4 * LZX compression routines
8 * Copyright (C) 2012, 2013 Eric Biggers
10 * This file is part of wimlib, a library for working with WIM files.
12 * wimlib is free software; you can redistribute it and/or modify it under the
13 * terms of the GNU General Public License as published by the Free
14 * Software Foundation; either version 3 of the License, or (at your option)
17 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
18 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
19 * A PARTICULAR PURPOSE. See the GNU General Public License for more
22 * You should have received a copy of the GNU General Public License
23 * along with wimlib; if not, see http://www.gnu.org/licenses/.
28 * This file contains a compressor for the LZX compression format, as used in
29 * the WIM file format.
34 * First, the primary reference for the LZX compression format is the
35 * specification released by Microsoft.
37 * Second, the comments in lzx-decompress.c provide some more information about
38 * the LZX compression format, including errors in the Microsoft specification.
40 * Do note that LZX shares many similarities with DEFLATE, the algorithm used by
41 * zlib and gzip. Both LZX and DEFLATE use LZ77 matching and Huffman coding,
42 * and certain other details are quite similar, such as the method for storing
43 * Huffman codes. However, some of the main differences are:
45 * - LZX preprocesses the data to attempt to make x86 machine code slightly more
46 * compressible before attempting to compress it further.
47 * - LZX uses a "main" alphabet which combines literals and matches, with the
48 * match symbols containing a "length header" (giving all or part of the match
49 * length) and a "position slot" (giving, roughly speaking, the order of
50 * magnitude of the match offset).
51 * - LZX does not have static Huffman blocks; however it does have two types of
52 * dynamic Huffman blocks ("aligned offset" and "verbatim").
53 * - LZX has a minimum match length of 2 rather than 3.
54 * - In LZX, match offsets 0 through 2 actually represent entries in an LRU
55 * queue of match offsets. This is very useful for certain types of files,
56 * such as binary files that have repeating records.
61 * There are actually two distinct overall algorithms implemented here. We
62 * shall refer to them as the "slow" algorithm and the "fast" algorithm. The
63 * "slow" algorithm spends more time compressing to achieve a higher compression
64 * ratio compared to the "fast" algorithm. More details are presented below.
69 * The "slow" algorithm to generate LZX-compressed data is roughly as follows:
71 * 1. Preprocess the input data to translate the targets of x86 call instructions
72 * to absolute offsets.
74 * 2. Build the suffix array and inverse suffix array for the input data. The
75 * suffix array contains the indices of all suffixes of the input data,
76 * sorted lexcographically by the corresponding suffixes. The "position" of
77 * a suffix is the index of that suffix in the original string, whereas the
78 * "rank" of a suffix is the index at which that suffix's position is found
79 * in the suffix array.
81 * 3. Build the longest common prefix array corresponding to the suffix array.
83 * 4. For each suffix, find the highest lower ranked suffix that has a
84 * lower position, the lowest higher ranked suffix that has a lower position,
85 * and the length of the common prefix shared between each. This
86 * information is later used to link suffix ranks into a doubly-linked list
87 * for searching the suffix array.
89 * 5. Set a default cost model for matches/literals.
91 * 6. Determine the lowest cost sequence of LZ77 matches ((offset, length) pairs)
92 * and literal bytes to divide the input into. Raw match-finding is done by
93 * searching the suffix array using a linked list to avoid considering any
94 * suffixes that start after the current position. Each run of the
95 * match-finder returns the approximate lowest-cost longest match as well as
96 * any shorter matches that have even lower approximate costs. Each such run
97 * also adds the suffix rank of the current position into the linked list
98 * being used to search the suffix array. Parsing, or match-choosing, is
99 * solved as a minimum-cost path problem using a forward "optimal parsing"
100 * algorithm based on the Deflate encoder from 7-Zip. This algorithm moves
101 * forward calculating the minimum cost to reach each byte until either a
102 * very long match is found or until a position is found at which no matches
105 * 7. Build the Huffman codes needed to output the matches/literals.
107 * 8. Up to a certain number of iterations, use the resulting Huffman codes to
108 * refine a cost model and go back to Step #6 to determine an improved
109 * sequence of matches and literals.
111 * 9. Output the resulting block using the match/literal sequences and the
112 * Huffman codes that were computed for the block.
114 * Note: the algorithm does not yet attempt to split the input into multiple LZX
120 * The fast algorithm (and the only one available in wimlib v1.5.1 and earlier)
121 * spends much less time on the main bottlenecks of the compression process ---
122 * that is, the match finding and match choosing. Matches are found and chosen
123 * with hash chains using a greedy parse with one position of look-ahead. No
124 * block splitting is done; only compressing the full input into an aligned
125 * offset block is considered.
130 * The old API (retained for backward compatibility) consists of just one function:
132 * wimlib_lzx_compress()
134 * The new compressor has more potential parameters and needs more memory, so
135 * the new API ties up memory allocations and compression parameters into a
138 * wimlib_lzx_alloc_context()
139 * wimlib_lzx_compress2()
140 * wimlib_lzx_free_context()
141 * wimlib_lzx_set_default_params()
143 * Both wimlib_lzx_compress() and wimlib_lzx_compress2() are designed to
144 * compress an in-memory buffer of up to 32768 bytes. There is no sliding
145 * window. This is suitable for the WIM format, which uses fixed-size chunks
146 * that are seemingly always 32768 bytes. If needed, the compressor potentially
147 * could be extended to support a larger and/or sliding window.
149 * Both wimlib_lzx_compress() and wimlib_lzx_compress2() return 0 if the data
150 * could not be compressed to less than the size of the uncompressed data.
151 * Again, this is suitable for the WIM format, which stores such data chunks
154 * The functions in this LZX compression API are exported from the library,
155 * although with the possible exception of wimlib_lzx_set_default_params(), this
156 * is only in case other programs happen to have uses for it other than WIM
157 * reading/writing as already handled through the rest of the library.
162 * Acknowledgments to several open-source projects and research papers that made
163 * it possible to implement this code:
165 * - divsufsort (author: Yuta Mori), for the suffix array construction code,
166 * located in a separate directory (divsufsort/).
168 * - "Linear-Time Longest-Common-Prefix Computation in Suffix Arrays and Its
169 * Applications" (Kasai et al. 2001), for the LCP array computation.
171 * - "LPF computation revisited" (Crochemore et al. 2009) for the prev and next
172 * array computations.
174 * - 7-Zip (author: Igor Pavlov) for the algorithm for forward optimal parsing
177 * - zlib (author: Jean-loup Gailly and Mark Adler), for the hash table
178 * match-finding algorithm (used in lz77.c).
180 * - lzx-compress (author: Matthew T. Russotto), on which some parts of this
181 * code were originally based.
189 #include "wimlib/compress.h"
190 #include "wimlib/error.h"
191 #include "wimlib/lzx.h"
192 #include "wimlib/util.h"
197 #ifdef ENABLE_LZX_DEBUG
198 # include "wimlib/decompress.h"
201 #include "divsufsort/divsufsort.h"
203 typedef u32 block_cost_t;
204 #define INFINITE_BLOCK_COST ((block_cost_t)~0U)
206 #define LZX_OPTIM_ARRAY_SIZE 4096
208 /* Currently, this constant can't simply be changed because the code currently
209 * uses a static number of position slots (and may make other assumptions as
211 #define LZX_MAX_WINDOW_SIZE 32768
213 /* This may be WIM-specific */
214 #define LZX_DEFAULT_BLOCK_SIZE 32768
216 #define LZX_MAX_CACHE_PER_POS 10
218 /* Codewords for the LZX main, length, and aligned offset Huffman codes */
219 struct lzx_codewords {
220 u16 main[LZX_MAINCODE_NUM_SYMBOLS];
221 u16 len[LZX_LENCODE_NUM_SYMBOLS];
222 u16 aligned[LZX_ALIGNEDCODE_NUM_SYMBOLS];
225 /* Codeword lengths (in bits) for the LZX main, length, and aligned offset
228 * A 0 length means the codeword has zero frequency.
231 u8 main[LZX_MAINCODE_NUM_SYMBOLS];
232 u8 len[LZX_LENCODE_NUM_SYMBOLS];
233 u8 aligned[LZX_ALIGNEDCODE_NUM_SYMBOLS];
236 /* Costs for the LZX main, length, and aligned offset Huffman symbols.
238 * If a codeword has zero frequency, it must still be assigned some nonzero cost
239 * --- generally a high cost, since even if it gets used in the next iteration,
240 * it probably will not be used very times. */
242 u8 main[LZX_MAINCODE_NUM_SYMBOLS];
243 u8 len[LZX_LENCODE_NUM_SYMBOLS];
244 u8 aligned[LZX_ALIGNEDCODE_NUM_SYMBOLS];
247 /* The LZX main, length, and aligned offset Huffman codes */
249 struct lzx_codewords codewords;
250 struct lzx_lens lens;
253 /* Tables for tallying symbol frequencies in the three LZX alphabets */
255 freq_t main[LZX_MAINCODE_NUM_SYMBOLS];
256 freq_t len[LZX_LENCODE_NUM_SYMBOLS];
257 freq_t aligned[LZX_ALIGNEDCODE_NUM_SYMBOLS];
260 /* LZX intermediate match/literal format */
264 * 31 1 if a match, 0 if a literal.
266 * 30-25 position slot. This can be at most 50, so it will fit in 6
269 * 8-24 position footer. This is the offset of the real formatted
270 * offset from the position base. This can be at most 17 bits
271 * (since lzx_extra_bits[LZX_NUM_POSITION_SLOTS - 1] is 17).
273 * 0-7 length of match, minus 2. This can be at most
274 * (LZX_MAX_MATCH_LEN - 2) == 255, so it will fit in 8 bits. */
278 /* Raw LZ match/literal format: just a length and offset.
280 * The length is the number of bytes of the match, and the offset is the number
281 * of bytes back in the input the match is from the current position.
283 * If @len < LZX_MIN_MATCH_LEN, then it's really just a literal byte and @offset is
290 /* Specification for an LZX block. */
291 struct lzx_block_spec {
293 /* One of the LZX_BLOCKTYPE_* constants indicating which type of this
297 /* 0-based position in the window at which this block starts. */
298 input_idx_t window_pos;
300 /* The number of bytes of uncompressed data this block represents. */
301 input_idx_t block_size;
303 /* The position in the 'chosen_matches' array in the `struct
304 * lzx_compressor' at which the match/literal specifications for
305 * this block begin. */
306 input_idx_t chosen_matches_start_pos;
308 /* The number of match/literal specifications for this block. */
309 input_idx_t num_chosen_matches;
311 /* Huffman codes for this block. */
312 struct lzx_codes codes;
316 * An array of these structures is used during the match-choosing algorithm.
317 * They correspond to consecutive positions in the window and are used to keep
318 * track of the cost to reach each position, and the match/literal choices that
319 * need to be chosen to reach that position.
