6 * Copyright (C) 2013, 2014 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: inserting anything to the front of the queue must be
70 * delayed by one LZMS item. The reason for this is presumably that there is
71 * almost no reason to code the same match offset twice in a row, since you
72 * might as well have coded a longer match at that offset. For this same
73 * reason, it also is a requirement that when an offset in the queue is used,
74 * that offset is removed from the queue immediately (and made pending for
75 * front-insertion after the following decoded item), and everything to the
76 * right is shifted left one queue slot. This creates a need for an "overflow"
77 * fourth entry in the queue, even though it is only possible to decode
78 * references to the first 3 entries at any given time. The queue must be
79 * initialized to the offsets {1, 2, 3, 4}.
81 * Repeat delta matches are handled similarly, but for them there are two queues
82 * updated in lock-step: one for powers and one for raw offsets. The power
83 * queue must be initialized to {0, 0, 0, 0}, and the raw offset queue must be
84 * initialized to {1, 2, 3, 4}.
86 * Bits from the range decoder must be used to disambiguate item types. The
87 * range decoder must hold two state variables: the range, which must initially
88 * be set to 0xffffffff, and the current code, which must initially be set to
89 * the first 32 bits read from the forwards bitstream. The range must be
90 * maintained above 0xffff; when it falls below 0xffff, both the range and code
91 * must be left-shifted by 16 bits and the low 16 bits of the code must be
92 * filled in with the next 16 bits from the forwards bitstream.
94 * To decode each bit, the range decoder requires a probability that is
95 * logically a real number between 0 and 1. Multiplying this probability by the
96 * current range and taking the floor gives the bound between the 0-bit region
97 * of the range and the 1-bit region of the range. However, in LZMS,
98 * probabilities are restricted to values of n/64 where n is an integer is
99 * between 1 and 63 inclusively, so the implementation may use integer
100 * operations instead. Following calculation of the bound, if the current code
101 * is in the 0-bit region, the new range becomes the current code and the
102 * decoded bit is 0; otherwise, the bound must be subtracted from both the range
103 * and the code, and the decoded bit is 1. More information about range coding
104 * can be found at https://en.wikipedia.org/wiki/Range_encoding. Furthermore,
105 * note that the LZMA format also uses range coding and has public domain code
108 * The probability used to range-decode each bit must be taken from a table, of
109 * which one instance must exist for each distinct context in which a
110 * range-decoded bit is needed. At each call of the range decoder, the
111 * appropriate probability must be obtained by indexing the appropriate
112 * probability table with the last 4 (in the context disambiguating literals
113 * from matches), 5 (in the context disambiguating LZ matches from delta
114 * matches), or 6 (in all other contexts) bits recently range-decoded in that
115 * context, ordered such that the most recently decoded bit is the low-order bit
118 * Furthermore, each probability entry itself is variable, as its value must be
119 * maintained as n/64 where n is the number of 0 bits in the most recently
120 * decoded 64 bits with that same entry. This allows the compressed
121 * representation to adapt to the input and use fewer bits to represent the most
122 * likely data; note that LZMA uses a similar scheme. Initially, the most
123 * recently 64 decoded bits for each probability entry are assumed to be
124 * 0x0000000055555555 (high order to low order); therefore, all probabilities
125 * are initially 48/64. During the course of decoding, each probability may be
126 * updated to as low as 0/64 (as a result of reading many consecutive 1 bits
127 * with that entry) or as high as 64/64 (as a result of reading many consecutive
128 * 0 bits with that entry); however, probabilities of 0/64 and 64/64 cannot be
129 * used as-is but rather must be adjusted to 1/64 and 63/64, respectively,
130 * before being used for range decoding.
132 * Representations of the LZMS items themselves must be read from the backwards
133 * bitstream. For this, there are 5 different Huffman codes used:
135 * - The literal code, used for decoding literal bytes. Each of the 256
136 * symbols represents a literal byte. This code must be rebuilt whenever
137 * 1024 symbols have been decoded with it.
139 * - The LZ offset code, used for decoding the offsets of standard LZ77
140 * matches. Each symbol represents an offset slot, which corresponds to a
141 * base value and some number of extra bits which must be read and added to
142 * the base value to reconstitute the full offset. The number of symbols in
143 * this code is the number of offset slots needed to represent all possible
144 * offsets in the uncompressed block. This code must be rebuilt whenever
145 * 1024 symbols have been decoded with it.
147 * - The length code, used for decoding length symbols. Each of the 54 symbols
148 * represents a length slot, which corresponds to a base value and some
149 * number of extra bits which must be read and added to the base value to
150 * reconstitute the full length. This code must be rebuilt whenever 512
151 * symbols have been decoded with it.
