6 * Copyright (C) 2013 Eric Biggers
8 * This file is part of wimlib, a library for working with WIM files.
10 * wimlib is free software; you can redistribute it and/or modify it under the
11 * terms of the GNU General Public License as published by the Free
12 * Software Foundation; either version 3 of the License, or (at your option)
15 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
16 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
17 * A PARTICULAR PURPOSE. See the GNU General Public License for more
20 * You should have received a copy of the GNU General Public License
21 * along with wimlib; if not, see http://www.gnu.org/licenses/.
25 * This is a decompressor for the LZMS compression format used by Microsoft.
26 * This format is not documented, but it is one of the formats supported by the
27 * compression API available in Windows 8, and as of Windows 8 it is one of the
28 * formats that can be used in WIM files.
30 * This decompressor only implements "raw" decompression, which decompresses a
31 * single LZMS-compressed block. This behavior is the same as that of
32 * Decompress() in the Windows 8 compression API when using a compression handle
33 * created with CreateDecompressor() with the Algorithm parameter specified as
34 * COMPRESS_ALGORITHM_LZMS | COMPRESS_RAW. Presumably, non-raw LZMS data
35 * is a container format from which the locations and sizes (both compressed and
36 * uncompressed) of the constituent blocks can be determined.
38 * An LZMS-compressed block must be read in 16-bit little endian units from both
39 * directions. One logical bitstream starts at the front of the block and
40 * proceeds forwards. Another logical bitstream starts at the end of the block
41 * and proceeds backwards. Bits read from the forwards bitstream constitute
42 * range-encoded data, whereas bits read from the backwards bitstream constitute
43 * Huffman-encoded symbols or verbatim bits. For both bitstreams, the ordering
44 * of the bits within the 16-bit coding units is such that the first bit is the
45 * high-order bit and the last bit is the low-order bit.
47 * From these two logical bitstreams, an LZMS decompressor can reconstitute the
48 * series of items that make up the LZMS data representation. Each such item
49 * may be a literal byte or a match. Matches may be either traditional LZ77
50 * matches or "delta" matches, either of which can have its offset encoded
51 * explicitly or encoded via a reference to a recently used (repeat) offset.
53 * A traditional LZ match consists of a length and offset; it asserts that the
54 * sequence of bytes beginning at the current position and extending for the
55 * length is exactly equal to the equal-length sequence of bytes at the offset
56 * back in the window. On the other hand, a delta match consists of a length,
57 * raw offset, and power. It asserts that the sequence of bytes beginning at
58 * the current position and extending for the length is equal to the bytewise
59 * sum of the two equal-length sequences of bytes (2**power) and (raw_offset *
60 * 2**power) bytes before the current position, minus bytewise the sequence of
61 * bytes beginning at (2**power + raw_offset * 2**power) bytes before the
62 * current position. Although not generally as useful as traditional LZ
63 * matches, delta matches can be helpful on some types of data. Both LZ and
64 * delta matches may overlap with the current position; in fact, the minimum
65 * offset is 1, regardless of match length.
67 * For LZ matches, up to 3 repeat offsets are allowed, similar to some other
68 * LZ-based formats such as LZX and LZMA. They must updated in an LRU fashion,
69 * except for a quirk: updates to the queue must be delayed by one LZMS item,
70 * except for the removal of a repeat match. As a result, 4 entries are
71 * actually needed in the queue, even though it is only possible to decode
72 * references to the first 3 at any given time. The queue must be initialized
73 * to the offsets {1, 2, 3, 4}.
75 * Repeat delta matches are handled similarly, but for them there are two queues
76 * updated in lock-step: one for powers and one for raw offsets. The power
77 * queue must be initialized to {0, 0, 0, 0}, and the raw offset queue must be
78 * initialized to {1, 2, 3, 4}.
80 * Bits from the range decoder must be used to disambiguate item types. The
81 * range decoder must hold two state variables: the range, which must initially
82 * be set to 0xffffffff, and the current code, which must initially be set to
83 * the first 32 bits read from the forwards bitstream. The range must be
84 * maintained above 0xffff; when it falls below 0xffff, both the range and code
85 * must be left-shifted by 16 bits and the low 16 bits of the code must be
86 * filled in with the next 16 bits from the forwards bitstream.
