4 * A compressor for the LZX compression format, as used in WIM archives.
8 * Copyright (C) 2012-2017 Eric Biggers
10 * This file is free software; you can redistribute it and/or modify it under
11 * the terms of the GNU Lesser General Public License as published by the Free
12 * Software Foundation; either version 3 of the License, or (at your option) any
15 * This file is distributed in the hope that it will be useful, but WITHOUT
16 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
17 * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
20 * You should have received a copy of the GNU Lesser General Public License
21 * along with this file; if not, see http://www.gnu.org/licenses/.
26 * This file contains a compressor for the LZX ("Lempel-Ziv eXtended")
27 * compression format, as used in the WIM (Windows IMaging) file format.
29 * Two different LZX-compatible algorithms are implemented: "near-optimal" and
30 * "lazy". "Near-optimal" is significantly slower than "lazy", but results in a
31 * better compression ratio. The "near-optimal" algorithm is used at the
32 * default compression level.
34 * This file may need some slight modifications to be used outside of the WIM
35 * format. In particular, in other situations the LZX block header might be
36 * slightly different, and sliding window support might be required.
38 * LZX is a compression format derived from DEFLATE, the format used by zlib and
39 * gzip. Both LZX and DEFLATE use LZ77 matching and Huffman coding. Certain
40 * details are quite similar, such as the method for storing Huffman codes.
41 * However, the main differences are:
43 * - LZX preprocesses the data to attempt to make x86 machine code slightly more
44 * compressible before attempting to compress it further.
46 * - LZX uses a "main" alphabet which combines literals and matches, with the
47 * match symbols containing a "length header" (giving all or part of the match
48 * length) and an "offset slot" (giving, roughly speaking, the order of
49 * magnitude of the match offset).
51 * - LZX does not have static Huffman blocks (that is, the kind with preset
52 * Huffman codes); however it does have two types of dynamic Huffman blocks
53 * ("verbatim" and "aligned").
55 * - LZX has a minimum match length of 2 rather than 3. Length 2 matches can be
56 * useful, but generally only if the compressor is smart about choosing them.
58 * - In LZX, offset slots 0 through 2 actually represent entries in an LRU queue
59 * of match offsets. This is very useful for certain types of files, such as
60 * binary files that have repeating records.
63 /******************************************************************************/
64 /* General parameters */
65 /*----------------------------------------------------------------------------*/
68 * The compressor uses the faster algorithm at levels <= MAX_FAST_LEVEL. It
69 * uses the slower algorithm at levels > MAX_FAST_LEVEL.
71 #define MAX_FAST_LEVEL 34
74 * The compressor-side limits on the codeword lengths (in bits) for each Huffman
75 * code. To make outputting bits slightly faster, some of these limits are
76 * lower than the limits defined by the LZX format. This does not significantly
77 * affect the compression ratio.
79 #define MAIN_CODEWORD_LIMIT 16
80 #define LENGTH_CODEWORD_LIMIT 12
81 #define ALIGNED_CODEWORD_LIMIT 7
82 #define PRE_CODEWORD_LIMIT 7
85 /******************************************************************************/
86 /* Block splitting parameters */
87 /*----------------------------------------------------------------------------*/
90 * The compressor always outputs blocks of at least this size in bytes, except
91 * for the last block which may need to be smaller.
93 #define MIN_BLOCK_SIZE 6500
96 * The compressor attempts to end a block when it reaches this size in bytes.
97 * The final size might be slightly larger due to matches extending beyond the
98 * end of the block. Specifically:
100 * - The near-optimal compressor may choose a match of up to LZX_MAX_MATCH_LEN
101 * bytes starting at position 'SOFT_MAX_BLOCK_SIZE - 1'.
103 * - The lazy compressor may choose a sequence of literals starting at position
104 * 'SOFT_MAX_BLOCK_SIZE - 1' when it sees a sequence of increasingly better
105 * matches. The final match may be up to LZX_MAX_MATCH_LEN bytes. The
106 * length of the literal sequence is approximately limited by the "nice match
109 #define SOFT_MAX_BLOCK_SIZE 100000
112 * The number of observed items (matches and literals) that represents
113 * sufficient data for the compressor to decide whether the current block should
116 #define NUM_OBSERVATIONS_PER_BLOCK_CHECK 400
119 /******************************************************************************/
120 /* Parameters for slower algorithm */
121 /*----------------------------------------------------------------------------*/
124 * The log base 2 of the number of entries in the hash table for finding length
125 * 2 matches. This could be as high as 16, but using a smaller hash table
126 * speeds up compression due to reduced cache pressure.
128 #define BT_MATCHFINDER_HASH2_ORDER 12
131 * The number of lz_match structures in the match cache, excluding the extra
132 * "overflow" entries. This value should be high enough so that nearly the
133 * time, all matches found in a given block can fit in the match cache.
134 * However, fallback behavior (immediately terminating the block) on cache
135 * overflow is still required.
137 #define CACHE_LENGTH (SOFT_MAX_BLOCK_SIZE * 5)
140 * An upper bound on the number of matches that can ever be saved in the match
141 * cache for a single position. Since each match we save for a single position
142 * has a distinct length, we can use the number of possible match lengths in LZX
143 * as this bound. This bound is guaranteed to be valid in all cases, although
144 * if 'nice_match_length < LZX_MAX_MATCH_LEN', then it will never actually be
147 #define MAX_MATCHES_PER_POS LZX_NUM_LENS
150 * A scaling factor that makes it possible to consider fractional bit costs. A
151 * single bit has a cost of BIT_COST.
153 * Note: this is only useful as a statistical trick for when the true costs are
154 * unknown. Ultimately, each token in LZX requires a whole number of bits to
160 * Should the compressor take into account the costs of aligned offset symbols
161 * instead of assuming that all are equally likely?
163 #define CONSIDER_ALIGNED_COSTS 1
166 * Should the "minimum" cost path search algorithm consider "gap" matches, where
167 * a normal match is followed by a literal, then by a match with the same
168 * offset? This is one specific, somewhat common situation in which the true
169 * minimum cost path is often different from the path found by looking only one
172 #define CONSIDER_GAP_MATCHES 1
174 /******************************************************************************/
176 /*----------------------------------------------------------------------------*/
182 #include "wimlib/compress_common.h"
183 #include "wimlib/compressor_ops.h"
184 #include "wimlib/error.h"
185 #include "wimlib/lzx_common.h"
186 #include "wimlib/unaligned.h"
187 #include "wimlib/util.h"
189 /* Note: BT_MATCHFINDER_HASH2_ORDER must be defined before including
190 * bt_matchfinder.h. */
192 /* Matchfinders with 16-bit positions */
194 #define MF_SUFFIX _16
195 #include "wimlib/bt_matchfinder.h"
196 #include "wimlib/hc_matchfinder.h"
198 /* Matchfinders with 32-bit positions */
202 #define MF_SUFFIX _32
203 #include "wimlib/bt_matchfinder.h"
204 #include "wimlib/hc_matchfinder.h"
206 /******************************************************************************/
207 /* Compressor structure */
208 /*----------------------------------------------------------------------------*/
210 /* Codewords for the Huffman codes */
211 struct lzx_codewords {
212 u32 main[LZX_MAINCODE_MAX_NUM_SYMBOLS];
213 u32 len[LZX_LENCODE_NUM_SYMBOLS];
214 u32 aligned[LZX_ALIGNEDCODE_NUM_SYMBOLS];
218 * Codeword lengths, in bits, for the Huffman codes.
220 * A codeword length of 0 means the corresponding codeword has zero frequency.
222 * The main and length codes each have one extra entry for use as a sentinel.
223 * See lzx_write_compressed_code().
226 u8 main[LZX_MAINCODE_MAX_NUM_SYMBOLS + 1];
227 u8 len[LZX_LENCODE_NUM_SYMBOLS + 1];
228 u8 aligned[LZX_ALIGNEDCODE_NUM_SYMBOLS];
231 /* Codewords and lengths for the Huffman codes */
233 struct lzx_codewords codewords;
234 struct lzx_lens lens;
237 /* Symbol frequency counters for the Huffman-encoded alphabets */
239 u32 main[LZX_MAINCODE_MAX_NUM_SYMBOLS];
240 u32 len[LZX_LENCODE_NUM_SYMBOLS];
241 u32 aligned[LZX_ALIGNEDCODE_NUM_SYMBOLS];
244 /* Block split statistics. See the "Block splitting algorithm" section later in
245 * this file for details. */
246 #define NUM_LITERAL_OBSERVATION_TYPES 8
247 #define NUM_MATCH_OBSERVATION_TYPES 2
248 #define NUM_OBSERVATION_TYPES (NUM_LITERAL_OBSERVATION_TYPES + \
249 NUM_MATCH_OBSERVATION_TYPES)
250 struct lzx_block_split_stats {
251 u32 new_observations[NUM_OBSERVATION_TYPES];
252 u32 observations[NUM_OBSERVATION_TYPES];
253 u32 num_new_observations;
254 u32 num_observations;
258 * Represents a run of literals followed by a match or end-of-block. This
259 * structure is needed to temporarily store items chosen by the compressor,
260 * since items cannot be written until all items for the block have been chosen
261 * and the block's Huffman codes have been computed.
263 struct lzx_sequence {
266 * Bits 9..31: the number of literals in this run. This may be 0 and
267 * can be at most about SOFT_MAX_BLOCK_LENGTH. The literals are not
268 * stored explicitly in this structure; instead, they are read directly
269 * from the uncompressed data.
271 * Bits 0..8: the length of the match which follows the literals, or 0
272 * if this literal run was the last in the block, so there is no match
273 * which follows it. This can be at most LZX_MAX_MATCH_LEN.
275 u32 litrunlen_and_matchlen;
276 #define SEQ_MATCHLEN_BITS 9
277 #define SEQ_MATCHLEN_MASK (((u32)1 << SEQ_MATCHLEN_BITS) - 1)
280 * If 'matchlen' doesn't indicate end-of-block, then this contains:
282 * Bits 10..31: either the offset plus LZX_OFFSET_ADJUSTMENT or a recent
283 * offset code, depending on the offset slot encoded in the main symbol.
285 * Bits 0..9: the main symbol.
287 u32 adjusted_offset_and_mainsym;
288 #define SEQ_MAINSYM_BITS 10
289 #define SEQ_MAINSYM_MASK (((u32)1 << SEQ_MAINSYM_BITS) - 1)
290 } _aligned_attribute(8);
293 * This structure represents a byte position in the input buffer and a node in
294 * the graph of possible match/literal choices.
296 * Logically, each incoming edge to this node is labeled with a literal or a
297 * match that can be taken to reach this position from an earlier position; and
298 * each outgoing edge from this node is labeled with a literal or a match that
299 * can be taken to advance from this position to a later position.
301 struct lzx_optimum_node {
303 /* The cost, in bits, of the lowest-cost path that has been found to
304 * reach this position. This can change as progressively lower cost
305 * paths are found to reach this position. */
309 * The best arrival to this node, i.e. the match or literal that was
310 * used to arrive to this position at the given 'cost'. This can change
311 * as progressively lower cost paths are found to reach this position.
313 * For non-gap matches, this variable is divided into two bitfields
314 * whose meanings depend on the item type:
317 * Low bits are 0, high bits are the literal.
319 * Explicit offset matches:
320 * Low bits are the match length, high bits are the offset plus
321 * LZX_OFFSET_ADJUSTMENT.
323 * Repeat offset matches:
324 * Low bits are the match length, high bits are the queue index.