322 /* The approximate minimum cost, in bits, to reach this position in the
323 * window which has been found so far. */
326 /* The union here is just for clarity, since the fields are used in two
327 * slightly different ways. Initially, the @prev structure is filled in
328 * first, and links go from later in the window to earlier in the
329 * window. Later, @next structure is filled in and links go from
330 * earlier in the window to later in the window. */
333 /* Position of the start of the match or literal that
334 * was taken to get to this position in the approximate
335 * minimum-cost parse. */
338 /* Offset (as in an LZ (length, offset) pair) of the
339 * match or literal that was taken to get to this
340 * position in the approximate minimum-cost parse. */
341 input_idx_t match_offset;
344 /* Position at which the match or literal starting at
345 * this position ends in the minimum-cost parse. */
348 /* Offset (as in an LZ (length, offset) pair) of the
349 * match or literal starting at this position in the
350 * approximate minimum-cost parse. */
351 input_idx_t match_offset;
355 /* The match offset LRU queue that will exist when the approximate
356 * minimum-cost path to reach this position is taken. */
357 struct lzx_lru_queue queue;
360 /* Suffix array link */
362 /* Rank of highest ranked suffix that has rank lower than the suffix
363 * corresponding to this structure and either has a lower position
364 * (initially) or has a position lower than the highest position at
365 * which matches have been searched for so far, or -1 if there is no
369 /* Rank of lowest ranked suffix that has rank greater than the suffix
370 * corresponding to this structure and either has a lower position
371 * (intially) or has a position lower than the highest position at which
372 * matches have been searched for so far, or -1 if there is no such
376 /* Length of longest common prefix between the suffix corresponding to
377 * this structure and the suffix with rank @prev, or 0 if @prev is -1.
381 /* Length of longest common prefix between the suffix corresponding to
382 * this structure and the suffix with rank @next, or 0 if @next is -1.
387 /* State of the LZX compressor. */
388 struct lzx_compressor {
390 /* The parameters that were used to create the compressor. */
391 struct wimlib_lzx_params params;
393 /* The buffer of data to be compressed.
395 * 0xe8 byte preprocessing is done directly on the data here before
396 * further compression.
398 * Note that this compressor does *not* use a sliding window!!!! It's
399 * not needed in the WIM format, since every chunk is compressed
400 * independently. This is by design, to allow random access to the
403 * We reserve a few extra bytes to potentially allow reading off the end
404 * of the array in the match-finding code for optimization purposes.
406 u8 window[LZX_MAX_WINDOW_SIZE + 12];
408 /* Number of bytes of data to be compressed, which is the number of
409 * bytes of data in @window that are actually valid. */
410 input_idx_t window_size;
412 /* The current match offset LRU queue. */
413 struct lzx_lru_queue queue;
415 /* Space for the sequences of matches/literals that were chosen for each
417 struct lzx_match *chosen_matches;
419 /* Information about the LZX blocks the preprocessed input was divided
421 struct lzx_block_spec *block_specs;
423 /* Number of LZX blocks the input was divided into; a.k.a. the number of
424 * elements of @block_specs that are valid. */
427 /* This is simply filled in with zeroes and used to avoid special-casing
428 * the output of the first compressed Huffman code, which conceptually
429 * has a delta taken from a code with all symbols having zero-length
431 struct lzx_codes zero_codes;
433 /* The current cost model. */
434 struct lzx_costs costs;
436 /* Suffix array for window.
437 * This is a mapping from suffix rank to suffix position. */
440 /* Inverse suffix array for window.
441 * This is a mapping from suffix position to suffix rank.
442 * If 0 <= r < window_size, then ISA[SA[r]] == r. */
445 /* Suffix array links.
447 * During a linear scan of the input string to find matches, this array
448 * used to keep track of which rank suffixes in the suffix array appear
449 * before the current position. Instead of searching in the original
450 * suffix array, scans for matches at a given position traverse a linked
451 * list containing only suffixes that appear before that position. */
452 struct salink *salink;
454 /* Position in window of next match to return.
455 * Note: This cannot simply be modified, as the match-finder must still
456 * be synchronized on the same position. To seek forwards or backwards,
457 * use lzx_lz_skip_bytes() or lzx_lz_rewind_matchfinder(), respectively.
459 input_idx_t match_window_pos;
461 /* The match-finder shall ensure the length of matches does not exceed
462 * this position in the input. */
463 input_idx_t match_window_end;
465 /* Matches found by the match-finder are cached in the following array
466 * to achieve a slight speedup when the same matches are needed on
467 * subsequent passes. This is suboptimal because different matches may
468 * be preferred with different cost models, but seems to be a worthwhile
470 struct raw_match *cached_matches;
471 unsigned cached_matches_pos;
474 /* Slow algorithm only: Temporary space used for match-choosing
477 * The size of this array must be at least LZX_MAX_MATCH_LEN but
478 * otherwise is arbitrary. More space simply allows the match-choosing
479 * algorithm to potentially find better matches (depending on the input,
481 struct lzx_optimal *optimum;
483 /* Slow algorithm only: Variables used by the match-choosing algorithm.
485 * When matches have been chosen, optimum_cur_idx is set to the position
486 * in the window of the next match/literal to return and optimum_end_idx
487 * is set to the position in the window at the end of the last
488 * match/literal to return. */
493 /* Returns the LZX position slot that corresponds to a given formatted offset.
495 * Logically, this returns the smallest i such that
496 * formatted_offset >= lzx_position_base[i].
498 * The actual implementation below takes advantage of the regularity of the
499 * numbers in the lzx_position_base array to calculate the slot directly from
500 * the formatted offset without actually looking at the array.
503 lzx_get_position_slot_raw(unsigned formatted_offset)
507 * Slots 36-49 (formatted_offset >= 262144) can be found by
508 * (formatted_offset/131072) + 34 == (formatted_offset >> 17) + 34;
509 * however, this check for formatted_offset >= 262144 is commented out
510 * because WIM chunks cannot be that large.
512 if (formatted_offset >= 262144) {
513 return (formatted_offset >> 17) + 34;
517 /* Note: this part here only works if:
519 * 2 <= formatted_offset < 655360
521 * It is < 655360 because the frequency of the position bases
522 * increases starting at the 655360 entry, and it is >= 2
523 * because the below calculation fails if the most significant
524 * bit is lower than the 2's place. */
525 LZX_ASSERT(2 <= formatted_offset && formatted_offset < 655360);
526 unsigned mssb_idx = bsr32(formatted_offset);
527 return (mssb_idx << 1) |
528 ((formatted_offset >> (mssb_idx - 1)) & 1);
533 /* Returns the LZX position slot that corresponds to a given match offset,
534 * taking into account the recent offset queue and updating it if the offset is
537 lzx_get_position_slot(unsigned offset, struct lzx_lru_queue *queue)
539 unsigned position_slot;
541 /* See if the offset was recently used. */
542 for (unsigned i = 0; i < LZX_NUM_RECENT_OFFSETS; i++) {
543 if (offset == queue->R[i]) {
546 /* Bring the repeat offset to the front of the
547 * queue. Note: this is, in fact, not a real
548 * LRU queue because repeat matches are simply
549 * swapped to the front. */
550 swap(queue->R[0], queue->R[i]);
552 /* The resulting position slot is simply the first index
553 * at which the offset was found in the queue. */
558 /* The offset was not recently used; look up its real position slot. */
559 position_slot = lzx_get_position_slot_raw(offset + LZX_OFFSET_OFFSET);
561 /* Bring the new offset to the front of the queue. */
562 for (unsigned i = LZX_NUM_RECENT_OFFSETS - 1; i > 0; i--)
563 queue->R[i] = queue->R[i - 1];
564 queue->R[0] = offset;
566 return position_slot;
569 /* Build the main, length, and aligned offset Huffman codes used in LZX.
571 * This takes as input the frequency tables for each code and produces as output
572 * a set of tables that map symbols to codewords and codeword lengths. */
574 lzx_make_huffman_codes(const struct lzx_freqs *freqs,
575 struct lzx_codes *codes)
577 make_canonical_huffman_code(LZX_MAINCODE_NUM_SYMBOLS,
578 LZX_MAX_MAIN_CODEWORD_LEN,
581 codes->codewords.main);
583 make_canonical_huffman_code(LZX_LENCODE_NUM_SYMBOLS,
584 LZX_MAX_LEN_CODEWORD_LEN,
587 codes->codewords.len);
589 make_canonical_huffman_code(LZX_ALIGNEDCODE_NUM_SYMBOLS,
590 LZX_MAX_ALIGNED_CODEWORD_LEN,
593 codes->codewords.aligned);
597 * Output an LZX match.
599 * @out: The bitstream to write the match to.
600 * @block_type: The type of the LZX block (LZX_BLOCKTYPE_ALIGNED or LZX_BLOCKTYPE_VERBATIM)
602 * @codes: Pointer to a structure that contains the codewords for the
603 * main, length, and aligned offset Huffman codes.
606 lzx_write_match(struct output_bitstream *out, int block_type,
607 struct lzx_match match, const struct lzx_codes *codes)
609 /* low 8 bits are the match length minus 2 */
610 unsigned match_len_minus_2 = match.data & 0xff;
611 /* Next 17 bits are the position footer */
612 unsigned position_footer = (match.data >> 8) & 0x1ffff; /* 17 bits */
613 /* Next 6 bits are the position slot. */
614 unsigned position_slot = (match.data >> 25) & 0x3f; /* 6 bits */
617 unsigned main_symbol;
618 unsigned num_extra_bits;
619 unsigned verbatim_bits;
620 unsigned aligned_bits;
622 /* If the match length is less than MIN_MATCH_LEN (= 2) +
623 * NUM_PRIMARY_LENS (= 7), the length header contains
624 * the match length minus MIN_MATCH_LEN, and there is no
627 * Otherwise, the length header contains
628 * NUM_PRIMARY_LENS, and the length footer contains
629 * the match length minus NUM_PRIMARY_LENS minus
631 if (match_len_minus_2 < LZX_NUM_PRIMARY_LENS) {
632 len_header = match_len_minus_2;
633 /* No length footer-- mark it with a special
635 len_footer = (unsigned)(-1);
637 len_header = LZX_NUM_PRIMARY_LENS;
638 len_footer = match_len_minus_2 - LZX_NUM_PRIMARY_LENS;
641 /* Combine the position slot with the length header into a single symbol
642 * that will be encoded with the main tree.