88 * To decode each bit, the range decoder requires a probability that is
89 * logically a real number between 0 and 1. Multiplying this probability by the
90 * current range and taking the floor gives the bound between the 0-bit region
91 * of the range and the 1-bit region of the range. However, in LZMS,
92 * probabilities are restricted to values of n/64 where n is an integer is
93 * between 1 and 63 inclusively, so the implementation may use integer
94 * operations instead. Following calculation of the bound, if the current code
95 * is in the 0-bit region, the new range becomes the current code and the
96 * decoded bit is 0; otherwise, the bound must be subtracted from both the range
97 * and the code, and the decoded bit is 1. More information about range coding
98 * can be found at https://en.wikipedia.org/wiki/Range_encoding. Furthermore,
99 * note that the LZMA format also uses range coding and has public domain code
102 * The probability used to range-decode each bit must be taken from a table, of
103 * which one instance must exist for each distinct context in which a
104 * range-decoded bit is needed. At each call of the range decoder, the
105 * appropriate probability must be obtained by indexing the appropriate
106 * probability table with the last 4 (in the context disambiguating literals
107 * from matches), 5 (in the context disambiguating LZ matches from delta
108 * matches), or 6 (in all other contexts) bits recently range-decoded in that
109 * context, ordered such that the most recently decoded bit is the low-order bit
112 * Furthermore, each probability entry itself is variable, as its value must be
113 * maintained as n/64 where n is the number of 0 bits in the most recently
114 * decoded 64 bits with that same entry. This allows the compressed
115 * representation to adapt to the input and use fewer bits to represent the most
116 * likely data; note that LZMA uses a similar scheme. Initially, the most
117 * recently 64 decoded bits for each probability entry are assumed to be
118 * 0x0000000055555555 (high order to low order); therefore, all probabilities
119 * are initially 48/64. During the course of decoding, each probability may be
120 * updated to as low as 0/64 (as a result of reading many consecutive 1 bits
121 * with that entry) or as high as 64/64 (as a result of reading many consecutive
122 * 0 bits with that entry); however, probabilities of 0/64 and 64/64 cannot be
123 * used as-is but rather must be adjusted to 1/64 and 63/64, respectively,
124 * before being used for range decoding.
126 * Representations of the LZMS items themselves must be read from the backwards
127 * bitstream. For this, there are 5 different Huffman codes used:
129 * - The literal code, used for decoding literal bytes. Each of the 256
130 * symbols represents a literal byte. This code must be rebuilt whenever
131 * 1024 symbols have been decoded with it.
133 * - The LZ offset code, used for decoding the offsets of standard LZ77
134 * matches. Each symbol represents a position slot, which corresponds to a
135 * base value and some number of extra bits which must be read and added to
136 * the base value to reconstitute the full offset. The number of symbols in
137 * this code is the number of position slots needed to represent all possible
138 * offsets in the uncompressed block. This code must be rebuilt whenever
139 * 1024 symbols have been decoded with it.
141 * - The length code, used for decoding length symbols. Each of the 54 symbols
142 * represents a length slot, which corresponds to a base value and some
143 * number of extra bits which must be read and added to the base value to
144 * reconstitute the full length. This code must be rebuilt whenever 512
145 * symbols have been decoded with it.
147 * - The delta offset code, used for decoding the offsets of delta matches.
148 * Each symbol corresponds to a position slot, which corresponds to a base
149 * value and some number of extra bits which must be read and added to the
150 * base value to reconstitute the full offset. The number of symbols in this
151 * code is equal to the number of symbols in the LZ offset code. This code
152 * must be rebuilt whenever 1024 symbols have been decoded with it.
154 * - The delta power code, used for decoding the powers of delta matches. Each
155 * of the 8 symbols corresponds to a power. This code must be rebuilt
156 * whenever 512 symbols have been decoded with it.
158 * All the LZMS Huffman codes must be built adaptively based on symbol
159 * frequencies. Initially, each code must be built assuming that all symbols
160 * have equal frequency. Following that, each code must be rebuilt whenever a
161 * certain number of symbols has been decoded with it.
163 * Like other compression formats such as XPRESS, LZX, and DEFLATE, the LZMS
164 * format requires that all Huffman codes be constructed in canonical form.
165 * This form requires that same-length codewords be lexicographically ordered
166 * the same way as the corresponding symbols and that all shorter codewords
167 * lexicographically precede longer codewords. Such a code can be constructed
168 * directly from codeword lengths, although in LZMS this is not actually
169 * necessary because the codes are built using adaptive symbol frequencies.
171 * Even with the canonical code restriction, the same frequencies can be used to
172 * construct multiple valid Huffman codes. Therefore, the decompressor needs to
173 * construct the right one. Specifically, the LZMS format requires that the
174 * Huffman code be constructed as if the well-known priority queue algorithm is
175 * used and frequency ties are always broken in favor of leaf nodes. See
176 * make_canonical_huffman_code() in compress_common.c for more information.
178 * Codewords in LZMS are guaranteed to not exceed 15 bits. The format otherwise
179 * places no restrictions on codeword length. Therefore, the Huffman code
180 * construction algorithm that a correct LZMS decompressor uses need not
181 * implement length-limited code construction. But if it does (e.g. by virtue
182 * of being shared among multiple compression algorithms), the details of how it
183 * does so are unimportant, provided that the maximum codeword length parameter
184 * is set to at least 15 bits.
186 * An LZMS-compressed block seemingly cannot have a compressed size greater than
187 * or equal to the uncompressed size. In such cases the block must be stored
190 * After all LZMS items have been decoded, the data must be postprocessed to
191 * translate absolute address encoded in x86 instructions into their original
192 * relative addresses.
194 * Details omitted above can be found in the code. Note that in the absence of
195 * an official specification there is no guarantee that this decompressor
196 * handles all possible cases.
