pd.if.org Git - zpackage/blob - lzma/lzma/lzma_decoder.c

   1 ///////////////////////////////////////////////////////////////////////////////
   2 //
   3 /// \file       lzma_decoder.c
   4 /// \brief      LZMA decoder
   5 ///
   6 //  Authors:    Igor Pavlov
   7 //              Lasse Collin
   8 //
   9 //  This file has been put into the public domain.
  10 //  You can do whatever you want with this file.
  11 //
  12 ///////////////////////////////////////////////////////////////////////////////
  13
  14 #include "lz_decoder.h"
  15 #include "lzma_common.h"
  16 #include "lzma_decoder.h"
  17 #include "range_decoder.h"
  18
  19 // The macros unroll loops with switch statements.
  20 // Silence warnings about missing fall-through comments.
  21 #if TUKLIB_GNUC_REQ(7, 0)
  22 #       pragma GCC diagnostic ignored "-Wimplicit-fallthrough"
  23 #endif
  24
  25
  26 #ifdef HAVE_SMALL
  27
  28 // Macros for (somewhat) size-optimized code.
  29 #define seq_4(seq) seq
  30
  31 #define seq_6(seq) seq
  32
  33 #define seq_8(seq) seq
  34
  35 #define seq_len(seq) \
  36         seq ## _CHOICE, \
  37         seq ## _CHOICE2, \
  38         seq ## _BITTREE
  39
  40 #define len_decode(target, ld, pos_state, seq) \
  41 do { \
  42 case seq ## _CHOICE: \
  43         rc_if_0(ld.choice, seq ## _CHOICE) { \
  44                 rc_update_0(ld.choice); \
  45                 probs = ld.low[pos_state];\
  46                 limit = LEN_LOW_SYMBOLS; \
  47                 target = MATCH_LEN_MIN; \
  48         } else { \
  49                 rc_update_1(ld.choice); \
  50 case seq ## _CHOICE2: \
  51                 rc_if_0(ld.choice2, seq ## _CHOICE2) { \
  52                         rc_update_0(ld.choice2); \
  53                         probs = ld.mid[pos_state]; \
  54                         limit = LEN_MID_SYMBOLS; \
  55                         target = MATCH_LEN_MIN + LEN_LOW_SYMBOLS; \
  56                 } else { \
  57                         rc_update_1(ld.choice2); \
  58                         probs = ld.high; \
  59                         limit = LEN_HIGH_SYMBOLS; \
  60                         target = MATCH_LEN_MIN + LEN_LOW_SYMBOLS \
  61                                         + LEN_MID_SYMBOLS; \
  62                 } \
  63         } \
  64         symbol = 1; \
  65 case seq ## _BITTREE: \
  66         do { \
  67                 rc_bit(probs[symbol], , , seq ## _BITTREE); \
  68         } while (symbol < limit); \
  69         target += symbol - limit; \
  70 } while (0)
  71
  72 #else // HAVE_SMALL
  73
  74 // Unrolled versions
  75 #define seq_4(seq) \
  76         seq ## 0, \
  77         seq ## 1, \
  78         seq ## 2, \
  79         seq ## 3
  80
  81 #define seq_6(seq) \
  82         seq ## 0, \
  83         seq ## 1, \
  84         seq ## 2, \
  85         seq ## 3, \
  86         seq ## 4, \
  87         seq ## 5
  88
  89 #define seq_8(seq) \
  90         seq ## 0, \
  91         seq ## 1, \
  92         seq ## 2, \
  93         seq ## 3, \
  94         seq ## 4, \
  95         seq ## 5, \
  96         seq ## 6, \
  97         seq ## 7
  98
  99 #define seq_len(seq) \
 100         seq ## _CHOICE, \
 101         seq ## _LOW0, \
 102         seq ## _LOW1, \
 103         seq ## _LOW2, \
 104         seq ## _CHOICE2, \
 105         seq ## _MID0, \
 106         seq ## _MID1, \
 107         seq ## _MID2, \
 108         seq ## _HIGH0, \
 109         seq ## _HIGH1, \
 110         seq ## _HIGH2, \
 111         seq ## _HIGH3, \
 112         seq ## _HIGH4, \
 113         seq ## _HIGH5, \
 114         seq ## _HIGH6, \
 115         seq ## _HIGH7
 116
 117 #define len_decode(target, ld, pos_state, seq) \
 118 do { \
 119         symbol = 1; \
 120 case seq ## _CHOICE: \
 121         rc_if_0(ld.choice, seq ## _CHOICE) { \
 122                 rc_update_0(ld.choice); \
 123                 rc_bit_case(ld.low[pos_state][symbol], , , seq ## _LOW0); \
 124                 rc_bit_case(ld.low[pos_state][symbol], , , seq ## _LOW1); \
 125                 rc_bit_case(ld.low[pos_state][symbol], , , seq ## _LOW2); \
 126                 target = symbol - LEN_LOW_SYMBOLS + MATCH_LEN_MIN; \
 127         } else { \
 128                 rc_update_1(ld.choice); \
 129 case seq ## _CHOICE2: \
 130                 rc_if_0(ld.choice2, seq ## _CHOICE2) { \
 131                         rc_update_0(ld.choice2); \
 132                         rc_bit_case(ld.mid[pos_state][symbol], , , \
 133                                         seq ## _MID0); \
 134                         rc_bit_case(ld.mid[pos_state][symbol], , , \
 135                                         seq ## _MID1); \
 136                         rc_bit_case(ld.mid[pos_state][symbol], , , \
 137                                         seq ## _MID2); \
 138                         target = symbol - LEN_MID_SYMBOLS \
 139                                         + MATCH_LEN_MIN + LEN_LOW_SYMBOLS; \
 140                 } else { \
 141                         rc_update_1(ld.choice2); \
 142                         rc_bit_case(ld.high[symbol], , , seq ## _HIGH0); \
 143                         rc_bit_case(ld.high[symbol], , , seq ## _HIGH1); \
 144                         rc_bit_case(ld.high[symbol], , , seq ## _HIGH2); \
 145                         rc_bit_case(ld.high[symbol], , , seq ## _HIGH3); \
 146                         rc_bit_case(ld.high[symbol], , , seq ## _HIGH4); \
 147                         rc_bit_case(ld.high[symbol], , , seq ## _HIGH5); \
 148                         rc_bit_case(ld.high[symbol], , , seq ## _HIGH6); \
 149                         rc_bit_case(ld.high[symbol], , , seq ## _HIGH7); \
 150                         target = symbol - LEN_HIGH_SYMBOLS \
 151                                         + MATCH_LEN_MIN \
 152                                         + LEN_LOW_SYMBOLS + LEN_MID_SYMBOLS; \
 153                 } \
 154         } \
 155 } while (0)
 156
 157 #endif // HAVE_SMALL
 158
 159
 160 /// Length decoder probabilities; see comments in lzma_common.h.