644 * The actual main symbol is offset by LZX_NUM_CHARS because values
645 * under LZX_NUM_CHARS are used to indicate a literal byte rather than a
647 main_symbol = ((position_slot << 3) | len_header) + LZX_NUM_CHARS;
649 /* Output main symbol. */
650 bitstream_put_bits(out, codes->codewords.main[main_symbol],
651 codes->lens.main[main_symbol]);
653 /* If there is a length footer, output it using the
654 * length Huffman code. */
655 if (len_footer != (unsigned)(-1)) {
656 bitstream_put_bits(out, codes->codewords.len[len_footer],
657 codes->lens.len[len_footer]);
660 num_extra_bits = lzx_get_num_extra_bits(position_slot);
662 /* For aligned offset blocks with at least 3 extra bits, output the
663 * verbatim bits literally, then the aligned bits encoded using the
664 * aligned offset tree. Otherwise, only the verbatim bits need to be
666 if ((block_type == LZX_BLOCKTYPE_ALIGNED) && (num_extra_bits >= 3)) {
668 verbatim_bits = position_footer >> 3;
669 bitstream_put_bits(out, verbatim_bits,
672 aligned_bits = (position_footer & 7);
673 bitstream_put_bits(out,
674 codes->codewords.aligned[aligned_bits],
675 codes->lens.aligned[aligned_bits]);
677 /* verbatim bits is the same as the position
678 * footer, in this case. */
679 bitstream_put_bits(out, position_footer, num_extra_bits);
684 lzx_build_precode(const u8 lens[restrict],
685 const u8 prev_lens[restrict],
686 const unsigned num_syms,
687 freq_t precode_freqs[restrict LZX_PRECODE_NUM_SYMBOLS],
688 u8 output_syms[restrict num_syms],
689 u8 precode_lens[restrict LZX_PRECODE_NUM_SYMBOLS],
690 u16 precode_codewords[restrict LZX_PRECODE_NUM_SYMBOLS],
691 unsigned *num_additional_bits_ret)
693 memset(precode_freqs, 0,
694 LZX_PRECODE_NUM_SYMBOLS * sizeof(precode_freqs[0]));
696 /* Since the code word lengths use a form of RLE encoding, the goal here
697 * is to find each run of identical lengths when going through them in
698 * symbol order (including runs of length 1). For each run, as many
699 * lengths are encoded using RLE as possible, and the rest are output
702 * output_syms[] will be filled in with the length symbols that will be
703 * output, including RLE codes, not yet encoded using the pre-tree.
705 * cur_run_len keeps track of how many code word lengths are in the
706 * current run of identical lengths. */
707 unsigned output_syms_idx = 0;
708 unsigned cur_run_len = 1;
709 unsigned num_additional_bits = 0;
710 for (unsigned i = 1; i <= num_syms; i++) {
712 if (i != num_syms && lens[i] == lens[i - 1]) {
713 /* Still in a run--- keep going. */
718 /* Run ended! Check if it is a run of zeroes or a run of
721 /* The symbol that was repeated in the run--- not to be confused
722 * with the length *of* the run (cur_run_len) */
723 unsigned len_in_run = lens[i - 1];
725 if (len_in_run == 0) {
726 /* A run of 0's. Encode it in as few length
727 * codes as we can. */
729 /* The magic length 18 indicates a run of 20 + n zeroes,
730 * where n is an uncompressed literal 5-bit integer that
731 * follows the magic length. */
732 while (cur_run_len >= 20) {
733 unsigned additional_bits;
735 additional_bits = min(cur_run_len - 20, 0x1f);
736 num_additional_bits += 5;
738 output_syms[output_syms_idx++] = 18;
739 output_syms[output_syms_idx++] = additional_bits;
740 cur_run_len -= 20 + additional_bits;
743 /* The magic length 17 indicates a run of 4 + n zeroes,
744 * where n is an uncompressed literal 4-bit integer that
745 * follows the magic length. */
746 while (cur_run_len >= 4) {
747 unsigned additional_bits;
749 additional_bits = min(cur_run_len - 4, 0xf);
750 num_additional_bits += 4;
752 output_syms[output_syms_idx++] = 17;
753 output_syms[output_syms_idx++] = additional_bits;
754 cur_run_len -= 4 + additional_bits;
759 /* A run of nonzero lengths. */
761 /* The magic length 19 indicates a run of 4 + n
762 * nonzeroes, where n is a literal bit that follows the
763 * magic length, and where the value of the lengths in
764 * the run is given by an extra length symbol, encoded
765 * with the precode, that follows the literal bit.
767 * The extra length symbol is encoded as a difference
768 * from the length of the codeword for the first symbol
769 * in the run in the previous tree.
771 while (cur_run_len >= 4) {
772 unsigned additional_bits;
775 additional_bits = (cur_run_len > 4);
776 num_additional_bits += 1;
777 delta = (signed char)prev_lens[i - cur_run_len] -
778 (signed char)len_in_run;
782 precode_freqs[(unsigned char)delta]++;
783 output_syms[output_syms_idx++] = 19;
784 output_syms[output_syms_idx++] = additional_bits;
785 output_syms[output_syms_idx++] = delta;
786 cur_run_len -= 4 + additional_bits;
790 /* Any remaining lengths in the run are outputted without RLE,
791 * as a difference from the length of that codeword in the
793 while (cur_run_len > 0) {
796 delta = (signed char)prev_lens[i - cur_run_len] -
797 (signed char)len_in_run;
801 precode_freqs[(unsigned char)delta]++;
802 output_syms[output_syms_idx++] = delta;
809 /* Build the precode from the frequencies of the length symbols. */
811 make_canonical_huffman_code(LZX_PRECODE_NUM_SYMBOLS,
812 LZX_MAX_PRE_CODEWORD_LEN,
813 precode_freqs, precode_lens,
816 *num_additional_bits_ret = num_additional_bits;
818 return output_syms_idx;
822 * Writes a compressed Huffman code to the output, preceded by the precode for
825 * The Huffman code is represented in the output as a series of path lengths
826 * from which the canonical Huffman code can be reconstructed. The path lengths
827 * themselves are compressed using a separate Huffman code, the precode, which
828 * consists of LZX_PRECODE_NUM_SYMBOLS (= 20) symbols that cover all possible
829 * code lengths, plus extra codes for repeated lengths. The path lengths of the
830 * precode precede the path lengths of the larger code and are uncompressed,
831 * consisting of 20 entries of 4 bits each.
833 * @out: Bitstream to write the code to.
834 * @lens: The code lengths for the Huffman code, indexed by symbol.
835 * @prev_lens: Code lengths for this Huffman code, indexed by symbol,
836 * in the *previous block*, or all zeroes if this is the
838 * @num_syms: The number of symbols in the code.
841 lzx_write_compressed_code(struct output_bitstream *out,
842 const u8 lens[restrict],
843 const u8 prev_lens[restrict],
846 freq_t precode_freqs[LZX_PRECODE_NUM_SYMBOLS];
847 u8 output_syms[num_syms];
848 u8 precode_lens[LZX_PRECODE_NUM_SYMBOLS];
849 u16 precode_codewords[LZX_PRECODE_NUM_SYMBOLS];
851 unsigned num_output_syms;
855 num_output_syms = lzx_build_precode(lens,
864 /* Write the lengths of the precode codes to the output. */
865 for (i = 0; i < LZX_PRECODE_NUM_SYMBOLS; i++)
866 bitstream_put_bits(out, precode_lens[i],
867 LZX_PRECODE_ELEMENT_SIZE);
869 /* Write the length symbols, encoded with the precode, to the output. */
871 for (i = 0; i < num_output_syms; ) {
872 precode_sym = output_syms[i++];
874 bitstream_put_bits(out, precode_codewords[precode_sym],
875 precode_lens[precode_sym]);
876 switch (precode_sym) {
878 bitstream_put_bits(out, output_syms[i++], 4);
881 bitstream_put_bits(out, output_syms[i++], 5);
884 bitstream_put_bits(out, output_syms[i++], 1);
885 bitstream_put_bits(out,
886 precode_codewords[output_syms[i]],
887 precode_lens[output_syms[i]]);
897 * Writes all compressed matches and literal bytes in an LZX block to the the
901 * The output bitstream.
903 * The type of the block (LZX_BLOCKTYPE_ALIGNED or LZX_BLOCKTYPE_VERBATIM).
905 * The array of matches/literals that will be output (length @match_count).
907 * Number of matches/literals to be output.
909 * Pointer to a structure that contains the codewords for the main, length,
910 * and aligned offset Huffman codes.
913 lzx_write_matches_and_literals(struct output_bitstream *ostream,
915 const struct lzx_match match_tab[],
916 unsigned match_count,
917 const struct lzx_codes *codes)
919 for (unsigned i = 0; i < match_count; i++) {
920 struct lzx_match match = match_tab[i];
922 /* High bit of the match indicates whether the match is an
923 * actual match (1) or a literal uncompressed byte (0) */
924 if (match.data & 0x80000000) {
926 lzx_write_match(ostream, block_type,
930 bitstream_put_bits(ostream,
931 codes->codewords.main[match.data],
932 codes->lens.main[match.data]);
938 lzx_assert_codes_valid(const struct lzx_codes * codes)
940 #ifdef ENABLE_LZX_DEBUG
943 for (i = 0; i < LZX_MAINCODE_NUM_SYMBOLS; i++)
944 LZX_ASSERT(codes->lens.main[i] <= LZX_MAX_MAIN_CODEWORD_LEN);
946 for (i = 0; i < LZX_LENCODE_NUM_SYMBOLS; i++)
947 LZX_ASSERT(codes->lens.len[i] <= LZX_MAX_LEN_CODEWORD_LEN);
949 for (i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++)
950 LZX_ASSERT(codes->lens.aligned[i] <= LZX_MAX_ALIGNED_CODEWORD_LEN);
952 const unsigned tablebits = 10;
953 u16 decode_table[(1 << tablebits) +
954 (2 * max(LZX_MAINCODE_NUM_SYMBOLS, LZX_LENCODE_NUM_SYMBOLS))]
955 _aligned_attribute(DECODE_TABLE_ALIGNMENT);
956 LZX_ASSERT(0 == make_huffman_decode_table(decode_table,
957 LZX_MAINCODE_NUM_SYMBOLS,
958 min(tablebits, LZX_MAINCODE_TABLEBITS),
960 LZX_MAX_MAIN_CODEWORD_LEN));
961 LZX_ASSERT(0 == make_huffman_decode_table(decode_table,
962 LZX_LENCODE_NUM_SYMBOLS,
963 min(tablebits, LZX_LENCODE_TABLEBITS),
965 LZX_MAX_LEN_CODEWORD_LEN));
966 LZX_ASSERT(0 == make_huffman_decode_table(decode_table,
967 LZX_ALIGNEDCODE_NUM_SYMBOLS,
968 min(tablebits, LZX_ALIGNEDCODE_TABLEBITS),
970 LZX_MAX_ALIGNED_CODEWORD_LEN));
971 #endif /* ENABLE_LZX_DEBUG */
974 /* Write an LZX aligned offset or verbatim block to the output. */
976 lzx_write_compressed_block(int block_type,
978 struct lzx_match * chosen_matches,
979 unsigned num_chosen_matches,
980 const struct lzx_codes * codes,
981 const struct lzx_codes * prev_codes,
982 struct output_bitstream * ostream)
986 LZX_ASSERT(block_type == LZX_BLOCKTYPE_ALIGNED ||
987 block_type == LZX_BLOCKTYPE_VERBATIM);
988 LZX_ASSERT(block_size <= LZX_MAX_WINDOW_SIZE);
989 LZX_ASSERT(num_chosen_matches <= LZX_MAX_WINDOW_SIZE);
990 lzx_assert_codes_valid(codes);
992 /* The first three bits indicate the type of block and are one of the
993 * LZX_BLOCKTYPE_* constants. */
994 bitstream_put_bits(ostream, block_type, LZX_BLOCKTYPE_NBITS);
996 /* The next bit indicates whether the block size is the default (32768),