203 #include "wimlib/compress_common.h"
204 #include "wimlib/decompressor_ops.h"
205 #include "wimlib/decompress_common.h"
206 #include "wimlib/error.h"
207 #include "wimlib/lzms.h"
208 #include "wimlib/util.h"
212 #define LZMS_DECODE_TABLE_BITS 10
214 /* Structure used for range decoding, reading bits forwards. This is the first
215 * logical bitstream mentioned above. */
216 struct lzms_range_decoder_raw {
217 /* The relevant part of the current range. Although the logical range
218 * for range decoding is a very large integer, only a small portion
219 * matters at any given time, and it can be normalized (shifted left)
220 * whenever it gets too small. */
223 /* The current position in the range encoded by the portion of the input
227 /* Pointer to the next little-endian 16-bit integer in the compressed
228 * input data (reading forwards). */
231 /* Number of 16-bit integers remaining in the compressed input data
232 * (reading forwards). */
233 size_t num_le16_remaining;
236 /* Structure used for reading raw bits backwards. This is the second logical
237 * bitstream mentioned above. */
238 struct lzms_input_bitstream {
239 /* Holding variable for bits that have been read from the compressed
240 * data. The bits are ordered from high-order to low-order. */
241 /* XXX: Without special-case code to handle reading more than 17 bits
242 * at a time, this needs to be 64 bits rather than 32 bits. */
245 /* Number of bits in @bitbuf that are used. */
246 unsigned num_filled_bits;
248 /* Pointer to the one past the next little-endian 16-bit integer in the
249 * compressed input data (reading backwards). */
252 /* Number of 16-bit integers remaining in the compressed input data
253 * (reading backwards). */
254 size_t num_le16_remaining;
257 /* Structure used for range decoding. This wraps around `struct
258 * lzms_range_decoder_raw' to use and maintain probability entries. */
259 struct lzms_range_decoder {
260 /* Pointer to the raw range decoder, which has no persistent knowledge
261 * of probabilities. Multiple lzms_range_decoder's share the same
262 * lzms_range_decoder_raw. */
263 struct lzms_range_decoder_raw *rd;
265 /* Bits recently decoded by this range decoder. This are used as in
266 * index into @prob_entries. */
269 /* Bitmask for @state to prevent its value from exceeding the number of
270 * probability entries. */
273 /* Probability entries being used for this range decoder. */
274 struct lzms_probability_entry prob_entries[LZMS_MAX_NUM_STATES];
277 /* Structure used for Huffman decoding, optionally using the decoded symbols as
278 * slots into a base table to determine how many extra bits need to be read to
279 * reconstitute the full value. */
280 struct lzms_huffman_decoder {
282 /* Bitstream to read Huffman-encoded symbols and verbatim bits from.
283 * Multiple lzms_huffman_decoder's share the same lzms_input_bitstream.
285 struct lzms_input_bitstream *is;
287 /* Pointer to the slot base table to use. It is indexed by the decoded
288 * Huffman symbol that specifies the slot. The entry specifies the base
289 * value to use, and the position of its high bit is the number of
290 * additional bits that must be read to reconstitute the full value.
292 * This member need not be set if only raw Huffman symbols are being
293 * read using this decoder. */
294 const u32 *slot_base_tab;
296 const u8 *extra_bits_tab;
298 /* Number of symbols that have been read using this code far. Reset to
299 * 0 whenever the code is rebuilt. */
302 /* When @num_syms_read reaches this number, the Huffman code must be
306 /* Number of symbols in the represented Huffman code. */
309 /* Running totals of symbol frequencies. These are diluted slightly
310 * whenever the code is rebuilt. */
311 u32 sym_freqs[LZMS_MAX_NUM_SYMS];
313 /* The length, in bits, of each symbol in the Huffman code. */
314 u8 lens[LZMS_MAX_NUM_SYMS];
316 /* The codeword of each symbol in the Huffman code. */
317 u32 codewords[LZMS_MAX_NUM_SYMS];
319 /* A table for quickly decoding symbols encoded using the Huffman code.