153 * - The delta offset code, used for decoding the offsets of delta matches.
154 * Each symbol corresponds to an offset slot, which corresponds to a base
155 * value and some number of extra bits which must be read and added to the
156 * base value to reconstitute the full offset. The number of symbols in this
157 * code is equal to the number of symbols in the LZ offset code. This code
158 * must be rebuilt whenever 1024 symbols have been decoded with it.
160 * - The delta power code, used for decoding the powers of delta matches. Each
161 * of the 8 symbols corresponds to a power. This code must be rebuilt
162 * whenever 512 symbols have been decoded with it.
164 * All the LZMS Huffman codes must be built adaptively based on symbol
165 * frequencies. Initially, each code must be built assuming that all symbols
166 * have equal frequency. Following that, each code must be rebuilt whenever a
167 * certain number of symbols has been decoded with it.
169 * Like other compression formats such as XPRESS, LZX, and DEFLATE, the LZMS
170 * format requires that all Huffman codes be constructed in canonical form.
171 * This form requires that same-length codewords be lexicographically ordered
172 * the same way as the corresponding symbols and that all shorter codewords
173 * lexicographically precede longer codewords. Such a code can be constructed
174 * directly from codeword lengths, although in LZMS this is not actually
175 * necessary because the codes are built using adaptive symbol frequencies.
177 * Even with the canonical code restriction, the same frequencies can be used to
178 * construct multiple valid Huffman codes. Therefore, the decompressor needs to
179 * construct the right one. Specifically, the LZMS format requires that the
180 * Huffman code be constructed as if the well-known priority queue algorithm is
181 * used and frequency ties are always broken in favor of leaf nodes. See
182 * make_canonical_huffman_code() in compress_common.c for more information.
184 * Codewords in LZMS are guaranteed to not exceed 15 bits. The format otherwise
185 * places no restrictions on codeword length. Therefore, the Huffman code
186 * construction algorithm that a correct LZMS decompressor uses need not
187 * implement length-limited code construction. But if it does (e.g. by virtue
188 * of being shared among multiple compression algorithms), the details of how it
189 * does so are unimportant, provided that the maximum codeword length parameter
190 * is set to at least 15 bits.
192 * An LZMS-compressed block seemingly cannot have a compressed size greater than
193 * or equal to the uncompressed size. In such cases the block must be stored
196 * After all LZMS items have been decoded, the data must be postprocessed to
197 * translate absolute address encoded in x86 instructions into their original
198 * relative addresses.
200 * Details omitted above can be found in the code. Note that in the absence of
201 * an official specification there is no guarantee that this decompressor
202 * handles all possible cases.
209 #include "wimlib/compress_common.h"
210 #include "wimlib/decompressor_ops.h"
211 #include "wimlib/decompress_common.h"
212 #include "wimlib/error.h"
213 #include "wimlib/lzms.h"
214 #include "wimlib/util.h"
218 #define LZMS_DECODE_TABLE_BITS 10
220 /* Structure used for range decoding, reading bits forwards. This is the first
221 * logical bitstream mentioned above. */
222 struct lzms_range_decoder_raw {
223 /* The relevant part of the current range. Although the logical range
224 * for range decoding is a very large integer, only a small portion
225 * matters at any given time, and it can be normalized (shifted left)
226 * whenever it gets too small. */
229 /* The current position in the range encoded by the portion of the input
233 /* Pointer to the next little-endian 16-bit integer in the compressed
234 * input data (reading forwards). */
237 /* Number of 16-bit integers remaining in the compressed input data
238 * (reading forwards). */
239 size_t num_le16_remaining;
242 /* Structure used for reading raw bits backwards. This is the second logical
243 * bitstream mentioned above. */
244 struct lzms_input_bitstream {
245 /* Holding variable for bits that have been read from the compressed
246 * data. The bits are ordered from high-order to low-order. */
247 /* XXX: Without special-case code to handle reading more than 17 bits
248 * at a time, this needs to be 64 bits rather than 32 bits. */
251 /* Number of bits in @bitbuf that are used. */
252 unsigned num_filled_bits;
254 /* Pointer to the one past the next little-endian 16-bit integer in the
255 * compressed input data (reading backwards). */
258 /* Number of 16-bit integers remaining in the compressed input data
259 * (reading backwards). */
260 size_t num_le16_remaining;
263 /* Structure used for range decoding. This wraps around `struct
264 * lzms_range_decoder_raw' to use and maintain probability entries. */
265 struct lzms_range_decoder {
266 /* Pointer to the raw range decoder, which has no persistent knowledge
267 * of probabilities. Multiple lzms_range_decoder's share the same
268 * lzms_range_decoder_raw. */
269 struct lzms_range_decoder_raw *rd;
271 /* Bits recently decoded by this range decoder. This are used as in
272 * index into @prob_entries. */
275 /* Bitmask for @state to prevent its value from exceeding the number of
276 * probability entries. */
279 /* Probability entries being used for this range decoder. */
280 struct lzms_probability_entry prob_entries[LZMS_MAX_NUM_STATES];
283 /* Structure used for Huffman decoding, optionally using the decoded symbols as
284 * slots into a base table to determine how many extra bits need to be read to
285 * reconstitute the full value. */
286 struct lzms_huffman_decoder {
288 /* Bitstream to read Huffman-encoded symbols and verbatim bits from.
