108, 66, 36, 20, 10, 5},
  128. {102, 62, 36, 20, 10, 5},
  129. { 88, 58, 34, 19, 10, 5},
  130. {162, 89, 49, 25, 13, 7},
  131. {156, 87, 49, 26, 14, 7},
  132. {150, 86, 47, 26, 14, 7},
  133. {142, 84, 47, 26, 14, 7},
  134. {131, 79, 46, 26, 14, 7}
  135. };
  136. enum RA_Flag {
  137. RA_FLAG_NONE,
  138. RA_FLAG_FRAMES,
  139. RA_FLAG_HEADER
  140. };
  141. typedef struct ALSSpecificConfig {
  142. uint32_t samples; ///< number of samples, 0xFFFFFFFF if unknown
  143. int resolution; ///< 000 = 8-bit; 001 = 16-bit; 010 = 24-bit; 011 = 32-bit
  144. int floating; ///< 1 = IEEE 32-bit floating-point, 0 = integer
  145. int msb_first; ///< 1 = original CRC calculated on big-endian system, 0 = little-endian
  146. int frame_length; ///< frame length for each frame (last frame may differ)
  147. int ra_distance; ///< distance between RA frames (in frames, 0...255)
  148. enum RA_Flag ra_flag; ///< indicates where the size of ra units is stored
  149. int adapt_order; ///< adaptive order: 1 = on, 0 = off
  150. int coef_table; ///< table index of Rice code parameters
  151. int long_term_prediction; ///< long term prediction (LTP): 1 = on, 0 = off
  152. int max_order; ///< maximum prediction order (0..1023)
  153. int block_switching; ///< number of block switching levels
  154. int bgmc; ///< "Block Gilbert-Moore Code": 1 = on, 0 = off (Rice coding only)
  155. int sb_part; ///< sub-block partition
  156. int joint_stereo; ///< joint stereo: 1 = on, 0 = off
  157. int mc_coding; ///< extended inter-channel coding (multi channel coding): 1 = on, 0 = off
  158. int chan_config; ///< indicates that a chan_config_info field is present
  159. int chan_sort; ///< channel rearrangement: 1 = on, 0 = off
  160. int rlslms; ///< use "Recursive Least Square-Least Mean Square" predictor: 1 = on, 0 = off
  161. int chan_config_info; ///< mapping of channels to loudspeaker locations. Unused until setting channel configuration is implemented.
  162. int *chan_pos; ///< original channel positions
  163. int crc_enabled; ///< enable Cyclic Redundancy Checksum
  164. } ALSSpecificConfig;
  165. typedef struct ALSChannelData {
  166. int stop_flag;
  167. int master_channel;
  168. int time_diff_flag;
  169. int time_diff_sign;
  170. int time_diff_index;
  171. int weighting[6];
  172. } ALSChannelData;
  173. typedef struct ALSDecContext {
  174. AVCodecContext *avctx;
  175. ALSSpecificConfig sconf;
  176. GetBitContext gb;
  177. BswapDSPContext bdsp;
  178. const AVCRC *crc_table;
  179. uint32_t crc_org; ///< CRC value of the original input data
  180. uint32_t crc; ///< CRC value calculated from decoded data
  181. unsigned int cur_frame_length; ///< length of the current frame to decode
  182. unsigned int frame_id; ///< the frame ID / number of the current frame
  183. unsigned int js_switch; ///< if true, joint-stereo decoding is enforced
  184. unsigned int cs_switch; ///< if true, channel rearrangement is done
  185. unsigned int num_blocks; ///< number of blocks used in the current frame
  186. unsigned int s_max; ///< maximum Rice parameter allowed in entropy coding
  187. uint8_t *bgmc_lut; ///< pointer at lookup tables used for BGMC
  188. int *bgmc_lut_status; ///< pointer at lookup table status flags used for BGMC
  189. int ltp_lag_length; ///< number of bits used for ltp lag value
  190. int *const_block; ///< contains const_block flags for all channels
  191. unsigned int *shift_lsbs; ///< contains shift_lsbs flags for all channels
  192. unsigned int *opt_order; ///< contains opt_order flags for all channels
  193. int *store_prev_samples; ///< contains store_prev_samples flags for all channels
  194. int *use_ltp; ///< contains use_ltp flags for all channels
  195. int *ltp_lag; ///< contains ltp lag values for all channels
  196. int **ltp_gain; ///< gain values for ltp 5-tap filter for a channel
  197. int *ltp_gain_buffer; ///< contains all gain values for ltp 5-tap filter
  198. int32_t **quant_cof; ///< quantized parcor coefficients for a channel
  199. int32_t *quant_cof_buffer; ///< contains all quantized parcor coefficients
  200. int32_t **lpc_cof; ///< coefficients of the direct form prediction filter for a channel
  201. int32_t *lpc_cof_buffer; ///< contains all coefficients of the direct form prediction filter
  202. int32_t *lpc_cof_reversed_buffer; ///< temporary buffer to set up a reversed versio of lpc_cof_buffer
  203. ALSChannelData **chan_data; ///< channel data for multi-channel correlation
  204. ALSChannelData *chan_data_buffer; ///< contains channel data for all channels
  205. int *reverted_channels; ///< stores a flag for each reverted channel
  206. int32_t *prev_raw_samples; ///< contains unshifted raw samples from the previous block
  207. int32_t **raw_samples; ///< decoded raw samples for each channel
  208. int32_t *raw_buffer; ///< contains all decoded raw samples including carryover samples
  209. uint8_t *crc_buffer; ///< buffer of byte order corrected samples used for CRC check
  210. MLZ* mlz; ///< masked lz decompression structure
  211. SoftFloat_IEEE754 *acf; ///< contains common multiplier for all channels
  212. int *last_acf_mantissa; ///< contains the last acf mantissa data of common multiplier for all channels
  213. int *shift_value; ///< value by which the binary point is to be shifted for all channels
  214. int *last_shift_value; ///< contains last shift value for all channels
  215. int **raw_mantissa; ///< decoded mantissa bits of the difference signal
  216. unsigned char *larray; ///< buffer to store the output of masked lz decompression
  217. int *nbits; ///< contains the number of bits to read for masked lz decompression for all samples
  218. } ALSDecContext;
  219. typedef struct ALSBlockData {
  220. unsigned int block_length; ///< number of samples within the block
  221. unsigned int ra_block; ///< if true, this is a random access block
  222. int *const_block; ///< if true, this is a constant value block
  223. int js_blocks; ///< true if this block contains a difference signal
  224. unsigned int *shift_lsbs; ///< shift of values for this block
  225. unsigned int *opt_order; ///< prediction order of this block
  226. int *store_prev_samples;///< if true, carryover samples have to be stored
  227. int *use_ltp; ///< if true, long-term prediction is used
  228. int *ltp_lag; ///< lag value for long-term prediction
  229. int *ltp_gain; ///< gain values for ltp 5-tap filter
  230. int32_t *quant_cof; ///< quantized parcor coefficients
  231. int32_t *lpc_cof; ///< coefficients of the direct form prediction
  232. int32_t *raw_samples; ///< decoded raw samples / residuals for this block
  233. int32_t *prev_raw_samples; ///< contains unshifted raw samples from the previous block
  234. int32_t *raw_other; ///< decoded raw samples of the other channel of a channel pair
  235. } ALSBlockData;
  236. static av_cold void dprint_specific_config(ALSDecContext *ctx)
  237. {
  238. #ifdef DEBUG
  239. AVCodecContext *avctx = ctx->avctx;
  240. ALSSpecificConfig *sconf = &ctx->sconf;
  241. ff_dlog(avctx, "resolution = %i\n", sconf->resolution);
  242. ff_dlog(avctx, "floating = %i\n", sconf->floating);
  243. ff_dlog(avctx, "frame_length = %i\n", sconf->frame_length);
  244. ff_dlog(avctx, "ra_distance = %i\n", sconf->ra_distance);
  245. ff_dlog(avctx, "ra_flag = %i\n", sconf->ra_flag);
  246. ff_dlog(avctx, "adapt_order = %i\n", sconf->adapt_order);
  247. ff_dlog(avctx, "coef_table = %i\n", sconf->coef_table);
  248. ff_dlog(avctx, "long_term_prediction = %i\n", sconf->long_term_prediction);
  249. ff_dlog(avctx, "max_order = %i\n", sconf->max_order);
  250. ff_dlog(avctx, "block_switching = %i\n", sconf->block_switching);
  251. ff_dlog(avctx, "bgmc = %i\n", sconf->bgmc);
  252. ff_dlog(avctx, "sb_part = %i\n", sconf->sb_part);
  253. ff_dlog(avctx, "joint_stereo = %i\n", sconf->joint_stereo);
  254. ff_dlog(avctx, "mc_coding = %i\n", sconf->mc_coding);
  255. ff_dlog(avctx, "chan_config = %i\n", sconf->chan_config);
  256. ff_dlog(avctx, "chan_sort = %i\n", sconf->chan_sort);
  257. ff_dlog(avctx, "RLSLMS = %i\n", sconf->rlslms);
  258. ff_dlog(avctx, "chan_config_info = %i\n", sconf->chan_config_info);
  259. #endif
  260. }
  261. /** Read an ALSSpecificConfig from a buffer into the output struct.