326 * For gap matches, identified by OPTIMUM_GAP_MATCH set, special
327 * behavior applies --- see the code.
330 #define OPTIMUM_OFFSET_SHIFT SEQ_MATCHLEN_BITS
331 #define OPTIMUM_LEN_MASK SEQ_MATCHLEN_MASK
332 #if CONSIDER_GAP_MATCHES
333 # define OPTIMUM_GAP_MATCH 0x80000000
336 } _aligned_attribute(8);
338 /* The cost model for near-optimal parsing */
342 * 'match_cost[offset_slot][len - LZX_MIN_MATCH_LEN]' is the cost of a
343 * length 'len' match which has an offset belonging to 'offset_slot'.
344 * The cost includes the main symbol, the length symbol if required, and
345 * the extra offset bits if any, excluding any entropy-coded bits
346 * (aligned offset bits). It does *not* include the cost of the aligned
347 * offset symbol which may be required.
349 u16 match_cost[LZX_MAX_OFFSET_SLOTS][LZX_NUM_LENS];
351 /* Cost of each symbol in the main code */
352 u32 main[LZX_MAINCODE_MAX_NUM_SYMBOLS];
354 /* Cost of each symbol in the length code */
355 u32 len[LZX_LENCODE_NUM_SYMBOLS];
357 #if CONSIDER_ALIGNED_COSTS
358 /* Cost of each symbol in the aligned offset code */
359 u32 aligned[LZX_ALIGNEDCODE_NUM_SYMBOLS];
363 struct lzx_output_bitstream;
365 /* The main LZX compressor structure */
366 struct lzx_compressor {
368 /* The buffer for preprocessed input data, if not using destructive
372 /* If true, then the compressor need not preserve the input buffer if it
373 * compresses the data successfully */
376 /* Pointer to the compress() implementation chosen at allocation time */
377 void (*impl)(struct lzx_compressor *, const u8 *, size_t,
378 struct lzx_output_bitstream *);
380 /* The log base 2 of the window size for match offset encoding purposes.
381 * This will be >= LZX_MIN_WINDOW_ORDER and <= LZX_MAX_WINDOW_ORDER. */
382 unsigned window_order;
384 /* The number of symbols in the main alphabet. This depends on the
385 * window order, since the window order determines the maximum possible
387 unsigned num_main_syms;
389 /* The "nice" match length: if a match of this length is found, then it
390 * is chosen immediately without further consideration. */
391 unsigned nice_match_length;
393 /* The maximum search depth: at most this many potential matches are
394 * considered at each position. */
395 unsigned max_search_depth;
397 /* The number of optimization passes per block */
398 unsigned num_optim_passes;
400 /* The symbol frequency counters for the current block */
401 struct lzx_freqs freqs;
403 /* Block split statistics for the current block */
404 struct lzx_block_split_stats split_stats;
406 /* The Huffman codes for the current and previous blocks. The one with
407 * index 'codes_index' is for the current block, and the other one is
408 * for the previous block. */
409 struct lzx_codes codes[2];
410 unsigned codes_index;
412 /* The matches and literals that the compressor has chosen for the
413 * current block. The required length of this array is limited by the
414 * maximum number of matches that can ever be chosen for a single block,
415 * plus one for the special entry at the end. */
416 struct lzx_sequence chosen_sequences[
417 DIV_ROUND_UP(SOFT_MAX_BLOCK_SIZE, LZX_MIN_MATCH_LEN) + 1];
419 /* Tables for mapping adjusted offsets to offset slots */
420 u8 offset_slot_tab_1[32768]; /* offset slots [0, 29] */
421 u8 offset_slot_tab_2[128]; /* offset slots [30, 49] */
424 /* Data for lzx_compress_lazy() */
426 /* Hash chains matchfinder (MUST BE LAST!!!) */
428 struct hc_matchfinder_16 hc_mf_16;
429 struct hc_matchfinder_32 hc_mf_32;
433 /* Data for lzx_compress_near_optimal() */
436 * Array of nodes, one per position, for running the
437 * minimum-cost path algorithm.
439 * This array must be large enough to accommodate the
440 * worst-case number of nodes, which occurs if the
441 * compressor finds a match of length LZX_MAX_MATCH_LEN
442 * at position 'SOFT_MAX_BLOCK_SIZE - 1', producing a
443 * block of size 'SOFT_MAX_BLOCK_SIZE - 1 +
444 * LZX_MAX_MATCH_LEN'. Add one for the end-of-block
447 struct lzx_optimum_node optimum_nodes[
448 SOFT_MAX_BLOCK_SIZE - 1 +
449 LZX_MAX_MATCH_LEN + 1];
451 /* The cost model for the current optimization pass */
452 struct lzx_costs costs;
455 * Cached matches for the current block. This array
456 * contains the matches that were found at each position
457 * in the block. Specifically, for each position, there
458 * is a special 'struct lz_match' whose 'length' field
459 * contains the number of matches that were found at
460 * that position; this is followed by the matches
461 * themselves, if any, sorted by strictly increasing
464 * Note: in rare cases, there will be a very high number
465 * of matches in the block and this array will overflow.
466 * If this happens, we force the end of the current
467 * block. CACHE_LENGTH is the length at which we
468 * actually check for overflow. The extra slots beyond
469 * this are enough to absorb the worst case overflow,
470 * which occurs if starting at &match_cache[CACHE_LENGTH
471 * - 1], we write the match count header, then write
472 * MAX_MATCHES_PER_POS matches, then skip searching for
473 * matches at 'LZX_MAX_MATCH_LEN - 1' positions and
474 * write the match count header for each.
476 struct lz_match match_cache[CACHE_LENGTH +
477 MAX_MATCHES_PER_POS +
478 LZX_MAX_MATCH_LEN - 1];
480 /* Binary trees matchfinder (MUST BE LAST!!!) */
482 struct bt_matchfinder_16 bt_mf_16;
483 struct bt_matchfinder_32 bt_mf_32;
489 /******************************************************************************/
490 /* Matchfinder utilities */
491 /*----------------------------------------------------------------------------*/
494 * Will a matchfinder using 16-bit positions be sufficient for compressing
495 * buffers of up to the specified size? The limit could be 65536 bytes, but we
496 * also want to optimize out the use of offset_slot_tab_2 in the 16-bit case.
497 * This requires that the limit be no more than the length of offset_slot_tab_1
500 static forceinline bool
501 lzx_is_16_bit(size_t max_bufsize)
503 STATIC_ASSERT(ARRAY_LEN(((struct lzx_compressor *)0)->offset_slot_tab_1) == 32768);
504 return max_bufsize <= 32768;
508 * Return the offset slot for the specified adjusted match offset.
510 static forceinline unsigned
511 lzx_get_offset_slot(struct lzx_compressor *c, u32 adjusted_offset,
514 if (__builtin_constant_p(adjusted_offset) &&
515 adjusted_offset < LZX_NUM_RECENT_OFFSETS)
516 return adjusted_offset;
517 if (is_16_bit || adjusted_offset < ARRAY_LEN(c->offset_slot_tab_1))
518 return c->offset_slot_tab_1[adjusted_offset];
519 return c->offset_slot_tab_2[adjusted_offset >> 14];
523 * For a match that has the specified length and adjusted offset, tally its main
524 * symbol, and if needed its length symbol; then return its main symbol.
526 static forceinline unsigned
527 lzx_tally_main_and_lensyms(struct lzx_compressor *c, unsigned length,
528 u32 adjusted_offset, bool is_16_bit)
532 if (length >= LZX_MIN_SECONDARY_LEN) {
533 /* Length symbol needed */
534 c->freqs.len[length - LZX_MIN_SECONDARY_LEN]++;
535 mainsym = LZX_NUM_CHARS + LZX_NUM_PRIMARY_LENS;
537 /* No length symbol needed */
538 mainsym = LZX_NUM_CHARS + length - LZX_MIN_MATCH_LEN;
541 mainsym += LZX_NUM_LEN_HEADERS *
542 lzx_get_offset_slot(c, adjusted_offset, is_16_bit);
543 c->freqs.main[mainsym]++;
548 * The following macros call either the 16-bit or the 32-bit version of a
549 * matchfinder function based on the value of 'is_16_bit', which will be known
550 * at compilation time.
553 #define CALL_HC_MF(is_16_bit, c, funcname, ...) \
554 ((is_16_bit) ? CONCAT(funcname, _16)(&(c)->hc_mf_16, ##__VA_ARGS__) : \
555 CONCAT(funcname, _32)(&(c)->hc_mf_32, ##__VA_ARGS__));
557 #define CALL_BT_MF(is_16_bit, c, funcname, ...) \
558 ((is_16_bit) ? CONCAT(funcname, _16)(&(c)->bt_mf_16, ##__VA_ARGS__) : \
559 CONCAT(funcname, _32)(&(c)->bt_mf_32, ##__VA_ARGS__));
561 /******************************************************************************/
562 /* Output bitstream */
563 /*----------------------------------------------------------------------------*/
566 * The LZX bitstream is encoded as a sequence of little endian 16-bit coding
567 * units. Bits are ordered from most significant to least significant within
572 * Structure to keep track of the current state of sending bits to the
573 * compressed output buffer.
575 struct lzx_output_bitstream {
577 /* Bits that haven't yet been written to the output buffer */
578 machine_word_t bitbuf;
580 /* Number of bits currently held in @bitbuf */
581 machine_word_t bitcount;
583 /* Pointer to the start of the output buffer */
586 /* Pointer to the position in the output buffer at which the next coding
587 * unit should be written */
590 /* Pointer to just past the end of the output buffer, rounded down by
591 * one byte if needed to make 'end - start' a multiple of 2 */
595 /* Can the specified number of bits always be added to 'bitbuf' after all
596 * pending 16-bit coding units have been flushed? */
597 #define CAN_BUFFER(n) ((n) <= WORDBITS - 15)
599 /* Initialize the output bitstream to write to the specified buffer. */
601 lzx_init_output(struct lzx_output_bitstream *os, void *buffer, size_t size)
607 os->end = (u8 *)buffer + (size & ~1);
611 * Add some bits to the bitbuffer variable of the output bitstream. The caller
612 * must make sure there is enough room.
614 static forceinline void
615 lzx_add_bits(struct lzx_output_bitstream *os, u32 bits, unsigned num_bits)
617 os->bitbuf = (os->bitbuf << num_bits) | bits;
618 os->bitcount += num_bits;
622 * Flush bits from the bitbuffer variable to the output buffer. 'max_num_bits'
623 * specifies the maximum number of bits that may have been added since the last
626 static forceinline void
627 lzx_flush_bits(struct lzx_output_bitstream *os, unsigned max_num_bits)
629 /* Masking the number of bits to shift is only needed to avoid undefined
630 * behavior; we don't actually care about the results of bad shifts. On
631 * x86, the explicit masking generates no extra code. */
632 const u32 shift_mask = WORDBITS - 1;
634 if (os->end - os->next < 6)
636 put_unaligned_le16(os->bitbuf >> ((os->bitcount - 16) &
637 shift_mask), os->next + 0);
638 if (max_num_bits > 16)
639 put_unaligned_le16(os->bitbuf >> ((os->bitcount - 32) &
640 shift_mask), os->next + 2);
641 if (max_num_bits > 32)
642 put_unaligned_le16(os->bitbuf >> ((os->bitcount - 48) &
643 shift_mask), os->next + 4);
644 os->next += (os->bitcount >> 4) << 1;
648 /* Add at most 16 bits to the bitbuffer and flush it. */
649 static forceinline void
650 lzx_write_bits(struct lzx_output_bitstream *os, u32 bits, unsigned num_bits)
652 lzx_add_bits(os, bits, num_bits);
653 lzx_flush_bits(os, 16);
657 * Flush the last coding unit to the output buffer if needed. Return the total
658 * number of bytes written to the output buffer, or 0 if an overflow occurred.