321 u16 decode_table[(1U << LZMS_DECODE_TABLE_BITS) + 2 * LZMS_MAX_NUM_SYMS]
322 _aligned_attribute(DECODE_TABLE_ALIGNMENT);
325 /* State of the LZMS decompressor. */
326 struct lzms_decompressor {
328 /* Pointer to the beginning of the uncompressed data buffer. */
331 /* Pointer to the next position in the uncompressed data buffer. */
334 /* Pointer to one past the end of the uncompressed data buffer. */
337 /* Range decoder, which reads bits from the beginning of the compressed
338 * block, going forwards. */
339 struct lzms_range_decoder_raw rd;
341 /* Input bitstream, which reads from the end of the compressed block,
342 * going backwards. */
343 struct lzms_input_bitstream is;
345 /* Range decoders. */
346 struct lzms_range_decoder main_range_decoder;
347 struct lzms_range_decoder match_range_decoder;
348 struct lzms_range_decoder lz_match_range_decoder;
349 struct lzms_range_decoder lz_repeat_match_range_decoders[LZMS_NUM_RECENT_OFFSETS - 1];
350 struct lzms_range_decoder delta_match_range_decoder;
351 struct lzms_range_decoder delta_repeat_match_range_decoders[LZMS_NUM_RECENT_OFFSETS - 1];
353 /* Huffman decoders. */
354 struct lzms_huffman_decoder literal_decoder;
355 struct lzms_huffman_decoder lz_offset_decoder;
356 struct lzms_huffman_decoder length_decoder;
357 struct lzms_huffman_decoder delta_power_decoder;
358 struct lzms_huffman_decoder delta_offset_decoder;
360 /* LRU (least-recently-used) queues for match information. */
361 struct lzms_lru_queues lru;
363 /* Used for postprocessing. */
364 s32 last_target_usages[65536];
367 /* Initialize the input bitstream @is to read forwards from the specified
368 * compressed data buffer @in that is @in_limit 16-bit integers long. */
370 lzms_input_bitstream_init(struct lzms_input_bitstream *is,
371 const le16 *in, size_t in_limit)
374 is->num_filled_bits = 0;
375 is->in = in + in_limit;
376 is->num_le16_remaining = in_limit;
379 /* Ensures that @num_bits bits are buffered in the input bitstream. */
381 lzms_input_bitstream_ensure_bits(struct lzms_input_bitstream *is,
384 while (is->num_filled_bits < num_bits) {
387 LZMS_ASSERT(is->num_filled_bits + 16 <= sizeof(is->bitbuf) * 8);
389 if (unlikely(is->num_le16_remaining == 0))
392 next = le16_to_cpu(*--is->in);
393 is->num_le16_remaining--;
395 is->bitbuf |= next << (sizeof(is->bitbuf) * 8 - is->num_filled_bits - 16);
396 is->num_filled_bits += 16;
402 /* Returns the next @num_bits bits that are buffered in the input bitstream. */
404 lzms_input_bitstream_peek_bits(struct lzms_input_bitstream *is,
407 LZMS_ASSERT(is->num_filled_bits >= num_bits);
408 return is->bitbuf >> (sizeof(is->bitbuf) * 8 - num_bits);
411 /* Removes the next @num_bits bits that are buffered in the input bitstream. */
413 lzms_input_bitstream_remove_bits(struct lzms_input_bitstream *is,
416 LZMS_ASSERT(is->num_filled_bits >= num_bits);
417 is->bitbuf <<= num_bits;
418 is->num_filled_bits -= num_bits;
421 /* Removes and returns the next @num_bits bits that are buffered in the input
424 lzms_input_bitstream_pop_bits(struct lzms_input_bitstream *is,
427 u32 bits = lzms_input_bitstream_peek_bits(is, num_bits);
428 lzms_input_bitstream_remove_bits(is, num_bits);
432 /* Reads the next @num_bits from the input bitstream. */
434 lzms_input_bitstream_read_bits(struct lzms_input_bitstream *is,
437 if (unlikely(lzms_input_bitstream_ensure_bits(is, num_bits)))
439 return lzms_input_bitstream_pop_bits(is, num_bits);
442 /* Initialize the range decoder @rd to read forwards from the specified
443 * compressed data buffer @in that is @in_limit 16-bit integers long. */
445 lzms_range_decoder_raw_init(struct lzms_range_decoder_raw *rd,
446 const le16 *in, size_t in_limit)
448 rd->range = 0xffffffff;
449 rd->code = ((u32)le16_to_cpu(in[0]) << 16) |
450 ((u32)le16_to_cpu(in[1]) << 0);
452 rd->num_le16_remaining = in_limit - 2;
455 /* Ensures the current range of the range decoder has at least 16 bits of
458 lzms_range_decoder_raw_normalize(struct lzms_range_decoder_raw *rd)
460 if (rd->range <= 0xffff) {
462 if (unlikely(rd->num_le16_remaining == 0))
464 rd->code = (rd->code << 16) | le16_to_cpu(*rd->in++);
465 rd->num_le16_remaining--;
470 /* Decode and return the next bit from the range decoder (raw version).
472 * @prob is the chance out of LZMS_PROBABILITY_MAX that the next bit is 0.