  262. */
  263. static av_cold int read_specific_config(ALSDecContext *ctx)
  264. {
  265. GetBitContext gb;
  266. uint64_t ht_size;
  267. int i, config_offset;
  268. MPEG4AudioConfig m4ac = {0};
  269. ALSSpecificConfig *sconf = &ctx->sconf;
  270. AVCodecContext *avctx = ctx->avctx;
  271. uint32_t als_id, header_size, trailer_size;
  272. int ret;
  273. if ((ret = init_get_bits8(&gb, avctx->extradata, avctx->extradata_size)) < 0)
  274. return ret;
  275. config_offset = avpriv_mpeg4audio_get_config(&m4ac, avctx->extradata,
  276. avctx->extradata_size * 8, 1);
  277. if (config_offset < 0)
  278. return AVERROR_INVALIDDATA;
  279. skip_bits_long(&gb, config_offset);
  280. if (get_bits_left(&gb) < (30 << 3))
  281. return AVERROR_INVALIDDATA;
  282. // read the fixed items
  283. als_id = get_bits_long(&gb, 32);
  284. avctx->sample_rate = m4ac.sample_rate;
  285. skip_bits_long(&gb, 32); // sample rate already known
  286. sconf->samples = get_bits_long(&gb, 32);
  287. avctx->channels = m4ac.channels;
  288. skip_bits(&gb, 16); // number of channels already known
  289. skip_bits(&gb, 3); // skip file_type
  290. sconf->resolution = get_bits(&gb, 3);
  291. sconf->floating = get_bits1(&gb);
  292. sconf->msb_first = get_bits1(&gb);
  293. sconf->frame_length = get_bits(&gb, 16) + 1;
  294. sconf->ra_distance = get_bits(&gb, 8);
  295. sconf->ra_flag = get_bits(&gb, 2);
  296. sconf->adapt_order = get_bits1(&gb);
  297. sconf->coef_table = get_bits(&gb, 2);
  298. sconf->long_term_prediction = get_bits1(&gb);
  299. sconf->max_order = get_bits(&gb, 10);
  300. sconf->block_switching = get_bits(&gb, 2);
  301. sconf->bgmc = get_bits1(&gb);
  302. sconf->sb_part = get_bits1(&gb);
  303. sconf->joint_stereo = get_bits1(&gb);
  304. sconf->mc_coding = get_bits1(&gb);
  305. sconf->chan_config = get_bits1(&gb);
  306. sconf->chan_sort = get_bits1(&gb);
  307. sconf->crc_enabled = get_bits1(&gb);
  308. sconf->rlslms = get_bits1(&gb);
  309. skip_bits(&gb, 5); // skip 5 reserved bits
  310. skip_bits1(&gb); // skip aux_data_enabled
  311. // check for ALSSpecificConfig struct
  312. if (als_id != MKBETAG('A','L','S','\0'))
  313. return AVERROR_INVALIDDATA;
  314. ctx->cur_frame_length = sconf->frame_length;
  315. // read channel config
  316. if (sconf->chan_config)
  317. sconf->chan_config_info = get_bits(&gb, 16);
  318. // TODO: use this to set avctx->channel_layout
  319. // read channel sorting
  320. if (sconf->chan_sort && avctx->channels > 1) {
  321. int chan_pos_bits = av_ceil_log2(avctx->channels);
  322. int bits_needed = avctx->channels * chan_pos_bits + 7;
  323. if (get_bits_left(&gb) < bits_needed)
  324. return AVERROR_INVALIDDATA;
  325. if (!(sconf->chan_pos = av_malloc_array(avctx->channels, sizeof(*sconf->chan_pos))))
  326. return AVERROR(ENOMEM);
  327. ctx->cs_switch = 1;
  328. for (i = 0; i < avctx->channels; i++) {
  329. sconf->chan_pos[i] = -1;
  330. }
  331. for (i = 0; i < avctx->channels; i++) {
  332. int idx;
  333. idx = get_bits(&gb, chan_pos_bits);
  334. if (idx >= avctx->channels || sconf->chan_pos[idx] != -1) {
  335. av_log(avctx, AV_LOG_WARNING, "Invalid channel reordering.\n");
  336. ctx->cs_switch = 0;
  337. break;
  338. }
  339. sconf->chan_pos[idx] = i;
  340. }
  341. align_get_bits(&gb);
  342. }
  343. // read fixed header and trailer sizes,
  344. // if size = 0xFFFFFFFF then there is no data field!
  345. if (get_bits_left(&gb) < 64)
  346. return AVERROR_INVALIDDATA;
  347. header_size = get_bits_long(&gb, 32);
  348. trailer_size = get_bits_long(&gb, 32);
  349. if (header_size == 0xFFFFFFFF)
  350. header_size = 0;
  351. if (trailer_size == 0xFFFFFFFF)
  352. trailer_size = 0;
  353. ht_size = ((int64_t)(header_size) + (int64_t)(trailer_size)) << 3;
  354. // skip the header and trailer data
  355. if (get_bits_left(&gb) < ht_size)
  356. return AVERROR_INVALIDDATA;
  357. if (ht_size > INT32_MAX)
  358. return AVERROR_PATCHWELCOME;
  359. skip_bits_long(&gb, ht_size);
  360. // initialize CRC calculation
  361. if (sconf->crc_enabled) {
  362. if (get_bits_left(&gb) < 32)
  363. return AVERROR_INVALIDDATA;
  364. if (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL)) {
  365. ctx->crc_table = av_crc_get_table(AV_CRC_32_IEEE_LE);
  366. ctx->crc = 0xFFFFFFFF;
  367. ctx->crc_org = ~get_bits_long(&gb, 32);
  368. } else
  369. skip_bits_long(&gb, 32);
  370. }
  371. // no need to read the rest of ALSSpecificConfig (ra_unit_size & aux data)
  372. dprint_specific_config(ctx);
  373. return 0;
  374. }
  375. /** Check the ALSSpecificConfig for unsupported features.
  376. */
  377. static int check_specific_config(ALSDecContext *ctx)
  378. {
  379. ALSSpecificConfig *sconf = &ctx->sconf;
  380. int error = 0;
  381. // report unsupported feature and set error value
  382. #define MISSING_ERR(cond, str, errval) \
  383. { \
  384. if (cond) { \
  385. avpriv_report_missing_feature(ctx->avctx, \
  386. str); \
  387. error = errval; \
  388. } \
  389. }
  390. MISSING_ERR(sconf->rlslms, "Adaptive RLS-LMS prediction", AVERROR_PATCHWELCOME);
  391. return error;
  392. }
  393. /** Parse the bs_info field to extract the block partitioning used in
  394. * block switching mode, refer to ISO/IEC 14496-3, section 11.6.2.
  395. */
  396. static void parse_bs_info(const uint32_t bs_info, unsigned int n,
  397. unsigned int div, unsigned int **div_blocks,
  398. unsigned int *num_blocks)
  399. {
  400. if (n < 31 && ((bs_info << n) & 0x40000000)) {
  401. // if the level is valid and the investigated bit n is set
  402. // then recursively check both children at bits (2n+1) and (2n+2)
  403. n *= 2;
  404. div += 1;
  405. parse_bs_info(bs_info, n + 1, div, div_blocks, num_blocks);
  406. parse_bs_info(bs_info, n + 2, div, div_blocks, num_blocks);
  407. } else {
  408. // else the bit is not set or the last level has been reached
  409. // (bit implicitly not set)
  410. **div_blocks = div;
  411. (*div_blocks)++;
  412. (*num_blocks)++;
  413. }
  414. }
  415. /** Read and decode a Rice codeword.
  416. */
  417. static int32_t decode_rice(GetBitContext *gb, unsigned int k)
  418. {
  419. int max = get_bits_left(gb) - k;
  420. int q = get_unary(gb, 0, max);
  421. int r = k ? get_bits1(gb) : !(q & 1);
  422. if (k > 1) {
  423. q <<= (k - 1);
  424. q += get_bits_long(gb, k - 1);
  425. } else if (!k) {
  426. q >>= 1;
  427. }
  428. return r ? q : ~q;
  429. }
  430. /** Convert PARCOR coefficient k to direct filter coefficient.
  431. */
  432. static void parcor_to_lpc(unsigned int k, const int32_t *par, int32_t *cof)
  433. {
  434. int i, j;
  435. for (i = 0, j = k - 1; i < j; i++, j--) {
  436. int tmp1 = ((MUL64(par[k], cof[j]) + (1 << 19)) >> 20);
  437. cof[j] += ((MUL64(par[k], cof[i]) + (1 << 19)) >> 20);
  438. cof[i] += tmp1;
  439. }
  440. if (i == j)
  441. cof[i] += ((MUL64(par[k], cof[j]) + (1 << 19)) >> 20);
  442. cof[k] = par[k];
  443. }
  444. /** Read block switching field if necessary and set actual block sizes.
  445. * Also assure that the block sizes of the last frame correspond to the
  446. * actual number of samples.
  447. */
  448. static void get_block_sizes(ALSDecContext *ctx, unsigned int *div_blocks,
  449. uint32_t *bs_info)
  450. {
  451. ALSSpecificConfig *sconf = &ctx->sconf;
  452. GetBitContext *gb = &ctx->gb;
  453. unsigned int *ptr_div_blocks = div_blocks;
  454. unsigned int b;
  455. if (sconf->block_switching) {
  456. unsigned int bs_info_len = 1 << (sconf->block_switching + 2);
  457. *bs_info = get_bits_long(gb, bs_info_len);
  458. *bs_info <<= (32 - bs_info_len);
  459. }
  460. ctx->num_blocks = 0;
  461. parse_bs_info(*bs_info, 0, 0, &ptr_div_blocks, &ctx->num_blocks);
  462. // The last frame may have an overdetermined block structure given in
  463. // the bitstream. In that case the defined block structure would need
  464. // more samples than available to be consistent.
  465. // The block structure is actually used but the block sizes are adapted
  466. // to fit the actual number of available samples.
  467. // Example: 5 samples, 2nd level block sizes: 2 2 2 2.
  468. // This results in the actual block sizes: 2 2 1 0.
  469. // This is not specified in 14496-3 but actually done by the reference
  470. // codec RM22 revision 2.
  471. // This appears to happen in case of an odd number of samples in the last
  472. // frame which is actually not allowed by the block length switching part
  473. // of 14496-3.
  474. // The ALS conformance files feature an odd number of samples in the last
  475. // frame.