997 * indicated by a 1 bit, or whether the block size is given by the next
998 * 16 bits, indicated by a 0 bit. */
999 if (block_size == LZX_DEFAULT_BLOCK_SIZE) {
1000 bitstream_put_bits(ostream, 1, 1);
1002 bitstream_put_bits(ostream, 0, 1);
1003 bitstream_put_bits(ostream, block_size, LZX_BLOCKSIZE_NBITS);
1006 /* Write out lengths of the main code. Note that the LZX specification
1007 * incorrectly states that the aligned offset code comes after the
1008 * length code, but in fact it is the very first tree to be written
1009 * (before the main code). */
1010 if (block_type == LZX_BLOCKTYPE_ALIGNED)
1011 for (i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++)
1012 bitstream_put_bits(ostream, codes->lens.aligned[i],
1013 LZX_ALIGNEDCODE_ELEMENT_SIZE);
1015 LZX_DEBUG("Writing main code...");
1017 /* Write the pre-tree and lengths for the first LZX_NUM_CHARS symbols in
1018 * the main code, which are the codewords for literal bytes. */
1019 lzx_write_compressed_code(ostream,
1021 prev_codes->lens.main,
1024 /* Write the pre-tree and lengths for the rest of the main code, which
1025 * are the codewords for match headers. */
1026 lzx_write_compressed_code(ostream,
1027 codes->lens.main + LZX_NUM_CHARS,
1028 prev_codes->lens.main + LZX_NUM_CHARS,
1029 LZX_MAINCODE_NUM_SYMBOLS - LZX_NUM_CHARS);
1031 LZX_DEBUG("Writing length code...");
1033 /* Write the pre-tree and lengths for the length code. */
1034 lzx_write_compressed_code(ostream,
1036 prev_codes->lens.len,
1037 LZX_LENCODE_NUM_SYMBOLS);
1039 LZX_DEBUG("Writing matches and literals...");
1041 /* Write the actual matches and literals. */
1042 lzx_write_matches_and_literals(ostream, block_type,
1043 chosen_matches, num_chosen_matches,
1046 LZX_DEBUG("Done writing block.");
1049 /* Write out the LZX blocks that were computed. */
1051 lzx_write_all_blocks(struct lzx_compressor *ctx, struct output_bitstream *ostream)
1053 const struct lzx_codes *prev_codes = &ctx->zero_codes;
1054 for (unsigned i = 0; i < ctx->num_blocks; i++) {
1055 const struct lzx_block_spec *spec = &ctx->block_specs[i];
1057 LZX_DEBUG("Writing block %u/%u (type=%d, size=%u, num_chosen_matches=%u)...",
1058 i + 1, ctx->num_blocks,
1059 spec->block_type, spec->block_size,
1060 spec->num_chosen_matches);
1062 lzx_write_compressed_block(spec->block_type,
1064 &ctx->chosen_matches[spec->chosen_matches_start_pos],
1065 spec->num_chosen_matches,
1069 prev_codes = &spec->codes;
1073 /* Constructs an LZX match from a literal byte and updates the main code symbol
1076 lzx_tally_literal(u8 lit, struct lzx_freqs *freqs)
1082 /* Constructs an LZX match from an offset and a length, and updates the LRU
1083 * queue and the frequency of symbols in the main, length, and aligned offset
1084 * alphabets. The return value is a 32-bit number that provides the match in an
1085 * intermediate representation documented below. */
1087 lzx_tally_match(unsigned match_len, unsigned match_offset,
1088 struct lzx_freqs *freqs, struct lzx_lru_queue *queue)
1090 unsigned position_slot;
1091 unsigned position_footer;
1093 unsigned main_symbol;
1094 unsigned len_footer;
1095 unsigned adjusted_match_len;
1097 LZX_ASSERT(match_len >= LZX_MIN_MATCH_LEN && match_len <= LZX_MAX_MATCH_LEN);
1099 /* The match offset shall be encoded as a position slot (itself encoded
1100 * as part of the main symbol) and a position footer. */
1101 position_slot = lzx_get_position_slot(match_offset, queue);
1102 position_footer = (match_offset + LZX_OFFSET_OFFSET) &
1103 ((1U << lzx_get_num_extra_bits(position_slot)) - 1);
1105 /* The match length shall be encoded as a length header (itself encoded
1106 * as part of the main symbol) and an optional length footer. */
1107 adjusted_match_len = match_len - LZX_MIN_MATCH_LEN;
1108 if (adjusted_match_len < LZX_NUM_PRIMARY_LENS) {
1109 /* No length footer needed. */
1110 len_header = adjusted_match_len;
1112 /* Length footer needed. It will be encoded using the length
1114 len_header = LZX_NUM_PRIMARY_LENS;
1115 len_footer = adjusted_match_len - LZX_NUM_PRIMARY_LENS;
1116 freqs->len[len_footer]++;
1119 /* Account for the main symbol. */
1120 main_symbol = ((position_slot << 3) | len_header) + LZX_NUM_CHARS;
1122 freqs->main[main_symbol]++;
1124 /* In an aligned offset block, 3 bits of the position footer are output
1125 * as an aligned offset symbol. Account for this, although we may
1126 * ultimately decide to output the block as verbatim. */
1128 /* The following check is equivalent to:
1130 * if (lzx_extra_bits[position_slot] >= 3)
1132 * Note that this correctly excludes position slots that correspond to
1133 * recent offsets. */
1134 if (position_slot >= 8)
1135 freqs->aligned[position_footer & 7]++;
1137 /* Pack the position slot, position footer, and match length into an
1138 * intermediate representation.
1141 * ---- -----------------------------------------------------------
1143 * 31 1 if a match, 0 if a literal.
1145 * 30-25 position slot. This can be at most 50, so it will fit in 6
1148 * 8-24 position footer. This is the offset of the real formatted
1149 * offset from the position base. This can be at most 17 bits
1150 * (since lzx_extra_bits[LZX_NUM_POSITION_SLOTS - 1] is 17).
1152 * 0-7 length of match, offset by 2. This can be at most
1153 * (LZX_MAX_MATCH_LEN - 2) == 255, so it will fit in 8 bits. */
1154 BUILD_BUG_ON(LZX_NUM_POSITION_SLOTS > 64);
1155 LZX_ASSERT(lzx_get_num_extra_bits(LZX_NUM_POSITION_SLOTS - 1) <= 17);
1156 BUILD_BUG_ON(LZX_MAX_MATCH_LEN - LZX_MIN_MATCH_LEN + 1 > 256);
1158 (position_slot << 25) |
1159 (position_footer << 8) |
1160 (adjusted_match_len);
1163 struct lzx_record_ctx {
1164 struct lzx_freqs freqs;
1165 struct lzx_lru_queue queue;
1166 struct lzx_match *matches;
1170 lzx_record_match(unsigned len, unsigned offset, void *_ctx)
1172 struct lzx_record_ctx *ctx = _ctx;
1174 (ctx->matches++)->data = lzx_tally_match(len, offset, &ctx->freqs, &ctx->queue);
1178 lzx_record_literal(u8 lit, void *_ctx)
1180 struct lzx_record_ctx *ctx = _ctx;
1182 (ctx->matches++)->data = lzx_tally_literal(lit, &ctx->freqs);
1185 /* Returns the cost, in bits, to output a literal byte using the specified cost
1188 lzx_literal_cost(u8 c, const struct lzx_costs * costs)
1190 return costs->main[c];
1193 /* Given a (length, offset) pair that could be turned into a valid LZX match as
1194 * well as costs for the codewords in the main, length, and aligned Huffman
1195 * codes, return the approximate number of bits it will take to represent this
1196 * match in the compressed output. Take into account the match offset LRU
1197 * queue and optionally update it. */
1199 lzx_match_cost(unsigned length, unsigned offset, const struct lzx_costs *costs,
1200 struct lzx_lru_queue *queue)
1202 unsigned position_slot;
1203 unsigned len_header, main_symbol;
1206 position_slot = lzx_get_position_slot(offset, queue);
1208 len_header = min(length - LZX_MIN_MATCH_LEN, LZX_NUM_PRIMARY_LENS);
1209 main_symbol = ((position_slot << 3) | len_header) + LZX_NUM_CHARS;
1211 /* Account for main symbol. */
1212 cost += costs->main[main_symbol];
1214 /* Account for extra position information. */
1215 unsigned num_extra_bits = lzx_get_num_extra_bits(position_slot);
1216 if (num_extra_bits >= 3) {
1217 cost += num_extra_bits - 3;
1218 cost += costs->aligned[(offset + LZX_OFFSET_OFFSET) & 7];
1220 cost += num_extra_bits;
1223 /* Account for extra length information. */
1224 if (len_header == LZX_NUM_PRIMARY_LENS)
1225 cost += costs->len[length - LZX_MIN_MATCH_LEN - LZX_NUM_PRIMARY_LENS];
1231 /* Fast heuristic cost evaluation to use in the inner loop of the match-finder.
1232 * Unlike lzx_match_cost() which does a true cost evaluation, this simply
1233 * prioritize matches based on their offset. */
1235 lzx_match_cost_fast(unsigned offset, const struct lzx_lru_queue *queue)
1237 /* It seems well worth it to take the time to give priority to recently
1239 for (unsigned i = 0; i < LZX_NUM_RECENT_OFFSETS; i++)
1240 if (offset == queue->R[i])
1243 BUILD_BUG_ON(LZX_MAX_WINDOW_SIZE >= (block_cost_t)~0U);
1247 /* Set the cost model @ctx->costs from the Huffman codeword lengths specified in
1250 * The cost model and codeword lengths are almost the same thing, but the
1251 * Huffman codewords with length 0 correspond to symbols with zero frequency
1252 * that still need to be assigned actual costs. The specific values assigned
1253 * are arbitrary, but they should be fairly high (near the maximum codeword
1254 * length) to take into account the fact that uses of these symbols are expected
1257 lzx_set_costs(struct lzx_compressor * ctx, const struct lzx_lens * lens)
1262 for (i = 0; i < LZX_MAINCODE_NUM_SYMBOLS; i++) {
1263 ctx->costs.main[i] = lens->main[i];
1264 if (ctx->costs.main[i] == 0)
1265 ctx->costs.main[i] = ctx->params.alg_params.slow.main_nostat_cost;
1269 for (i = 0; i < LZX_LENCODE_NUM_SYMBOLS; i++) {
1270 ctx->costs.len[i] = lens->len[i];
1271 if (ctx->costs.len[i] == 0)
1272 ctx->costs.len[i] = ctx->params.alg_params.slow.len_nostat_cost;
1275 /* Aligned offset code */
1276 for (i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) {
1277 ctx->costs.aligned[i] = lens->aligned[i];
1278 if (ctx->costs.aligned[i] == 0)
1279 ctx->costs.aligned[i] = ctx->params.alg_params.slow.aligned_nostat_cost;
1283 /* Advance the suffix array match-finder to the next position. */
1285 lzx_lz_update_salink(input_idx_t i,
1286 const input_idx_t SA[restrict],
1287 const input_idx_t ISA[restrict],
1288 struct salink link[restrict])
1290 /* r = Rank of the suffix at the current position. */
1291 const input_idx_t r = ISA[i];
1293 /* next = rank of LOWEST ranked suffix that is ranked HIGHER than the
1294 * current suffix AND has a LOWER position, or -1 if none exists. */
1295 const input_idx_t next = link[r].next;
1297 /* prev = rank of HIGHEST ranked suffix that is ranked LOWER than the
1298 * current suffix AND has a LOWER position, or -1 if none exists. */
1299 const input_idx_t prev = link[r].prev;
1301 /* Link the suffix at the current position into the linked list that
1302 * contains all suffixes in the suffix array that are appear at or
1303 * before the current position, sorted by rank.