475 lzms_range_decoder_raw_decode_bit(struct lzms_range_decoder_raw *rd, u32 prob)
479 /* Ensure the range has at least 16 bits of precision. */
480 lzms_range_decoder_raw_normalize(rd);
482 /* Based on the probability, calculate the bound between the 0-bit
483 * region and the 1-bit region of the range. */
484 bound = (rd->range >> LZMS_PROBABILITY_BITS) * prob;
486 if (rd->code < bound) {
487 /* Current code is in the 0-bit region of the range. */
491 /* Current code is in the 1-bit region of the range. */
498 /* Decode and return the next bit from the range decoder. This wraps around
499 * lzms_range_decoder_raw_decode_bit() to handle using and updating the
500 * appropriate probability table. */
502 lzms_range_decode_bit(struct lzms_range_decoder *dec)
504 struct lzms_probability_entry *prob_entry;
508 /* Load the probability entry corresponding to the current state. */
509 prob_entry = &dec->prob_entries[dec->state];
511 /* Treat the number of zero bits in the most recently decoded
512 * LZMS_PROBABILITY_MAX bits with this probability entry as the chance,
513 * out of LZMS_PROBABILITY_MAX, that the next bit will be a 0. However,
514 * don't allow 0% or 100% probabilities. */
515 prob = prob_entry->num_recent_zero_bits;
516 if (prob == LZMS_PROBABILITY_MAX)
517 prob = LZMS_PROBABILITY_MAX - 1;
521 /* Decode the next bit. */
522 bit = lzms_range_decoder_raw_decode_bit(dec->rd, prob);
524 /* Update the state based on the newly decoded bit. */
525 dec->state = (((dec->state << 1) | bit) & dec->mask);
527 /* Update the recent bits, including the cached count of 0's. */
528 BUILD_BUG_ON(LZMS_PROBABILITY_MAX > sizeof(prob_entry->recent_bits) * 8);
530 if (prob_entry->recent_bits & (1ULL << (LZMS_PROBABILITY_MAX - 1))) {
531 /* Replacing 1 bit with 0 bit; increment the zero count.
533 prob_entry->num_recent_zero_bits++;
536 if (!(prob_entry->recent_bits & (1ULL << (LZMS_PROBABILITY_MAX - 1)))) {
537 /* Replacing 0 bit with 1 bit; decrement the zero count.
539 prob_entry->num_recent_zero_bits--;
542 prob_entry->recent_bits = (prob_entry->recent_bits << 1) | bit;
544 /* Return the decoded bit. */
549 /* Build the decoding table for a new adaptive Huffman code using the alphabet
550 * used in the specified Huffman decoder, with the symbol frequencies
553 lzms_rebuild_adaptive_huffman_code(struct lzms_huffman_decoder *dec)
556 /* XXX: This implementation makes use of code already implemented for
557 * the XPRESS and LZX compression formats. However, since for the
558 * adaptive codes used in LZMS we don't actually need the explicit codes
559 * themselves, only the decode tables, it may be possible to optimize
560 * this by somehow directly building or updating the Huffman decode
561 * table. This may be a worthwhile optimization because the adaptive
562 * codes change many times throughout a decompression run. */
563 LZMS_DEBUG("Rebuilding adaptive Huffman code (num_syms=%u)",
565 make_canonical_huffman_code(dec->num_syms, LZMS_MAX_CODEWORD_LEN,
566 dec->sym_freqs, dec->lens, dec->codewords);
567 #if defined(ENABLE_LZMS_DEBUG)
570 make_huffman_decode_table(dec->decode_table, dec->num_syms,
571 LZMS_DECODE_TABLE_BITS, dec->lens,
572 LZMS_MAX_CODEWORD_LEN);
573 LZMS_ASSERT(ret == 0);
576 /* Decode and return the next Huffman-encoded symbol from the LZMS-compressed
577 * block using the specified Huffman decoder. */
579 lzms_huffman_decode_symbol(struct lzms_huffman_decoder *dec)
581 const u16 *decode_table = dec->decode_table;
582 struct lzms_input_bitstream *is = dec->is;
587 /* The Huffman codes used in LZMS are adaptive and must be rebuilt
588 * whenever a certain number of symbols have been read. Each such
589 * rebuild uses the current symbol frequencies, but the format also
590 * requires that the symbol frequencies be halved after each code
591 * rebuild. This diminishes the effect of old symbols on the current
592 * Huffman codes, thereby causing the Huffman codes to be more locally
594 if (dec->num_syms_read == dec->rebuild_freq) {
595 lzms_rebuild_adaptive_huffman_code(dec);
596 for (unsigned i = 0; i < dec->num_syms; i++) {
597 dec->sym_freqs[i] >>= 1;
598 dec->sym_freqs[i] += 1;
600 dec->num_syms_read = 0;
603 /* XXX: Copied from read_huffsym() (decompress_common.h), since this
604 * uses a different input bitstream type. Should unify the
605 * implementations. */
606 lzms_input_bitstream_ensure_bits(is, LZMS_MAX_CODEWORD_LEN);
608 /* Index the decode table by the next table_bits bits of the input. */
609 key_bits = lzms_input_bitstream_peek_bits(is, LZMS_DECODE_TABLE_BITS);
610 entry = decode_table[key_bits];
611 if (likely(entry < 0xC000)) {
612 /* Fast case: The decode table directly provided the symbol and
613 * codeword length. The low 11 bits are the symbol, and the
614 * high 5 bits are the codeword length. */
615 lzms_input_bitstream_remove_bits(is, entry >> 11);
618 /* Slow case: The codeword for the symbol is longer than
619 * table_bits, so the symbol does not have an entry directly in
620 * the first (1 << table_bits) entries of the decode table.