  476. for (b = 0; b < ctx->num_blocks; b++)
  477. div_blocks[b] = ctx->sconf.frame_length >> div_blocks[b];
  478. if (ctx->cur_frame_length != ctx->sconf.frame_length) {
  479. unsigned int remaining = ctx->cur_frame_length;
  480. for (b = 0; b < ctx->num_blocks; b++) {
  481. if (remaining <= div_blocks[b]) {
  482. div_blocks[b] = remaining;
  483. ctx->num_blocks = b + 1;
  484. break;
  485. }
  486. remaining -= div_blocks[b];
  487. }
  488. }
  489. }
  490. /** Read the block data for a constant block
  491. */
  492. static int read_const_block_data(ALSDecContext *ctx, ALSBlockData *bd)
  493. {
  494. ALSSpecificConfig *sconf = &ctx->sconf;
  495. AVCodecContext *avctx = ctx->avctx;
  496. GetBitContext *gb = &ctx->gb;
  497. if (bd->block_length <= 0)
  498. return AVERROR_INVALIDDATA;
  499. *bd->raw_samples = 0;
  500. *bd->const_block = get_bits1(gb); // 1 = constant value, 0 = zero block (silence)
  501. bd->js_blocks = get_bits1(gb);
  502. // skip 5 reserved bits
  503. skip_bits(gb, 5);
  504. if (*bd->const_block) {
  505. unsigned int const_val_bits = sconf->floating ? 24 : avctx->bits_per_raw_sample;
  506. *bd->raw_samples = get_sbits_long(gb, const_val_bits);
  507. }
  508. // ensure constant block decoding by reusing this field
  509. *bd->const_block = 1;
  510. return 0;
  511. }
  512. /** Decode the block data for a constant block
  513. */
  514. static void decode_const_block_data(ALSDecContext *ctx, ALSBlockData *bd)
  515. {
  516. int smp = bd->block_length - 1;
  517. int32_t val = *bd->raw_samples;
  518. int32_t *dst = bd->raw_samples + 1;
  519. // write raw samples into buffer
  520. for (; smp; smp--)
  521. *dst++ = val;
  522. }
  523. /** Read the block data for a non-constant block
  524. */
  525. static int read_var_block_data(ALSDecContext *ctx, ALSBlockData *bd)
  526. {
  527. ALSSpecificConfig *sconf = &ctx->sconf;
  528. AVCodecContext *avctx = ctx->avctx;
  529. GetBitContext *gb = &ctx->gb;
  530. unsigned int k;
  531. unsigned int s[8];
  532. unsigned int sx[8];
  533. unsigned int sub_blocks, log2_sub_blocks, sb_length;
  534. unsigned int start = 0;
  535. unsigned int opt_order;
  536. int sb;
  537. int32_t *quant_cof = bd->quant_cof;
  538. int32_t *current_res;
  539. // ensure variable block decoding by reusing this field
  540. *bd->const_block = 0;
  541. *bd->opt_order = 1;
  542. bd->js_blocks = get_bits1(gb);
  543. opt_order = *bd->opt_order;
  544. // determine the number of subblocks for entropy decoding
  545. if (!sconf->bgmc && !sconf->sb_part) {
  546. log2_sub_blocks = 0;
  547. } else {
  548. if (sconf->bgmc && sconf->sb_part)
  549. log2_sub_blocks = get_bits(gb, 2);
  550. else
  551. log2_sub_blocks = 2 * get_bits1(gb);
  552. }
  553. sub_blocks = 1 << log2_sub_blocks;
  554. // do not continue in case of a damaged stream since
  555. // block_length must be evenly divisible by sub_blocks
  556. if (bd->block_length & (sub_blocks - 1)) {
  557. av_log(avctx, AV_LOG_WARNING,
  558. "Block length is not evenly divisible by the number of subblocks.\n");
  559. return AVERROR_INVALIDDATA;
  560. }
  561. sb_length = bd->block_length >> log2_sub_blocks;
  562. if (sconf->bgmc) {
  563. s[0] = get_bits(gb, 8 + (sconf->resolution > 1));
  564. for (k = 1; k < sub_blocks; k++)
  565. s[k] = s[k - 1] + decode_rice(gb, 2);
  566. for (k = 0; k < sub_blocks; k++) {
  567. sx[k] = s[k] & 0x0F;
  568. s [k] >>= 4;
  569. }
  570. } else {
  571. s[0] = get_bits(gb, 4 + (sconf->resolution > 1));
  572. for (k = 1; k < sub_blocks; k++)
  573. s[k] = s[k - 1] + decode_rice(gb, 0);
  574. }
  575. for (k = 1; k < sub_blocks; k++)
  576. if (s[k] > 32) {
  577. av_log(avctx, AV_LOG_ERROR, "k invalid for rice code.\n");
  578. return AVERROR_INVALIDDATA;
  579. }
  580. if (get_bits1(gb))
  581. *bd->shift_lsbs = get_bits(gb, 4) + 1;
  582. *bd->store_prev_samples = (bd->js_blocks && bd->raw_other) || *bd->shift_lsbs;
  583. if (!sconf->rlslms) {
  584. if (sconf->adapt_order && sconf->max_order) {
  585. int opt_order_length = av_ceil_log2(av_clip((bd->block_length >> 3) - 1,
  586. 2, sconf->max_order + 1));
  587. *bd->opt_order = get_bits(gb, opt_order_length);
  588. if (*bd->opt_order > sconf->max_order) {
  589. *bd->opt_order = sconf->max_order;
  590. av_log(avctx, AV_LOG_ERROR, "Predictor order too large.\n");
  591. return AVERROR_INVALIDDATA;
  592. }
  593. } else {
  594. *bd->opt_order = sconf->max_order;
  595. }
  596. opt_order = *bd->opt_order;
  597. if (opt_order) {
  598. int add_base;
  599. if (sconf->coef_table == 3) {
  600. add_base = 0x7F;
  601. // read coefficient 0
  602. quant_cof[0] = 32 * parcor_scaled_values[get_bits(gb, 7)];
  603. // read coefficient 1
  604. if (opt_order > 1)
  605. quant_cof[1] = -32 * parcor_scaled_values[get_bits(gb, 7)];
  606. // read coefficients 2 to opt_order
  607. for (k = 2; k < opt_order; k++)
  608. quant_cof[k] = get_bits(gb, 7);
  609. } else {
  610. int k_max;
  611. add_base = 1;
  612. // read coefficient 0 to 19
  613. k_max = FFMIN(opt_order, 20);
  614. for (k = 0; k < k_max; k++) {
  615. int rice_param = parcor_rice_table[sconf->coef_table][k][1];
  616. int offset = parcor_rice_table[sconf->coef_table][k][0];
  617. quant_cof[k] = decode_rice(gb, rice_param) + offset;
  618. if (quant_cof[k] < -64 || quant_cof[k] > 63) {
  619. av_log(avctx, AV_LOG_ERROR,
  620. "quant_cof %"PRId32" is out of range.\n",
  621. quant_cof[k]);
  622. return AVERROR_INVALIDDATA;
  623. }
  624. }
  625. // read coefficients 20 to 126
  626. k_max = FFMIN(opt_order, 127);
  627. for (; k < k_max; k++)
  628. quant_cof[k] = decode_rice(gb, 2) + (k & 1);
  629. // read coefficients 127 to opt_order
  630. for (; k < opt_order; k++)
  631. quant_cof[k] = decode_rice(gb, 1);
  632. quant_cof[0] = 32 * parcor_scaled_values[quant_cof[0] + 64];
  633. if (opt_order > 1)
  634. quant_cof[1] = -32 * parcor_scaled_values[quant_cof[1] + 64];
  635. }
  636. for (k = 2; k < opt_order; k++)
  637. quant_cof[k] = (quant_cof[k] * (1 << 14)) + (add_base << 13);
  638. }
  639. }
  640. // read LTP gain and lag values
  641. if (sconf->long_term_prediction) {
  642. *bd->use_ltp = get_bits1(gb);
  643. if (*bd->use_ltp) {
  644. int r, c;
  645. bd->ltp_gain[0] = decode_rice(gb, 1) << 3;
  646. bd->ltp_gain[1] = decode_rice(gb, 2) << 3;
  647. r = get_unary(gb, 0, 4);
  648. c = get_bits(gb, 2);
  649. if (r >= 4) {
  650. av_log(avctx, AV_LOG_ERROR, "r overflow\n");
  651. return AVERROR_INVALIDDATA;
  652. }
  653. bd->ltp_gain[2] = ltp_gain_values[r][c];
  654. bd->ltp_gain[3] = decode_rice(gb, 2) << 3;
  655. bd->ltp_gain[4] = decode_rice(gb, 1) << 3;
  656. *bd->ltp_lag = get_bits(gb, ctx->ltp_lag_length);
  657. *bd->ltp_lag += FFMAX(4, opt_order + 1);
  658. }
  659. }
  660. // read first value and residuals in case of a random access block
  661. if (bd->ra_block) {
  662. if (opt_order)
  663. bd->raw_samples[0] = decode_rice(gb, avctx->bits_per_raw_sample - 4);
  664. if (opt_order > 1)
  665. bd->raw_samples[1] = decode_rice(gb, FFMIN(s[0] + 3, ctx->s_max));
  666. if (opt_order > 2)
  667. bd->raw_samples[2] = decode_rice(gb, FFMIN(s[0] + 1, ctx->s_max));
  668. start = FFMIN(opt_order, 3);
  669. }
  670. // read all residuals
  671. if (sconf->bgmc) {
  672. int delta[8];
  673. unsigned int k [8];
  674. unsigned int b = av_clip((av_ceil_log2(bd->block_length) - 3) >> 1, 0, 5);
  675. // read most significant bits
  676. unsigned int high;
  677. unsigned int low;
  678. unsigned int value;
  679. ff_bgmc_decode_init(gb, &high, &low, &value);
  680. current_res = bd->raw_samples + start;
  681. for (sb = 0; sb < sub_blocks; sb++) {
  682. unsigned int sb_len = sb_length - (sb ? 0 : start);
  683. k [sb] = s[sb] > b ? s[sb] - b : 0;
  684. delta[sb] = 5 - s[sb] + k[sb];
  685. ff_bgmc_decode(gb, sb_len, current_res,
  686. delta[sb], sx[sb], &high, &low, &value, ctx->bgmc_lut, ctx->bgmc_lut_status);
  687. current_res += sb_len;
  688. }
  689. ff_bgmc_decode_end(gb);
  690. // read least significant bits and tails
  691. current_res = bd->raw_samples + start;
  692. for (sb = 0; sb < sub_blocks; sb++, start = 0) {
  693. unsigned int cur_tail_code = tail_code[sx[sb]][delta[sb]];
  694. unsigned int cur_k = k[sb];
  695. unsigned int cur_s = s[sb];
  696. for (; start < sb_length; start++) {
  697. int32_t res = *current_res;
  698. if (res == cur_tail_code) {
  699. unsigned int max_msb = (2 + (sx[sb] > 2) + (sx[sb] > 10))
  700. << (5 - delta[sb]);
  701. res = decode_rice(gb, cur_s);
  702. if (res >= 0) {
  703. res += (max_msb ) << cur_k;
  704. } else {
  705. res -= (max_msb - 1) << cur_k;
  706. }
  707. } else {
  708. if (res > cur_tail_code)
  709. res--;
  710. if (res & 1)
  711. res = -res;
  712. res >>= 1;
  713. if (cur_k) {
  714. res *= 1 << cur_k;
  715. res |= get_bits_long(gb, cur_k);
  716. }
  717. }
  718. *current_res++ = res;
  719. }
  720. }
  721. } else {
  722. current_res = bd->raw_samples + start;
  723. for (sb = 0; sb < sub_blocks; sb++, start = 0)