1305 * Save the values of all fields we overwrite so that rollback is
1307 if (next != (input_idx_t)~0U) {
1309 link[next].prev = r;
1310 link[next].lcpprev = link[r].lcpnext;
1313 if (prev != (input_idx_t)~0U) {
1315 link[prev].next = r;
1316 link[prev].lcpnext = link[r].lcpprev;
1320 /* Rewind the suffix array match-finder to the specified position.
1322 * This undoes a series of updates by lzx_lz_update_salink(). */
1324 lzx_lz_rewind_matchfinder(struct lzx_compressor *ctx,
1325 const unsigned orig_pos)
1327 LZX_DEBUG("Rewind match-finder %u => %u", ctx->match_window_pos, orig_pos);
1329 if (ctx->match_window_pos == orig_pos)
1332 /* NOTE: this has been optimized for the current algorithm where no
1333 * block-splitting is done and matches are cached, so that the suffix
1334 * array match-finder only runs through the input one time. Generalized
1335 * rewinds of the suffix array match-finder are possible, but require
1336 * incrementally saving fields being overwritten in
1337 * lzx_lz_update_salink(), then restoring them here in reverse order.
1340 LZX_ASSERT(ctx->match_window_pos > orig_pos);
1341 LZX_ASSERT(orig_pos == 0);
1342 ctx->matches_cached = true;
1343 ctx->cached_matches_pos = 0;
1344 ctx->match_window_pos = orig_pos;
1348 * Use the suffix array match-finder to retrieve a list of LZ matches at the
1351 * [in] @i Current position in the window.
1352 * [in] @SA Suffix array for the window.
1353 * [in] @ISA Inverse suffix array for the window.
1354 * [inout] @link Suffix array links used internally by the match-finder.
1355 * [out] @matches The (length, offset) pairs of the resulting matches will
1356 * be written here, sorted in decreasing order by
1357 * length. All returned lengths will be unique.
1358 * [in] @queue Recently used match offsets, used when evaluating the
1360 * [in] @min_match_len Minimum match length to return.
1361 * [in] @max_matches_to_consider Maximum number of matches to consider at
1363 * [in] @max_matches_to_return Maximum number of matches to return.
1365 * The return value is the number of matches found and written to @matches.
1368 lzx_lz_get_matches(const input_idx_t i,
1369 const input_idx_t SA[const restrict],
1370 const input_idx_t ISA[const restrict],
1371 struct salink link[const restrict],
1372 struct raw_match matches[const restrict],
1373 const struct lzx_lru_queue * const restrict queue,
1374 const unsigned min_match_len,
1375 const uint32_t max_matches_to_consider,
1376 const uint32_t max_matches_to_return)
1378 /* r = Rank of the suffix at the current position. */
1379 const input_idx_t r = ISA[i];
1381 /* Prepare for searching the current position. */
1382 lzx_lz_update_salink(i, SA, ISA, link);
1384 /* L = rank of next suffix to the left;
1385 * R = rank of next suffix to the right;
1386 * lenL = length of match between current position and the suffix with rank L;
1387 * lenR = length of match between current position and the suffix with rank R.
1389 * This is left and right relative to the rank of the current suffix.
1390 * Since the suffixes in the suffix array are sorted, the longest
1391 * matches are immediately to the left and right (using the linked list
1392 * to ignore all suffixes that occur later in the window). The match
1393 * length decreases the farther left and right we go. We shall keep the
1394 * length on both sides in sync in order to choose the lowest-cost match
1397 input_idx_t L = link[r].prev;
1398 input_idx_t R = link[r].next;
1399 input_idx_t lenL = link[r].lcpprev;
1400 input_idx_t lenR = link[r].lcpnext;
1402 /* nmatches = number of matches found so far. */
1403 unsigned nmatches = 0;
1405 /* best_cost = cost of lowest-cost match found so far.
1407 * We keep track of this so that we can ignore shorter matches that do
1408 * not have lower costs than a longer matches already found.
1410 block_cost_t best_cost = INFINITE_BLOCK_COST;
1412 /* count_remaining = maximum number of possible matches remaining to be
1414 uint32_t count_remaining = max_matches_to_consider;
1416 /* pending = match currently being considered for a specific length. */
1417 struct raw_match pending;
1418 block_cost_t pending_cost;
1420 while (lenL >= min_match_len || lenR >= min_match_len)
1423 pending_cost = INFINITE_BLOCK_COST;
1427 if (lenL >= min_match_len && lenL >= lenR) {
1430 if (--count_remaining == 0)
1431 goto out_save_pending;
1433 input_idx_t offset = i - SA[L];
1435 /* Save match if it has smaller cost. */
1436 cost = lzx_match_cost_fast(offset, queue);
1437 if (cost < pending_cost) {
1438 pending.offset = offset;
1439 pending_cost = cost;
1442 if (link[L].lcpprev < lenL) {
1443 /* Match length decreased. */
1445 lenL = link[L].lcpprev;
1447 /* Save the pending match unless the
1448 * right side still may have matches of
1449 * this length to be scanned, or if a
1450 * previous (longer) match had lower
1452 if (pending.len > lenR) {
1453 if (pending_cost < best_cost) {
1454 best_cost = pending_cost;
1455 matches[nmatches++] = pending;
1456 if (nmatches == max_matches_to_return)
1460 pending_cost = INFINITE_BLOCK_COST;
1462 if (lenL < min_match_len || lenL < lenR)
1472 if (lenR >= min_match_len && lenR > lenL) {
1475 if (--count_remaining == 0)
1476 goto out_save_pending;
1478 input_idx_t offset = i - SA[R];
1480 /* Save match if it has smaller cost. */
1481 cost = lzx_match_cost_fast(offset, queue);
1482 if (cost < pending_cost) {
1483 pending.offset = offset;
1484 pending_cost = cost;
1487 if (link[R].lcpnext < lenR) {
1488 /* Match length decreased. */
1490 lenR = link[R].lcpnext;
1492 /* Save the pending match unless a
1493 * previous (longer) match had lower
1495 if (pending_cost < best_cost) {
1496 matches[nmatches++] = pending;
1497 best_cost = pending_cost;
1498 if (nmatches == max_matches_to_return)
1502 if (lenR < min_match_len || lenR <= lenL)
1506 pending_cost = INFINITE_BLOCK_COST;
1515 if (pending_cost != INFINITE_BLOCK_COST)
1516 matches[nmatches++] = pending;
1523 /* Tell the match-finder to skip the specified number of bytes (@n) in the
1526 lzx_lz_skip_bytes(struct lzx_compressor *ctx, unsigned n)
1528 LZX_ASSERT(n <= ctx->match_window_end - ctx->match_window_pos);
1529 if (ctx->matches_cached) {
1530 ctx->match_window_pos += n;
1532 ctx->cached_matches_pos +=
1533 ctx->cached_matches[ctx->cached_matches_pos].len + 1;
1537 ctx->cached_matches[ctx->cached_matches_pos++].len = 0;
1538 lzx_lz_update_salink(ctx->match_window_pos++, ctx->SA,
1539 ctx->ISA, ctx->salink);
1544 /* Retrieve a list of matches available at the next position in the input.
1546 * The matches are written to ctx->matches in decreasing order of length, and
1547 * the return value is the number of matches found. */
1549 lzx_lz_get_matches_caching(struct lzx_compressor *ctx,
1550 const struct lzx_lru_queue *queue,
1551 struct raw_match **matches_ret)
1553 unsigned num_matches;
1554 struct raw_match *matches;
1556 LZX_ASSERT(ctx->match_window_pos <= ctx->match_window_end);
1558 matches = &ctx->cached_matches[ctx->cached_matches_pos + 1];
1560 if (ctx->matches_cached) {
1561 num_matches = matches[-1].len;
1563 unsigned min_match_len = LZX_MIN_MATCH_LEN;
1564 if (!ctx->params.alg_params.slow.use_len2_matches)
1565 min_match_len = max(min_match_len, 3);
1566 const uint32_t max_search_depth = ctx->params.alg_params.slow.max_search_depth;
1567 const uint32_t max_matches_per_pos = ctx->params.alg_params.slow.max_matches_per_pos;
1569 if (unlikely(max_search_depth == 0 || max_matches_per_pos == 0))
1572 num_matches = lzx_lz_get_matches(ctx->match_window_pos,
1580 max_matches_per_pos);
1581 matches[-1].len = num_matches;
1583 ctx->cached_matches_pos += num_matches + 1;
1584 *matches_ret = matches;
1586 /* Cap the length of returned matches to the number of bytes remaining,
1587 * if it is not the whole window. */
1588 if (ctx->match_window_end < ctx->window_size) {
1589 unsigned maxlen = ctx->match_window_end - ctx->match_window_pos;
1590 for (unsigned i = 0; i < num_matches; i++)
1591 if (matches[i].len > maxlen)
1592 matches[i].len = maxlen;
1595 fprintf(stderr, "Pos %u/%u: %u matches\n",
1596 ctx->match_window_pos, ctx->match_window_end, num_matches);
1597 for (unsigned i = 0; i < num_matches; i++)
1598 fprintf(stderr, "\tLen %u Offset %u\n", matches[i].len, matches[i].offset);
1601 #ifdef ENABLE_LZX_DEBUG
1602 for (unsigned i = 0; i < num_matches; i++) {
1603 LZX_ASSERT(matches[i].len >= LZX_MIN_MATCH_LEN);
1604 LZX_ASSERT(matches[i].len <= LZX_MAX_MATCH_LEN);
1605 LZX_ASSERT(matches[i].len <= ctx->match_window_end - ctx->match_window_pos);
1606 LZX_ASSERT(matches[i].offset > 0);
1607 LZX_ASSERT(matches[i].offset <= ctx->match_window_pos);
1608 LZX_ASSERT(!memcmp(&ctx->window[ctx->match_window_pos],
1609 &ctx->window[ctx->match_window_pos - matches[i].offset],
1614 ctx->match_window_pos++;
1619 * Reverse the linked list of near-optimal matches so that they can be returned
1620 * in forwards order.
1622 * Returns the first match in the list.