 161 typedef struct {
 162         probability choice;
 163         probability choice2;
 164         probability low[POS_STATES_MAX][LEN_LOW_SYMBOLS];
 165         probability mid[POS_STATES_MAX][LEN_MID_SYMBOLS];
 166         probability high[LEN_HIGH_SYMBOLS];
 167 } lzma_length_decoder;
 168
 169
 170 typedef struct {
 171         ///////////////////
 172         // Probabilities //
 173         ///////////////////
 174
 175         /// Literals; see comments in lzma_common.h.
 176         probability literal[LITERAL_CODERS_MAX][LITERAL_CODER_SIZE];
 177
 178         /// If 1, it's a match. Otherwise it's a single 8-bit literal.
 179         probability is_match[STATES][POS_STATES_MAX];
 180
 181         /// If 1, it's a repeated match. The distance is one of rep0 .. rep3.
 182         probability is_rep[STATES];
 183
 184         /// If 0, distance of a repeated match is rep0.
 185         /// Otherwise check is_rep1.
 186         probability is_rep0[STATES];
 187
 188         /// If 0, distance of a repeated match is rep1.
 189         /// Otherwise check is_rep2.
 190         probability is_rep1[STATES];
 191
 192         /// If 0, distance of a repeated match is rep2. Otherwise it is rep3.
 193         probability is_rep2[STATES];
 194
 195         /// If 1, the repeated match has length of one byte. Otherwise
 196         /// the length is decoded from rep_len_decoder.
 197         probability is_rep0_long[STATES][POS_STATES_MAX];
 198
 199         /// Probability tree for the highest two bits of the match distance.
 200         /// There is a separate probability tree for match lengths of
 201         /// 2 (i.e. MATCH_LEN_MIN), 3, 4, and [5, 273].
 202         probability dist_slot[DIST_STATES][DIST_SLOTS];
 203
 204         /// Probability trees for additional bits for match distance when the
 205         /// distance is in the range [4, 127].
 206         probability pos_special[FULL_DISTANCES - DIST_MODEL_END];
 207
 208         /// Probability tree for the lowest four bits of a match distance
 209         /// that is equal to or greater than 128.
 210         probability pos_align[ALIGN_SIZE];
 211
 212         /// Length of a normal match
 213         lzma_length_decoder match_len_decoder;
 214
 215         /// Length of a repeated match
 216         lzma_length_decoder rep_len_decoder;
 217
 218         ///////////////////
 219         // Decoder state //
 220         ///////////////////
 221
 222         // Range coder
 223         lzma_range_decoder rc;
 224
 225         // Types of the most recently seen LZMA symbols
 226         lzma_lzma_state state;
 227
 228         uint32_t rep0;      ///< Distance of the latest match
 229         uint32_t rep1;      ///< Distance of second latest match
 230         uint32_t rep2;      ///< Distance of third latest match
 231         uint32_t rep3;      ///< Distance of fourth latest match
 232
 233         uint32_t pos_mask; // (1U << pb) - 1
 234         uint32_t literal_context_bits;
 235         uint32_t literal_pos_mask;
 236
 237         /// Uncompressed size as bytes, or LZMA_VLI_UNKNOWN if end of
 238         /// payload marker is expected.