  724. for (; start < sb_length; start++)
  725. *current_res++ = decode_rice(gb, s[sb]);
  726. }
  727. return 0;
  728. }
  729. /** Decode the block data for a non-constant block
  730. */
  731. static int decode_var_block_data(ALSDecContext *ctx, ALSBlockData *bd)
  732. {
  733. ALSSpecificConfig *sconf = &ctx->sconf;
  734. unsigned int block_length = bd->block_length;
  735. unsigned int smp = 0;
  736. unsigned int k;
  737. int opt_order = *bd->opt_order;
  738. int sb;
  739. int64_t y;
  740. int32_t *quant_cof = bd->quant_cof;
  741. int32_t *lpc_cof = bd->lpc_cof;
  742. int32_t *raw_samples = bd->raw_samples;
  743. int32_t *raw_samples_end = bd->raw_samples + bd->block_length;
  744. int32_t *lpc_cof_reversed = ctx->lpc_cof_reversed_buffer;
  745. // reverse long-term prediction
  746. if (*bd->use_ltp) {
  747. int ltp_smp;
  748. for (ltp_smp = FFMAX(*bd->ltp_lag - 2, 0); ltp_smp < block_length; ltp_smp++) {
  749. int center = ltp_smp - *bd->ltp_lag;
  750. int begin = FFMAX(0, center - 2);
  751. int end = center + 3;
  752. int tab = 5 - (end - begin);
  753. int base;
  754. y = 1 << 6;
  755. for (base = begin; base < end; base++, tab++)
  756. y += MUL64(bd->ltp_gain[tab], raw_samples[base]);
  757. raw_samples[ltp_smp] += y >> 7;
  758. }
  759. }
  760. // reconstruct all samples from residuals
  761. if (bd->ra_block) {
  762. for (smp = 0; smp < opt_order; smp++) {
  763. y = 1 << 19;
  764. for (sb = 0; sb < smp; sb++)
  765. y += MUL64(lpc_cof[sb], raw_samples[-(sb + 1)]);
  766. *raw_samples++ -= y >> 20;
  767. parcor_to_lpc(smp, quant_cof, lpc_cof);
  768. }
  769. } else {
  770. for (k = 0; k < opt_order; k++)
  771. parcor_to_lpc(k, quant_cof, lpc_cof);
  772. // store previous samples in case that they have to be altered
  773. if (*bd->store_prev_samples)
  774. memcpy(bd->prev_raw_samples, raw_samples - sconf->max_order,
  775. sizeof(*bd->prev_raw_samples) * sconf->max_order);
  776. // reconstruct difference signal for prediction (joint-stereo)
  777. if (bd->js_blocks && bd->raw_other) {
  778. int32_t *left, *right;
  779. if (bd->raw_other > raw_samples) { // D = R - L
  780. left = raw_samples;
  781. right = bd->raw_other;
  782. } else { // D = R - L
  783. left = bd->raw_other;
  784. right = raw_samples;
  785. }
  786. for (sb = -1; sb >= -sconf->max_order; sb--)
  787. raw_samples[sb] = right[sb] - left[sb];
  788. }
  789. // reconstruct shifted signal
  790. if (*bd->shift_lsbs)
  791. for (sb = -1; sb >= -sconf->max_order; sb--)
  792. raw_samples[sb] >>= *bd->shift_lsbs;
  793. }
  794. // reverse linear prediction coefficients for efficiency
  795. lpc_cof = lpc_cof + opt_order;
  796. for (sb = 0; sb < opt_order; sb++)
  797. lpc_cof_reversed[sb] = lpc_cof[-(sb + 1)];
  798. // reconstruct raw samples
  799. raw_samples = bd->raw_samples + smp;
  800. lpc_cof = lpc_cof_reversed + opt_order;
  801. for (; raw_samples < raw_samples_end; raw_samples++) {
  802. y = 1 << 19;
  803. for (sb = -opt_order; sb < 0; sb++)
  804. y += MUL64(lpc_cof[sb], raw_samples[sb]);
  805. *raw_samples -= y >> 20;
  806. }
  807. raw_samples = bd->raw_samples;
  808. // restore previous samples in case that they have been altered
  809. if (*bd->store_prev_samples)
  810. memcpy(raw_samples - sconf->max_order, bd->prev_raw_samples,
  811. sizeof(*raw_samples) * sconf->max_order);
  812. return 0;
  813. }
  814. /** Read the block data.
  815. */
  816. static int read_block(ALSDecContext *ctx, ALSBlockData *bd)
  817. {
  818. int ret;
  819. GetBitContext *gb = &ctx->gb;
  820. ALSSpecificConfig *sconf = &ctx->sconf;
  821. *bd->shift_lsbs = 0;
  822. // read block type flag and read the samples accordingly
  823. if (get_bits1(gb)) {
  824. ret = read_var_block_data(ctx, bd);
  825. } else {
  826. ret = read_const_block_data(ctx, bd);
  827. }
  828. if (!sconf->mc_coding || ctx->js_switch)
  829. align_get_bits(gb);
  830. return ret;
  831. }
  832. /** Decode the block data.
  833. */
  834. static int decode_block(ALSDecContext *ctx, ALSBlockData *bd)
  835. {
  836. unsigned int smp;
  837. int ret = 0;
  838. // read block type flag and read the samples accordingly
  839. if (*bd->const_block)
  840. decode_const_block_data(ctx, bd);
  841. else
  842. ret = decode_var_block_data(ctx, bd); // always return 0
  843. if (ret < 0)
  844. return ret;
  845. // TODO: read RLSLMS extension data
  846. if (*bd->shift_lsbs)
  847. for (smp = 0; smp < bd->block_length; smp++)
  848. bd->raw_samples[smp] <<= *bd->shift_lsbs;
  849. return 0;
  850. }
  851. /** Read and decode block data successively.
  852. */
  853. static int read_decode_block(ALSDecContext *ctx, ALSBlockData *bd)
  854. {
  855. int ret;
  856. if ((ret = read_block(ctx, bd)) < 0)
  857. return ret;
  858. return decode_block(ctx, bd);
  859. }
  860. /** Compute the number of samples left to decode for the current frame and
  861. * sets these samples to zero.
  862. */
  863. static void zero_remaining(unsigned int b, unsigned int b_max,
  864. const unsigned int *div_blocks, int32_t *buf)
  865. {
  866. unsigned int count = 0;
  867. while (b < b_max)
  868. count += div_blocks[b++];
  869. if (count)
  870. memset(buf, 0, sizeof(*buf) * count);
  871. }
  872. /** Decode blocks independently.
  873. */
  874. static int decode_blocks_ind(ALSDecContext *ctx, unsigned int ra_frame,
  875. unsigned int c, const unsigned int *div_blocks,
  876. unsigned int *js_blocks)
  877. {
  878. int ret;
  879. unsigned int b;
  880. ALSBlockData bd = { 0 };
  881. bd.ra_block = ra_frame;
  882. bd.const_block = ctx->const_block;
  883. bd.shift_lsbs = ctx->shift_lsbs;
  884. bd.opt_order = ctx->opt_order;
  885. bd.store_prev_samples = ctx->store_prev_samples;
  886. bd.use_ltp = ctx->use_ltp;
  887. bd.ltp_lag = ctx->ltp_lag;
  888. bd.ltp_gain = ctx->ltp_gain[0];
  889. bd.quant_cof = ctx->quant_cof[0];
  890. bd.lpc_cof = ctx->lpc_cof[0];
  891. bd.prev_raw_samples = ctx->prev_raw_samples;
  892. bd.raw_samples = ctx->raw_samples[c];
  893. for (b = 0; b < ctx->num_blocks; b++) {
  894. bd.block_length = div_blocks[b];
  895. if ((ret = read_decode_block(ctx, &bd)) < 0) {
  896. // damaged block, write zero for the rest of the frame
  897. zero_remaining(b, ctx->num_blocks, div_blocks, bd.raw_samples);
  898. return ret;
  899. }
  900. bd.raw_samples += div_blocks[b];
  901. bd.ra_block = 0;
  902. }
  903. return 0;
  904. }
  905. /** Decode blocks dependently.
  906. */
  907. static int decode_blocks(ALSDecContext *ctx, unsigned int ra_frame,
  908. unsigned int c, const unsigned int *div_blocks,
  909. unsigned int *js_blocks)
  910. {
  911. ALSSpecificConfig *sconf = &ctx->sconf;
  912. unsigned int offset = 0;
  913. unsigned int b;
  914. int ret;
  915. ALSBlockData bd[2] = { { 0 } };
  916. bd[0].ra_block = ra_frame;
  917. bd[0].const_block = ctx->const_block;
  918. bd[0].shift_lsbs = ctx->shift_lsbs;
  919. bd[0].opt_order = ctx->opt_order;
  920. bd[0].store_prev_samples = ctx->store_prev_samples;
  921. bd[0].use_ltp = ctx->use_ltp;
  922. bd[0].ltp_lag = ctx->ltp_lag;
  923. bd[0].ltp_gain = ctx->ltp_gain[0];
  924. bd[0].quant_cof = ctx->quant_cof[0];
  925. bd[0].lpc_cof = ctx->lpc_cof[0];
  926. bd[0].prev_raw_samples = ctx->prev_raw_samples;
  927. bd[0].js_blocks = *js_blocks;
  928. bd[1].ra_block = ra_frame;
  929. bd[1].const_block = ctx->const_block;
  930. bd[1].shift_lsbs = ctx->shift_lsbs;
  931. bd[1].opt_order = ctx->opt_order;
  932. bd[1].store_prev_samples = ctx->store_prev_samples;
  933. bd[1].use_ltp = ctx->use_ltp;
  934. bd[1].ltp_lag = ctx->ltp_lag;
  935. bd[1].ltp_gain = ctx->ltp_gain[0];
  936. bd[1].quant_cof = ctx->quant_cof[0];
  937. bd[1].lpc_cof = ctx->lpc_cof[0];
  938. bd[1].prev_raw_samples = ctx->prev_raw_samples;
  939. bd[1].js_blocks = *(js_blocks + 1);
  940. // decode all blocks
  941. for (b = 0; b < ctx->num_blocks; b++) {
  942. unsigned int s;
  943. bd[0].block_length = div_blocks[b];
  944. bd[1].block_length = div_blocks[b];
  945. bd[0].raw_samples = ctx->raw_samples[c ] + offset;
  946. bd[1].raw_samples = ctx->raw_samples[c + 1] + offset;
  947. bd[0].raw_other = bd[1].raw_samples;
  948. bd[1].raw_other = bd[0].raw_samples;
  949. if ((ret = read_decode_block(ctx, &bd[0])) < 0 ||
  950. (ret = read_decode_block(ctx, &bd[1])) < 0)
  951. goto fail;
  952. // reconstruct joint-stereo blocks
  953. if (bd[0].js_blocks) {
  954. if (bd[1].js_blocks)
  955. av_log(ctx->avctx, AV_LOG_WARNING, "Invalid channel pair.\n");
  956. for (s = 0; s < div_blocks[b]; s++)
  957. bd[0].raw_samples[s] = bd[1].raw_samples[s] - bd[0].raw_samples[s];
  958. } else if (bd[1].js_blocks) {
  959. for (s = 0; s < div_blocks[b]; s++)
  960. bd[1].raw_samples[s] = bd[1].raw_samples[s] + bd[0].raw_samples[s];
  961. }
  962. offset += div_blocks[b];
  963. bd[0].ra_block = 0;
  964. bd[1].ra_block = 0;
  965. }
  966. // store carryover raw samples,
  967. // the others channel raw samples are stored by the calling function.