289 * Multiple lzms_huffman_decoder's share the same lzms_input_bitstream.
291 struct lzms_input_bitstream *is;
293 /* Pointer to the slot base table to use. It is indexed by the decoded
294 * Huffman symbol that specifies the slot. The entry specifies the base
295 * value to use, and the position of its high bit is the number of
296 * additional bits that must be read to reconstitute the full value.
298 * This member need not be set if only raw Huffman symbols are being
299 * read using this decoder. */
300 const u32 *slot_base_tab;
302 const u8 *extra_bits_tab;
304 /* Number of symbols that have been read using this code far. Reset to
305 * 0 whenever the code is rebuilt. */
308 /* When @num_syms_read reaches this number, the Huffman code must be
312 /* Number of symbols in the represented Huffman code. */
315 /* Running totals of symbol frequencies. These are diluted slightly
316 * whenever the code is rebuilt. */
317 u32 sym_freqs[LZMS_MAX_NUM_SYMS];
319 /* The length, in bits, of each symbol in the Huffman code. */
320 u8 lens[LZMS_MAX_NUM_SYMS];
322 /* The codeword of each symbol in the Huffman code. */
323 u32 codewords[LZMS_MAX_NUM_SYMS];
325 /* A table for quickly decoding symbols encoded using the Huffman code.
327 u16 decode_table[(1U << LZMS_DECODE_TABLE_BITS) + 2 * LZMS_MAX_NUM_SYMS]
328 _aligned_attribute(DECODE_TABLE_ALIGNMENT);
331 /* State of the LZMS decompressor. */
332 struct lzms_decompressor {
334 /* Pointer to the beginning of the uncompressed data buffer. */
337 /* Pointer to the next position in the uncompressed data buffer. */
340 /* Pointer to one past the end of the uncompressed data buffer. */
343 /* Range decoder, which reads bits from the beginning of the compressed
344 * block, going forwards. */
345 struct lzms_range_decoder_raw rd;
347 /* Input bitstream, which reads from the end of the compressed block,
348 * going backwards. */
349 struct lzms_input_bitstream is;
351 /* Range decoders. */
352 struct lzms_range_decoder main_range_decoder;
353 struct lzms_range_decoder match_range_decoder;
354 struct lzms_range_decoder lz_match_range_decoder;
355 struct lzms_range_decoder lz_repeat_match_range_decoders[LZMS_NUM_RECENT_OFFSETS - 1];
356 struct lzms_range_decoder delta_match_range_decoder;
357 struct lzms_range_decoder delta_repeat_match_range_decoders[LZMS_NUM_RECENT_OFFSETS - 1];
359 /* Huffman decoders. */
360 struct lzms_huffman_decoder literal_decoder;
361 struct lzms_huffman_decoder lz_offset_decoder;
362 struct lzms_huffman_decoder length_decoder;
363 struct lzms_huffman_decoder delta_power_decoder;
364 struct lzms_huffman_decoder delta_offset_decoder;
366 /* LRU (least-recently-used) queues for match information. */
367 struct lzms_lru_queues lru;
369 /* Used for postprocessing. */
370 s32 last_target_usages[65536];
373 /* Initialize the input bitstream @is to read forwards from the specified
374 * compressed data buffer @in that is @in_limit 16-bit integers long. */
376 lzms_input_bitstream_init(struct lzms_input_bitstream *is,
377 const le16 *in, size_t in_limit)
380 is->num_filled_bits = 0;
381 is->in = in + in_limit;
382 is->num_le16_remaining = in_limit;
385 /* Ensures that @num_bits bits are buffered in the input bitstream. */
387 lzms_input_bitstream_ensure_bits(struct lzms_input_bitstream *is,
390 while (is->num_filled_bits < num_bits) {
393 LZMS_ASSERT(is->num_filled_bits + 16 <= sizeof(is->bitbuf) * 8);
395 if (unlikely(is->num_le16_remaining == 0))
398 next = le16_to_cpu(*--is->in);
399 is->num_le16_remaining--;
401 is->bitbuf |= next << (sizeof(is->bitbuf) * 8 - is->num_filled_bits - 16);