661 lzx_flush_output(struct lzx_output_bitstream *os)
663 if (os->end - os->next < 6)
666 if (os->bitcount != 0) {
667 put_unaligned_le16(os->bitbuf << (16 - os->bitcount), os->next);
671 return os->next - os->start;
674 /******************************************************************************/
675 /* Preparing Huffman codes */
676 /*----------------------------------------------------------------------------*/
679 * Build the Huffman codes. This takes as input the frequency tables for each
680 * code and produces as output a set of tables that map symbols to codewords and
684 lzx_build_huffman_codes(struct lzx_compressor *c)
686 const struct lzx_freqs *freqs = &c->freqs;
687 struct lzx_codes *codes = &c->codes[c->codes_index];
689 STATIC_ASSERT(MAIN_CODEWORD_LIMIT >= 9 &&
690 MAIN_CODEWORD_LIMIT <= LZX_MAX_MAIN_CODEWORD_LEN);
691 make_canonical_huffman_code(c->num_main_syms,
695 codes->codewords.main);
697 STATIC_ASSERT(LENGTH_CODEWORD_LIMIT >= 8 &&
698 LENGTH_CODEWORD_LIMIT <= LZX_MAX_LEN_CODEWORD_LEN);
699 make_canonical_huffman_code(LZX_LENCODE_NUM_SYMBOLS,
700 LENGTH_CODEWORD_LIMIT,
703 codes->codewords.len);
705 STATIC_ASSERT(ALIGNED_CODEWORD_LIMIT >= LZX_NUM_ALIGNED_OFFSET_BITS &&
706 ALIGNED_CODEWORD_LIMIT <= LZX_MAX_ALIGNED_CODEWORD_LEN);
707 make_canonical_huffman_code(LZX_ALIGNEDCODE_NUM_SYMBOLS,
708 ALIGNED_CODEWORD_LIMIT,
711 codes->codewords.aligned);
714 /* Reset the symbol frequencies for the current block. */
716 lzx_reset_symbol_frequencies(struct lzx_compressor *c)
718 memset(&c->freqs, 0, sizeof(c->freqs));
722 lzx_compute_precode_items(const u8 lens[restrict],
723 const u8 prev_lens[restrict],
724 u32 precode_freqs[restrict],
725 unsigned precode_items[restrict])
734 itemptr = precode_items;
737 while (!((len = lens[run_start]) & 0x80)) {
739 /* len = the length being repeated */
741 /* Find the next run of codeword lengths. */
743 run_end = run_start + 1;
745 /* Fast case for a single length. */
746 if (likely(len != lens[run_end])) {
747 delta = prev_lens[run_start] - len;
750 precode_freqs[delta]++;
756 /* Extend the run. */
759 } while (len == lens[run_end]);
764 /* Symbol 18: RLE 20 to 51 zeroes at a time. */
765 while ((run_end - run_start) >= 20) {
766 extra_bits = min((run_end - run_start) - 20, 0x1F);
768 *itemptr++ = 18 | (extra_bits << 5);
769 run_start += 20 + extra_bits;
772 /* Symbol 17: RLE 4 to 19 zeroes at a time. */
773 if ((run_end - run_start) >= 4) {
774 extra_bits = min((run_end - run_start) - 4, 0xF);
776 *itemptr++ = 17 | (extra_bits << 5);
777 run_start += 4 + extra_bits;
781 /* A run of nonzero lengths. */
783 /* Symbol 19: RLE 4 to 5 of any length at a time. */
784 while ((run_end - run_start) >= 4) {
785 extra_bits = (run_end - run_start) > 4;
786 delta = prev_lens[run_start] - len;
790 precode_freqs[delta]++;
791 *itemptr++ = 19 | (extra_bits << 5) | (delta << 6);
792 run_start += 4 + extra_bits;
796 /* Output any remaining lengths without RLE. */
797 while (run_start != run_end) {
798 delta = prev_lens[run_start] - len;
801 precode_freqs[delta]++;
807 return itemptr - precode_items;
810 /******************************************************************************/
811 /* Outputting compressed data */
812 /*----------------------------------------------------------------------------*/
815 * Output a Huffman code in the compressed form used in LZX.
817 * The Huffman code is represented in the output as a logical series of codeword
818 * lengths from which the Huffman code, which must be in canonical form, can be
821 * The codeword lengths are themselves compressed using a separate Huffman code,
822 * the "precode", which contains a symbol for each possible codeword length in
823 * the larger code as well as several special symbols to represent repeated
824 * codeword lengths (a form of run-length encoding). The precode is itself
825 * constructed in canonical form, and its codeword lengths are represented
826 * literally in 20 4-bit fields that immediately precede the compressed codeword
827 * lengths of the larger code.
829 * Furthermore, the codeword lengths of the larger code are actually represented
830 * as deltas from the codeword lengths of the corresponding code in the previous
834 * Bitstream to which to write the compressed Huffman code.
836 * The codeword lengths, indexed by symbol, in the Huffman code.
838 * The codeword lengths, indexed by symbol, in the corresponding Huffman
839 * code in the previous block, or all zeroes if this is the first block.
841 * The number of symbols in the Huffman code.
844 lzx_write_compressed_code(struct lzx_output_bitstream *os,
845 const u8 lens[restrict],
846 const u8 prev_lens[restrict],
849 u32 precode_freqs[LZX_PRECODE_NUM_SYMBOLS];
850 u8 precode_lens[LZX_PRECODE_NUM_SYMBOLS];
851 u32 precode_codewords[LZX_PRECODE_NUM_SYMBOLS];
852 unsigned precode_items[num_lens];
853 unsigned num_precode_items;
854 unsigned precode_item;
855 unsigned precode_sym;
857 u8 saved = lens[num_lens];
858 *(u8 *)(lens + num_lens) = 0x80;
860 for (i = 0; i < LZX_PRECODE_NUM_SYMBOLS; i++)
861 precode_freqs[i] = 0;
863 /* Compute the "items" (RLE / literal tokens and extra bits) with which
864 * the codeword lengths in the larger code will be output. */
865 num_precode_items = lzx_compute_precode_items(lens,
870 /* Build the precode. */
871 STATIC_ASSERT(PRE_CODEWORD_LIMIT >= 5 &&
872 PRE_CODEWORD_LIMIT <= LZX_MAX_PRE_CODEWORD_LEN);
873 make_canonical_huffman_code(LZX_PRECODE_NUM_SYMBOLS, PRE_CODEWORD_LIMIT,
874 precode_freqs, precode_lens,
877 /* Output the lengths of the codewords in the precode. */
878 for (i = 0; i < LZX_PRECODE_NUM_SYMBOLS; i++)
879 lzx_write_bits(os, precode_lens[i], LZX_PRECODE_ELEMENT_SIZE);
881 /* Output the encoded lengths of the codewords in the larger code. */
882 for (i = 0; i < num_precode_items; i++) {
883 precode_item = precode_items[i];
884 precode_sym = precode_item & 0x1F;
885 lzx_add_bits(os, precode_codewords[precode_sym],
886 precode_lens[precode_sym]);
887 if (precode_sym >= 17) {
888 if (precode_sym == 17) {
889 lzx_add_bits(os, precode_item >> 5, 4);
890 } else if (precode_sym == 18) {
891 lzx_add_bits(os, precode_item >> 5, 5);
893 lzx_add_bits(os, (precode_item >> 5) & 1, 1);
894 precode_sym = precode_item >> 6;
895 lzx_add_bits(os, precode_codewords[precode_sym],
896 precode_lens[precode_sym]);
899 STATIC_ASSERT(CAN_BUFFER(2 * PRE_CODEWORD_LIMIT + 1));
900 lzx_flush_bits(os, 2 * PRE_CODEWORD_LIMIT + 1);
903 *(u8 *)(lens + num_lens) = saved;
907 * Write all matches and literal bytes (which were precomputed) in an LZX
908 * compressed block to the output bitstream in the final compressed
912 * The output bitstream.
914 * The chosen type of the LZX compressed block (LZX_BLOCKTYPE_ALIGNED or
915 * LZX_BLOCKTYPE_VERBATIM).
917 * The uncompressed data of the block.
919 * The matches and literals to output, given as a series of sequences.
921 * The main, length, and aligned offset Huffman codes for the block.