  968. memmove(ctx->raw_samples[c] - sconf->max_order,
  969. ctx->raw_samples[c] - sconf->max_order + sconf->frame_length,
  970. sizeof(*ctx->raw_samples[c]) * sconf->max_order);
  971. return 0;
  972. fail:
  973. // damaged block, write zero for the rest of the frame
  974. zero_remaining(b, ctx->num_blocks, div_blocks, bd[0].raw_samples);
  975. zero_remaining(b, ctx->num_blocks, div_blocks, bd[1].raw_samples);
  976. return ret;
  977. }
  978. static inline int als_weighting(GetBitContext *gb, int k, int off)
  979. {
  980. int idx = av_clip(decode_rice(gb, k) + off,
  981. 0, FF_ARRAY_ELEMS(mcc_weightings) - 1);
  982. return mcc_weightings[idx];
  983. }
  984. /** Read the channel data.
  985. */
  986. static int read_channel_data(ALSDecContext *ctx, ALSChannelData *cd, int c)
  987. {
  988. GetBitContext *gb = &ctx->gb;
  989. ALSChannelData *current = cd;
  990. unsigned int channels = ctx->avctx->channels;
  991. int entries = 0;
  992. while (entries < channels && !(current->stop_flag = get_bits1(gb))) {
  993. current->master_channel = get_bits_long(gb, av_ceil_log2(channels));
  994. if (current->master_channel >= channels) {
  995. av_log(ctx->avctx, AV_LOG_ERROR, "Invalid master channel.\n");
  996. return AVERROR_INVALIDDATA;
  997. }
  998. if (current->master_channel != c) {
  999. current->time_diff_flag = get_bits1(gb);
  1000. current->weighting[0] = als_weighting(gb, 1, 16);
  1001. current->weighting[1] = als_weighting(gb, 2, 14);
  1002. current->weighting[2] = als_weighting(gb, 1, 16);
  1003. if (current->time_diff_flag) {
  1004. current->weighting[3] = als_weighting(gb, 1, 16);
  1005. current->weighting[4] = als_weighting(gb, 1, 16);
  1006. current->weighting[5] = als_weighting(gb, 1, 16);
  1007. current->time_diff_sign = get_bits1(gb);
  1008. current->time_diff_index = get_bits(gb, ctx->ltp_lag_length - 3) + 3;
  1009. }
  1010. }
  1011. current++;
  1012. entries++;
  1013. }
  1014. if (entries == channels) {
  1015. av_log(ctx->avctx, AV_LOG_ERROR, "Damaged channel data.\n");
  1016. return AVERROR_INVALIDDATA;
  1017. }
  1018. align_get_bits(gb);
  1019. return 0;
  1020. }
  1021. /** Recursively reverts the inter-channel correlation for a block.
  1022. */
  1023. static int revert_channel_correlation(ALSDecContext *ctx, ALSBlockData *bd,
  1024. ALSChannelData **cd, int *reverted,
  1025. unsigned int offset, int c)
  1026. {
  1027. ALSChannelData *ch = cd[c];
  1028. unsigned int dep = 0;
  1029. unsigned int channels = ctx->avctx->channels;
  1030. unsigned int channel_size = ctx->sconf.frame_length + ctx->sconf.max_order;
  1031. if (reverted[c])
  1032. return 0;
  1033. reverted[c] = 1;
  1034. while (dep < channels && !ch[dep].stop_flag) {
  1035. revert_channel_correlation(ctx, bd, cd, reverted, offset,
  1036. ch[dep].master_channel);
  1037. dep++;
  1038. }
  1039. if (dep == channels) {
  1040. av_log(ctx->avctx, AV_LOG_WARNING, "Invalid channel correlation.\n");
  1041. return AVERROR_INVALIDDATA;
  1042. }
  1043. bd->const_block = ctx->const_block + c;
  1044. bd->shift_lsbs = ctx->shift_lsbs + c;
  1045. bd->opt_order = ctx->opt_order + c;
  1046. bd->store_prev_samples = ctx->store_prev_samples + c;
  1047. bd->use_ltp = ctx->use_ltp + c;
  1048. bd->ltp_lag = ctx->ltp_lag + c;
  1049. bd->ltp_gain = ctx->ltp_gain[c];
  1050. bd->lpc_cof = ctx->lpc_cof[c];
  1051. bd->quant_cof = ctx->quant_cof[c];
  1052. bd->raw_samples = ctx->raw_samples[c] + offset;
  1053. for (dep = 0; !ch[dep].stop_flag; dep++) {
  1054. ptrdiff_t smp;
  1055. ptrdiff_t begin = 1;
  1056. ptrdiff_t end = bd->block_length - 1;
  1057. int64_t y;
  1058. int32_t *master = ctx->raw_samples[ch[dep].master_channel] + offset;
  1059. if (ch[dep].master_channel == c)
  1060. continue;
  1061. if (ch[dep].time_diff_flag) {
  1062. int t = ch[dep].time_diff_index;
  1063. if (ch[dep].time_diff_sign) {
  1064. t = -t;
  1065. if (begin < t) {
  1066. av_log(ctx->avctx, AV_LOG_ERROR, "begin %"PTRDIFF_SPECIFIER" smaller than time diff index %d.\n", begin, t);
  1067. return AVERROR_INVALIDDATA;
  1068. }
  1069. begin -= t;
  1070. } else {
  1071. if (end < t) {
  1072. av_log(ctx->avctx, AV_LOG_ERROR, "end %"PTRDIFF_SPECIFIER" smaller than time diff index %d.\n", end, t);
  1073. return AVERROR_INVALIDDATA;
  1074. }
  1075. end -= t;
  1076. }
  1077. if (FFMIN(begin - 1, begin - 1 + t) < ctx->raw_buffer - master ||
  1078. FFMAX(end + 1, end + 1 + t) > ctx->raw_buffer + channels * channel_size - master) {
  1079. av_log(ctx->avctx, AV_LOG_ERROR,
  1080. "sample pointer range [%p, %p] not contained in raw_buffer [%p, %p].\n",
  1081. master + FFMIN(begin - 1, begin - 1 + t), master + FFMAX(end + 1, end + 1 + t),
  1082. ctx->raw_buffer, ctx->raw_buffer + channels * channel_size);
  1083. return AVERROR_INVALIDDATA;
  1084. }
  1085. for (smp = begin; smp < end; smp++) {
  1086. y = (1 << 6) +
  1087. MUL64(ch[dep].weighting[0], master[smp - 1 ]) +
  1088. MUL64(ch[dep].weighting[1], master[smp ]) +
  1089. MUL64(ch[dep].weighting[2], master[smp + 1 ]) +
  1090. MUL64(ch[dep].weighting[3], master[smp - 1 + t]) +
  1091. MUL64(ch[dep].weighting[4], master[smp + t]) +
  1092. MUL64(ch[dep].weighting[5], master[smp + 1 + t]);
  1093. bd->raw_samples[smp] += y >> 7;
  1094. }
  1095. } else {
  1096. if (begin - 1 < ctx->raw_buffer - master ||
  1097. end + 1 > ctx->raw_buffer + channels * channel_size - master) {
  1098. av_log(ctx->avctx, AV_LOG_ERROR,
  1099. "sample pointer range [%p, %p] not contained in raw_buffer [%p, %p].\n",
  1100. master + begin - 1, master + end + 1,
  1101. ctx->raw_buffer, ctx->raw_buffer + channels * channel_size);
  1102. return AVERROR_INVALIDDATA;
  1103. }
  1104. for (smp = begin; smp < end; smp++) {
  1105. y = (1 << 6) +
  1106. MUL64(ch[dep].weighting[0], master[smp - 1]) +
  1107. MUL64(ch[dep].weighting[1], master[smp ]) +
  1108. MUL64(ch[dep].weighting[2], master[smp + 1]);
  1109. bd->raw_samples[smp] += y >> 7;
  1110. }
  1111. }
  1112. }
  1113. return 0;
  1114. }
  1115. /** multiply two softfloats and handle the rounding off
  1116. */
  1117. static SoftFloat_IEEE754 multiply(SoftFloat_IEEE754 a, SoftFloat_IEEE754 b) {
  1118. uint64_t mantissa_temp;
  1119. uint64_t mask_64;
  1120. int cutoff_bit_count;
  1121. unsigned char last_2_bits;
  1122. unsigned int mantissa;
  1123. int32_t sign;
  1124. uint32_t return_val = 0;
  1125. int bit_count = 48;
  1126. sign = a.sign ^ b.sign;
  1127. // Multiply mantissa bits in a 64-bit register
  1128. mantissa_temp = (uint64_t)a.mant * (uint64_t)b.mant;
  1129. mask_64 = (uint64_t)0x1 << 47;
  1130. // Count the valid bit count
  1131. while (!(mantissa_temp & mask_64) && mask_64) {
  1132. bit_count--;
  1133. mask_64 >>= 1;
  1134. }
  1135. // Round off
  1136. cutoff_bit_count = bit_count - 24;
  1137. if (cutoff_bit_count > 0) {
  1138. last_2_bits = (unsigned char)(((unsigned int)mantissa_temp >> (cutoff_bit_count - 1)) & 0x3 );
  1139. if ((last_2_bits == 0x3) || ((last_2_bits == 0x1) && ((unsigned int)mantissa_temp & ((0x1UL << (cutoff_bit_count - 1)) - 1)))) {
  1140. // Need to round up
  1141. mantissa_temp += (uint64_t)0x1 << cutoff_bit_count;
  1142. }
  1143. }
  1144. mantissa = (unsigned int)(mantissa_temp >> cutoff_bit_count);
  1145. // Need one more shift?