402 is->num_filled_bits += 16;
408 /* Returns the next @num_bits bits that are buffered in the input bitstream. */
410 lzms_input_bitstream_peek_bits(struct lzms_input_bitstream *is,
413 LZMS_ASSERT(is->num_filled_bits >= num_bits);
414 return is->bitbuf >> (sizeof(is->bitbuf) * 8 - num_bits);
417 /* Removes the next @num_bits bits that are buffered in the input bitstream. */
419 lzms_input_bitstream_remove_bits(struct lzms_input_bitstream *is,
422 LZMS_ASSERT(is->num_filled_bits >= num_bits);
423 is->bitbuf <<= num_bits;
424 is->num_filled_bits -= num_bits;
427 /* Removes and returns the next @num_bits bits that are buffered in the input
430 lzms_input_bitstream_pop_bits(struct lzms_input_bitstream *is,
433 u32 bits = lzms_input_bitstream_peek_bits(is, num_bits);
434 lzms_input_bitstream_remove_bits(is, num_bits);
438 /* Reads the next @num_bits from the input bitstream. */
440 lzms_input_bitstream_read_bits(struct lzms_input_bitstream *is,
443 if (unlikely(lzms_input_bitstream_ensure_bits(is, num_bits)))
445 return lzms_input_bitstream_pop_bits(is, num_bits);
448 /* Initialize the range decoder @rd to read forwards from the specified
449 * compressed data buffer @in that is @in_limit 16-bit integers long. */
451 lzms_range_decoder_raw_init(struct lzms_range_decoder_raw *rd,
452 const le16 *in, size_t in_limit)
454 rd->range = 0xffffffff;
455 rd->code = ((u32)le16_to_cpu(in[0]) << 16) |
456 ((u32)le16_to_cpu(in[1]) << 0);
458 rd->num_le16_remaining = in_limit - 2;
461 /* Ensures the current range of the range decoder has at least 16 bits of
464 lzms_range_decoder_raw_normalize(struct lzms_range_decoder_raw *rd)
466 if (rd->range <= 0xffff) {
468 if (unlikely(rd->num_le16_remaining == 0))
470 rd->code = (rd->code << 16) | le16_to_cpu(*rd->in++);
471 rd->num_le16_remaining--;
476 /* Decode and return the next bit from the range decoder (raw version).
478 * @prob is the chance out of LZMS_PROBABILITY_MAX that the next bit is 0.
481 lzms_range_decoder_raw_decode_bit(struct lzms_range_decoder_raw *rd, u32 prob)
485 /* Ensure the range has at least 16 bits of precision. */
486 lzms_range_decoder_raw_normalize(rd);
488 /* Based on the probability, calculate the bound between the 0-bit
489 * region and the 1-bit region of the range. */
490 bound = (rd->range >> LZMS_PROBABILITY_BITS) * prob;
492 if (rd->code < bound) {
493 /* Current code is in the 0-bit region of the range. */
497 /* Current code is in the 1-bit region of the range. */
504 /* Decode and return the next bit from the range decoder. This wraps around
505 * lzms_range_decoder_raw_decode_bit() to handle using and updating the
506 * appropriate probability table. */
508 lzms_range_decode_bit(struct lzms_range_decoder *dec)
510 struct lzms_probability_entry *prob_entry;
514 /* Load the probability entry corresponding to the current state. */
515 prob_entry = &dec->prob_entries[dec->state];
517 /* Get the probability that the next bit is 0. */
518 prob = lzms_get_probability(prob_entry);
520 /* Decode the next bit. */
521 bit = lzms_range_decoder_raw_decode_bit(dec->rd, prob);
523 /* Update the state and probability entry based on the decoded bit. */
524 dec->state = (((dec->state << 1) | bit) & dec->mask);
525 lzms_update_probability_entry(prob_entry, bit);
527 /* Return the decoded bit. */
532 /* Build the decoding table for a new adaptive Huffman code using the alphabet
533 * used in the specified Huffman decoder, with the symbol frequencies
536 lzms_rebuild_adaptive_huffman_code(struct lzms_huffman_decoder *dec)
539 /* XXX: This implementation makes use of code already implemented for