1624 static struct raw_match
1625 lzx_lz_reverse_near_optimal_match_list(struct lzx_compressor *ctx,
1628 unsigned prev_link, saved_prev_link;
1629 unsigned prev_match_offset, saved_prev_match_offset;
1631 ctx->optimum_end_idx = cur_pos;
1633 saved_prev_link = ctx->optimum[cur_pos].prev.link;
1634 saved_prev_match_offset = ctx->optimum[cur_pos].prev.match_offset;
1637 prev_link = saved_prev_link;
1638 prev_match_offset = saved_prev_match_offset;
1640 saved_prev_link = ctx->optimum[prev_link].prev.link;
1641 saved_prev_match_offset = ctx->optimum[prev_link].prev.match_offset;
1643 ctx->optimum[prev_link].next.link = cur_pos;
1644 ctx->optimum[prev_link].next.match_offset = prev_match_offset;
1646 cur_pos = prev_link;
1647 } while (cur_pos != 0);
1649 ctx->optimum_cur_idx = ctx->optimum[0].next.link;
1651 return (struct raw_match)
1652 { .len = ctx->optimum_cur_idx,
1653 .offset = ctx->optimum[0].next.match_offset,
1658 * lzx_lz_get_near_optimal_match() -
1660 * Choose the optimal match or literal to use at the next position in the input.
1662 * Unlike a greedy parser that always takes the longest match, or even a
1663 * parser with one match/literal look-ahead like zlib, the algorithm used here
1664 * may look ahead many matches/literals to determine the optimal match/literal to
1665 * output next. The motivation is that the compression ratio is improved if the
1666 * compressor can do things like use a shorter-than-possible match in order to
1667 * allow a longer match later, and also take into account the Huffman code cost
1668 * model rather than simply assuming that longer is better.
1670 * Still, this is not truly an optimal parser because very long matches are
1671 * taken immediately, and the raw match-finder takes some shortcuts. This is
1672 * done to avoid considering many different alternatives that are unlikely to
1673 * be significantly better.
1675 * This algorithm is based on that used in 7-Zip's DEFLATE encoder.
1677 * Each call to this function does one of two things:
1679 * 1. Build a near-optimal sequence of matches/literals, up to some point, that
1680 * will be returned by subsequent calls to this function, then return the
1685 * 2. Return the next match/literal previously computed by a call to this
1688 * This function relies on the following state in the compressor context:
1690 * ctx->window (read-only: preprocessed data being compressed)
1691 * ctx->cost (read-only: cost model to use)
1692 * ctx->optimum (internal state; leave uninitialized)
1693 * ctx->optimum_cur_idx (must set to 0 before first call)
1694 * ctx->optimum_end_idx (must set to 0 before first call)
1696 * Plus any state used by the raw match-finder.
1698 * The return value is a (length, offset) pair specifying the match or literal
1699 * chosen. For literals, the length is less than LZX_MIN_MATCH_LEN and the
1700 * offset is meaningless.
1702 static struct raw_match
1703 lzx_lz_get_near_optimal_match(struct lzx_compressor * ctx)
1705 unsigned num_possible_matches;
1706 struct raw_match *possible_matches;
1707 struct raw_match match;
1708 unsigned longest_match_len;
1710 if (ctx->optimum_cur_idx != ctx->optimum_end_idx) {
1711 /* Case 2: Return the next match/literal already found. */
1712 match.len = ctx->optimum[ctx->optimum_cur_idx].next.link -
1713 ctx->optimum_cur_idx;
1714 match.offset = ctx->optimum[ctx->optimum_cur_idx].next.match_offset;
1716 ctx->optimum_cur_idx = ctx->optimum[ctx->optimum_cur_idx].next.link;
1720 /* Case 1: Compute a new list of matches/literals to return. */
1722 ctx->optimum_cur_idx = 0;
1723 ctx->optimum_end_idx = 0;
1725 /* Get matches at this position. */
1726 num_possible_matches = lzx_lz_get_matches_caching(ctx, &ctx->queue, &possible_matches);
1728 /* If no matches found, return literal. */
1729 if (num_possible_matches == 0)
1730 return (struct raw_match){ .len = 0 };
1732 /* The matches that were found are sorted in decreasing order by length.
1733 * Get the length of the longest one. */
1734 longest_match_len = possible_matches[0].len;
1736 /* Greedy heuristic: if the longest match that was found is greater
1737 * than the number of fast bytes, return it immediately; don't both
1738 * doing more work. */
1739 if (longest_match_len > ctx->params.alg_params.slow.num_fast_bytes) {
1740 lzx_lz_skip_bytes(ctx, longest_match_len - 1);
1741 return possible_matches[0];
1744 /* Calculate the cost to reach the next position by outputting a
1746 ctx->optimum[0].queue = ctx->queue;
1747 ctx->optimum[1].queue = ctx->optimum[0].queue;
1748 ctx->optimum[1].cost = lzx_literal_cost(ctx->window[ctx->match_window_pos],
1750 ctx->optimum[1].prev.link = 0;
1752 /* Calculate the cost to reach any position up to and including that
1753 * reached by the longest match, using the shortest (i.e. closest) match
1754 * that reaches each position. */
1755 BUILD_BUG_ON(LZX_MIN_MATCH_LEN != 2);
1756 for (unsigned len = LZX_MIN_MATCH_LEN, match_idx = num_possible_matches - 1;
1757 len <= longest_match_len; len++) {
1759 LZX_ASSERT(match_idx < num_possible_matches);
1761 ctx->optimum[len].queue = ctx->optimum[0].queue;
1762 ctx->optimum[len].prev.link = 0;
1763 ctx->optimum[len].prev.match_offset = possible_matches[match_idx].offset;
1764 ctx->optimum[len].cost = lzx_match_cost(len,
1765 possible_matches[match_idx].offset,
1767 &ctx->optimum[len].queue);
1768 if (len == possible_matches[match_idx].len)
1772 unsigned cur_pos = 0;
1774 /* len_end: greatest index forward at which costs have been calculated
1776 unsigned len_end = longest_match_len;
1779 /* Advance to next position. */
1782 if (cur_pos == len_end || cur_pos == LZX_OPTIM_ARRAY_SIZE)
1783 return lzx_lz_reverse_near_optimal_match_list(ctx, cur_pos);
1785 /* retrieve the number of matches available at this position */
1786 num_possible_matches = lzx_lz_get_matches_caching(ctx, &ctx->optimum[cur_pos].queue,
1789 unsigned new_len = 0;
1791 if (num_possible_matches != 0) {
1792 new_len = possible_matches[0].len;
1794 /* Greedy heuristic: if we found a match greater than
1795 * the number of fast bytes, stop immediately. */
1796 if (new_len > ctx->params.alg_params.slow.num_fast_bytes) {
1798 /* Build the list of matches to return and get
1800 match = lzx_lz_reverse_near_optimal_match_list(ctx, cur_pos);
1802 /* Append the long match to the end of the list. */
1803 ctx->optimum[cur_pos].next.match_offset =
1804 possible_matches[0].offset;
1805 ctx->optimum[cur_pos].next.link = cur_pos + new_len;
1806 ctx->optimum_end_idx = cur_pos + new_len;
1808 /* Skip over the remaining bytes of the long match. */
1809 lzx_lz_skip_bytes(ctx, new_len - 1);
1811 /* Return first match in the list */
1816 /* Consider proceeding with a literal byte. */
1817 block_cost_t cur_cost = ctx->optimum[cur_pos].cost;
1818 block_cost_t cur_plus_literal_cost = cur_cost +
1819 lzx_literal_cost(ctx->window[ctx->match_window_pos - 1],
1821 if (cur_plus_literal_cost < ctx->optimum[cur_pos + 1].cost) {
1822 ctx->optimum[cur_pos + 1].cost = cur_plus_literal_cost;
1823 ctx->optimum[cur_pos + 1].prev.link = cur_pos;
1824 ctx->optimum[cur_pos + 1].queue = ctx->optimum[cur_pos].queue;
1827 if (num_possible_matches == 0)
1830 /* Consider proceeding with a match. */
1832 while (len_end < cur_pos + new_len)
1833 ctx->optimum[++len_end].cost = INFINITE_BLOCK_COST;
1835 for (unsigned len = LZX_MIN_MATCH_LEN, match_idx = num_possible_matches - 1;
1836 len <= new_len; len++) {
1837 LZX_ASSERT(match_idx < num_possible_matches);
1838 struct lzx_lru_queue q = ctx->optimum[cur_pos].queue;
1839 block_cost_t cost = cur_cost + lzx_match_cost(len,
1840 possible_matches[match_idx].offset,
1844 if (cost < ctx->optimum[cur_pos + len].cost) {
1845 ctx->optimum[cur_pos + len].cost = cost;
1846 ctx->optimum[cur_pos + len].prev.link = cur_pos;
1847 ctx->optimum[cur_pos + len].prev.match_offset =
1848 possible_matches[match_idx].offset;
1849 ctx->optimum[cur_pos + len].queue = q;
1852 if (len == possible_matches[match_idx].len)
1859 * Set default symbol costs.