  1146. if (mantissa & 0x01000000ul) {
  1147. bit_count++;
  1148. mantissa >>= 1;
  1149. }
  1150. if (!sign) {
  1151. return_val = 0x80000000U;
  1152. }
  1153. return_val |= (a.exp + b.exp + bit_count - 47) << 23;
  1154. return_val |= mantissa;
  1155. return av_bits2sf_ieee754(return_val);
  1156. }
  1157. /** Read and decode the floating point sample data
  1158. */
  1159. static int read_diff_float_data(ALSDecContext *ctx, unsigned int ra_frame) {
  1160. AVCodecContext *avctx = ctx->avctx;
  1161. GetBitContext *gb = &ctx->gb;
  1162. SoftFloat_IEEE754 *acf = ctx->acf;
  1163. int *shift_value = ctx->shift_value;
  1164. int *last_shift_value = ctx->last_shift_value;
  1165. int *last_acf_mantissa = ctx->last_acf_mantissa;
  1166. int **raw_mantissa = ctx->raw_mantissa;
  1167. int *nbits = ctx->nbits;
  1168. unsigned char *larray = ctx->larray;
  1169. int frame_length = ctx->cur_frame_length;
  1170. SoftFloat_IEEE754 scale = av_int2sf_ieee754(0x1u, 23);
  1171. unsigned int partA_flag;
  1172. unsigned int highest_byte;
  1173. unsigned int shift_amp;
  1174. uint32_t tmp_32;
  1175. int use_acf;
  1176. int nchars;
  1177. int i;
  1178. int c;
  1179. long k;
  1180. long nbits_aligned;
  1181. unsigned long acc;
  1182. unsigned long j;
  1183. uint32_t sign;
  1184. uint32_t e;
  1185. uint32_t mantissa;
  1186. skip_bits_long(gb, 32); //num_bytes_diff_float
  1187. use_acf = get_bits1(gb);
  1188. if (ra_frame) {
  1189. memset(last_acf_mantissa, 0, avctx->channels * sizeof(*last_acf_mantissa));
  1190. memset(last_shift_value, 0, avctx->channels * sizeof(*last_shift_value) );
  1191. ff_mlz_flush_dict(ctx->mlz);
  1192. }
  1193. for (c = 0; c < avctx->channels; ++c) {
  1194. if (use_acf) {
  1195. //acf_flag
  1196. if (get_bits1(gb)) {
  1197. tmp_32 = get_bits(gb, 23);
  1198. last_acf_mantissa[c] = tmp_32;
  1199. } else {
  1200. tmp_32 = last_acf_mantissa[c];
  1201. }
  1202. acf[c] = av_bits2sf_ieee754(tmp_32);
  1203. } else {
  1204. acf[c] = FLOAT_1;
  1205. }
  1206. highest_byte = get_bits(gb, 2);
  1207. partA_flag = get_bits1(gb);
  1208. shift_amp = get_bits1(gb);
  1209. if (shift_amp) {
  1210. shift_value[c] = get_bits(gb, 8);
  1211. last_shift_value[c] = shift_value[c];
  1212. } else {
  1213. shift_value[c] = last_shift_value[c];
  1214. }
  1215. if (partA_flag) {
  1216. if (!get_bits1(gb)) { //uncompressed
  1217. for (i = 0; i < frame_length; ++i) {
  1218. if (ctx->raw_samples[c][i] == 0) {
  1219. ctx->raw_mantissa[c][i] = get_bits_long(gb, 32);
  1220. }
  1221. }
  1222. } else { //compressed
  1223. nchars = 0;
  1224. for (i = 0; i < frame_length; ++i) {
  1225. if (ctx->raw_samples[c][i] == 0) {
  1226. nchars += 4;
  1227. }
  1228. }
  1229. tmp_32 = ff_mlz_decompression(ctx->mlz, gb, nchars, larray);
  1230. if(tmp_32 != nchars) {
  1231. av_log(ctx->avctx, AV_LOG_ERROR, "Error in MLZ decompression (%"PRId32", %d).\n", tmp_32, nchars);
  1232. return AVERROR_INVALIDDATA;
  1233. }
  1234. for (i = 0; i < frame_length; ++i) {
  1235. ctx->raw_mantissa[c][i] = AV_RB32(larray);
  1236. }
  1237. }
  1238. }
  1239. //decode part B
  1240. if (highest_byte) {
  1241. for (i = 0; i < frame_length; ++i) {
  1242. if (ctx->raw_samples[c][i] != 0) {
  1243. //The following logic is taken from Tabel 14.45 and 14.46 from the ISO spec
  1244. if (av_cmp_sf_ieee754(acf[c], FLOAT_1)) {
  1245. nbits[i] = 23 - av_log2(abs(ctx->raw_samples[c][i]));
  1246. } else {
  1247. nbits[i] = 23;
  1248. }
  1249. nbits[i] = FFMIN(nbits[i], highest_byte*8);
  1250. }
  1251. }
  1252. if (!get_bits1(gb)) { //uncompressed
  1253. for (i = 0; i < frame_length; ++i) {
  1254. if (ctx->raw_samples[c][i] != 0) {
  1255. raw_mantissa[c][i] = get_bitsz(gb, nbits[i]);
  1256. }
  1257. }
  1258. } else { //compressed
  1259. nchars = 0;
  1260. for (i = 0; i < frame_length; ++i) {
  1261. if (ctx->raw_samples[c][i]) {
  1262. nchars += (int) nbits[i] / 8;
  1263. if (nbits[i] & 7) {
  1264. ++nchars;
  1265. }
  1266. }
  1267. }
  1268. tmp_32 = ff_mlz_decompression(ctx->mlz, gb, nchars, larray);
  1269. if(tmp_32 != nchars) {
  1270. av_log(ctx->avctx, AV_LOG_ERROR, "Error in MLZ decompression (%"PRId32", %d).\n", tmp_32, nchars);
  1271. return AVERROR_INVALIDDATA;
  1272. }
  1273. j = 0;
  1274. for (i = 0; i < frame_length; ++i) {
  1275. if (ctx->raw_samples[c][i]) {
  1276. if (nbits[i] & 7) {
  1277. nbits_aligned = 8 * ((unsigned int)(nbits[i] / 8) + 1);
  1278. } else {
  1279. nbits_aligned = nbits[i];
  1280. }
  1281. acc = 0;
  1282. for (k = 0; k < nbits_aligned/8; ++k) {
  1283. acc = (acc << 8) + larray[j++];
  1284. }
  1285. acc >>= (nbits_aligned - nbits[i]);
  1286. raw_mantissa[c][i] = acc;
  1287. }
  1288. }
  1289. }
  1290. }
  1291. for (i = 0; i < frame_length; ++i) {
  1292. SoftFloat_IEEE754 pcm_sf = av_int2sf_ieee754(ctx->raw_samples[c][i], 0);
  1293. pcm_sf = av_div_sf_ieee754(pcm_sf, scale);
  1294. if (ctx->raw_samples[c][i] != 0) {
  1295. if (!av_cmp_sf_ieee754(acf[c], FLOAT_1)) {
  1296. pcm_sf = multiply(acf[c], pcm_sf);
  1297. }
  1298. sign = pcm_sf.sign;
  1299. e = pcm_sf.exp;
  1300. mantissa = (pcm_sf.mant | 0x800000) + raw_mantissa[c][i];
  1301. while(mantissa >= 0x1000000) {
  1302. e++;
  1303. mantissa >>= 1;
  1304. }
  1305. if (mantissa) e += (shift_value[c] - 127);
  1306. mantissa &= 0x007fffffUL;
  1307. tmp_32 = (sign << 31) | ((e + EXP_BIAS) << 23) | (mantissa);
  1308. ctx->raw_samples[c][i] = tmp_32;
  1309. } else {
  1310. ctx->raw_samples[c][i] = raw_mantissa[c][i] & 0x007fffffUL;
  1311. }
  1312. }
  1313. align_get_bits(gb);
  1314. }
  1315. return 0;
  1316. }
  1317. /** Read the frame data.
  1318. */
  1319. static int read_frame_data(ALSDecContext *ctx, unsigned int ra_frame)
  1320. {
  1321. ALSSpecificConfig *sconf = &ctx->sconf;
  1322. AVCodecContext *avctx = ctx->avctx;
  1323. GetBitContext *gb = &ctx->gb;
  1324. unsigned int div_blocks[32]; ///< block sizes.
  1325. unsigned int c;
  1326. unsigned int js_blocks[2];
  1327. uint32_t bs_info = 0;
  1328. int ret;
  1329. // skip the size of the ra unit if present in the frame
  1330. if (sconf->ra_flag == RA_FLAG_FRAMES && ra_frame)
  1331. skip_bits_long(gb, 32);
  1332. if (sconf->mc_coding && sconf->joint_stereo) {
  1333. ctx->js_switch = get_bits1(gb);
  1334. align_get_bits(gb);
  1335. }
  1336. if (!sconf->mc_coding || ctx->js_switch) {
  1337. int independent_bs = !sconf->joint_stereo;
  1338. for (c = 0; c < avctx->channels; c++) {
  1339. js_blocks[0] = 0;
  1340. js_blocks[1] = 0;
  1341. get_block_sizes(ctx, div_blocks, &bs_info);
  1342. // if joint_stereo and block_switching is set, independent decoding
  1343. // is signaled via the first bit of bs_info
  1344. if (sconf->joint_stereo && sconf->block_switching)
  1345. if (bs_info >> 31)
  1346. independent_bs = 2;
  1347. // if this is the last channel, it has to be decoded independently
  1348. if (c == avctx->channels - 1 || (c & 1))
  1349. independent_bs = 1;
  1350. if (independent_bs) {
  1351. ret = decode_blocks_ind(ctx, ra_frame, c,
  1352. div_blocks, js_blocks);
  1353. if (ret < 0)
  1354. return ret;
  1355. independent_bs--;
  1356. } else {
  1357. ret = decode_blocks(ctx, ra_frame, c, div_blocks, js_blocks);
  1358. if (ret < 0)
  1359. return ret;
  1360. c++;
  1361. }
  1362. // store carryover raw samples
  1363. memmove(ctx->raw_samples[c] - sconf->max_order,
  1364. ctx->raw_samples[c] - sconf->max_order + sconf->frame_length,
  1365. sizeof(*ctx->raw_samples[c]) * sconf->max_order);
  1366. }
  1367. } else { // multi-channel coding
  1368. ALSBlockData bd = { 0 };
  1369. int b, ret;
  1370. int *reverted_channels = ctx->reverted_channels;
  1371. unsigned int offset = 0;
  1372. for (c = 0; c < avctx->channels; c++)
  1373. if (ctx->chan_data[c] < ctx->chan_data_buffer) {
  1374. av_log(ctx->avctx, AV_LOG_ERROR, "Invalid channel data.\n");
  1375. return AVERROR_INVALIDDATA;
  1376. }
  1377. memset(reverted_channels, 0, sizeof(*reverted_channels) * avctx->channels);
  1378. bd.ra_block = ra_frame;
  1379. bd.prev_raw_samples = ctx->prev_raw_samples;
  1380. get_block_sizes(ctx, div_blocks, &bs_info);
  1381. for (b = 0; b < ctx->num_blocks; b++) {
  1382. bd.block_length = div_blocks[b];
  1383. if (bd.block_length <= 0) {
  1384. av_log(ctx->avctx, AV_LOG_WARNING,
  1385. "Invalid block length %u in channel data!\n",
  1386. bd.block_length);
  1387. continue;
  1388. }
  1389. for (c = 0; c < avctx->channels; c++) {
  1390. bd.const_block = ctx->const_block + c;
  1391. bd.shift_lsbs = ctx->shift_lsbs + c;
  1392. bd.opt_order = ctx->opt_order + c;
  1393. bd.store_prev_samples = ctx->store_prev_samples + c;
  1394. bd.use_ltp = ctx->use_ltp + c;
  1395. bd.ltp_lag = ctx->ltp_lag + c;
  1396. bd.ltp_gain = ctx->ltp_gain[c];
  1397. bd.lpc_cof = ctx->lpc_cof[c];
  1398. bd.quant_cof = ctx->quant_cof[c];
  1399. bd.raw_samples = ctx->raw_samples[c] + offset;
  1400. bd.raw_other = NULL;
  1401. if ((ret = read_block(ctx, &bd)) < 0)
  1402. return ret;
  1403. if ((ret = read_channel_data(ctx, ctx->chan_data[c], c)) < 0)
  1404. return ret;
  1405. }
  1406. for (c = 0; c < avctx->channels; c++) {
  1407. ret = revert_channel_correlation(ctx, &bd, ctx->chan_data,
  1408. reverted_channels, offset, c);
  1409. if (ret < 0)
  1410. return ret;
  1411. }
  1412. for (c = 0; c < avctx->channels; c++) {
  1413. bd.const_block = ctx->const_block + c;
  1414. bd.shift_lsbs = ctx->shift_lsbs + c;
  1415. bd.opt_order = ctx->opt_order + c;
  1416. bd.store_prev_samples = ctx->store_prev_samples + c;
  1417. bd.use_ltp = ctx->use_ltp + c;
  1418. bd.ltp_lag = ctx->ltp_lag + c;
  1419. bd.ltp_gain = ctx->ltp_gain[c];
  1420. bd.lpc_cof = ctx->lpc_cof[c];
  1421. bd.quant_cof = ctx->quant_cof[c];
  1422. bd.raw_samples = ctx->raw_samples[c] + offset;
  1423. if ((ret = decode_block(ctx, &bd)) < 0)
  1424. return ret;
  1425. }
  1426. memset(reverted_channels, 0, avctx->channels * sizeof(*reverted_channels));
  1427. offset += div_blocks[b];
  1428. bd.ra_block = 0;
  1429. }
  1430. // store carryover raw samples
  1431. for (c = 0; c < avctx->channels; c++)
  1432. memmove(ctx->raw_samples[c] - sconf->max_order,
  1433. ctx->raw_samples[c] - sconf->max_order + sconf->frame_length,
  1434. sizeof(*ctx->raw_samples[c]) * sconf->max_order);
  1435. }
  1436. if (sconf->floating) {
  1437. read_diff_float_data(ctx, ra_frame);
  1438. }
  1439. if (get_bits_left(gb) < 0) {
  1440. av_log(ctx->avctx, AV_LOG_ERROR, "Overread %d\n", -get_bits_left(gb));
  1441. return AVERROR_INVALIDDATA;
  1442. }
  1443. return 0;
  1444. }
  1445. /** Decode an ALS frame.