540 * the XPRESS and LZX compression formats. However, since for the
541 * adaptive codes used in LZMS we don't actually need the explicit codes
542 * themselves, only the decode tables, it may be possible to optimize
543 * this by somehow directly building or updating the Huffman decode
544 * table. This may be a worthwhile optimization because the adaptive
545 * codes change many times throughout a decompression run. */
546 LZMS_DEBUG("Rebuilding adaptive Huffman code (num_syms=%u)",
548 make_canonical_huffman_code(dec->num_syms, LZMS_MAX_CODEWORD_LEN,
549 dec->sym_freqs, dec->lens, dec->codewords);
550 #if defined(ENABLE_LZMS_DEBUG)
553 make_huffman_decode_table(dec->decode_table, dec->num_syms,
554 LZMS_DECODE_TABLE_BITS, dec->lens,
555 LZMS_MAX_CODEWORD_LEN);
556 LZMS_ASSERT(ret == 0);
559 /* Decode and return the next Huffman-encoded symbol from the LZMS-compressed
560 * block using the specified Huffman decoder. */
562 lzms_huffman_decode_symbol(struct lzms_huffman_decoder *dec)
564 const u16 *decode_table = dec->decode_table;
565 struct lzms_input_bitstream *is = dec->is;
570 /* The Huffman codes used in LZMS are adaptive and must be rebuilt
571 * whenever a certain number of symbols have been read. Each such
572 * rebuild uses the current symbol frequencies, but the format also
573 * requires that the symbol frequencies be halved after each code
574 * rebuild. This diminishes the effect of old symbols on the current
575 * Huffman codes, thereby causing the Huffman codes to be more locally
577 if (dec->num_syms_read == dec->rebuild_freq) {
578 lzms_rebuild_adaptive_huffman_code(dec);
579 for (unsigned i = 0; i < dec->num_syms; i++) {
580 dec->sym_freqs[i] >>= 1;
581 dec->sym_freqs[i] += 1;
583 dec->num_syms_read = 0;
586 /* XXX: Copied from read_huffsym() (decompress_common.h), since this
587 * uses a different input bitstream type. Should unify the
588 * implementations. */
589 lzms_input_bitstream_ensure_bits(is, LZMS_MAX_CODEWORD_LEN);
591 /* Index the decode table by the next table_bits bits of the input. */
592 key_bits = lzms_input_bitstream_peek_bits(is, LZMS_DECODE_TABLE_BITS);
593 entry = decode_table[key_bits];
594 if (likely(entry < 0xC000)) {
595 /* Fast case: The decode table directly provided the symbol and
596 * codeword length. The low 11 bits are the symbol, and the
597 * high 5 bits are the codeword length. */
598 lzms_input_bitstream_remove_bits(is, entry >> 11);
601 /* Slow case: The codeword for the symbol is longer than
602 * table_bits, so the symbol does not have an entry directly in
603 * the first (1 << table_bits) entries of the decode table.
604 * Traverse the appropriate binary tree bit-by-bit in order to
605 * decode the symbol. */
606 lzms_input_bitstream_remove_bits(is, LZMS_DECODE_TABLE_BITS);
608 key_bits = (entry & 0x3FFF) + lzms_input_bitstream_pop_bits(is, 1);
609 } while ((entry = decode_table[key_bits]) >= 0xC000);
613 /* Tally and return the decoded symbol. */
614 ++dec->sym_freqs[sym];
615 ++dec->num_syms_read;
619 /* Decode a number from the LZMS bitstream, encoded as a Huffman-encoded symbol
620 * specifying a "slot" (whose corresponding value is looked up in a static
621 * table) plus the number specified by a number of extra bits depending on the
624 lzms_decode_value(struct lzms_huffman_decoder *dec)
627 unsigned num_extra_bits;
630 LZMS_ASSERT(dec->slot_base_tab != NULL);
631 LZMS_ASSERT(dec->extra_bits_tab != NULL);
633 /* Read the slot (offset slot, length slot, etc.), which is encoded as a
635 slot = lzms_huffman_decode_symbol(dec);
637 /* Get the number of extra bits needed to represent the range of values
638 * that share the slot. */