924 lzx_write_sequences(struct lzx_output_bitstream *os, int block_type,
925 const u8 *block_data, const struct lzx_sequence sequences[],
926 const struct lzx_codes *codes)
928 const struct lzx_sequence *seq = sequences;
929 unsigned min_aligned_offset_slot;
931 if (block_type == LZX_BLOCKTYPE_ALIGNED)
932 min_aligned_offset_slot = LZX_MIN_ALIGNED_OFFSET_SLOT;
934 min_aligned_offset_slot = LZX_MAX_OFFSET_SLOTS;
937 /* Output the next sequence. */
939 u32 litrunlen = seq->litrunlen_and_matchlen >> SEQ_MATCHLEN_BITS;
940 unsigned matchlen = seq->litrunlen_and_matchlen & SEQ_MATCHLEN_MASK;
941 STATIC_ASSERT((u32)~SEQ_MATCHLEN_MASK >> SEQ_MATCHLEN_BITS >=
942 SOFT_MAX_BLOCK_SIZE);
944 unsigned main_symbol;
945 unsigned offset_slot;
946 unsigned num_extra_bits;
949 /* Output the literal run of the sequence. */
951 if (litrunlen) { /* Is the literal run nonempty? */
953 /* Verify optimization is enabled on 64-bit */
954 STATIC_ASSERT(WORDBITS < 64 ||
955 CAN_BUFFER(3 * MAIN_CODEWORD_LIMIT));
957 if (CAN_BUFFER(3 * MAIN_CODEWORD_LIMIT)) {
959 /* 64-bit: write 3 literals at a time. */
960 while (litrunlen >= 3) {
961 unsigned lit0 = block_data[0];
962 unsigned lit1 = block_data[1];
963 unsigned lit2 = block_data[2];
964 lzx_add_bits(os, codes->codewords.main[lit0],
965 codes->lens.main[lit0]);
966 lzx_add_bits(os, codes->codewords.main[lit1],
967 codes->lens.main[lit1]);
968 lzx_add_bits(os, codes->codewords.main[lit2],
969 codes->lens.main[lit2]);
970 lzx_flush_bits(os, 3 * MAIN_CODEWORD_LIMIT);
975 unsigned lit = *block_data++;
976 lzx_add_bits(os, codes->codewords.main[lit],
977 codes->lens.main[lit]);
979 unsigned lit = *block_data++;
980 lzx_add_bits(os, codes->codewords.main[lit],
981 codes->lens.main[lit]);
982 lzx_flush_bits(os, 2 * MAIN_CODEWORD_LIMIT);
984 lzx_flush_bits(os, 1 * MAIN_CODEWORD_LIMIT);
988 /* 32-bit: write 1 literal at a time. */
990 unsigned lit = *block_data++;
991 lzx_add_bits(os, codes->codewords.main[lit],
992 codes->lens.main[lit]);
993 lzx_flush_bits(os, MAIN_CODEWORD_LIMIT);
994 } while (--litrunlen);
998 /* Was this the last literal run? */
1002 /* Nope; output the match. */
1004 block_data += matchlen;
1006 adjusted_offset = seq->adjusted_offset_and_mainsym >> SEQ_MAINSYM_BITS;
1007 main_symbol = seq->adjusted_offset_and_mainsym & SEQ_MAINSYM_MASK;
1009 offset_slot = (main_symbol - LZX_NUM_CHARS) / LZX_NUM_LEN_HEADERS;
1010 num_extra_bits = lzx_extra_offset_bits[offset_slot];
1011 extra_bits = adjusted_offset - (lzx_offset_slot_base[offset_slot] +
1012 LZX_OFFSET_ADJUSTMENT);
1014 #define MAX_MATCH_BITS (MAIN_CODEWORD_LIMIT + \
1015 LENGTH_CODEWORD_LIMIT + \
1016 LZX_MAX_NUM_EXTRA_BITS - \
1017 LZX_NUM_ALIGNED_OFFSET_BITS + \
1018 ALIGNED_CODEWORD_LIMIT)
1020 /* Verify optimization is enabled on 64-bit */
1021 STATIC_ASSERT(WORDBITS < 64 || CAN_BUFFER(MAX_MATCH_BITS));
1023 /* Output the main symbol for the match. */
1025 lzx_add_bits(os, codes->codewords.main[main_symbol],
1026 codes->lens.main[main_symbol]);
1027 if (!CAN_BUFFER(MAX_MATCH_BITS))
1028 lzx_flush_bits(os, MAIN_CODEWORD_LIMIT);
1030 /* If needed, output the length symbol for the match. */
1032 if (matchlen >= LZX_MIN_SECONDARY_LEN) {
1033 lzx_add_bits(os, codes->codewords.len[matchlen -
1034 LZX_MIN_SECONDARY_LEN],
1035 codes->lens.len[matchlen -
1036 LZX_MIN_SECONDARY_LEN]);
1037 if (!CAN_BUFFER(MAX_MATCH_BITS))
1038 lzx_flush_bits(os, LENGTH_CODEWORD_LIMIT);
1041 /* Output the extra offset bits for the match. In aligned
1042 * offset blocks, the lowest 3 bits of the adjusted offset are
1043 * Huffman-encoded using the aligned offset code, provided that
1044 * there are at least extra 3 offset bits required. All other
1045 * extra offset bits are output verbatim. */
1047 if (offset_slot >= min_aligned_offset_slot) {
1049 lzx_add_bits(os, extra_bits >> LZX_NUM_ALIGNED_OFFSET_BITS,
1050 num_extra_bits - LZX_NUM_ALIGNED_OFFSET_BITS);
1051 if (!CAN_BUFFER(MAX_MATCH_BITS))
1052 lzx_flush_bits(os, LZX_MAX_NUM_EXTRA_BITS -
1053 LZX_NUM_ALIGNED_OFFSET_BITS);
1055 lzx_add_bits(os, codes->codewords.aligned[adjusted_offset &
1056 LZX_ALIGNED_OFFSET_BITMASK],
1057 codes->lens.aligned[adjusted_offset &
1058 LZX_ALIGNED_OFFSET_BITMASK]);
1059 if (!CAN_BUFFER(MAX_MATCH_BITS))
1060 lzx_flush_bits(os, ALIGNED_CODEWORD_LIMIT);
1062 STATIC_ASSERT(CAN_BUFFER(LZX_MAX_NUM_EXTRA_BITS));
1064 lzx_add_bits(os, extra_bits, num_extra_bits);
1065 if (!CAN_BUFFER(MAX_MATCH_BITS))
1066 lzx_flush_bits(os, LZX_MAX_NUM_EXTRA_BITS);
1069 if (CAN_BUFFER(MAX_MATCH_BITS))
1070 lzx_flush_bits(os, MAX_MATCH_BITS);
1072 /* Advance to the next sequence. */
1078 lzx_write_compressed_block(const u8 *block_begin,
1081 unsigned window_order,
1082 unsigned num_main_syms,
1083 const struct lzx_sequence sequences[],
1084 const struct lzx_codes * codes,
1085 const struct lzx_lens * prev_lens,
1086 struct lzx_output_bitstream * os)
1088 /* The first three bits indicate the type of block and are one of the
1089 * LZX_BLOCKTYPE_* constants. */
1090 lzx_write_bits(os, block_type, 3);
1093 * Output the block size.
1095 * The original LZX format encoded the block size in 24 bits. However,
1096 * the LZX format used in WIM archives uses 1 bit to specify whether the
1097 * block has the default size of 32768 bytes, then optionally 16 bits to
1098 * specify a non-default size. This works fine for Microsoft's WIM
1099 * software (WIMGAPI), which never compresses more than 32768 bytes at a
1100 * time with LZX. However, as an extension, our LZX compressor supports
1101 * compressing up to 2097152 bytes, with a corresponding increase in
1102 * window size. It is possible for blocks in these larger buffers to
1103 * exceed 65535 bytes; such blocks cannot have their size represented in
1106 * The chosen solution was to use 24 bits for the block size when
1107 * possibly required --- specifically, when the compressor has been
1108 * allocated to be capable of compressing more than 32768 bytes at once
1109 * (which also causes the number of main symbols to be increased).
1111 if (block_size == LZX_DEFAULT_BLOCK_SIZE) {
1112 lzx_write_bits(os, 1, 1);
1114 lzx_write_bits(os, 0, 1);
1116 if (window_order >= 16)
1117 lzx_write_bits(os, block_size >> 16, 8);
1119 lzx_write_bits(os, block_size & 0xFFFF, 16);
1122 /* If it's an aligned offset block, output the aligned offset code. */
1123 if (block_type == LZX_BLOCKTYPE_ALIGNED) {
1124 for (int i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) {
1125 lzx_write_bits(os, codes->lens.aligned[i],
1126 LZX_ALIGNEDCODE_ELEMENT_SIZE);
1130 /* Output the main code (two parts). */
1131 lzx_write_compressed_code(os, codes->lens.main,
1134 lzx_write_compressed_code(os, codes->lens.main + LZX_NUM_CHARS,
1135 prev_lens->main + LZX_NUM_CHARS,
1136 num_main_syms - LZX_NUM_CHARS);
1138 /* Output the length code. */
1139 lzx_write_compressed_code(os, codes->lens.len,
1141 LZX_LENCODE_NUM_SYMBOLS);
1143 /* Output the compressed matches and literals. */
1144 lzx_write_sequences(os, block_type, block_begin, sequences, codes);
1148 * Given the frequencies of symbols in an LZX-compressed block and the
1149 * corresponding Huffman codes, return LZX_BLOCKTYPE_ALIGNED or
1150 * LZX_BLOCKTYPE_VERBATIM if an aligned offset or verbatim block, respectively,
1151 * will take fewer bits to output.
1154 lzx_choose_verbatim_or_aligned(const struct lzx_freqs * freqs,
1155 const struct lzx_codes * codes)
1157 u32 verbatim_cost = 0;
1158 u32 aligned_cost = 0;
1160 /* A verbatim block requires 3 bits in each place that an aligned offset
1161 * symbol would be used in an aligned offset block. */
1162 for (unsigned i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) {
1163 verbatim_cost += LZX_NUM_ALIGNED_OFFSET_BITS * freqs->aligned[i];
1164 aligned_cost += codes->lens.aligned[i] * freqs->aligned[i];
1167 /* Account for the cost of sending the codeword lengths of the aligned
1169 aligned_cost += LZX_ALIGNEDCODE_ELEMENT_SIZE *
1170 LZX_ALIGNEDCODE_NUM_SYMBOLS;
1172 if (aligned_cost < verbatim_cost)
1173 return LZX_BLOCKTYPE_ALIGNED;
1175 return LZX_BLOCKTYPE_VERBATIM;
1179 * Flush an LZX block:
1181 * 1. Build the Huffman codes.
1182 * 2. Decide whether to output the block as VERBATIM or ALIGNED.
1183 * 3. Write the block.
1184 * 4. Swap the indices of the current and previous Huffman codes.
1186 * Note: we never output UNCOMPRESSED blocks. This probably should be
1187 * implemented sometime, but it doesn't make much difference.
1190 lzx_flush_block(struct lzx_compressor *c, struct lzx_output_bitstream *os,
1191 const u8 *block_begin, u32 block_size, u32 seq_idx)
1195 lzx_build_huffman_codes(c);
1197 block_type = lzx_choose_verbatim_or_aligned(&c->freqs,
1198 &c->codes[c->codes_index]);
1199 lzx_write_compressed_block(block_begin,
1204 &c->chosen_sequences[seq_idx],
1205 &c->codes[c->codes_index],
1206 &c->codes[c->codes_index ^ 1].lens,
1208 c->codes_index ^= 1;
1211 /******************************************************************************/
1212 /* Block splitting algorithm */
1213 /*----------------------------------------------------------------------------*/
1216 * The problem of block splitting is to decide when it is worthwhile to start a
1217 * new block with new entropy codes. There is a theoretically optimal solution:
1218 * recursively consider every possible block split, considering the exact cost
1219 * of each block, and choose the minimum cost approach. But this is far too
1220 * slow. Instead, as an approximation, we can count symbols and after every N
1221 * symbols, compare the expected distribution of symbols based on the previous
1222 * data with the actual distribution. If they differ "by enough", then start a
1225 * As an optimization and heuristic, we don't distinguish between every symbol
1226 * but rather we combine many symbols into a single "observation type". For
1227 * literals we only look at the high bits and low bits, and for matches we only
1228 * look at whether the match is long or not. The assumption is that for typical
1229 * "real" data, places that are good block boundaries will tend to be noticeable
1230 * based only on changes in these aggregate frequencies, without looking for
1231 * subtle differences in individual symbols. For example, a change from ASCII
1232 * bytes to non-ASCII bytes, or from few matches (generally less compressible)
1233 * to many matches (generally more compressible), would be easily noticed based
1234 * on the aggregates.
1236 * For determining whether the frequency distributions are "different enough" to
1237 * start a new block, the simply heuristic of splitting when the sum of absolute
1238 * differences exceeds a constant seems to be good enough.
1240 * Finally, for an approximation, it is not strictly necessary that the exact
1241 * symbols being used are considered. With "near-optimal parsing", for example,
1242 * the actual symbols that will be used are unknown until after the block
1243 * boundary is chosen and the block has been optimized. Since the final choices
1244 * cannot be used, we can use preliminary "greedy" choices instead.