  1446. */
  1447. static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame_ptr,
  1448. AVPacket *avpkt)
  1449. {
  1450. ALSDecContext *ctx = avctx->priv_data;
  1451. AVFrame *frame = data;
  1452. ALSSpecificConfig *sconf = &ctx->sconf;
  1453. const uint8_t *buffer = avpkt->data;
  1454. int buffer_size = avpkt->size;
  1455. int invalid_frame, ret;
  1456. unsigned int c, sample, ra_frame, bytes_read, shift;
  1457. if ((ret = init_get_bits8(&ctx->gb, buffer, buffer_size)) < 0)
  1458. return ret;
  1459. // In the case that the distance between random access frames is set to zero
  1460. // (sconf->ra_distance == 0) no frame is treated as a random access frame.
  1461. // For the first frame, if prediction is used, all samples used from the
  1462. // previous frame are assumed to be zero.
  1463. ra_frame = sconf->ra_distance && !(ctx->frame_id % sconf->ra_distance);
  1464. // the last frame to decode might have a different length
  1465. if (sconf->samples != 0xFFFFFFFF)
  1466. ctx->cur_frame_length = FFMIN(sconf->samples - ctx->frame_id * (uint64_t) sconf->frame_length,
  1467. sconf->frame_length);
  1468. else
  1469. ctx->cur_frame_length = sconf->frame_length;
  1470. // decode the frame data
  1471. if ((invalid_frame = read_frame_data(ctx, ra_frame)) < 0)
  1472. av_log(ctx->avctx, AV_LOG_WARNING,
  1473. "Reading frame data failed. Skipping RA unit.\n");
  1474. ctx->frame_id++;
  1475. /* get output buffer */
  1476. frame->nb_samples = ctx->cur_frame_length;
  1477. if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
  1478. return ret;
  1479. // transform decoded frame into output format
  1480. #define INTERLEAVE_OUTPUT(bps) \
  1481. { \
  1482. int##bps##_t *dest = (int##bps##_t*)frame->data[0]; \
  1483. shift = bps - ctx->avctx->bits_per_raw_sample; \
  1484. if (!ctx->cs_switch) { \
  1485. for (sample = 0; sample < ctx->cur_frame_length; sample++) \
  1486. for (c = 0; c < avctx->channels; c++) \
  1487. *dest++ = ctx->raw_samples[c][sample] << shift; \
  1488. } else { \
  1489. for (sample = 0; sample < ctx->cur_frame_length; sample++) \
  1490. for (c = 0; c < avctx->channels; c++) \
  1491. *dest++ = ctx->raw_samples[sconf->chan_pos[c]][sample] << shift; \
  1492. } \
  1493. }
  1494. if (ctx->avctx->bits_per_raw_sample <= 16) {
  1495. INTERLEAVE_OUTPUT(16)
  1496. } else {
  1497. INTERLEAVE_OUTPUT(32)
  1498. }
  1499. // update CRC
  1500. if (sconf->crc_enabled && (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL))) {
  1501. int swap = HAVE_BIGENDIAN != sconf->msb_first;
  1502. if (ctx->avctx->bits_per_raw_sample == 24) {
  1503. int32_t *src = (int32_t *)frame->data[0];
  1504. for (sample = 0;
  1505. sample < ctx->cur_frame_length * avctx->channels;
  1506. sample++) {
  1507. int32_t v;
  1508. if (swap)
  1509. v = av_bswap32(src[sample]);
  1510. else
  1511. v = src[sample];
  1512. if (!HAVE_BIGENDIAN)
  1513. v >>= 8;
  1514. ctx->crc = av_crc(ctx->crc_table, ctx->crc, (uint8_t*)(&v), 3);
  1515. }
  1516. } else {
  1517. uint8_t *crc_source;
  1518. if (swap) {
  1519. if (ctx->avctx->bits_per_raw_sample <= 16) {
  1520. int16_t *src = (int16_t*) frame->data[0];
  1521. int16_t *dest = (int16_t*) ctx->crc_buffer;
  1522. for (sample = 0;
  1523. sample < ctx->cur_frame_length * avctx->channels;
  1524. sample++)
  1525. *dest++ = av_bswap16(src[sample]);
  1526. } else {
  1527. ctx->bdsp.bswap_buf((uint32_t *) ctx->crc_buffer,
  1528. (uint32_t *) frame->data[0],
  1529. ctx->cur_frame_length * avctx->channels);
  1530. }
  1531. crc_source = ctx->crc_buffer;
  1532. } else {
  1533. crc_source = frame->data[0];
  1534. }
  1535. ctx->crc = av_crc(ctx->crc_table, ctx->crc, crc_source,
  1536. ctx->cur_frame_length * avctx->channels *
  1537. av_get_bytes_per_sample(avctx->sample_fmt));
  1538. }
  1539. // check CRC sums if this is the last frame
  1540. if (ctx->cur_frame_length != sconf->frame_length &&
  1541. ctx->crc_org != ctx->crc) {
  1542. av_log(avctx, AV_LOG_ERROR, "CRC error.\n");
  1543. if (avctx->err_recognition & AV_EF_EXPLODE)
  1544. return AVERROR_INVALIDDATA;
  1545. }
  1546. }
  1547. *got_frame_ptr = 1;
  1548. bytes_read = invalid_frame ? buffer_size :
  1549. (get_bits_count(&ctx->gb) + 7) >> 3;
  1550. return bytes_read;
  1551. }
  1552. /** Uninitialize the ALS decoder.
  1553. */
  1554. static av_cold int decode_end(AVCodecContext *avctx)
  1555. {
  1556. ALSDecContext *ctx = avctx->priv_data;
  1557. int i;
  1558. av_freep(&ctx->sconf.chan_pos);
  1559. ff_bgmc_end(&ctx->bgmc_lut, &ctx->bgmc_lut_status);
  1560. av_freep(&ctx->const_block);
  1561. av_freep(&ctx->shift_lsbs);
  1562. av_freep(&ctx->opt_order);
  1563. av_freep(&ctx->store_prev_samples);
  1564. av_freep(&ctx->use_ltp);
  1565. av_freep(&ctx->ltp_lag);
  1566. av_freep(&ctx->ltp_gain);
  1567. av_freep(&ctx->ltp_gain_buffer);
  1568. av_freep(&ctx->quant_cof);
  1569. av_freep(&ctx->lpc_cof);
  1570. av_freep(&ctx->quant_cof_buffer);
  1571. av_freep(&ctx->lpc_cof_buffer);
  1572. av_freep(&ctx->lpc_cof_reversed_buffer);
  1573. av_freep(&ctx->prev_raw_samples);
  1574. av_freep(&ctx->raw_samples);
  1575. av_freep(&ctx->raw_buffer);
  1576. av_freep(&ctx->chan_data);
  1577. av_freep(&ctx->chan_data_buffer);
  1578. av_freep(&ctx->reverted_channels);
  1579. av_freep(&ctx->crc_buffer);
  1580. if (ctx->mlz) {
  1581. av_freep(&ctx->mlz->dict);
  1582. av_freep(&ctx->mlz);
  1583. }
  1584. av_freep(&ctx->acf);
  1585. av_freep(&ctx->last_acf_mantissa);
  1586. av_freep(&ctx->shift_value);
  1587. av_freep(&ctx->last_shift_value);
  1588. if (ctx->raw_mantissa) {
  1589. for (i = 0; i < avctx->channels; i++) {
  1590. av_freep(&ctx->raw_mantissa[i]);
  1591. }
  1592. av_freep(&ctx->raw_mantissa);
  1593. }
  1594. av_freep(&ctx->larray);
  1595. av_freep(&ctx->nbits);
  1596. return 0;
  1597. }
  1598. /** Initialize the ALS decoder.