1862 lzx_set_default_costs(struct lzx_costs * costs)
1866 /* Literal symbols */
1867 for (i = 0; i < LZX_NUM_CHARS; i++)
1870 /* Match header symbols */
1871 for (; i < LZX_MAINCODE_NUM_SYMBOLS; i++)
1872 costs->main[i] = 10;
1874 /* Length symbols */
1875 for (i = 0; i < LZX_LENCODE_NUM_SYMBOLS; i++)
1878 /* Aligned offset symbols */
1879 for (i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++)
1880 costs->aligned[i] = 3;
1883 /* Given the frequencies of symbols in a compressed block and the corresponding
1884 * Huffman codes, return LZX_BLOCKTYPE_ALIGNED or LZX_BLOCKTYPE_VERBATIM if an
1885 * aligned offset or verbatim block, respectively, will take fewer bits to
1888 lzx_choose_verbatim_or_aligned(const struct lzx_freqs * freqs,
1889 const struct lzx_codes * codes)
1891 unsigned aligned_cost = 0;
1892 unsigned verbatim_cost = 0;
1894 /* Verbatim blocks have a constant 3 bits per position footer. Aligned
1895 * offset blocks have an aligned offset symbol per position footer, plus
1896 * an extra 24 bits to output the lengths necessary to reconstruct the
1897 * aligned offset code itself. */
1898 for (unsigned i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) {
1899 verbatim_cost += 3 * freqs->aligned[i];
1900 aligned_cost += codes->lens.aligned[i] * freqs->aligned[i];
1902 aligned_cost += LZX_ALIGNEDCODE_ELEMENT_SIZE * LZX_ALIGNEDCODE_NUM_SYMBOLS;
1903 if (aligned_cost < verbatim_cost)
1904 return LZX_BLOCKTYPE_ALIGNED;
1906 return LZX_BLOCKTYPE_VERBATIM;
1909 /* Find a near-optimal sequence of matches/literals with which to output the
1910 * specified LZX block, then set its type to that which has the minimum cost to
1913 lzx_optimize_block(struct lzx_compressor *ctx, struct lzx_block_spec *spec,
1914 unsigned num_passes)
1916 const struct lzx_lru_queue orig_queue = ctx->queue;
1917 struct lzx_freqs freqs;
1919 ctx->match_window_end = spec->window_pos + spec->block_size;
1920 spec->chosen_matches_start_pos = spec->window_pos;
1922 LZX_ASSERT(num_passes >= 1);
1924 /* The first optimal parsing pass is done using the cost model already
1925 * set in ctx->costs. Each later pass is done using a cost model
1926 * computed from the previous pass. */
1927 for (unsigned pass = 0; pass < num_passes; pass++) {
1929 lzx_lz_rewind_matchfinder(ctx, spec->window_pos);
1930 ctx->queue = orig_queue;
1931 spec->num_chosen_matches = 0;
1932 memset(&freqs, 0, sizeof(freqs));
1934 for (unsigned i = spec->window_pos; i < spec->window_pos + spec->block_size; ) {
1935 struct raw_match raw_match;
1936 struct lzx_match lzx_match;
1938 raw_match = lzx_lz_get_near_optimal_match(ctx);
1939 if (raw_match.len >= LZX_MIN_MATCH_LEN) {
1940 lzx_match.data = lzx_tally_match(raw_match.len, raw_match.offset,
1941 &freqs, &ctx->queue);
1944 lzx_match.data = lzx_tally_literal(ctx->window[i], &freqs);
1947 ctx->chosen_matches[spec->chosen_matches_start_pos +
1948 spec->num_chosen_matches++] = lzx_match;
1951 lzx_make_huffman_codes(&freqs, &spec->codes);
1952 if (pass < num_passes - 1)
1953 lzx_set_costs(ctx, &spec->codes.lens);
1955 spec->block_type = lzx_choose_verbatim_or_aligned(&freqs, &spec->codes);
1959 lzx_optimize_blocks(struct lzx_compressor *ctx)
1961 lzx_lru_queue_init(&ctx->queue);
1962 ctx->optimum_cur_idx = 0;
1963 ctx->optimum_end_idx = 0;
1965 const unsigned num_passes = ctx->params.alg_params.slow.num_optim_passes;
1967 for (unsigned i = 0; i < ctx->num_blocks; i++)
1968 lzx_optimize_block(ctx, &ctx->block_specs[i], num_passes);
1971 /* Initialize the suffix array match-finder for the specified input. */
1973 lzx_lz_init_matchfinder(const u8 T[const restrict],
1974 const input_idx_t n,
1975 input_idx_t SA[const restrict],
1976 input_idx_t ISA[const restrict],
1977 struct salink link[const restrict],
1978 const unsigned max_match_len)
1980 /* Compute SA (Suffix Array). */
1984 /* ISA and link are used as temporary space. */
1985 BUILD_BUG_ON(LZX_MAX_WINDOW_SIZE * sizeof(ISA[0]) < 256 * sizeof(saidx_t));
1986 BUILD_BUG_ON(LZX_MAX_WINDOW_SIZE * 2 * sizeof(link[0]) < 256 * 256 * sizeof(saidx_t));
1987 divsufsort(T, sa, n, (saidx_t*)ISA, (saidx_t*)link);
1988 for (input_idx_t i = 0; i < n; i++)
1992 #ifdef ENABLE_LZX_DEBUG
1996 /* Verify suffix array. */
2000 for (input_idx_t r = 0; r < n; r++) {
2001 input_idx_t i = SA[r];
2003 LZX_ASSERT(!found[i]);
2008 for (input_idx_t r = 0; r < n - 1; r++) {
2010 input_idx_t i1 = SA[r];
2011 input_idx_t i2 = SA[r + 1];
2013 input_idx_t n1 = n - i1;
2014 input_idx_t n2 = n - i2;
2016 LZX_ASSERT(memcmp(&T[i1], &T[i2], min(n1, n2)) <= 0);
2018 LZX_DEBUG("Verified SA (len %u)", n);
2019 #endif /* ENABLE_LZX_DEBUG */
2021 /* Compute ISA (Inverse Suffix Array) */
2022 for (input_idx_t r = 0; r < n; r++)
2027 /* Compute LCP (longest common prefix) array.
2029 * Algorithm adapted from Kasai et al. 2001: "Linear-Time
2030 * Longest-Common-Prefix Computation in Suffix Arrays and Its
2034 for (input_idx_t i = 0; i < n; i++) {
2035 input_idx_t r = ISA[i];
2037 input_idx_t j = SA[r - 1];
2039 input_idx_t lim = min(n - i, n - j);
2041 while (h < lim && T[i + h] == T[j + h])
2050 #ifdef ENABLE_LZX_DEBUG
2051 /* Verify LCP array. */
2052 for (input_idx_t r = 0; r < n - 1; r++) {
2053 LZX_ASSERT(ISA[SA[r]] == r);
2054 LZX_ASSERT(ISA[SA[r + 1]] == r + 1);
2056 input_idx_t i1 = SA[r];
2057 input_idx_t i2 = SA[r + 1];
2058 input_idx_t lcp = LCP[r + 1];
2060 input_idx_t n1 = n - i1;
2061 input_idx_t n2 = n - i2;
2063 LZX_ASSERT(lcp <= min(n1, n2));
2065 LZX_ASSERT(memcmp(&T[i1], &T[i2], lcp) == 0);
2066 if (lcp < min(n1, n2))
2067 LZX_ASSERT(T[i1 + lcp] != T[i2 + lcp]);
2069 #endif /* ENABLE_LZX_DEBUG */
2071 /* Compute salink.next and salink.lcpnext.
2073 * Algorithm adapted from Crochemore et al. 2009:
2074 * "LPF computation revisited".
2076 * Note: we cap lcpnext to the maximum match length so that the
2077 * match-finder need not worry about it later. */
2078 link[n - 1].next = (input_idx_t)~0U;
2079 link[n - 1].prev = (input_idx_t)~0U;
2080 link[n - 1].lcpnext = 0;
2081 link[n - 1].lcpprev = 0;
2082 for (input_idx_t r = n - 2; r != (input_idx_t)~0U; r--) {
2083 input_idx_t t = r + 1;
2084 input_idx_t l = LCP[t];
2085 while (t != (input_idx_t)~0 && SA[t] > SA[r]) {
2086 l = min(l, link[t].lcpnext);
2090 link[r].lcpnext = min(l, max_match_len);
2091 LZX_ASSERT(t == (input_idx_t)~0U || l <= n - SA[t]);
2092 LZX_ASSERT(l <= n - SA[r]);
2093 LZX_ASSERT(memcmp(&T[SA[r]], &T[SA[t]], l) == 0);
2096 /* Compute salink.prev and salink.lcpprev.
2098 * Algorithm adapted from Crochemore et al. 2009:
2099 * "LPF computation revisited".
2101 * Note: we cap lcpprev to the maximum match length so that the
2102 * match-finder need not worry about it later. */
2103 link[0].prev = (input_idx_t)~0;
2104 link[0].next = (input_idx_t)~0;
2105 link[0].lcpprev = 0;
2106 link[0].lcpnext = 0;
2107 for (input_idx_t r = 1; r < n; r++) {
2108 input_idx_t t = r - 1;
2109 input_idx_t l = LCP[r];
2110 while (t != (input_idx_t)~0 && SA[t] > SA[r]) {
2111 l = min(l, link[t].lcpprev);
2115 link[r].lcpprev = min(l, max_match_len);
2116 LZX_ASSERT(t == (input_idx_t)~0 || l <= n - SA[t]);
2117 LZX_ASSERT(l <= n - SA[r]);
2118 LZX_ASSERT(memcmp(&T[SA[r]], &T[SA[t]], l) == 0);
2123 /* Prepare the input window into one or more LZX blocks ready to be output. */
2125 lzx_prepare_blocks(struct lzx_compressor * ctx)
2127 /* Initialize the match-finder. */
2128 lzx_lz_init_matchfinder(ctx->window, ctx->window_size,
2129 ctx->SA, ctx->ISA, ctx->salink,
2131 ctx->cached_matches_pos = 0;
2132 ctx->matches_cached = false;
2133 ctx->match_window_pos = 0;
2135 /* Set up a default cost model. */
2136 lzx_set_default_costs(&ctx->costs);
2138 /* Assume that the entire input will be one LZX block. */
2139 ctx->block_specs[0].window_pos = 0;
2140 ctx->block_specs[0].block_size = ctx->window_size;
2141 ctx->num_blocks = 1;
2143 /* Determine sequence of matches/literals to output for each block. */
2144 lzx_optimize_blocks(ctx);
2148 * This is the fast version of lzx_prepare_blocks(). This version "quickly"
2149 * prepares a single compressed block containing the entire input. See the
2150 * description of the "Fast algorithm" at the beginning of this file for more
2153 * Input --- the preprocessed data:
2158 * Output --- the block specification and the corresponding match/literal data:
2160 * ctx->block_specs[]
2162 * ctx->chosen_matches[]
2165 lzx_prepare_block_fast(struct lzx_compressor * ctx)
2167 struct lzx_record_ctx record_ctx;
2168 struct lzx_block_spec *spec;
2170 /* Parameters to hash chain LZ match finder
2171 * (lazy with 1 match lookahead) */
2172 static const struct lz_params lzx_lz_params = {
2173 /* Although LZX_MIN_MATCH_LEN == 2, length 2 matches typically
2174 * aren't worth choosing when using greedy or lazy parsing. */
2176 .max_match = LZX_MAX_MATCH_LEN,
2177 .max_offset = 32768,
2178 .good_match = LZX_MAX_MATCH_LEN,
2179 .nice_match = LZX_MAX_MATCH_LEN,
2180 .max_chain_len = LZX_MAX_MATCH_LEN,
2181 .max_lazy_match = LZX_MAX_MATCH_LEN,
2185 /* Initialize symbol frequencies and match offset LRU queue. */
2186 memset(&record_ctx.freqs, 0, sizeof(struct lzx_freqs));
2187 lzx_lru_queue_init(&record_ctx.queue);
2188 record_ctx.matches = ctx->chosen_matches;
2190 /* Determine series of matches/literals to output. */
2192 input_idx_t prev_tab[ctx->window_size];
2193 lz_analyze_block(ctx->window,
2203 /* Set up block specification. */
2204 spec = &ctx->block_specs[0];
2205 spec->block_type = LZX_BLOCKTYPE_ALIGNED;
2206 spec->window_pos = 0;
2207 spec->block_size = ctx->window_size;
2208 spec->num_chosen_matches = (record_ctx.matches - ctx->chosen_matches);
2209 spec->chosen_matches_start_pos = 0;
2210 lzx_make_huffman_codes(&record_ctx.freqs, &spec->codes);
2211 ctx->num_blocks = 1;
2215 do_call_insn_translation(u32 *call_insn_target, int input_pos,
2221 rel_offset = le32_to_cpu(*call_insn_target);
2222 if (rel_offset >= -input_pos && rel_offset < file_size) {
2223 if (rel_offset < file_size - input_pos) {
2224 /* "good translation" */
2225 abs_offset = rel_offset + input_pos;
2227 /* "compensating translation" */
2228 abs_offset = rel_offset - file_size;
2230 *call_insn_target = cpu_to_le32(abs_offset);
2234 /* This is the reverse of undo_call_insn_preprocessing() in lzx-decompress.c.