639 num_extra_bits = dec->extra_bits_tab[slot];
641 /* Read the number of extra bits and add them to the slot base to form
642 * the final decoded value. */
643 extra_bits = lzms_input_bitstream_read_bits(dec->is, num_extra_bits);
644 return dec->slot_base_tab[slot] + extra_bits;
647 /* Copy a literal to the output buffer. */
649 lzms_copy_literal(struct lzms_decompressor *ctx, u8 literal)
651 *ctx->out_next++ = literal;
655 /* Validate an LZ match and copy it to the output buffer. */
657 lzms_copy_lz_match(struct lzms_decompressor *ctx, u32 length, u32 offset)
661 if (length > ctx->out_end - ctx->out_next) {
662 LZMS_DEBUG("Match overrun!");
665 if (offset > ctx->out_next - ctx->out_begin) {
666 LZMS_DEBUG("Match underrun!");
670 out_next = ctx->out_next;
672 lz_copy(out_next, length, offset, ctx->out_end);
673 ctx->out_next = out_next + length;
678 /* Validate a delta match and copy it to the output buffer. */
680 lzms_copy_delta_match(struct lzms_decompressor *ctx, u32 length,
681 u32 power, u32 raw_offset)
683 u32 offset1 = 1U << power;
684 u32 offset2 = raw_offset << power;
685 u32 offset = offset1 + offset2;
691 if (length > ctx->out_end - ctx->out_next) {
692 LZMS_DEBUG("Match overrun!");
695 if (offset > ctx->out_next - ctx->out_begin) {
696 LZMS_DEBUG("Match underrun!");
700 out_next = ctx->out_next;
701 matchptr1 = out_next - offset1;
702 matchptr2 = out_next - offset2;
703 matchptr = out_next - offset;
706 *out_next++ = *matchptr1++ + *matchptr2++ - *matchptr++;
708 ctx->out_next = out_next;
712 /* Decode a (length, offset) pair from the input. */
714 lzms_decode_lz_match(struct lzms_decompressor *ctx)
719 /* Decode the match offset. The next range-encoded bit indicates
720 * whether it's a repeat offset or an explicit offset. */
722 bit = lzms_range_decode_bit(&ctx->lz_match_range_decoder);
724 /* Explicit offset. */
725 offset = lzms_decode_value(&ctx->lz_offset_decoder);
730 for (i = 0; i < LZMS_NUM_RECENT_OFFSETS - 1; i++)
731 if (!lzms_range_decode_bit(&ctx->lz_repeat_match_range_decoders[i]))
734 offset = ctx->lru.lz.recent_offsets[i];
736 for (; i < LZMS_NUM_RECENT_OFFSETS; i++)
737 ctx->lru.lz.recent_offsets[i] = ctx->lru.lz.recent_offsets[i + 1];
740 /* Decode match length, which is always given explicitly (there is no
741 * LRU queue for repeat lengths). */
742 length = lzms_decode_value(&ctx->length_decoder);
744 ctx->lru.lz.upcoming_offset = offset;
746 LZMS_DEBUG("Decoded %s LZ match: length=%u, offset=%u",
747 (bit ? "repeat" : "explicit"), length, offset);
749 /* Validate the match and copy it to the output. */
750 return lzms_copy_lz_match(ctx, length, offset);
753 /* Decodes a "delta" match from the input. */
755 lzms_decode_delta_match(struct lzms_decompressor *ctx)
758 u32 length, power, raw_offset;
760 /* Decode the match power and raw offset. The next range-encoded bit
761 * indicates whether these data are a repeat, or given explicitly. */
763 bit = lzms_range_decode_bit(&ctx->delta_match_range_decoder);
765 power = lzms_huffman_decode_symbol(&ctx->delta_power_decoder);
766 raw_offset = lzms_decode_value(&ctx->delta_offset_decoder);
770 for (i = 0; i < LZMS_NUM_RECENT_OFFSETS - 1; i++)
771 if (!lzms_range_decode_bit(&ctx->delta_repeat_match_range_decoders[i]))
774 power = ctx->lru.delta.recent_powers[i];
775 raw_offset = ctx->lru.delta.recent_offsets[i];
777 for (; i < LZMS_NUM_RECENT_OFFSETS; i++) {
778 ctx->lru.delta.recent_powers[i] = ctx->lru.delta.recent_powers[i + 1];
779 ctx->lru.delta.recent_offsets[i] = ctx->lru.delta.recent_offsets[i + 1];
783 length = lzms_decode_value(&ctx->length_decoder);