 239         lzma_vli uncompressed_size;
 240
 241         ////////////////////////////////
 242         // State of incomplete symbol //
 243         ////////////////////////////////
 244
 245         /// Position where to continue the decoder loop
 246         enum {
 247                 SEQ_NORMALIZE,
 248                 SEQ_IS_MATCH,
 249                 seq_8(SEQ_LITERAL),
 250                 seq_8(SEQ_LITERAL_MATCHED),
 251                 SEQ_LITERAL_WRITE,
 252                 SEQ_IS_REP,
 253                 seq_len(SEQ_MATCH_LEN),
 254                 seq_6(SEQ_DIST_SLOT),
 255                 SEQ_DIST_MODEL,
 256                 SEQ_DIRECT,
 257                 seq_4(SEQ_ALIGN),
 258                 SEQ_EOPM,
 259                 SEQ_IS_REP0,
 260                 SEQ_SHORTREP,
 261                 SEQ_IS_REP0_LONG,
 262                 SEQ_IS_REP1,
 263                 SEQ_IS_REP2,
 264                 seq_len(SEQ_REP_LEN),
 265                 SEQ_COPY,
 266         } sequence;
 267
 268         /// Base of the current probability tree
 269         probability *probs;
 270
 271         /// Symbol being decoded. This is also used as an index variable in
 272         /// bittree decoders: probs[symbol]
 273         uint32_t symbol;
 274
 275         /// Used as a loop termination condition on bittree decoders and
 276         /// direct bits decoder.
 277         uint32_t limit;
 278
 279         /// Matched literal decoder: 0x100 or 0 to help avoiding branches.
 280         /// Bittree reverse decoders: Offset of the next bit: 1 << offset
 281         uint32_t offset;
 282
 283         /// If decoding a literal: match byte.
 284         /// If decoding a match: length of the match.
 285         uint32_t len;
 286 } lzma_lzma1_decoder;
 287
 288
 289 static lzma_ret
 290 lzma_decode(void *coder_ptr, lzma_dict *restrict dictptr,
 291                 const uint8_t *restrict in,
 292                 size_t *restrict in_pos, size_t in_size)
 293 {
 294         lzma_lzma1_decoder *restrict coder = coder_ptr;
 295
 296         ////////////////////
 297         // Initialization //
 298         ////////////////////
 299
 300         {
 301                 const lzma_ret ret = rc_read_init(
 302                                 &coder->rc, in, in_pos, in_size);
 303                 if (ret != LZMA_STREAM_END)
 304                         return ret;
 305         }
 306
 307         ///////////////
 308         // Variables //
 309         ///////////////
 310
 311         // Making local copies of often-used variables improves both
 312         // speed and readability.
 313
 314         lzma_dict dict = *dictptr;
 315
 316         const size_t dict_start = dict.pos;
 317
 318         // Range decoder
 319         rc_to_local(coder->rc, *in_pos);
 320
 321         // State
 322         uint32_t state = coder->state;
 323         uint32_t rep0 = coder->rep0;
 324         uint32_t rep1 = coder->rep1;
 325         uint32_t rep2 = coder->rep2;
 326         uint32_t rep3 = coder->rep3;
 327
 328         const uint32_t pos_mask = coder->pos_mask;
 329
 330         // These variables are actually needed only if we last time ran
 331         // out of input in the middle of the decoder loop.
 332         probability *probs = coder->probs;
 333         uint32_t symbol = coder->symbol;
 334         uint32_t limit = coder->limit;
 335         uint32_t offset = coder->offset;
 336         uint32_t len = coder->len;
 337
 338         const uint32_t literal_pos_mask = coder->literal_pos_mask;
 339         const uint32_t literal_context_bits = coder->literal_context_bits;
 340
 341         // Temporary variables
 342         uint32_t pos_state = dict.pos & pos_mask;
 343
 344         lzma_ret ret = LZMA_OK;
 345
 346         // If uncompressed size is known, there must be no end of payload
 347         // marker.
 348         const bool no_eopm = coder->uncompressed_size
 349                         != LZMA_VLI_UNKNOWN;
 350         if (no_eopm && coder->uncompressed_size < dict.limit - dict.pos)
 351                 dict.limit = dict.pos + (size_t)(coder->uncompressed_size);
 352
 353         // The main decoder loop. The "switch" is used to restart the decoder at
 354         // correct location. Once restarted, the "switch" is no longer used.
 355         switch (coder->sequence)
 356         while (true) {
 357                 // Calculate new pos_state. This is skipped on the first loop
 358                 // since we already calculated it when setting up the local
 359                 // variables.
 360                 pos_state = dict.pos & pos_mask;
 361
 362         case SEQ_NORMALIZE:
 363         case SEQ_IS_MATCH:
 364                 if (unlikely(no_eopm && dict.pos == dict.limit))
 365                         break;
 366
 367                 rc_if_0(coder->is_match[state][pos_state], SEQ_IS_MATCH) {
 368                         rc_update_0(coder->is_match[state][pos_state]);
 369
 370                         // It's a literal i.e. a single 8-bit byte.
 371
 372                         probs = literal_subcoder(coder->literal,
 373                                         literal_context_bits, literal_pos_mask,
 374                                         dict.pos, dict_get(&dict, 0));
 375                         symbol = 1;
 376
 377                         if (is_literal_state(state)) {
 378                                 // Decode literal without match byte.
 379 #ifdef HAVE_SMALL
 380         case SEQ_LITERAL:
 381                                 do {
 382                                         rc_bit(probs[symbol], , , SEQ_LITERAL);
 383                                 } while (symbol < (1 << 8));
 384 #else
 385                                 rc_bit_case(probs[symbol], , , SEQ_LITERAL0);
 386                                 rc_bit_case(probs[symbol], , , SEQ_LITERAL1);
 387                                 rc_bit_case(probs[symbol], , , SEQ_LITERAL2);
 388                                 rc_bit_case(probs[symbol], , , SEQ_LITERAL3);
 389                                 rc_bit_case(probs[symbol], , , SEQ_LITERAL4);
 390                                 rc_bit_case(probs[symbol], , , SEQ_LITERAL5);
 391                                 rc_bit_case(probs[symbol], , , SEQ_LITERAL6);
 392                                 rc_bit_case(probs[symbol], , , SEQ_LITERAL7);
 393 #endif
 394                         } else {
 395                                 // Decode literal with match byte.