1247 /* Initialize the block split statistics when starting a new block. */
1249 lzx_init_block_split_stats(struct lzx_block_split_stats *stats)
1251 memset(stats, 0, sizeof(*stats));
1254 /* Literal observation. Heuristic: use the top 2 bits and low 1 bits of the
1255 * literal, for 8 possible literal observation types. */
1256 static forceinline void
1257 lzx_observe_literal(struct lzx_block_split_stats *stats, u8 lit)
1259 stats->new_observations[((lit >> 5) & 0x6) | (lit & 1)]++;
1260 stats->num_new_observations++;
1263 /* Match observation. Heuristic: use one observation type for "short match" and
1264 * one observation type for "long match". */
1265 static forceinline void
1266 lzx_observe_match(struct lzx_block_split_stats *stats, unsigned length)
1268 stats->new_observations[NUM_LITERAL_OBSERVATION_TYPES + (length >= 5)]++;
1269 stats->num_new_observations++;
1273 lzx_should_end_block(struct lzx_block_split_stats *stats)
1275 if (stats->num_observations > 0) {
1277 /* Note: to avoid slow divisions, we do not divide by
1278 * 'num_observations', but rather do all math with the numbers
1279 * multiplied by 'num_observations'. */
1280 u32 total_delta = 0;
1281 for (int i = 0; i < NUM_OBSERVATION_TYPES; i++) {
1282 u32 expected = stats->observations[i] *
1283 stats->num_new_observations;
1284 u32 actual = stats->new_observations[i] *
1285 stats->num_observations;
1286 u32 delta = (actual > expected) ? actual - expected :
1288 total_delta += delta;
1291 /* Ready to end the block? */
1293 stats->num_new_observations * 7 / 8 * stats->num_observations)
1297 for (int i = 0; i < NUM_OBSERVATION_TYPES; i++) {
1298 stats->num_observations += stats->new_observations[i];
1299 stats->observations[i] += stats->new_observations[i];
1300 stats->new_observations[i] = 0;
1302 stats->num_new_observations = 0;
1306 /******************************************************************************/
1307 /* Slower ("near-optimal") compression algorithm */
1308 /*----------------------------------------------------------------------------*/
1311 * Least-recently-used queue for match offsets.
1313 * This is represented as a 64-bit integer for efficiency. There are three
1314 * offsets of 21 bits each. Bit 64 is garbage.
1316 struct lzx_lru_queue {
1318 } _aligned_attribute(8);
1320 #define LZX_QUEUE_OFFSET_SHIFT 21
1321 #define LZX_QUEUE_OFFSET_MASK (((u64)1 << LZX_QUEUE_OFFSET_SHIFT) - 1)
1323 #define LZX_QUEUE_R0_SHIFT (0 * LZX_QUEUE_OFFSET_SHIFT)
1324 #define LZX_QUEUE_R1_SHIFT (1 * LZX_QUEUE_OFFSET_SHIFT)
1325 #define LZX_QUEUE_R2_SHIFT (2 * LZX_QUEUE_OFFSET_SHIFT)
1327 #define LZX_QUEUE_R0_MASK (LZX_QUEUE_OFFSET_MASK << LZX_QUEUE_R0_SHIFT)
1328 #define LZX_QUEUE_R1_MASK (LZX_QUEUE_OFFSET_MASK << LZX_QUEUE_R1_SHIFT)
1329 #define LZX_QUEUE_R2_MASK (LZX_QUEUE_OFFSET_MASK << LZX_QUEUE_R2_SHIFT)
1331 #define LZX_QUEUE_INITIALIZER { \
1332 ((u64)1 << LZX_QUEUE_R0_SHIFT) | \
1333 ((u64)1 << LZX_QUEUE_R1_SHIFT) | \
1334 ((u64)1 << LZX_QUEUE_R2_SHIFT) }
1336 static forceinline u64
1337 lzx_lru_queue_R0(struct lzx_lru_queue queue)
1339 return (queue.R >> LZX_QUEUE_R0_SHIFT) & LZX_QUEUE_OFFSET_MASK;
1342 static forceinline u64
1343 lzx_lru_queue_R1(struct lzx_lru_queue queue)
1345 return (queue.R >> LZX_QUEUE_R1_SHIFT) & LZX_QUEUE_OFFSET_MASK;
1348 static forceinline u64
1349 lzx_lru_queue_R2(struct lzx_lru_queue queue)
1351 return (queue.R >> LZX_QUEUE_R2_SHIFT) & LZX_QUEUE_OFFSET_MASK;
1354 /* Push a match offset onto the front (most recently used) end of the queue. */
1355 static forceinline struct lzx_lru_queue
1356 lzx_lru_queue_push(struct lzx_lru_queue queue, u32 offset)
1358 return (struct lzx_lru_queue) {
1359 .R = (queue.R << LZX_QUEUE_OFFSET_SHIFT) | offset,
1363 /* Swap a match offset to the front of the queue. */
1364 static forceinline struct lzx_lru_queue
1365 lzx_lru_queue_swap(struct lzx_lru_queue queue, unsigned idx)
1367 unsigned shift = idx * 21;
1368 const u64 mask = LZX_QUEUE_R0_MASK;
1369 const u64 mask_high = mask << shift;
1371 return (struct lzx_lru_queue) {
1372 (queue.R & ~(mask | mask_high)) |
1373 ((queue.R & mask_high) >> shift) |
1374 ((queue.R & mask) << shift)
1378 static forceinline u32
1379 lzx_walk_item_list(struct lzx_compressor *c, u32 block_size, bool is_16_bit,
1382 struct lzx_sequence *seq =
1383 &c->chosen_sequences[ARRAY_LEN(c->chosen_sequences) - 1];
1384 u32 node_idx = block_size;
1385 u32 litrun_end; /* if record=true: end of the current literal run */
1388 /* The last sequence has matchlen 0 */
1389 seq->litrunlen_and_matchlen = 0;
1390 litrun_end = node_idx;
1396 u32 adjusted_offset;
1399 /* Tally literals until either a match or the beginning of the
1400 * block is reached. Note: the item in the node at the
1401 * beginning of the block (c->optimum_nodes[0]) has all bits
1402 * set, causing this loop to end when it is reached. */
1404 item = c->optimum_nodes[node_idx].item;
1405 if (item & OPTIMUM_LEN_MASK)
1407 c->freqs.main[item >> OPTIMUM_OFFSET_SHIFT]++;
1411 #if CONSIDER_GAP_MATCHES
1412 if (item & OPTIMUM_GAP_MATCH) {
1415 /* Tally/record the rep0 match after the gap. */
1416 matchlen = item & OPTIMUM_LEN_MASK;
1417 mainsym = lzx_tally_main_and_lensyms(c, matchlen, 0,
1420 seq->litrunlen_and_matchlen |=
1421 (litrun_end - node_idx) <<
1424 seq->litrunlen_and_matchlen = matchlen;
1425 seq->adjusted_offset_and_mainsym = mainsym;
1426 litrun_end = node_idx - matchlen;
1429 /* Tally the literal in the gap. */
1430 c->freqs.main[(u8)(item >> OPTIMUM_OFFSET_SHIFT)]++;
1432 /* Fall through and tally the match before the gap.
1433 * (It was temporarily saved in the 'cost' field of the
1434 * previous node, which was free to reuse.) */
1435 item = c->optimum_nodes[--node_idx].cost;
1436 node_idx -= matchlen;
1438 #else /* CONSIDER_GAP_MATCHES */
1441 #endif /* !CONSIDER_GAP_MATCHES */
1443 /* Tally/record a match. */
1444 matchlen = item & OPTIMUM_LEN_MASK;
1445 adjusted_offset = item >> OPTIMUM_OFFSET_SHIFT;
1446 mainsym = lzx_tally_main_and_lensyms(c, matchlen,
1449 if (adjusted_offset >= LZX_MIN_ALIGNED_OFFSET +
1450 LZX_OFFSET_ADJUSTMENT)
1451 c->freqs.aligned[adjusted_offset &
1452 LZX_ALIGNED_OFFSET_BITMASK]++;
1454 seq->litrunlen_and_matchlen |=
1455 (litrun_end - node_idx) << SEQ_MATCHLEN_BITS;
1457 seq->litrunlen_and_matchlen = matchlen;
1458 seq->adjusted_offset_and_mainsym =
1459 (adjusted_offset << SEQ_MAINSYM_BITS) | mainsym;
1460 litrun_end = node_idx - matchlen;
1462 node_idx -= matchlen;
1465 /* Record the literal run length for the first sequence. */
1467 seq->litrunlen_and_matchlen |=
1468 (litrun_end - node_idx) << SEQ_MATCHLEN_BITS;
1471 /* Return the index in chosen_sequences at which the sequences begin. */
1472 return seq - &c->chosen_sequences[0];
1476 * Given the minimum-cost path computed through the item graph for the current
1477 * block, walk the path and count how many of each symbol in each Huffman-coded
1478 * alphabet would be required to output the items (matches and literals) along
1481 * Note that the path will be walked backwards (from the end of the block to the
1482 * beginning of the block), but this doesn't matter because this function only
1483 * computes frequencies.
1485 static forceinline void
1486 lzx_tally_item_list(struct lzx_compressor *c, u32 block_size, bool is_16_bit)
1488 lzx_walk_item_list(c, block_size, is_16_bit, false);
1492 * Like lzx_tally_item_list(), but this function also generates the list of
1493 * lzx_sequences for the minimum-cost path and writes it to c->chosen_sequences,
1494 * ready to be output to the bitstream after the Huffman codes are computed.
1495 * The lzx_sequences will be written to decreasing memory addresses as the path
1496 * is walked backwards, which means they will end up in the expected
1497 * first-to-last order. The return value is the index in c->chosen_sequences at
1498 * which the lzx_sequences begin.
1500 static forceinline u32
1501 lzx_record_item_list(struct lzx_compressor *c, u32 block_size, bool is_16_bit)
1503 return lzx_walk_item_list(c, block_size, is_16_bit, true);
1507 * Find an inexpensive path through the graph of possible match/literal choices
1508 * for the current block. The nodes of the graph are
1509 * c->optimum_nodes[0...block_size]. They correspond directly to the bytes in
1510 * the current block, plus one extra node for end-of-block. The edges of the
1511 * graph are matches and literals. The goal is to find the minimum cost path
1512 * from 'c->optimum_nodes[0]' to 'c->optimum_nodes[block_size]', given the cost
1515 * The algorithm works forwards, starting at 'c->optimum_nodes[0]' and
1516 * proceeding forwards one node at a time. At each node, a selection of matches
1517 * (len >= 2), as well as the literal byte (len = 1), is considered. An item of
1518 * length 'len' provides a new path to reach the node 'len' bytes later. If
1519 * such a path is the lowest cost found so far to reach that later node, then
1520 * that later node is updated with the new cost and the "arrival" which provided
1523 * Note that although this algorithm is based on minimum cost path search, due
1524 * to various simplifying assumptions the result is not guaranteed to be the
1525 * true minimum cost, or "optimal", path over the graph of all valid LZX
1526 * representations of this block.
1528 * Also, note that because of the presence of the recent offsets queue (which is
1529 * a type of adaptive state), the algorithm cannot work backwards and compute
1530 * "cost to end" instead of "cost to beginning". Furthermore, the way the
1531 * algorithm handles this adaptive state in the "minimum cost" parse is actually
1532 * only an approximation. It's possible for the globally optimal, minimum cost
1533 * path to contain a prefix, ending at a position, where that path prefix is
1534 * *not* the minimum cost path to that position. This can happen if such a path
1535 * prefix results in a different adaptive state which results in lower costs
1536 * later. The algorithm does not solve this problem in general; it only looks
1537 * one step ahead, with the exception of special consideration for "gap
1540 static forceinline struct lzx_lru_queue
1541 lzx_find_min_cost_path(struct lzx_compressor * const restrict c,
1542 const u8 * const restrict block_begin,
1543 const u32 block_size,
1544 const struct lzx_lru_queue initial_queue,
1547 struct lzx_optimum_node *cur_node = c->optimum_nodes;
1548 struct lzx_optimum_node * const end_node = cur_node + block_size;
1549 struct lz_match *cache_ptr = c->match_cache;
1550 const u8 *in_next = block_begin;
1551 const u8 * const block_end = block_begin + block_size;
1554 * Instead of storing the match offset LRU queues in the
1555 * 'lzx_optimum_node' structures, we save memory (and cache lines) by
1556 * storing them in a smaller array. This works because the algorithm
1557 * only requires a limited history of the adaptive state. Once a given
1558 * state is more than LZX_MAX_MATCH_LEN bytes behind the current node
1559 * (more if gap match consideration is enabled; we just round up to 512
1560 * so it's a power of 2), it is no longer needed.