785 ctx->lru.delta.upcoming_power = power;
786 ctx->lru.delta.upcoming_offset = raw_offset;
788 LZMS_DEBUG("Decoded %s delta match: length=%u, power=%u, raw_offset=%u",
789 (bit ? "repeat" : "explicit"), length, power, raw_offset);
791 /* Validate the match and copy it to the output. */
792 return lzms_copy_delta_match(ctx, length, power, raw_offset);
795 /* Decode an LZ or delta match. */
797 lzms_decode_match(struct lzms_decompressor *ctx)
799 if (!lzms_range_decode_bit(&ctx->match_range_decoder))
800 return lzms_decode_lz_match(ctx);
802 return lzms_decode_delta_match(ctx);
805 /* Decode a literal byte encoded using the literal Huffman code. */
807 lzms_decode_literal(struct lzms_decompressor *ctx)
809 u8 literal = lzms_huffman_decode_symbol(&ctx->literal_decoder);
810 LZMS_DEBUG("Decoded literal: 0x%02x", literal);
811 return lzms_copy_literal(ctx, literal);
814 /* Decode the next LZMS match or literal. */
816 lzms_decode_item(struct lzms_decompressor *ctx)
820 ctx->lru.lz.upcoming_offset = 0;
821 ctx->lru.delta.upcoming_power = 0;
822 ctx->lru.delta.upcoming_offset = 0;
824 if (lzms_range_decode_bit(&ctx->main_range_decoder))
825 ret = lzms_decode_match(ctx);
827 ret = lzms_decode_literal(ctx);
832 lzms_update_lru_queues(&ctx->lru);
837 lzms_init_range_decoder(struct lzms_range_decoder *dec,
838 struct lzms_range_decoder_raw *rd, u32 num_states)
842 dec->mask = num_states - 1;
843 for (u32 i = 0; i < num_states; i++) {
844 dec->prob_entries[i].num_recent_zero_bits = LZMS_INITIAL_PROBABILITY;
845 dec->prob_entries[i].recent_bits = LZMS_INITIAL_RECENT_BITS;
850 lzms_init_huffman_decoder(struct lzms_huffman_decoder *dec,
851 struct lzms_input_bitstream *is,
852 const u32 *slot_base_tab,
853 const u8 *extra_bits_tab,
855 unsigned rebuild_freq)
858 dec->slot_base_tab = slot_base_tab;
859 dec->extra_bits_tab = extra_bits_tab;
860 dec->num_syms = num_syms;
861 dec->num_syms_read = rebuild_freq;
862 dec->rebuild_freq = rebuild_freq;
863 for (unsigned i = 0; i < num_syms; i++)
864 dec->sym_freqs[i] = 1;
867 /* Prepare to decode items from an LZMS-compressed block. */
869 lzms_init_decompressor(struct lzms_decompressor *ctx,
870 const void *cdata, unsigned clen,
871 void *ubuf, unsigned ulen)
873 unsigned num_offset_slots;
875 LZMS_DEBUG("Initializing decompressor (clen=%u, ulen=%u)", clen, ulen);
877 /* Initialize output pointers. */
878 ctx->out_begin = ubuf;
879 ctx->out_next = ubuf;
880 ctx->out_end = (u8*)ubuf + ulen;
882 /* Initialize the raw range decoder (reading forwards). */
883 lzms_range_decoder_raw_init(&ctx->rd, cdata, clen / 2);
885 /* Initialize the input bitstream for Huffman symbols (reading
887 lzms_input_bitstream_init(&ctx->is, cdata, clen / 2);
889 /* Calculate the number of offset slots needed for this compressed
891 num_offset_slots = lzms_get_offset_slot(ulen - 1) + 1;
893 LZMS_DEBUG("Using %u offset slots", num_offset_slots);
895 /* Initialize Huffman decoders for each alphabet used in the compressed
897 lzms_init_huffman_decoder(&ctx->literal_decoder, &ctx->is,
898 NULL, NULL, LZMS_NUM_LITERAL_SYMS,
899 LZMS_LITERAL_CODE_REBUILD_FREQ);
901 lzms_init_huffman_decoder(&ctx->lz_offset_decoder, &ctx->is,
902 lzms_offset_slot_base,
903 lzms_extra_offset_bits,
905 LZMS_LZ_OFFSET_CODE_REBUILD_FREQ);
907 lzms_init_huffman_decoder(&ctx->length_decoder, &ctx->is,
908 lzms_length_slot_base,
909 lzms_extra_length_bits,
911 LZMS_LENGTH_CODE_REBUILD_FREQ);
913 lzms_init_huffman_decoder(&ctx->delta_offset_decoder, &ctx->is,
914 lzms_offset_slot_base,
915 lzms_extra_offset_bits,
917 LZMS_DELTA_OFFSET_CODE_REBUILD_FREQ);
919 lzms_init_huffman_decoder(&ctx->delta_power_decoder, &ctx->is,