 396                                 //
 397                                 // We store the byte we compare against
 398                                 // ("match byte") to "len" to minimize the
 399                                 // number of variables we need to store
 400                                 // between decoder calls.
 401                                 len = dict_get(&dict, rep0) << 1;
 402
 403                                 // The usage of "offset" allows omitting some
 404                                 // branches, which should give tiny speed
 405                                 // improvement on some CPUs. "offset" gets
 406                                 // set to zero if match_bit didn't match.
 407                                 offset = 0x100;
 408
 409 #ifdef HAVE_SMALL
 410         case SEQ_LITERAL_MATCHED:
 411                                 do {
 412                                         const uint32_t match_bit
 413                                                         = len & offset;
 414                                         const uint32_t subcoder_index
 415                                                         = offset + match_bit
 416                                                         + symbol;
 417
 418                                         rc_bit(probs[subcoder_index],
 419                                                         offset &= ~match_bit,
 420                                                         offset &= match_bit,
 421                                                         SEQ_LITERAL_MATCHED);
 422
 423                                         // It seems to be faster to do this
 424                                         // here instead of putting it to the
 425                                         // beginning of the loop and then
 426                                         // putting the "case" in the middle
 427                                         // of the loop.
 428                                         len <<= 1;
 429
 430                                 } while (symbol < (1 << 8));
 431 #else
 432                                 // Unroll the loop.
 433                                 uint32_t match_bit;
 434                                 uint32_t subcoder_index;
 435
 436 #       define d(seq) \
 437                 case seq: \
 438                         match_bit = len & offset; \
 439                         subcoder_index = offset + match_bit + symbol; \
 440                         rc_bit(probs[subcoder_index], \
 441                                         offset &= ~match_bit, \
 442                                         offset &= match_bit, \
 443                                         seq)
 444
 445                                 d(SEQ_LITERAL_MATCHED0);
 446                                 len <<= 1;
 447                                 d(SEQ_LITERAL_MATCHED1);
 448                                 len <<= 1;
 449                                 d(SEQ_LITERAL_MATCHED2);
 450                                 len <<= 1;
 451                                 d(SEQ_LITERAL_MATCHED3);
 452                                 len <<= 1;
 453                                 d(SEQ_LITERAL_MATCHED4);
 454                                 len <<= 1;
 455                                 d(SEQ_LITERAL_MATCHED5);
 456                                 len <<= 1;
 457                                 d(SEQ_LITERAL_MATCHED6);
 458                                 len <<= 1;
 459                                 d(SEQ_LITERAL_MATCHED7);
 460 #       undef d
 461 #endif
 462                         }
 463
 464                         //update_literal(state);
 465                         // Use a lookup table to update to literal state,
 466                         // since compared to other state updates, this would
 467                         // need two branches.
 468                         static const lzma_lzma_state next_state[] = {
 469                                 STATE_LIT_LIT,
 470                                 STATE_LIT_LIT,
 471                                 STATE_LIT_LIT,
 472                                 STATE_LIT_LIT,
 473                                 STATE_MATCH_LIT_LIT,
 474                                 STATE_REP_LIT_LIT,
 475                                 STATE_SHORTREP_LIT_LIT,
 476                                 STATE_MATCH_LIT,
 477                                 STATE_REP_LIT,
 478                                 STATE_SHORTREP_LIT,
 479                                 STATE_MATCH_LIT,
 480                                 STATE_REP_LIT
 481                         };
 482                         state = next_state[state];
 483
 484         case SEQ_LITERAL_WRITE:
 485                         if (unlikely(dict_put(&dict, symbol))) {
 486                                 coder->sequence = SEQ_LITERAL_WRITE;
 487                                 goto out;
 488                         }
 489
 490                         continue;
 491                 }
 492
 493                 // Instead of a new byte we are going to get a byte range
 494                 // (distance and length) which will be repeated from our
 495                 // output history.
 496
 497                 rc_update_1(coder->is_match[state][pos_state]);
 498
 499         case SEQ_IS_REP:
 500                 rc_if_0(coder->is_rep[state], SEQ_IS_REP) {
 501                         // Not a repeated match
 502                         rc_update_0(coder->is_rep[state]);
 503                         update_match(state);
 504
 505                         // The latest three match distances are kept in
 506                         // memory in case there are repeated matches.
 507                         rep3 = rep2;
 508                         rep2 = rep1;
 509                         rep1 = rep0;
 510
 511                         // Decode the length of the match.
 512                         len_decode(len, coder->match_len_decoder,
 513                                         pos_state, SEQ_MATCH_LEN);
 514
 515                         // Prepare to decode the highest two bits of the
 516                         // match distance.