621 * Traverse the appropriate binary tree bit-by-bit in order to
622 * decode the symbol. */
623 lzms_input_bitstream_remove_bits(is, LZMS_DECODE_TABLE_BITS);
625 key_bits = (entry & 0x3FFF) + lzms_input_bitstream_pop_bits(is, 1);
626 } while ((entry = decode_table[key_bits]) >= 0xC000);
630 /* Tally and return the decoded symbol. */
631 ++dec->sym_freqs[sym];
632 ++dec->num_syms_read;
636 /* Decode a number from the LZMS bitstream, encoded as a Huffman-encoded symbol
637 * specifying a "slot" (whose corresponding value is looked up in a static
638 * table) plus the number specified by a number of extra bits depending on the
641 lzms_decode_value(struct lzms_huffman_decoder *dec)
644 unsigned num_extra_bits;
647 LZMS_ASSERT(dec->slot_base_tab != NULL);
648 LZMS_ASSERT(dec->extra_bits_tab != NULL);
650 /* Read the slot (position slot, length slot, etc.), which is encoded as
651 * a Huffman symbol. */
652 slot = lzms_huffman_decode_symbol(dec);
654 /* Get the number of extra bits needed to represent the range of values
655 * that share the slot. */
656 num_extra_bits = dec->extra_bits_tab[slot];
658 /* Read the number of extra bits and add them to the slot base to form
659 * the final decoded value. */
660 extra_bits = lzms_input_bitstream_read_bits(dec->is, num_extra_bits);
661 return dec->slot_base_tab[slot] + extra_bits;
664 /* Copy a literal to the output buffer. */
666 lzms_copy_literal(struct lzms_decompressor *ctx, u8 literal)
668 *ctx->out_next++ = literal;
672 /* Validate an LZ match and copy it to the output buffer. */
674 lzms_copy_lz_match(struct lzms_decompressor *ctx, u32 length, u32 offset)
678 if (length > ctx->out_end - ctx->out_next) {
679 LZMS_DEBUG("Match overrun!");
682 if (offset > ctx->out_next - ctx->out_begin) {
683 LZMS_DEBUG("Match underrun!");
687 out_next = ctx->out_next;
689 lz_copy(out_next, length, offset, ctx->out_end);
690 ctx->out_next = out_next + length;
695 /* Validate a delta match and copy it to the output buffer. */
697 lzms_copy_delta_match(struct lzms_decompressor *ctx, u32 length,
698 u32 power, u32 raw_offset)
700 u32 offset1 = 1U << power;
701 u32 offset2 = raw_offset << power;
702 u32 offset = offset1 + offset2;
708 if (length > ctx->out_end - ctx->out_next) {
709 LZMS_DEBUG("Match overrun!");
712 if (offset > ctx->out_next - ctx->out_begin) {
713 LZMS_DEBUG("Match underrun!");
717 out_next = ctx->out_next;
718 matchptr1 = out_next - offset1;
719 matchptr2 = out_next - offset2;
720 matchptr = out_next - offset;
723 *out_next++ = *matchptr1++ + *matchptr2++ - *matchptr++;
725 ctx->out_next = out_next;
729 /* Decode a (length, offset) pair from the input. */
731 lzms_decode_lz_match(struct lzms_decompressor *ctx)
736 /* Decode the match offset. The next range-encoded bit indicates
737 * whether it's a repeat offset or an explicit offset. */
739 bit = lzms_range_decode_bit(&ctx->lz_match_range_decoder);
741 /* Explicit offset. */
742 offset = lzms_decode_value(&ctx->lz_offset_decoder);
747 for (i = 0; i < LZMS_NUM_RECENT_OFFSETS - 1; i++)
748 if (!lzms_range_decode_bit(&ctx->lz_repeat_match_range_decoders[i]))
751 offset = ctx->lru.lz.recent_offsets[i];
753 for (; i < LZMS_NUM_RECENT_OFFSETS; i++)
754 ctx->lru.lz.recent_offsets[i] = ctx->lru.lz.recent_offsets[i + 1];
757 /* Decode match length, which is always given explicitly (there is no
758 * LRU queue for repeat lengths). */
759 length = lzms_decode_value(&ctx->length_decoder);
761 ctx->lru.lz.upcoming_offset = offset;
763 LZMS_DEBUG("Decoded %s LZ match: length=%u, offset=%u",
764 (bit ? "repeat" : "explicit"), length, offset);
766 /* Validate the match and copy it to the output. */
767 return lzms_copy_lz_match(ctx, length, offset);
770 /* Decodes a "delta" match from the input. */
772 lzms_decode_delta_match(struct lzms_decompressor *ctx)
775 u32 length, power, raw_offset;
777 /* Decode the match power and raw offset. The next range-encoded bit
778 * indicates whether these data are a repeat, or given explicitly. */