1562 * The QUEUE() macro finds the queue for the given node. This macro has
1563 * been optimized by taking advantage of 'struct lzx_lru_queue' and
1564 * 'struct lzx_optimum_node' both being 8 bytes in size and alignment.
1566 struct lzx_lru_queue queues[512];
1567 STATIC_ASSERT(ARRAY_LEN(queues) >= LZX_MAX_MATCH_LEN + 1);
1568 STATIC_ASSERT(sizeof(c->optimum_nodes[0]) == sizeof(queues[0]));
1569 #define QUEUE(node) \
1570 (*(struct lzx_lru_queue *)((char *)queues + \
1571 ((uintptr_t)(node) % (ARRAY_LEN(queues) * sizeof(queues[0])))))
1572 /*(queues[(uintptr_t)(node) / sizeof(*(node)) % ARRAY_LEN(queues)])*/
1574 #if CONSIDER_GAP_MATCHES
1575 u32 matches_before_gap[ARRAY_LEN(queues)];
1576 #define MATCH_BEFORE_GAP(node) \
1577 (matches_before_gap[(uintptr_t)(node) / sizeof(*(node)) % \
1578 ARRAY_LEN(matches_before_gap)])
1582 * Initially, the cost to reach each node is "infinity".
1584 * The first node actually should have cost 0, but "infinity"
1585 * (0xFFFFFFFF) works just as well because it immediately overflows.
1587 * The following statement also intentionally sets the 'item' of the
1588 * first node, which would otherwise have no meaning, to 0xFFFFFFFF for
1589 * use as a sentinel. See lzx_walk_item_list().
1591 memset(c->optimum_nodes, 0xFF,
1592 (block_size + 1) * sizeof(c->optimum_nodes[0]));
1594 /* Initialize the recent offsets queue for the first node. */
1595 QUEUE(cur_node) = initial_queue;
1597 do { /* For each node in the block in position order... */
1599 unsigned num_matches;
1604 * A selection of matches for the block was already saved in
1605 * memory so that we don't have to run the uncompressed data
1606 * through the matchfinder on every optimization pass. However,
1607 * we still search for repeat offset matches during each
1608 * optimization pass because we cannot predict the state of the
1609 * recent offsets queue. But as a heuristic, we don't bother
1610 * searching for repeat offset matches if the general-purpose
1611 * matchfinder failed to find any matches.
1613 * Note that a match of length n at some offset implies there is
1614 * also a match of length l for LZX_MIN_MATCH_LEN <= l <= n at
1615 * that same offset. In other words, we don't necessarily need
1616 * to use the full length of a match. The key heuristic that
1617 * saves a significicant amount of time is that for each
1618 * distinct length, we only consider the smallest offset for
1619 * which that length is available. This heuristic also applies
1620 * to repeat offsets, which we order specially: R0 < R1 < R2 <
1621 * any explicit offset. Of course, this heuristic may be
1622 * produce suboptimal results because offset slots in LZX are
1623 * subject to entropy encoding, but in practice this is a useful
1627 num_matches = cache_ptr->length;
1631 struct lz_match *end_matches = cache_ptr + num_matches;
1632 unsigned next_len = LZX_MIN_MATCH_LEN;
1633 unsigned max_len = min(block_end - in_next, LZX_MAX_MATCH_LEN);
1636 /* Consider rep0 matches. */
1637 matchptr = in_next - lzx_lru_queue_R0(QUEUE(cur_node));
1638 if (load_u16_unaligned(matchptr) != load_u16_unaligned(in_next))
1640 STATIC_ASSERT(LZX_MIN_MATCH_LEN == 2);
1642 u32 cost = cur_node->cost +
1643 c->costs.match_cost[0][
1644 next_len - LZX_MIN_MATCH_LEN];
1645 if (cost <= (cur_node + next_len)->cost) {
1646 (cur_node + next_len)->cost = cost;
1647 (cur_node + next_len)->item =
1648 (0 << OPTIMUM_OFFSET_SHIFT) | next_len;
1650 if (unlikely(++next_len > max_len)) {
1651 cache_ptr = end_matches;
1654 } while (in_next[next_len - 1] == matchptr[next_len - 1]);
1658 /* Consider rep1 matches. */
1659 matchptr = in_next - lzx_lru_queue_R1(QUEUE(cur_node));
1660 if (load_u16_unaligned(matchptr) != load_u16_unaligned(in_next))
1662 if (matchptr[next_len - 1] != in_next[next_len - 1])
1664 for (unsigned len = 2; len < next_len - 1; len++)
1665 if (matchptr[len] != in_next[len])
1668 u32 cost = cur_node->cost +
1669 c->costs.match_cost[1][
1670 next_len - LZX_MIN_MATCH_LEN];
1671 if (cost <= (cur_node + next_len)->cost) {
1672 (cur_node + next_len)->cost = cost;
1673 (cur_node + next_len)->item =
1674 (1 << OPTIMUM_OFFSET_SHIFT) | next_len;
1676 if (unlikely(++next_len > max_len)) {
1677 cache_ptr = end_matches;
1680 } while (in_next[next_len - 1] == matchptr[next_len - 1]);
1684 /* Consider rep2 matches. */
1685 matchptr = in_next - lzx_lru_queue_R2(QUEUE(cur_node));
1686 if (load_u16_unaligned(matchptr) != load_u16_unaligned(in_next))
1688 if (matchptr[next_len - 1] != in_next[next_len - 1])
1690 for (unsigned len = 2; len < next_len - 1; len++)
1691 if (matchptr[len] != in_next[len])
1694 u32 cost = cur_node->cost +
1695 c->costs.match_cost[2][
1696 next_len - LZX_MIN_MATCH_LEN];
1697 if (cost <= (cur_node + next_len)->cost) {
1698 (cur_node + next_len)->cost = cost;
1699 (cur_node + next_len)->item =
1700 (2 << OPTIMUM_OFFSET_SHIFT) | next_len;
1702 if (unlikely(++next_len > max_len)) {
1703 cache_ptr = end_matches;
1706 } while (in_next[next_len - 1] == matchptr[next_len - 1]);
1710 while (next_len > cache_ptr->length)
1711 if (++cache_ptr == end_matches)
1714 /* Consider explicit offset matches. */
1716 u32 offset = cache_ptr->offset;
1717 u32 adjusted_offset = offset + LZX_OFFSET_ADJUSTMENT;
1718 unsigned offset_slot = lzx_get_offset_slot(c, adjusted_offset, is_16_bit);
1719 u32 base_cost = cur_node->cost;
1722 #if CONSIDER_ALIGNED_COSTS
1723 if (offset >= LZX_MIN_ALIGNED_OFFSET)
1724 base_cost += c->costs.aligned[adjusted_offset &
1725 LZX_ALIGNED_OFFSET_BITMASK];
1729 c->costs.match_cost[offset_slot][
1730 next_len - LZX_MIN_MATCH_LEN];
1731 if (cost < (cur_node + next_len)->cost) {
1732 (cur_node + next_len)->cost = cost;
1733 (cur_node + next_len)->item =
1734 (adjusted_offset << OPTIMUM_OFFSET_SHIFT) | next_len;
1736 } while (++next_len <= cache_ptr->length);
1738 if (++cache_ptr == end_matches) {
1739 #if CONSIDER_GAP_MATCHES
1740 /* Also consider the longest explicit
1741 * offset match as a "gap match": match
1743 s32 remaining = (block_end - in_next) - (s32)next_len;
1744 if (likely(remaining >= 2)) {
1745 const u8 *strptr = in_next + next_len;
1746 const u8 *matchptr = strptr - offset;
1747 if (load_u16_unaligned(strptr) == load_u16_unaligned(matchptr)) {
1748 STATIC_ASSERT(ARRAY_LEN(queues) - LZX_MAX_MATCH_LEN - 2 >= 250);
1749 STATIC_ASSERT(ARRAY_LEN(queues) == ARRAY_LEN(matches_before_gap));
1750 unsigned limit = min(remaining,
1751 min(ARRAY_LEN(queues) - LZX_MAX_MATCH_LEN - 2,
1752 LZX_MAX_MATCH_LEN));
1753 unsigned rep0_len = lz_extend(strptr, matchptr, 2, limit);
1754 u8 lit = strptr[-1];
1755 cost += c->costs.main[lit] +
1756 c->costs.match_cost[0][rep0_len - LZX_MIN_MATCH_LEN];
1757 unsigned total_len = next_len + rep0_len;
1758 if (cost < (cur_node + total_len)->cost) {
1759 (cur_node + total_len)->cost = cost;
1760 (cur_node + total_len)->item =
1762 ((u32)lit << OPTIMUM_OFFSET_SHIFT) |
1764 MATCH_BEFORE_GAP(cur_node + total_len) =
1765 (adjusted_offset << OPTIMUM_OFFSET_SHIFT) |
1770 #endif /* CONSIDER_GAP_MATCHES */
1778 /* Consider coding a literal.
1780 * To avoid an extra branch, actually checking the preferability
1781 * of coding the literal is integrated into the queue update
1783 literal = *in_next++;
1784 cost = cur_node->cost + c->costs.main[literal];
1786 /* Advance to the next position. */
1789 /* The lowest-cost path to the current position is now known.
1790 * Finalize the recent offsets queue that results from taking
1791 * this lowest-cost path. */
1793 if (cost <= cur_node->cost) {
1794 /* Literal: queue remains unchanged. */
1795 cur_node->cost = cost;
1796 cur_node->item = (u32)literal << OPTIMUM_OFFSET_SHIFT;
1797 QUEUE(cur_node) = QUEUE(cur_node - 1);
1799 /* Match: queue update is needed. */
1800 unsigned len = cur_node->item & OPTIMUM_LEN_MASK;
1801 #if CONSIDER_GAP_MATCHES
1802 s32 adjusted_offset = (s32)cur_node->item >> OPTIMUM_OFFSET_SHIFT;
1803 STATIC_ASSERT(OPTIMUM_GAP_MATCH == 0x80000000); /* assuming sign extension */
1805 u32 adjusted_offset = cur_node->item >> OPTIMUM_OFFSET_SHIFT;
1808 if (adjusted_offset >= LZX_NUM_RECENT_OFFSETS) {
1809 /* Explicit offset match: insert offset at front. */
1811 lzx_lru_queue_push(QUEUE(cur_node - len),
1812 adjusted_offset - LZX_OFFSET_ADJUSTMENT);
1814 #if CONSIDER_GAP_MATCHES
1815 else if (adjusted_offset < 0) {
1816 /* "Gap match": Explicit offset match, then a
1817 * literal, then rep0 match. Save the explicit
1818 * offset match information in the cost field of
1819 * the previous node, which isn't needed
1820 * anymore. Then insert the offset at the front
1822 u32 match_before_gap = MATCH_BEFORE_GAP(cur_node);
1823 (cur_node - 1)->cost = match_before_gap;
1825 lzx_lru_queue_push(QUEUE(cur_node - len - 1 -
1826 (match_before_gap & OPTIMUM_LEN_MASK)),
1827 (match_before_gap >> OPTIMUM_OFFSET_SHIFT) -
1828 LZX_OFFSET_ADJUSTMENT);
1832 /* Repeat offset match: swap offset to front. */
1834 lzx_lru_queue_swap(QUEUE(cur_node - len),
1838 } while (cur_node != end_node);
1840 /* Return the recent offsets queue at the end of the path. */
1841 return QUEUE(cur_node);
1845 * Given the costs for the main and length codewords (c->costs.main and
1846 * c->costs.len), initialize the match cost array (c->costs.match_cost) which
1847 * directly provides the cost of every possible (length, offset slot) pair.