 517                         probs = coder->dist_slot[get_dist_state(len)];
 518                         symbol = 1;
 519
 520 #ifdef HAVE_SMALL
 521         case SEQ_DIST_SLOT:
 522                         do {
 523                                 rc_bit(probs[symbol], , , SEQ_DIST_SLOT);
 524                         } while (symbol < DIST_SLOTS);
 525 #else
 526                         rc_bit_case(probs[symbol], , , SEQ_DIST_SLOT0);
 527                         rc_bit_case(probs[symbol], , , SEQ_DIST_SLOT1);
 528                         rc_bit_case(probs[symbol], , , SEQ_DIST_SLOT2);
 529                         rc_bit_case(probs[symbol], , , SEQ_DIST_SLOT3);
 530                         rc_bit_case(probs[symbol], , , SEQ_DIST_SLOT4);
 531                         rc_bit_case(probs[symbol], , , SEQ_DIST_SLOT5);
 532 #endif
 533                         // Get rid of the highest bit that was needed for
 534                         // indexing of the probability array.
 535                         symbol -= DIST_SLOTS;
 536                         assert(symbol <= 63);
 537
 538                         if (symbol < DIST_MODEL_START) {
 539                                 // Match distances [0, 3] have only two bits.
 540                                 rep0 = symbol;
 541                         } else {
 542                                 // Decode the lowest [1, 29] bits of
 543                                 // the match distance.
 544                                 limit = (symbol >> 1) - 1;
 545                                 assert(limit >= 1 && limit <= 30);
 546                                 rep0 = 2 + (symbol & 1);
 547
 548                                 if (symbol < DIST_MODEL_END) {
 549                                         // Prepare to decode the low bits for
 550                                         // a distance of [4, 127].
 551                                         assert(limit <= 5);
 552                                         rep0 <<= limit;
 553                                         assert(rep0 <= 96);
 554                                         // -1 is fine, because we start
 555                                         // decoding at probs[1], not probs[0].
 556                                         // NOTE: This violates the C standard,
 557                                         // since we are doing pointer
 558                                         // arithmetic past the beginning of
 559                                         // the array.
 560                                         assert((int32_t)(rep0 - symbol - 1)
 561                                                         >= -1);
 562                                         assert((int32_t)(rep0 - symbol - 1)
 563                                                         <= 82);
 564                                         probs = coder->pos_special + rep0
 565                                                         - symbol - 1;
 566                                         symbol = 1;
 567                                         offset = 0;
 568         case SEQ_DIST_MODEL:
 569 #ifdef HAVE_SMALL
 570                                         do {
 571                                                 rc_bit(probs[symbol], ,
 572                                                         rep0 += 1 << offset,
 573                                                         SEQ_DIST_MODEL);
 574                                         } while (++offset < limit);
 575 #else
 576                                         switch (limit) {
 577                                         case 5:
 578                                                 assert(offset == 0);
 579                                                 rc_bit(probs[symbol], ,
 580                                                         rep0 += 1,
 581                                                         SEQ_DIST_MODEL);
 582                                                 ++offset;
 583                                                 --limit;
 584                                         case 4:
 585                                                 rc_bit(probs[symbol], ,
 586                                                         rep0 += 1 << offset,
 587                                                         SEQ_DIST_MODEL);
 588                                                 ++offset;
 589                                                 --limit;
 590                                         case 3:
 591                                                 rc_bit(probs[symbol], ,
 592                                                         rep0 += 1 << offset,
 593                                                         SEQ_DIST_MODEL);
 594                                                 ++offset;
 595                                                 --limit;
 596                                         case 2:
 597                                                 rc_bit(probs[symbol], ,
 598                                                         rep0 += 1 << offset,
 599                                                         SEQ_DIST_MODEL);
 600                                                 ++offset;
 601                                                 --limit;
 602                                         case 1:
 603                                                 // We need "symbol" only for
 604                                                 // indexing the probability
 605                                                 // array, thus we can use
 606                                                 // rc_bit_last() here to omit
 607                                                 // the unneeded updating of
 608                                                 // "symbol".
 609                                                 rc_bit_last(probs[symbol], ,
 610                                                         rep0 += 1 << offset,
 611                                                         SEQ_DIST_MODEL);
 612                                         }
 613 #endif
 614                                 } else {
 615                                         // The distance is >= 128. Decode the
 616                                         // lower bits without probabilities
 617                                         // except the lowest four bits.
 618                                         assert(symbol >= 14);
 619                                         assert(limit >= 6);
 620                                         limit -= ALIGN_BITS;
 621                                         assert(limit >= 2);
 622         case SEQ_DIRECT:
 623                                         // Not worth manual unrolling
 624                                         do {
 625                                                 rc_direct(rep0, SEQ_DIRECT);
 626                                         } while (--limit > 0);
 627
 628                                         // Decode the lowest four bits using
 629                                         // probabilities.
 630                                         rep0 <<= ALIGN_BITS;
 631                                         symbol = 1;
 632 #ifdef HAVE_SMALL
 633                                         offset = 0;
 634         case SEQ_ALIGN:
 635                                         do {
 636                                                 rc_bit(coder->pos_align[
 637                                                                 symbol], ,
 638                                                         rep0 += 1 << offset,
 639                                                         SEQ_ALIGN);
 640                                         } while (++offset < ALIGN_BITS);
 641 #else
 642         case SEQ_ALIGN0:
 643                                         rc_bit(coder->pos_align[symbol], ,
 644                                                         rep0 += 1, SEQ_ALIGN0);
 645         case SEQ_ALIGN1:
 646                                         rc_bit(coder->pos_align[symbol], ,
 647                                                         rep0 += 2, SEQ_ALIGN1);
 648         case SEQ_ALIGN2:
 649                                         rc_bit(coder->pos_align[symbol], ,
 650                                                         rep0 += 4, SEQ_ALIGN2);
 651         case SEQ_ALIGN3:
 652                                         // Like in SEQ_DIST_MODEL, we don't
 653                                         // need "symbol" for anything else
 654                                         // than indexing the probability array.