780 bit = lzms_range_decode_bit(&ctx->delta_match_range_decoder);
782 power = lzms_huffman_decode_symbol(&ctx->delta_power_decoder);
783 raw_offset = lzms_decode_value(&ctx->delta_offset_decoder);
787 for (i = 0; i < LZMS_NUM_RECENT_OFFSETS - 1; i++)
788 if (!lzms_range_decode_bit(&ctx->delta_repeat_match_range_decoders[i]))
791 power = ctx->lru.delta.recent_powers[i];
792 raw_offset = ctx->lru.delta.recent_offsets[i];
794 for (; i < LZMS_NUM_RECENT_OFFSETS; i++) {
795 ctx->lru.delta.recent_powers[i] = ctx->lru.delta.recent_powers[i + 1];
796 ctx->lru.delta.recent_offsets[i] = ctx->lru.delta.recent_offsets[i + 1];
800 length = lzms_decode_value(&ctx->length_decoder);
802 ctx->lru.delta.upcoming_power = power;
803 ctx->lru.delta.upcoming_offset = raw_offset;
805 LZMS_DEBUG("Decoded %s delta match: length=%u, power=%u, raw_offset=%u",
806 (bit ? "repeat" : "explicit"), length, power, raw_offset);
808 /* Validate the match and copy it to the output. */
809 return lzms_copy_delta_match(ctx, length, power, raw_offset);
812 /* Decode an LZ or delta match. */
814 lzms_decode_match(struct lzms_decompressor *ctx)
816 if (!lzms_range_decode_bit(&ctx->match_range_decoder))
817 return lzms_decode_lz_match(ctx);
819 return lzms_decode_delta_match(ctx);
822 /* Decode a literal byte encoded using the literal Huffman code. */
824 lzms_decode_literal(struct lzms_decompressor *ctx)
826 u8 literal = lzms_huffman_decode_symbol(&ctx->literal_decoder);
827 LZMS_DEBUG("Decoded literal: 0x%02x", literal);
828 return lzms_copy_literal(ctx, literal);
831 /* Decode the next LZMS match or literal. */
833 lzms_decode_item(struct lzms_decompressor *ctx)
837 ctx->lru.lz.upcoming_offset = 0;
838 ctx->lru.delta.upcoming_power = 0;
839 ctx->lru.delta.upcoming_offset = 0;
841 if (lzms_range_decode_bit(&ctx->main_range_decoder))
842 ret = lzms_decode_match(ctx);
844 ret = lzms_decode_literal(ctx);
849 lzms_update_lru_queues(&ctx->lru);
854 lzms_init_range_decoder(struct lzms_range_decoder *dec,
855 struct lzms_range_decoder_raw *rd, u32 num_states)
859 dec->mask = num_states - 1;
860 for (u32 i = 0; i < num_states; i++) {
861 dec->prob_entries[i].num_recent_zero_bits = LZMS_INITIAL_PROBABILITY;
862 dec->prob_entries[i].recent_bits = LZMS_INITIAL_RECENT_BITS;
867 lzms_init_huffman_decoder(struct lzms_huffman_decoder *dec,
868 struct lzms_input_bitstream *is,
869 const u32 *slot_base_tab,
870 const u8 *extra_bits_tab,
872 unsigned rebuild_freq)
875 dec->slot_base_tab = slot_base_tab;
876 dec->extra_bits_tab = extra_bits_tab;
877 dec->num_syms = num_syms;
878 dec->num_syms_read = rebuild_freq;
879 dec->rebuild_freq = rebuild_freq;
880 for (unsigned i = 0; i < num_syms; i++)
881 dec->sym_freqs[i] = 1;
884 /* Prepare to decode items from an LZMS-compressed block. */
886 lzms_init_decompressor(struct lzms_decompressor *ctx,
887 const void *cdata, unsigned clen,
888 void *ubuf, unsigned ulen)
890 unsigned num_position_slots;
892 LZMS_DEBUG("Initializing decompressor (clen=%u, ulen=%u)", clen, ulen);
894 /* Initialize output pointers. */
895 ctx->out_begin = ubuf;
896 ctx->out_next = ubuf;
897 ctx->out_end = (u8*)ubuf + ulen;
899 /* Initialize the raw range decoder (reading forwards). */
900 lzms_range_decoder_raw_init(&ctx->rd, cdata, clen / 2);
902 /* Initialize the input bitstream for Huffman symbols (reading
904 lzms_input_bitstream_init(&ctx->is, cdata, clen / 2);
906 /* Calculate the number of position slots needed for this compressed
908 num_position_slots = lzms_get_position_slot(ulen - 1) + 1;
910 LZMS_DEBUG("Using %u position slots", num_position_slots);
912 /* Initialize Huffman decoders for each alphabet used in the compressed
914 lzms_init_huffman_decoder(&ctx->literal_decoder, &ctx->is,
915 NULL, NULL, LZMS_NUM_LITERAL_SYMS,
916 LZMS_LITERAL_CODE_REBUILD_FREQ);
918 lzms_init_huffman_decoder(&ctx->lz_offset_decoder, &ctx->is,
919 lzms_position_slot_base,
920 lzms_extra_position_bits,
922 LZMS_LZ_OFFSET_CODE_REBUILD_FREQ);
924 lzms_init_huffman_decoder(&ctx->length_decoder, &ctx->is,