1850 lzx_compute_match_costs(struct lzx_compressor *c)
1852 unsigned num_offset_slots = (c->num_main_syms - LZX_NUM_CHARS) /
1853 LZX_NUM_LEN_HEADERS;
1854 struct lzx_costs *costs = &c->costs;
1855 unsigned main_symbol = LZX_NUM_CHARS;
1857 for (unsigned offset_slot = 0; offset_slot < num_offset_slots;
1860 u32 extra_cost = lzx_extra_offset_bits[offset_slot] * BIT_COST;
1863 #if CONSIDER_ALIGNED_COSTS
1864 if (offset_slot >= LZX_MIN_ALIGNED_OFFSET_SLOT)
1865 extra_cost -= LZX_NUM_ALIGNED_OFFSET_BITS * BIT_COST;
1868 for (i = 0; i < LZX_NUM_PRIMARY_LENS; i++) {
1869 costs->match_cost[offset_slot][i] =
1870 costs->main[main_symbol++] + extra_cost;
1873 extra_cost += costs->main[main_symbol++];
1875 for (; i < LZX_NUM_LENS; i++) {
1876 costs->match_cost[offset_slot][i] =
1877 costs->len[i - LZX_NUM_PRIMARY_LENS] +
1884 * Fast approximation for log2f(x). This is not as accurate as the standard C
1885 * version. It does not need to be perfectly accurate because it is only used
1886 * for estimating symbol costs, which is very approximate anyway.
1896 /* Extract the exponent and subtract 127 to remove the bias. This gives
1897 * the integer part of the result. */
1898 float res = ((u.i >> 23) & 0xFF) - 127;
1900 /* Set the exponent to 0 (plus bias of 127). This transforms the number
1901 * to the range [1, 2) while retaining the same mantissa. */
1902 u.i = (u.i & ~(0xFF << 23)) | (127 << 23);
1905 * Approximate the log2 of the transformed number using a degree 2
1906 * interpolating polynomial for log2(x) over the interval [1, 2). Then
1907 * add this to the extracted exponent to produce the final approximation
1910 * The coefficients of the interpolating polynomial used here were found
1911 * using the script tools/log2_interpolation.r.
1913 return res - 1.653124006f + u.f * (1.9941812f - u.f * 0.3347490189f);
1918 * Return the estimated cost of a symbol which has been estimated to have the
1919 * given probability.
1922 lzx_cost_for_probability(float prob)
1925 * The basic formula is:
1927 * entropy = -log2(probability)
1929 * Use this to get the cost in fractional bits. Then multiply by our
1930 * scaling factor of BIT_COST and convert to an integer.
1932 * In addition, the minimum cost is BIT_COST (one bit) because the
1933 * entropy coding method will be Huffman codes.
1935 * Careful: even though 'prob' should be <= 1.0, 'log2f_fast(prob)' may
1936 * be positive due to inaccuracy in our log2 approximation. Therefore,
1937 * we cannot, in general, assume the computed cost is non-negative, and
1938 * we should make sure negative costs get rounded up correctly.
1940 s32 cost = -log2f_fast(prob) * BIT_COST;
1941 return max(cost, BIT_COST);
1945 * Mapping: number of used literals => heuristic probability of a literal times
1946 * 6870. Generated by running this R command:
1948 * cat(paste(round(6870*2^-((304+(0:256))/64)), collapse=", "))
1950 static const u8 literal_scaled_probs[257] = {
1951 255, 253, 250, 247, 244, 242, 239, 237, 234, 232, 229, 227, 224, 222,
1952 219, 217, 215, 212, 210, 208, 206, 203, 201, 199, 197, 195, 193, 191,
1953 189, 186, 184, 182, 181, 179, 177, 175, 173, 171, 169, 167, 166, 164,
1954 162, 160, 159, 157, 155, 153, 152, 150, 149, 147, 145, 144, 142, 141,
1955 139, 138, 136, 135, 133, 132, 130, 129, 128, 126, 125, 124, 122, 121,
1956 120, 118, 117, 116, 115, 113, 112, 111, 110, 109, 107, 106, 105, 104,
1957 103, 102, 101, 100, 98, 97, 96, 95, 94, 93, 92, 91, 90, 89, 88, 87, 86,
1958 86, 85, 84, 83, 82, 81, 80, 79, 78, 78, 77, 76, 75, 74, 73, 73, 72, 71,
1959 70, 70, 69, 68, 67, 67, 66, 65, 65, 64, 63, 62, 62, 61, 60, 60, 59, 59,
1960 58, 57, 57, 56, 55, 55, 54, 54, 53, 53, 52, 51, 51, 50, 50, 49, 49, 48,
1961 48, 47, 47, 46, 46, 45, 45, 44, 44, 43, 43, 42, 42, 41, 41, 40, 40, 40,
1962 39, 39, 38, 38, 38, 37, 37, 36, 36, 36, 35, 35, 34, 34, 34, 33, 33, 33,
1963 32, 32, 32, 31, 31, 31, 30, 30, 30, 29, 29, 29, 28, 28, 28, 27, 27, 27,
1964 27, 26, 26, 26, 25, 25, 25, 25, 24, 24, 24, 24, 23, 23, 23, 23, 22, 22,
1965 22, 22, 21, 21, 21, 21, 20, 20, 20, 20, 20, 19, 19, 19, 19, 19, 18, 18,
1966 18, 18, 18, 17, 17, 17, 17, 17, 16, 16, 16, 16
1970 * Mapping: length symbol => default cost of that symbol. This is derived from
1971 * sample data but has been slightly edited to add more bias towards the
1972 * shortest lengths, which are the most common.
1974 static const u16 lzx_default_len_costs[LZX_LENCODE_NUM_SYMBOLS] = {
1975 300, 310, 320, 330, 360, 396, 399, 416, 451, 448, 463, 466, 505, 492,
1976 503, 514, 547, 531, 566, 561, 589, 563, 592, 586, 623, 602, 639, 627,
1977 659, 643, 657, 650, 685, 662, 661, 672, 685, 686, 696, 680, 657, 682,
1978 666, 699, 674, 699, 679, 709, 688, 712, 692, 714, 694, 716, 698, 712,
1979 706, 727, 714, 727, 713, 723, 712, 718, 719, 719, 720, 735, 725, 735,
1980 728, 740, 727, 739, 727, 742, 716, 733, 733, 740, 738, 746, 737, 747,
1981 738, 745, 736, 748, 742, 749, 745, 749, 743, 748, 741, 752, 745, 752,
1982 747, 750, 747, 752, 748, 753, 750, 752, 753, 753, 749, 744, 752, 755,
1983 753, 756, 745, 748, 746, 745, 723, 757, 755, 758, 755, 758, 752, 757,
1984 754, 757, 755, 759, 755, 758, 753, 755, 755, 758, 757, 761, 755, 750,
1985 758, 759, 759, 760, 758, 751, 757, 757, 759, 759, 758, 759, 758, 761,
1986 750, 761, 758, 760, 759, 761, 758, 761, 760, 752, 759, 760, 759, 759,
1987 757, 762, 760, 761, 761, 748, 761, 760, 762, 763, 752, 762, 762, 763,
1988 762, 762, 763, 763, 762, 763, 762, 763, 762, 763, 763, 764, 763, 762,
1989 763, 762, 762, 762, 764, 764, 763, 764, 763, 763, 763, 762, 763, 763,
1990 762, 764, 764, 763, 762, 763, 763, 763, 763, 762, 764, 763, 762, 764,
1991 764, 763, 763, 765, 764, 764, 762, 763, 764, 765, 763, 764, 763, 764,
1992 762, 764, 764, 754, 763, 764, 763, 763, 762, 763, 584,
1995 /* Set default costs to bootstrap the iterative optimization algorithm. */
1997 lzx_set_default_costs(struct lzx_compressor *c)
2000 u32 num_literals = 0;
2001 u32 num_used_literals = 0;
2002 float inv_num_matches = 1.0f / c->freqs.main[LZX_NUM_CHARS];
2003 float inv_num_items;
2004 float prob_match = 1.0f;
2006 float base_literal_prob;
2008 /* Some numbers here have been hardcoded to assume a bit cost of 64. */
2009 STATIC_ASSERT(BIT_COST == 64);
2011 /* Estimate the number of literals that will used. 'num_literals' is
2012 * the total number, whereas 'num_used_literals' is the number of
2013 * distinct symbols. */
2014 for (i = 0; i < LZX_NUM_CHARS; i++) {
2015 num_literals += c->freqs.main[i];
2016 num_used_literals += (c->freqs.main[i] != 0);
2019 /* Note: all match headers were tallied as symbol 'LZX_NUM_CHARS'. We
2020 * don't attempt to estimate which ones will be used. */
2022 inv_num_items = 1.0f / (num_literals + c->freqs.main[LZX_NUM_CHARS]);
2023 base_literal_prob = literal_scaled_probs[num_used_literals] *
2026 /* Literal costs. We use two different methods to compute the
2027 * probability of each literal and mix together their results. */
2028 for (i = 0; i < LZX_NUM_CHARS; i++) {
2029 u32 freq = c->freqs.main[i];
2031 float prob = 0.5f * ((freq * inv_num_items) +
2033 c->costs.main[i] = lzx_cost_for_probability(prob);
2036 c->costs.main[i] = 11 * BIT_COST;
2040 /* Match header costs. We just assume that all match headers are
2041 * equally probable, but we do take into account the relative cost of a
2042 * match header vs. a literal depending on how common matches are
2043 * expected to be vs. literals. */
2044 prob_match = max(prob_match, 0.15f);
2045 match_cost = lzx_cost_for_probability(prob_match / (c->num_main_syms -
2047 for (; i < c->num_main_syms; i++)
2048 c->costs.main[i] = match_cost;
2050 /* Length symbol costs. These are just set to fixed values which
2051 * reflect the fact the smallest lengths are typically the most common,
2052 * and therefore are typically the cheapest. */
2053 for (i = 0; i < LZX_LENCODE_NUM_SYMBOLS; i++)
2054 c->costs.len[i] = lzx_default_len_costs[i];
2056 #if CONSIDER_ALIGNED_COSTS
2057 /* Aligned offset symbol costs. These are derived from the estimated
2058 * probability of each aligned offset symbol. */
2059 for (i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) {
2060 /* We intentionally tallied the frequencies in the wrong slots,
2061 * not accounting for LZX_OFFSET_ADJUSTMENT, since doing the
2062 * fixup here is faster: a constant 8 subtractions here vs. one
2063 * addition for every match. */
2064 unsigned j = (i - LZX_OFFSET_ADJUSTMENT) & LZX_ALIGNED_OFFSET_BITMASK;
2065 if (c->freqs.aligned[j] != 0) {
2066 float prob = c->freqs.aligned[j] * inv_num_matches;
2067 c->costs.aligned[i] = lzx_cost_for_probability(prob);
2069 c->costs.aligned[i] =
2070 (2 * LZX_NUM_ALIGNED_OFFSET_BITS) * BIT_COST;
2076 /* Update the current cost model to reflect the computed Huffman codes. */
2078 lzx_set_costs_from_codes(struct lzx_compressor *c)
2081 const struct lzx_lens *lens = &c->codes[c->codes_index].lens;
2083 for (i = 0; i < c->num_main_syms; i++) {
2084 c->costs.main[i] = (lens->main[i] ? lens->main[i] :
2085 MAIN_CODEWORD_LIMIT) * BIT_COST;
2088 for (i = 0; i < LZX_LENCODE_NUM_SYMBOLS; i++) {
2089 c->costs.len[i] = (lens->len[i] ? lens->len[i] :
2090 LENGTH_CODEWORD_LIMIT) * BIT_COST;
2093 #if CONSIDER_ALIGNED_COSTS
2094 for (i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) {
2095 c->costs.aligned[i] = (lens->aligned[i] ? lens->aligned[i] :
2096 ALIGNED_CODEWORD_LIMIT) * BIT_COST;
2102 * Choose a "near-optimal" literal/match sequence to use for the current block,
2103 * then flush the block. Because the cost of each Huffman symbol is unknown
2104 * until the Huffman codes have been built and the Huffman codes themselves
2105 * depend on the symbol frequencies, this uses an iterative optimization
2106 * algorithm to approximate an optimal solution. The first optimization pass
2107 * for the block uses default costs; additional passes use costs derived from
2108 * the Huffman codes computed in the previous pass.