 655                                         rc_bit_last(coder->pos_align[symbol], ,
 656                                                         rep0 += 8, SEQ_ALIGN3);
 657 #endif
 658
 659                                         if (rep0 == UINT32_MAX) {
 660                                                 // End of payload marker was
 661                                                 // found. It must not be
 662                                                 // present if uncompressed
 663                                                 // size is known.
 664                                                 if (coder->uncompressed_size
 665                                                 != LZMA_VLI_UNKNOWN) {
 666                                                         ret = LZMA_DATA_ERROR;
 667                                                         goto out;
 668                                                 }
 669
 670         case SEQ_EOPM:
 671                                                 // LZMA1 stream with
 672                                                 // end-of-payload marker.
 673                                                 rc_normalize(SEQ_EOPM);
 674                                                 ret = LZMA_STREAM_END;
 675                                                 goto out;
 676                                         }
 677                                 }
 678                         }
 679
 680                         // Validate the distance we just decoded.
 681                         if (unlikely(!dict_is_distance_valid(&dict, rep0))) {
 682                                 ret = LZMA_DATA_ERROR;
 683                                 goto out;
 684                         }
 685
 686                 } else {
 687                         rc_update_1(coder->is_rep[state]);
 688
 689                         // Repeated match
 690                         //
 691                         // The match distance is a value that we have had
 692                         // earlier. The latest four match distances are
 693                         // available as rep0, rep1, rep2 and rep3. We will
 694                         // now decode which of them is the new distance.
 695                         //
 696                         // There cannot be a match if we haven't produced
 697                         // any output, so check that first.
 698                         if (unlikely(!dict_is_distance_valid(&dict, 0))) {
 699                                 ret = LZMA_DATA_ERROR;
 700                                 goto out;
 701                         }
 702
 703         case SEQ_IS_REP0:
 704                         rc_if_0(coder->is_rep0[state], SEQ_IS_REP0) {
 705                                 rc_update_0(coder->is_rep0[state]);
 706                                 // The distance is rep0.
 707
 708         case SEQ_IS_REP0_LONG:
 709                                 rc_if_0(coder->is_rep0_long[state][pos_state],
 710                                                 SEQ_IS_REP0_LONG) {
 711                                         rc_update_0(coder->is_rep0_long[
 712                                                         state][pos_state]);
 713
 714                                         update_short_rep(state);
 715
 716         case SEQ_SHORTREP:
 717                                         if (unlikely(dict_put(&dict, dict_get(
 718                                                         &dict, rep0)))) {
 719                                                 coder->sequence = SEQ_SHORTREP;
 720                                                 goto out;
 721                                         }
 722
 723                                         continue;
 724                                 }
 725
 726                                 // Repeating more than one byte at
 727                                 // distance of rep0.
 728                                 rc_update_1(coder->is_rep0_long[
 729                                                 state][pos_state]);
 730
 731                         } else {
 732                                 rc_update_1(coder->is_rep0[state]);
 733
 734         case SEQ_IS_REP1:
 735                                 // The distance is rep1, rep2 or rep3. Once
 736                                 // we find out which one of these three, it
 737                                 // is stored to rep0 and rep1, rep2 and rep3
 738                                 // are updated accordingly.
 739                                 rc_if_0(coder->is_rep1[state], SEQ_IS_REP1) {
 740                                         rc_update_0(coder->is_rep1[state]);
 741
 742                                         const uint32_t distance = rep1;
 743                                         rep1 = rep0;
 744                                         rep0 = distance;
 745
 746                                 } else {
 747                                         rc_update_1(coder->is_rep1[state]);
 748         case SEQ_IS_REP2:
 749                                         rc_if_0(coder->is_rep2[state],
 750                                                         SEQ_IS_REP2) {
 751                                                 rc_update_0(coder->is_rep2[
 752                                                                 state]);
 753
 754                                                 const uint32_t distance = rep2;
 755                                                 rep2 = rep1;
 756                                                 rep1 = rep0;
 757                                                 rep0 = distance;
 758
 759                                         } else {
 760                                                 rc_update_1(coder->is_rep2[
 761                                                                 state]);
 762
 763                                                 const uint32_t distance = rep3;
 764                                                 rep3 = rep2;
 765                                                 rep2 = rep1;
 766                                                 rep1 = rep0;
 767                                                 rep0 = distance;
 768                                         }
 769                                 }
 770                         }
 771
 772                         update_long_rep(state);
 773
 774                         // Decode the length of the repeated match.
 775                         len_decode(len, coder->rep_len_decoder,
 776                                         pos_state, SEQ_REP_LEN);
 777                 }
 778
 779                 /////////////////////////////////
 780                 // Repeat from history buffer. //
 781                 /////////////////////////////////
 782
 783                 // The length is always between these limits. There is no way
 784                 // to trigger the algorithm to set len outside this range.
 785                 assert(len >= MATCH_LEN_MIN);
 786                 assert(len <= MATCH_LEN_MAX);
 787
 788         case SEQ_COPY:
 789                 // Repeat len bytes from distance of rep0.