2235 * See the comment above that function for more information. */
2237 do_call_insn_preprocessing(u8 data[], int size)
2239 for (int i = 0; i < size - 10; i++) {
2240 if (data[i] == 0xe8) {
2241 do_call_insn_translation((u32*)&data[i + 1], i,
2242 LZX_WIM_MAGIC_FILESIZE);
2248 /* API function documented in wimlib.h */
2250 wimlib_lzx_compress2(const void * const restrict uncompressed_data,
2251 unsigned const uncompressed_len,
2252 void * const restrict compressed_data,
2253 struct wimlib_lzx_context * const restrict lzx_ctx)
2255 struct lzx_compressor *ctx = (struct lzx_compressor*)lzx_ctx;
2256 struct output_bitstream ostream;
2257 input_idx_t compressed_len;
2259 if (uncompressed_len < 100) {
2260 LZX_DEBUG("Too small to bother compressing.");
2264 if (uncompressed_len > 32768) {
2265 LZX_DEBUG("Only up to 32768 bytes of uncompressed data are supported.");
2269 wimlib_assert(lzx_ctx != NULL);
2271 LZX_DEBUG("Attempting to compress %u bytes...", uncompressed_len);
2273 /* The input data must be preprocessed. To avoid changing the original
2274 * input, copy it to a temporary buffer. */
2275 memcpy(ctx->window, uncompressed_data, uncompressed_len);
2276 ctx->window_size = uncompressed_len;
2278 /* This line is unnecessary; it just avoids inconsequential accesses of
2279 * uninitialized memory that would show up in memory-checking tools such
2281 memset(&ctx->window[ctx->window_size], 0, 12);
2283 LZX_DEBUG("Preprocessing data...");
2285 /* Before doing any actual compression, do the call instruction (0xe8
2286 * byte) translation on the uncompressed data. */
2287 do_call_insn_preprocessing(ctx->window, ctx->window_size);
2289 LZX_DEBUG("Preparing blocks...");
2291 /* Prepare the compressed data. */
2292 if (ctx->params.algorithm == WIMLIB_LZX_ALGORITHM_FAST)
2293 lzx_prepare_block_fast(ctx);
2295 lzx_prepare_blocks(ctx);
2297 LZX_DEBUG("Writing compressed blocks...");
2299 /* Generate the compressed data. */
2300 init_output_bitstream(&ostream, compressed_data, ctx->window_size - 1);
2301 lzx_write_all_blocks(ctx, &ostream);
2303 LZX_DEBUG("Flushing bitstream...");
2304 compressed_len = flush_output_bitstream(&ostream);
2305 if (compressed_len == ~(input_idx_t)0) {
2306 LZX_DEBUG("Data did not compress to less than original length!");
2310 LZX_DEBUG("Done: compressed %u => %u bytes.",
2311 uncompressed_len, compressed_len);
2313 /* Verify that we really get the same thing back when decompressing.
2314 * Although this could be disabled by default in all cases, it only
2315 * takes around 2-3% of the running time of the slow algorithm to do the
2317 if (ctx->params.algorithm == WIMLIB_LZX_ALGORITHM_SLOW
2318 #if defined(ENABLE_LZX_DEBUG) || defined(ENABLE_VERIFY_COMPRESSION)
2323 u8 buf[uncompressed_len];
2325 if (wimlib_lzx_decompress(compressed_data, compressed_len,
2326 buf, uncompressed_len))
2328 ERROR("Failed to decompress data we "
2329 "compressed using LZX algorithm");
2334 if (memcmp(uncompressed_data, buf, uncompressed_len)) {
2335 ERROR("Data we compressed using LZX algorithm "
2336 "didn't decompress to original");
2341 return compressed_len;
2345 lzx_params_compatible(const struct wimlib_lzx_params *oldparams,
2346 const struct wimlib_lzx_params *newparams)
2348 return 0 == memcmp(oldparams, newparams, sizeof(struct wimlib_lzx_params));
2351 static struct wimlib_lzx_params lzx_user_default_params;
2352 static struct wimlib_lzx_params *lzx_user_default_params_ptr;
2355 lzx_params_valid(const struct wimlib_lzx_params *params)
2357 /* Validate parameters. */
2358 if (params->size_of_this != sizeof(struct wimlib_lzx_params)) {
2359 LZX_DEBUG("Invalid parameter structure size!");
2363 if (params->algorithm != WIMLIB_LZX_ALGORITHM_SLOW &&
2364 params->algorithm != WIMLIB_LZX_ALGORITHM_FAST)
2366 LZX_DEBUG("Invalid algorithm.");
2370 if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) {
2371 if (params->alg_params.slow.num_optim_passes < 1)
2373 LZX_DEBUG("Invalid number of optimization passes!");
2377 if (params->alg_params.slow.main_nostat_cost < 1 ||
2378 params->alg_params.slow.main_nostat_cost > 16)
2380 LZX_DEBUG("Invalid main_nostat_cost!");
2384 if (params->alg_params.slow.len_nostat_cost < 1 ||
2385 params->alg_params.slow.len_nostat_cost > 16)
2387 LZX_DEBUG("Invalid len_nostat_cost!");
2391 if (params->alg_params.slow.aligned_nostat_cost < 1 ||
2392 params->alg_params.slow.aligned_nostat_cost > 8)
2394 LZX_DEBUG("Invalid aligned_nostat_cost!");
2401 /* API function documented in wimlib.h */
2403 wimlib_lzx_set_default_params(const struct wimlib_lzx_params * params)
2406 if (!lzx_params_valid(params))
2407 return WIMLIB_ERR_INVALID_PARAM;
2408 lzx_user_default_params = *params;
2409 lzx_user_default_params_ptr = &lzx_user_default_params;
2411 lzx_user_default_params_ptr = NULL;
2416 /* API function documented in wimlib.h */
2418 wimlib_lzx_alloc_context(const struct wimlib_lzx_params *params,
2419 struct wimlib_lzx_context **ctx_pp)
2422 LZX_DEBUG("Allocating LZX context...");
2424 struct lzx_compressor *ctx;
2426 static const struct wimlib_lzx_params fast_default = {
2427 .size_of_this = sizeof(struct wimlib_lzx_params),
2428 .algorithm = WIMLIB_LZX_ALGORITHM_FAST,
2435 static const struct wimlib_lzx_params slow_default = {
2436 .size_of_this = sizeof(struct wimlib_lzx_params),
2437 .algorithm = WIMLIB_LZX_ALGORITHM_SLOW,
2441 .use_len2_matches = 1,
2442 .num_fast_bytes = 32,
2443 .num_optim_passes = 2,
2444 .max_search_depth = 50,
2445 .max_matches_per_pos = 3,
2446 .main_nostat_cost = 15,
2447 .len_nostat_cost = 15,
2448 .aligned_nostat_cost = 7,
2454 if (!lzx_params_valid(params))
2455 return WIMLIB_ERR_INVALID_PARAM;
2457 LZX_DEBUG("Using default algorithm and parameters.");
2458 if (lzx_user_default_params_ptr)
2459 params = lzx_user_default_params_ptr;
2461 params = &slow_default;
2464 if (params->use_defaults) {
2465 if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW)
2466 params = &slow_default;
2468 params = &fast_default;
2472 ctx = *(struct lzx_compressor**)ctx_pp;
2474 if (ctx && lzx_params_compatible(&ctx->params, params))
2477 LZX_DEBUG("Check parameters only.");
2481 LZX_DEBUG("Allocating memory.");
2483 ctx = MALLOC(sizeof(struct lzx_compressor));
2487 size_t block_specs_length;
2490 if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW)
2491 block_specs_length = 1U << params->alg_params.slow.num_split_passes;
2494 block_specs_length = 1U;
2495 ctx->block_specs = MALLOC(block_specs_length * sizeof(ctx->block_specs[0]));
2496 if (ctx->block_specs == NULL)
2499 if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) {
2500 ctx->SA = MALLOC(3U * LZX_MAX_WINDOW_SIZE * sizeof(ctx->SA[0]));
2501 if (ctx->SA == NULL)
2502 goto err_free_block_specs;
2503 ctx->ISA = ctx->SA + LZX_MAX_WINDOW_SIZE;
2504 ctx->salink = MALLOC(LZX_MAX_WINDOW_SIZE * sizeof(ctx->salink[0]));
2505 if (ctx->salink == NULL)
2513 if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) {
2514 ctx->optimum = MALLOC((LZX_OPTIM_ARRAY_SIZE + LZX_MAX_MATCH_LEN) *
2515 sizeof(ctx->optimum[0]));
2516 if (ctx->optimum == NULL)
2517 goto err_free_salink;
2519 ctx->optimum = NULL;
2522 if (params->algorithm == WIMLIB_LZX_ALGORITHM_SLOW) {
2523 uint32_t cache_per_pos;
2525 cache_per_pos = params->alg_params.slow.max_matches_per_pos;
2526 if (cache_per_pos > LZX_MAX_CACHE_PER_POS)
2527 cache_per_pos = LZX_MAX_CACHE_PER_POS;
2529 ctx->cached_matches = MALLOC(LZX_MAX_WINDOW_SIZE * (cache_per_pos + 1) *
2530 sizeof(ctx->cached_matches[0]));
2531 if (ctx->cached_matches == NULL)
2532 goto err_free_optimum;
2534 ctx->cached_matches = NULL;
2537 ctx->chosen_matches = MALLOC(LZX_MAX_WINDOW_SIZE *
2538 sizeof(ctx->chosen_matches[0]));
2539 if (ctx->chosen_matches == NULL)
2540 goto err_free_cached_matches;
2542 memcpy(&ctx->params, params, sizeof(struct wimlib_lzx_params));
2543 memset(&ctx->zero_codes, 0, sizeof(ctx->zero_codes));
2545 LZX_DEBUG("Successfully allocated new LZX context.");
2547 wimlib_lzx_free_context(*ctx_pp);
2548 *ctx_pp = (struct wimlib_lzx_context*)ctx;
2551 err_free_cached_matches:
2552 FREE(ctx->cached_matches);
2559 err_free_block_specs:
2560 FREE(ctx->block_specs);
2564 LZX_DEBUG("Ran out of memory.");
2565 return WIMLIB_ERR_NOMEM;
2568 /* API function documented in wimlib.h */
2570 wimlib_lzx_free_context(struct wimlib_lzx_context *_ctx)
2572 struct lzx_compressor *ctx = (struct lzx_compressor*)_ctx;
2575 FREE(ctx->chosen_matches);
2576 FREE(ctx->cached_matches);
2580 FREE(ctx->block_specs);
2585 /* API function documented in wimlib.h */
2587 wimlib_lzx_compress(const void * const restrict uncompressed_data,
2588 unsigned const uncompressed_len,
2589 void * const restrict compressed_data)
2592 struct wimlib_lzx_context *ctx = NULL;
2593 unsigned compressed_len;
2595 ret = wimlib_lzx_alloc_context(NULL, &ctx);
2597 wimlib_assert(ret != WIMLIB_ERR_INVALID_PARAM);
2598 WARNING("Couldn't allocate LZX compression context: %"TS"",
2599 wimlib_get_error_string(ret));
2603 compressed_len = wimlib_lzx_compress2(uncompressed_data,
2608 wimlib_lzx_free_context(ctx);
2610 return compressed_len;