920 NULL, NULL, LZMS_NUM_DELTA_POWER_SYMS,
921 LZMS_DELTA_POWER_CODE_REBUILD_FREQ);
924 /* Initialize range decoders, all of which wrap around the same
925 * lzms_range_decoder_raw. */
926 lzms_init_range_decoder(&ctx->main_range_decoder,
927 &ctx->rd, LZMS_NUM_MAIN_STATES);
929 lzms_init_range_decoder(&ctx->match_range_decoder,
930 &ctx->rd, LZMS_NUM_MATCH_STATES);
932 lzms_init_range_decoder(&ctx->lz_match_range_decoder,
933 &ctx->rd, LZMS_NUM_LZ_MATCH_STATES);
935 for (size_t i = 0; i < ARRAY_LEN(ctx->lz_repeat_match_range_decoders); i++)
936 lzms_init_range_decoder(&ctx->lz_repeat_match_range_decoders[i],
937 &ctx->rd, LZMS_NUM_LZ_REPEAT_MATCH_STATES);
939 lzms_init_range_decoder(&ctx->delta_match_range_decoder,
940 &ctx->rd, LZMS_NUM_DELTA_MATCH_STATES);
942 for (size_t i = 0; i < ARRAY_LEN(ctx->delta_repeat_match_range_decoders); i++)
943 lzms_init_range_decoder(&ctx->delta_repeat_match_range_decoders[i],
944 &ctx->rd, LZMS_NUM_DELTA_REPEAT_MATCH_STATES);
946 /* Initialize LRU match information. */
947 lzms_init_lru_queues(&ctx->lru);
949 LZMS_DEBUG("Decompressor successfully initialized");
952 /* Decode the series of literals and matches from the LZMS-compressed data.
953 * Returns 0 on success; nonzero if the compressed data is invalid. */
955 lzms_decode_items(const u8 *cdata, size_t clen, u8 *ubuf, size_t ulen,
956 struct lzms_decompressor *ctx)
958 /* Initialize the LZMS decompressor. */
959 lzms_init_decompressor(ctx, cdata, clen, ubuf, ulen);
961 /* Decode the sequence of items. */
962 while (ctx->out_next != ctx->out_end) {
963 LZMS_DEBUG("Position %u", ctx->out_next - ctx->out_begin);
964 if (lzms_decode_item(ctx))
971 lzms_decompress(const void *compressed_data, size_t compressed_size,
972 void *uncompressed_data, size_t uncompressed_size, void *_ctx)
974 struct lzms_decompressor *ctx = _ctx;
976 /* The range decoder requires that a minimum of 4 bytes of compressed
977 * data be initially available. */
978 if (compressed_size < 4) {
979 LZMS_DEBUG("Compressed size too small (got %zu, expected >= 4)",
984 /* An LZMS-compressed data block should be evenly divisible into 16-bit
986 if (compressed_size % 2 != 0) {
987 LZMS_DEBUG("Compressed size not divisible by 2 (got %zu)",
992 /* Handle the trivial case where nothing needs to be decompressed.
993 * (Necessary because a window of size 0 does not have a valid offset
995 if (uncompressed_size == 0)
998 /* Decode the literals and matches. */
999 if (lzms_decode_items(compressed_data, compressed_size,
1000 uncompressed_data, uncompressed_size, ctx))
1003 /* Postprocess the data. */
1004 lzms_x86_filter(uncompressed_data, uncompressed_size,
1005 ctx->last_target_usages, true);
1007 LZMS_DEBUG("Decompression successful.");
1012 lzms_free_decompressor(void *_ctx)
1014 struct lzms_decompressor *ctx = _ctx;
1020 lzms_create_decompressor(size_t max_block_size, void **ctx_ret)
1022 struct lzms_decompressor *ctx;
1024 /* The x86 post-processor requires that the uncompressed length fit into
1025 * a signed 32-bit integer. Also, the offset slot table cannot be
1026 * searched for an offset of INT32_MAX or greater. */
1027 if (max_block_size >= INT32_MAX)
1028 return WIMLIB_ERR_INVALID_PARAM;
1030 ctx = ALIGNED_MALLOC(sizeof(struct lzms_decompressor),
1031 DECODE_TABLE_ALIGNMENT);
1033 return WIMLIB_ERR_NOMEM;
1035 /* Initialize offset and length slot data if not done already. */
1042 const struct decompressor_ops lzms_decompressor_ops = {
1043 .create_decompressor = lzms_create_decompressor,
1044 .decompress = lzms_decompress,
1045 .free_decompressor = lzms_free_decompressor,