 790                 if (unlikely(dict_repeat(&dict, rep0, &len))) {
 791                         coder->sequence = SEQ_COPY;
 792                         goto out;
 793                 }
 794         }
 795
 796         rc_normalize(SEQ_NORMALIZE);
 797         coder->sequence = SEQ_IS_MATCH;
 798
 799 out:
 800         // Save state
 801
 802         // NOTE: Must not copy dict.limit.
 803         dictptr->pos = dict.pos;
 804         dictptr->full = dict.full;
 805
 806         rc_from_local(coder->rc, *in_pos);
 807
 808         coder->state = state;
 809         coder->rep0 = rep0;
 810         coder->rep1 = rep1;
 811         coder->rep2 = rep2;
 812         coder->rep3 = rep3;
 813
 814         coder->probs = probs;
 815         coder->symbol = symbol;
 816         coder->limit = limit;
 817         coder->offset = offset;
 818         coder->len = len;
 819
 820         // Update the remaining amount of uncompressed data if uncompressed
 821         // size was known.
 822         if (coder->uncompressed_size != LZMA_VLI_UNKNOWN) {
 823                 coder->uncompressed_size -= dict.pos - dict_start;
 824
 825                 // Since there cannot be end of payload marker if the
 826                 // uncompressed size was known, we check here if we
 827                 // finished decoding.
 828                 if (coder->uncompressed_size == 0 && ret == LZMA_OK
 829                                 && coder->sequence != SEQ_NORMALIZE)
 830                         ret = coder->sequence == SEQ_IS_MATCH
 831                                         ? LZMA_STREAM_END : LZMA_DATA_ERROR;
 832         }
 833
 834         // We can do an additional check in the range decoder to catch some
 835         // corrupted files.
 836         if (ret == LZMA_STREAM_END) {
 837                 if (!rc_is_finished(coder->rc))
 838                         ret = LZMA_DATA_ERROR;
 839
 840                 // Reset the range decoder so that it is ready to reinitialize
 841                 // for a new LZMA2 chunk.
 842                 rc_reset(coder->rc);
 843         }
 844
 845         return ret;
 846 }
 847
 848
 849
 850 static void
 851 lzma_decoder_uncompressed(void *coder_ptr, lzma_vli uncompressed_size)
 852 {
 853         lzma_lzma1_decoder *coder = coder_ptr;
 854         coder->uncompressed_size = uncompressed_size;
 855 }
 856
 857
 858 static void
 859 lzma_decoder_reset(void *coder_ptr, const void *opt)
 860 {
 861         lzma_lzma1_decoder *coder = coder_ptr;
 862         const lzma_options_lzma *options = opt;
 863
 864         // NOTE: We assume that lc/lp/pb are valid since they were
 865         // successfully decoded with lzma_lzma_decode_properties().
 866
 867         // Calculate pos_mask. We don't need pos_bits as is for anything.
 868         coder->pos_mask = (1U << options->pb) - 1;
 869
 870         // Initialize the literal decoder.
 871         literal_init(coder->literal, options->lc, options->lp);
 872
 873         coder->literal_context_bits = options->lc;
 874         coder->literal_pos_mask = (1U << options->lp) - 1;
 875
 876         // State
 877         coder->state = STATE_LIT_LIT;
 878         coder->rep0 = 0;
 879         coder->rep1 = 0;
 880         coder->rep2 = 0;
 881         coder->rep3 = 0;
 882         coder->pos_mask = (1U << options->pb) - 1;
 883
 884         // Range decoder
 885         rc_reset(coder->rc);
 886
 887         // Bit and bittree decoders
 888         for (uint32_t i = 0; i < STATES; ++i) {
 889                 for (uint32_t j = 0; j <= coder->pos_mask; ++j) {
 890                         bit_reset(coder->is_match[i][j]);
 891                         bit_reset(coder->is_rep0_long[i][j]);
 892                 }
 893
 894                 bit_reset(coder->is_rep[i]);
 895                 bit_reset(coder->is_rep0[i]);
 896                 bit_reset(coder->is_rep1[i]);
 897                 bit_reset(coder->is_rep2[i]);
 898         }
 899
 900         for (uint32_t i = 0; i < DIST_STATES; ++i)
 901                 bittree_reset(coder->dist_slot[i], DIST_SLOT_BITS);
 902
 903         for (uint32_t i = 0; i < FULL_DISTANCES - DIST_MODEL_END; ++i)
 904                 bit_reset(coder->pos_special[i]);
 905
 906         bittree_reset(coder->pos_align, ALIGN_BITS);
 907
 908         // Len decoders (also bit/bittree)
 909         const uint32_t num_pos_states = 1U << options->pb;
 910         bit_reset(coder->match_len_decoder.choice);
 911         bit_reset(coder->match_len_decoder.choice2);
 912         bit_reset(coder->rep_len_decoder.choice);
 913         bit_reset(coder->rep_len_decoder.choice2);
 914
 915         for (uint32_t pos_state = 0; pos_state < num_pos_states; ++pos_state) {
 916                 bittree_reset(coder->match_len_decoder.low[pos_state],
 917                                 LEN_LOW_BITS);
 918                 bittree_reset(coder->match_len_decoder.mid[pos_state],
 919                                 LEN_MID_BITS);
 920
 921                 bittree_reset(coder->rep_len_decoder.low[pos_state],
 922                                 LEN_LOW_BITS);
 923                 bittree_reset(coder->rep_len_decoder.mid[pos_state],
 924                                 LEN_MID_BITS);
 925         }
 926
 927         bittree_reset(coder->match_len_decoder.high, LEN_HIGH_BITS);
 928         bittree_reset(coder->rep_len_decoder.high, LEN_HIGH_BITS);
 929
 930         coder->sequence = SEQ_IS_MATCH;
 931         coder->probs = NULL;
 932         coder->symbol = 0;
 933         coder->limit = 0;
 934         coder->offset = 0;
 935         coder->len = 0;
 936
 937         return;
 938 }
 939
 940
 941 extern lzma_ret
 942 lzma_lzma_decoder_create(lzma_lz_decoder *lz, const lzma_allocator *allocator,
 943                 const void *opt, lzma_lz_options *lz_options)
 944 {
 945         if (lz->coder == NULL) {
 946                 lz->coder = lzma_alloc(sizeof(lzma_lzma1_decoder), allocator);
 947                 if (lz->coder == NULL)
 948                         return LZMA_MEM_ERROR;
 949
 950                 lz->code = &lzma_decode;
 951                 lz->reset = &lzma_decoder_reset;
 952                 lz->set_uncompressed = &lzma_decoder_uncompressed;
 953         }
 954
 955         // All dictionary sizes are OK here. LZ decoder will take care of
 956         // the special cases.