925 lzms_length_slot_base,
926 lzms_extra_length_bits,
928 LZMS_LENGTH_CODE_REBUILD_FREQ);
930 lzms_init_huffman_decoder(&ctx->delta_offset_decoder, &ctx->is,
931 lzms_position_slot_base,
932 lzms_extra_position_bits,
934 LZMS_DELTA_OFFSET_CODE_REBUILD_FREQ);
936 lzms_init_huffman_decoder(&ctx->delta_power_decoder, &ctx->is,
937 NULL, NULL, LZMS_NUM_DELTA_POWER_SYMS,
938 LZMS_DELTA_POWER_CODE_REBUILD_FREQ);
941 /* Initialize range decoders, all of which wrap around the same
942 * lzms_range_decoder_raw. */
943 lzms_init_range_decoder(&ctx->main_range_decoder,
944 &ctx->rd, LZMS_NUM_MAIN_STATES);
946 lzms_init_range_decoder(&ctx->match_range_decoder,
947 &ctx->rd, LZMS_NUM_MATCH_STATES);
949 lzms_init_range_decoder(&ctx->lz_match_range_decoder,
950 &ctx->rd, LZMS_NUM_LZ_MATCH_STATES);
952 for (size_t i = 0; i < ARRAY_LEN(ctx->lz_repeat_match_range_decoders); i++)
953 lzms_init_range_decoder(&ctx->lz_repeat_match_range_decoders[i],
954 &ctx->rd, LZMS_NUM_LZ_REPEAT_MATCH_STATES);
956 lzms_init_range_decoder(&ctx->delta_match_range_decoder,
957 &ctx->rd, LZMS_NUM_DELTA_MATCH_STATES);
959 for (size_t i = 0; i < ARRAY_LEN(ctx->delta_repeat_match_range_decoders); i++)
960 lzms_init_range_decoder(&ctx->delta_repeat_match_range_decoders[i],
961 &ctx->rd, LZMS_NUM_DELTA_REPEAT_MATCH_STATES);
963 /* Initialize LRU match information. */
964 lzms_init_lru_queues(&ctx->lru);
966 LZMS_DEBUG("Decompressor successfully initialized");
969 /* Decode the series of literals and matches from the LZMS-compressed data.
970 * Returns 0 on success; nonzero if the compressed data is invalid. */
972 lzms_decode_items(const u8 *cdata, size_t clen, u8 *ubuf, size_t ulen,
973 struct lzms_decompressor *ctx)
975 /* Initialize the LZMS decompressor. */
976 lzms_init_decompressor(ctx, cdata, clen, ubuf, ulen);
978 /* Decode the sequence of items. */
979 while (ctx->out_next != ctx->out_end) {
980 LZMS_DEBUG("Position %u", ctx->out_next - ctx->out_begin);
981 if (lzms_decode_item(ctx))
988 lzms_decompress(const void *compressed_data, size_t compressed_size,
989 void *uncompressed_data, size_t uncompressed_size, void *_ctx)
991 struct lzms_decompressor *ctx = _ctx;
993 /* The range decoder requires that a minimum of 4 bytes of compressed
994 * data be initially available. */
995 if (compressed_size < 4) {
996 LZMS_DEBUG("Compressed size too small (got %zu, expected >= 4)",
1001 /* An LZMS-compressed data block should be evenly divisible into 16-bit
1003 if (compressed_size % 2 != 0) {
1004 LZMS_DEBUG("Compressed size not divisible by 2 (got %zu)",
1009 /* Handle the trivial case where nothing needs to be decompressed.
1010 * (Necessary because a window of size 0 does not have a valid position
1012 if (uncompressed_size == 0)
1015 /* The x86 post-processor requires that the uncompressed length fit into
1016 * a signed 32-bit integer. Also, the position slot table cannot be
1017 * searched for a position of INT32_MAX or greater. */
1018 if (uncompressed_size >= INT32_MAX) {
1019 LZMS_DEBUG("Uncompressed length too large "
1020 "(got %zu, expected < INT32_MAX)",
1025 /* Decode the literals and matches. */
1026 if (lzms_decode_items(compressed_data, compressed_size,
1027 uncompressed_data, uncompressed_size, ctx))
1030 /* Postprocess the data. */
1031 lzms_x86_filter(uncompressed_data, uncompressed_size,
1032 ctx->last_target_usages, true);
1034 LZMS_DEBUG("Decompression successful.");
1039 lzms_free_decompressor(void *_ctx)
1041 struct lzms_decompressor *ctx = _ctx;
1047 lzms_create_decompressor(size_t max_block_size, void **ctx_ret)
1049 struct lzms_decompressor *ctx;
1051 ctx = ALIGNED_MALLOC(sizeof(struct lzms_decompressor),
1052 DECODE_TABLE_ALIGNMENT);
1054 return WIMLIB_ERR_NOMEM;
1056 /* Initialize position and length slot data if not done already. */
1063 const struct decompressor_ops lzms_decompressor_ops = {
1064 .create_decompressor = lzms_create_decompressor,
1065 .decompress = lzms_decompress,
1066 .free_decompressor = lzms_free_decompressor,