  1599. */
  1600. static av_cold int decode_init(AVCodecContext *avctx)
  1601. {
  1602. unsigned int c;
  1603. unsigned int channel_size;
  1604. int num_buffers, ret;
  1605. ALSDecContext *ctx = avctx->priv_data;
  1606. ALSSpecificConfig *sconf = &ctx->sconf;
  1607. ctx->avctx = avctx;
  1608. if (!avctx->extradata) {
  1609. av_log(avctx, AV_LOG_ERROR, "Missing required ALS extradata.\n");
  1610. return AVERROR_INVALIDDATA;
  1611. }
  1612. if ((ret = read_specific_config(ctx)) < 0) {
  1613. av_log(avctx, AV_LOG_ERROR, "Reading ALSSpecificConfig failed.\n");
  1614. goto fail;
  1615. }
  1616. if ((ret = check_specific_config(ctx)) < 0) {
  1617. goto fail;
  1618. }
  1619. if (sconf->bgmc) {
  1620. ret = ff_bgmc_init(avctx, &ctx->bgmc_lut, &ctx->bgmc_lut_status);
  1621. if (ret < 0)
  1622. goto fail;
  1623. }
  1624. if (sconf->floating) {
  1625. avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
  1626. avctx->bits_per_raw_sample = 32;
  1627. } else {
  1628. avctx->sample_fmt = sconf->resolution > 1
  1629. ? AV_SAMPLE_FMT_S32 : AV_SAMPLE_FMT_S16;
  1630. avctx->bits_per_raw_sample = (sconf->resolution + 1) * 8;
  1631. if (avctx->bits_per_raw_sample > 32) {
  1632. av_log(avctx, AV_LOG_ERROR, "Bits per raw sample %d larger than 32.\n",
  1633. avctx->bits_per_raw_sample);
  1634. ret = AVERROR_INVALIDDATA;
  1635. goto fail;
  1636. }
  1637. }
  1638. // set maximum Rice parameter for progressive decoding based on resolution
  1639. // This is not specified in 14496-3 but actually done by the reference
  1640. // codec RM22 revision 2.
  1641. ctx->s_max = sconf->resolution > 1 ? 31 : 15;
  1642. // set lag value for long-term prediction
  1643. ctx->ltp_lag_length = 8 + (avctx->sample_rate >= 96000) +
  1644. (avctx->sample_rate >= 192000);
  1645. // allocate quantized parcor coefficient buffer
  1646. num_buffers = sconf->mc_coding ? avctx->channels : 1;
  1647. ctx->quant_cof = av_malloc_array(num_buffers, sizeof(*ctx->quant_cof));
  1648. ctx->lpc_cof = av_malloc_array(num_buffers, sizeof(*ctx->lpc_cof));
  1649. ctx->quant_cof_buffer = av_malloc_array(num_buffers * sconf->max_order,
  1650. sizeof(*ctx->quant_cof_buffer));
  1651. ctx->lpc_cof_buffer = av_malloc_array(num_buffers * sconf->max_order,
  1652. sizeof(*ctx->lpc_cof_buffer));
  1653. ctx->lpc_cof_reversed_buffer = av_malloc_array(sconf->max_order,
  1654. sizeof(*ctx->lpc_cof_buffer));
  1655. if (!ctx->quant_cof || !ctx->lpc_cof ||
  1656. !ctx->quant_cof_buffer || !ctx->lpc_cof_buffer ||
  1657. !ctx->lpc_cof_reversed_buffer) {
  1658. av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
  1659. ret = AVERROR(ENOMEM);
  1660. goto fail;
  1661. }
  1662. // assign quantized parcor coefficient buffers
  1663. for (c = 0; c < num_buffers; c++) {
  1664. ctx->quant_cof[c] = ctx->quant_cof_buffer + c * sconf->max_order;
  1665. ctx->lpc_cof[c] = ctx->lpc_cof_buffer + c * sconf->max_order;
  1666. }
  1667. // allocate and assign lag and gain data buffer for ltp mode
  1668. ctx->const_block = av_malloc_array(num_buffers, sizeof(*ctx->const_block));
  1669. ctx->shift_lsbs = av_malloc_array(num_buffers, sizeof(*ctx->shift_lsbs));
  1670. ctx->opt_order = av_malloc_array(num_buffers, sizeof(*ctx->opt_order));
  1671. ctx->store_prev_samples = av_malloc_array(num_buffers, sizeof(*ctx->store_prev_samples));
  1672. ctx->use_ltp = av_mallocz_array(num_buffers, sizeof(*ctx->use_ltp));
  1673. ctx->ltp_lag = av_malloc_array(num_buffers, sizeof(*ctx->ltp_lag));
  1674. ctx->ltp_gain = av_malloc_array(num_buffers, sizeof(*ctx->ltp_gain));
  1675. ctx->ltp_gain_buffer = av_malloc_array(num_buffers * 5, sizeof(*ctx->ltp_gain_buffer));
  1676. if (!ctx->const_block || !ctx->shift_lsbs ||
  1677. !ctx->opt_order || !ctx->store_prev_samples ||
  1678. !ctx->use_ltp || !ctx->ltp_lag ||
  1679. !ctx->ltp_gain || !ctx->ltp_gain_buffer) {
  1680. av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
  1681. ret = AVERROR(ENOMEM);
  1682. goto fail;
  1683. }
  1684. for (c = 0; c < num_buffers; c++)
  1685. ctx->ltp_gain[c] = ctx->ltp_gain_buffer + c * 5;
  1686. // allocate and assign channel data buffer for mcc mode
  1687. if (sconf->mc_coding) {
  1688. ctx->chan_data_buffer = av_mallocz_array(num_buffers * num_buffers,
  1689. sizeof(*ctx->chan_data_buffer));
  1690. ctx->chan_data = av_mallocz_array(num_buffers,
  1691. sizeof(*ctx->chan_data));
  1692. ctx->reverted_channels = av_malloc_array(num_buffers,
  1693. sizeof(*ctx->reverted_channels));
  1694. if (!ctx->chan_data_buffer || !ctx->chan_data || !ctx->reverted_channels) {
  1695. av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
  1696. ret = AVERROR(ENOMEM);
  1697. goto fail;
  1698. }
  1699. for (c = 0; c < num_buffers; c++)
  1700. ctx->chan_data[c] = ctx->chan_data_buffer + c * num_buffers;
  1701. } else {
  1702. ctx->chan_data = NULL;
  1703. ctx->chan_data_buffer = NULL;
  1704. ctx->reverted_channels = NULL;
  1705. }
  1706. channel_size = sconf->frame_length + sconf->max_order;
  1707. ctx->prev_raw_samples = av_malloc_array(sconf->max_order, sizeof(*ctx->prev_raw_samples));
  1708. ctx->raw_buffer = av_mallocz_array(avctx->channels * channel_size, sizeof(*ctx->raw_buffer));
  1709. ctx->raw_samples = av_malloc_array(avctx->channels, sizeof(*ctx->raw_samples));
  1710. if (sconf->floating) {
  1711. ctx->acf = av_malloc_array(avctx->channels, sizeof(*ctx->acf));
  1712. ctx->shift_value = av_malloc_array(avctx->channels, sizeof(*ctx->shift_value));
  1713. ctx->last_shift_value = av_malloc_array(avctx->channels, sizeof(*ctx->last_shift_value));
  1714. ctx->last_acf_mantissa = av_malloc_array(avctx->channels, sizeof(*ctx->last_acf_mantissa));
  1715. ctx->raw_mantissa = av_mallocz_array(avctx->channels, sizeof(*ctx->raw_mantissa));
  1716. ctx->larray = av_malloc_array(ctx->cur_frame_length * 4, sizeof(*ctx->larray));
  1717. ctx->nbits = av_malloc_array(ctx->cur_frame_length, sizeof(*ctx->nbits));
  1718. ctx->mlz = av_mallocz(sizeof(*ctx->mlz));
  1719. if (!ctx->mlz || !ctx->acf || !ctx->shift_value || !ctx->last_shift_value
  1720. || !ctx->last_acf_mantissa || !ctx->raw_mantissa) {
  1721. av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
  1722. ret = AVERROR(ENOMEM);
  1723. goto fail;
  1724. }
  1725. ff_mlz_init_dict(avctx, ctx->mlz);
  1726. ff_mlz_flush_dict(ctx->mlz);
  1727. for (c = 0; c < avctx->channels; ++c) {
  1728. ctx->raw_mantissa[c] = av_mallocz_array(ctx->cur_frame_length, sizeof(**ctx->raw_mantissa));
  1729. }
  1730. }
  1731. // allocate previous raw sample buffer
  1732. if (!ctx->prev_raw_samples || !ctx->raw_buffer|| !ctx->raw_samples) {
  1733. av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
  1734. ret = AVERROR(ENOMEM);
  1735. goto fail;
  1736. }
  1737. // assign raw samples buffers
  1738. ctx->raw_samples[0] = ctx->raw_buffer + sconf->max_order;
  1739. for (c = 1; c < avctx->channels; c++)
  1740. ctx->raw_samples[c] = ctx->raw_samples[c - 1] + channel_size;
  1741. // allocate crc buffer
  1742. if (HAVE_BIGENDIAN != sconf->msb_first && sconf->crc_enabled &&
  1743. (avctx->err_recognition & (AV_EF_CRCCHECK|AV_EF_CAREFUL))) {
  1744. ctx->crc_buffer = av_malloc_array(ctx->cur_frame_length *
  1745. avctx->channels *
  1746. av_get_bytes_per_sample(avctx->sample_fmt),
  1747. sizeof(*ctx->crc_buffer));
  1748. if (!ctx->crc_buffer) {
  1749. av_log(avctx, AV_LOG_ERROR, "Allocating buffer memory failed.\n");
  1750. ret = AVERROR(ENOMEM);
  1751. goto fail;
  1752. }
  1753. }
  1754. ff_bswapdsp_init(&ctx->bdsp);
  1755. return 0;
  1756. fail:
  1757. decode_end(avctx);
  1758. return ret;
  1759. }
  1760. /** Flush (reset) the frame ID after seeking.
  1761. */
  1762. static av_cold void flush(AVCodecContext *avctx)
  1763. {
  1764. ALSDecContext *ctx = avctx->priv_data;
  1765. ctx->frame_id = 0;
  1766. }
  1767. AVCodec ff_als_decoder = {
  1768. .name = "als",
  1769. .long_name = NULL_IF_CONFIG_SMALL("MPEG-4 Audio Lossless Coding (ALS)"),
  1770. .type = AVMEDIA_TYPE_AUDIO,
  1771. .id = AV_CODEC_ID_MP4ALS,
  1772. .priv_data_size = sizeof(ALSDecContext),
  1773. .init = decode_init,
  1774. .close = decode_end,
  1775. .decode = decode_frame,
  1776. .flush = flush,
  1777. .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
  1778. };