 957         const lzma_options_lzma *options = opt;
 958         lz_options->dict_size = options->dict_size;
 959         lz_options->preset_dict = options->preset_dict;
 960         lz_options->preset_dict_size = options->preset_dict_size;
 961
 962         return LZMA_OK;
 963 }
 964
 965
 966 /// Allocate and initialize LZMA decoder. This is used only via LZ
 967 /// initialization (lzma_lzma_decoder_init() passes function pointer to
 968 /// the LZ initialization).
 969 static lzma_ret
 970 lzma_decoder_init(lzma_lz_decoder *lz, const lzma_allocator *allocator,
 971                 const void *options, lzma_lz_options *lz_options)
 972 {
 973         if (!is_lclppb_valid(options))
 974                 return LZMA_PROG_ERROR;
 975
 976         return_if_error(lzma_lzma_decoder_create(
 977                         lz, allocator, options, lz_options));
 978
 979         lzma_decoder_reset(lz->coder, options);
 980         lzma_decoder_uncompressed(lz->coder, LZMA_VLI_UNKNOWN);
 981
 982         return LZMA_OK;
 983 }
 984
 985
 986 extern lzma_ret
 987 lzma_lzma_decoder_init(lzma_next_coder *next, const lzma_allocator *allocator,
 988                 const lzma_filter_info *filters)
 989 {
 990         // LZMA can only be the last filter in the chain. This is enforced
 991         // by the raw_decoder initialization.
 992         assert(filters[1].init == NULL);
 993
 994         return lzma_lz_decoder_init(next, allocator, filters,
 995                         &lzma_decoder_init);
 996 }
 997
 998
 999 extern bool
1000 lzma_lzma_lclppb_decode(lzma_options_lzma *options, uint8_t byte)
1001 {
1002         if (byte > (4 * 5 + 4) * 9 + 8)
1003                 return true;
1004
1005         // See the file format specification to understand this.
1006         options->pb = byte / (9 * 5);
1007         byte -= options->pb * 9 * 5;
1008         options->lp = byte / 9;
1009         options->lc = byte - options->lp * 9;
1010
1011         return options->lc + options->lp > LZMA_LCLP_MAX;
1012 }
1013
1014
1015 extern uint64_t
1016 lzma_lzma_decoder_memusage_nocheck(const void *options)
1017 {
1018         const lzma_options_lzma *const opt = options;
1019         return sizeof(lzma_lzma1_decoder)
1020                         + lzma_lz_decoder_memusage(opt->dict_size);
1021 }
1022
1023
1024 extern uint64_t
1025 lzma_lzma_decoder_memusage(const void *options)
1026 {
1027         if (!is_lclppb_valid(options))
1028                 return UINT64_MAX;
1029
1030         return lzma_lzma_decoder_memusage_nocheck(options);
1031 }
1032
1033
1034 extern lzma_ret
1035 lzma_lzma_props_decode(void **options, const lzma_allocator *allocator,
1036                 const uint8_t *props, size_t props_size)
1037 {
1038         if (props_size != 5)
1039                 return LZMA_OPTIONS_ERROR;
1040
1041         lzma_options_lzma *opt
1042                         = lzma_alloc(sizeof(lzma_options_lzma), allocator);
1043         if (opt == NULL)
1044                 return LZMA_MEM_ERROR;
1045
1046         if (lzma_lzma_lclppb_decode(opt, props[0]))
1047                 goto error;
1048
1049         // All dictionary sizes are accepted, including zero. LZ decoder
1050         // will automatically use a dictionary at least a few KiB even if
1051         // a smaller dictionary is requested.
1052         opt->dict_size = unaligned_read32le(props + 1);
1053
1054         opt->preset_dict = NULL;
1055         opt->preset_dict_size = 0;
1056
1057         *options = opt;
1058
1059         return LZMA_OK;
1060
1061 error:
1062         lzma_free(opt, allocator);
1063         return LZMA_OPTIONS_ERROR;
1064 }