2110 static forceinline struct lzx_lru_queue
2111 lzx_optimize_and_flush_block(struct lzx_compressor * const restrict c,
2112 struct lzx_output_bitstream * const restrict os,
2113 const u8 * const restrict block_begin,
2114 const u32 block_size,
2115 const struct lzx_lru_queue initial_queue,
2118 unsigned num_passes_remaining = c->num_optim_passes;
2119 struct lzx_lru_queue new_queue;
2122 lzx_set_default_costs(c);
2125 lzx_compute_match_costs(c);
2126 new_queue = lzx_find_min_cost_path(c, block_begin, block_size,
2127 initial_queue, is_16_bit);
2129 if (--num_passes_remaining == 0)
2132 /* At least one optimization pass remains. Update the costs. */
2133 lzx_reset_symbol_frequencies(c);
2134 lzx_tally_item_list(c, block_size, is_16_bit);
2135 lzx_build_huffman_codes(c);
2136 lzx_set_costs_from_codes(c);
2139 /* Done optimizing. Generate the sequence list and flush the block. */
2140 lzx_reset_symbol_frequencies(c);
2141 seq_idx = lzx_record_item_list(c, block_size, is_16_bit);
2142 lzx_flush_block(c, os, block_begin, block_size, seq_idx);
2147 * This is the "near-optimal" LZX compressor.
2149 * For each block, it performs a relatively thorough graph search to find an
2150 * inexpensive (in terms of compressed size) way to output the block.
2152 * Note: there are actually many things this algorithm leaves on the table in
2153 * terms of compression ratio. So although it may be "near-optimal", it is
2154 * certainly not "optimal". The goal is not to produce the optimal compression
2155 * ratio, which for LZX is probably impossible within any practical amount of
2156 * time, but rather to produce a compression ratio significantly better than a
2157 * simpler "greedy" or "lazy" parse while still being relatively fast.
2159 static forceinline void
2160 lzx_compress_near_optimal(struct lzx_compressor * restrict c,
2161 const u8 * const restrict in_begin, size_t in_nbytes,
2162 struct lzx_output_bitstream * restrict os,
2165 const u8 * in_next = in_begin;
2166 const u8 * const in_end = in_begin + in_nbytes;
2167 u32 max_len = LZX_MAX_MATCH_LEN;
2168 u32 nice_len = min(c->nice_match_length, max_len);
2169 u32 next_hashes[2] = {0, 0};
2170 struct lzx_lru_queue queue = LZX_QUEUE_INITIALIZER;
2172 /* Initialize the matchfinder. */
2173 CALL_BT_MF(is_16_bit, c, bt_matchfinder_init);
2176 /* Starting a new block */
2178 const u8 * const in_block_begin = in_next;
2179 const u8 * const in_max_block_end =
2180 in_next + min(SOFT_MAX_BLOCK_SIZE, in_end - in_next);
2181 struct lz_match *cache_ptr = c->match_cache;
2182 const u8 *next_search_pos = in_next;
2183 const u8 *next_observation = in_next;
2184 const u8 *next_pause_point =
2185 min(in_next + min(MIN_BLOCK_SIZE,
2186 in_max_block_end - in_next),
2187 in_max_block_end - min(LZX_MAX_MATCH_LEN - 1,
2188 in_max_block_end - in_next));
2190 lzx_init_block_split_stats(&c->split_stats);
2191 lzx_reset_symbol_frequencies(c);
2193 if (in_next >= next_pause_point)
2197 * Run the input buffer through the matchfinder, caching the
2198 * matches, until we decide to end the block.
2200 * For a tighter matchfinding loop, we compute a "pause point",
2201 * which is the next position at which we may need to check
2202 * whether to end the block or to decrease max_len. We then
2203 * only do these extra checks upon reaching the pause point.
2205 resume_matchfinding:
2207 if (in_next >= next_search_pos) {
2208 /* Search for matches at this position. */
2209 struct lz_match *lz_matchptr;
2212 lz_matchptr = CALL_BT_MF(is_16_bit, c,
2213 bt_matchfinder_get_matches,
2218 c->max_search_depth,
2222 cache_ptr->length = lz_matchptr - (cache_ptr + 1);
2223 cache_ptr = lz_matchptr;
2225 /* Accumulate literal/match statistics for block
2226 * splitting and for generating the initial cost
2228 if (in_next >= next_observation) {
2229 best_len = cache_ptr[-1].length;
2230 if (best_len >= 3) {
2231 /* Match (len >= 3) */
2234 * Note: for performance reasons this has
2235 * been simplified significantly:
2237 * - We wait until later to account for
2238 * LZX_OFFSET_ADJUSTMENT.
2239 * - We don't account for repeat offsets.
2240 * - We don't account for different match headers.
2242 c->freqs.aligned[cache_ptr[-1].offset &
2243 LZX_ALIGNED_OFFSET_BITMASK]++;
2244 c->freqs.main[LZX_NUM_CHARS]++;
2246 lzx_observe_match(&c->split_stats, best_len);
2247 next_observation = in_next + best_len;
2250 c->freqs.main[*in_next]++;
2251 lzx_observe_literal(&c->split_stats, *in_next);
2252 next_observation = in_next + 1;
2257 * If there was a very long match found, then
2258 * don't cache any matches for the bytes covered
2259 * by that match. This avoids degenerate
2260 * behavior when compressing highly redundant
2261 * data, where the number of matches can be very
2264 * This heuristic doesn't actually hurt the
2265 * compression ratio *too* much. If there's a
2266 * long match, then the data must be highly
2267 * compressible, so it doesn't matter as much
2270 if (best_len >= nice_len)
2271 next_search_pos = in_next + best_len;
2273 /* Don't search for matches at this position. */
2274 CALL_BT_MF(is_16_bit, c,
2275 bt_matchfinder_skip_byte,
2279 c->max_search_depth,
2281 cache_ptr->length = 0;
2284 } while (++in_next < next_pause_point &&
2285 likely(cache_ptr < &c->match_cache[CACHE_LENGTH]));
2289 /* Adjust max_len and nice_len if we're nearing the end of the
2290 * input buffer. In addition, if we are so close to the end of
2291 * the input buffer that there cannot be any more matches, then
2292 * just advance through the last few positions and record no
2294 if (unlikely(max_len > in_end - in_next)) {
2295 max_len = in_end - in_next;
2296 nice_len = min(max_len, nice_len);
2297 if (max_len < BT_MATCHFINDER_REQUIRED_NBYTES) {
2298 while (in_next != in_end) {
2299 cache_ptr->length = 0;
2306 /* End the block if the match cache may overflow. */
2307 if (unlikely(cache_ptr >= &c->match_cache[CACHE_LENGTH]))
2310 /* End the block if the soft maximum size has been reached. */
2311 if (in_next >= in_max_block_end)
2314 /* End the block if the block splitting algorithm thinks this is
2315 * a good place to do so. */
2316 if (c->split_stats.num_new_observations >=
2317 NUM_OBSERVATIONS_PER_BLOCK_CHECK &&
2318 in_max_block_end - in_next >= MIN_BLOCK_SIZE &&
2319 lzx_should_end_block(&c->split_stats))
2322 /* It's not time to end the block yet. Compute the next pause
2323 * point and resume matchfinding. */
2325 min(in_next + min(NUM_OBSERVATIONS_PER_BLOCK_CHECK * 2 -
2326 c->split_stats.num_new_observations,
2327 in_max_block_end - in_next),
2328 in_max_block_end - min(LZX_MAX_MATCH_LEN - 1,
2329 in_max_block_end - in_next));
2330 goto resume_matchfinding;
2333 /* We've decided on a block boundary and cached matches. Now
2334 * choose a match/literal sequence and flush the block. */
2335 queue = lzx_optimize_and_flush_block(c, os, in_block_begin,
2336 in_next - in_block_begin,
2338 } while (in_next != in_end);
2342 lzx_compress_near_optimal_16(struct lzx_compressor *c, const u8 *in,
2343 size_t in_nbytes, struct lzx_output_bitstream *os)
2345 lzx_compress_near_optimal(c, in, in_nbytes, os, true);
2349 lzx_compress_near_optimal_32(struct lzx_compressor *c, const u8 *in,
2350 size_t in_nbytes, struct lzx_output_bitstream *os)
2352 lzx_compress_near_optimal(c, in, in_nbytes, os, false);
2355 /******************************************************************************/
2356 /* Faster ("lazy") compression algorithm */
2357 /*----------------------------------------------------------------------------*/
2360 * Called when the compressor chooses to use a literal. This tallies the
2361 * Huffman symbol for the literal, increments the current literal run length,
2362 * and "observes" the literal for the block split statistics.
2364 static forceinline void
2365 lzx_choose_literal(struct lzx_compressor *c, unsigned literal, u32 *litrunlen_p)
2367 lzx_observe_literal(&c->split_stats, literal);
2368 c->freqs.main[literal]++;
2373 * Called when the compressor chooses to use a match. This tallies the Huffman
2374 * symbol(s) for a match, saves the match data and the length of the preceding
2375 * literal run, updates the recent offsets queue, and "observes" the match for
2376 * the block split statistics.
2378 static forceinline void
2379 lzx_choose_match(struct lzx_compressor *c, unsigned length, u32 adjusted_offset,
2380 u32 recent_offsets[LZX_NUM_RECENT_OFFSETS], bool is_16_bit,
2381 u32 *litrunlen_p, struct lzx_sequence **next_seq_p)
2383 struct lzx_sequence *next_seq = *next_seq_p;
2386 lzx_observe_match(&c->split_stats, length);
2388 mainsym = lzx_tally_main_and_lensyms(c, length, adjusted_offset,
2390 next_seq->litrunlen_and_matchlen =
2391 (*litrunlen_p << SEQ_MATCHLEN_BITS) | length;
2392 next_seq->adjusted_offset_and_mainsym =
2393 (adjusted_offset << SEQ_MAINSYM_BITS) | mainsym;
2395 /* Update the recent offsets queue. */
2396 if (adjusted_offset < LZX_NUM_RECENT_OFFSETS) {
2397 /* Repeat offset match. */
2398 swap(recent_offsets[0], recent_offsets[adjusted_offset]);
2400 /* Explicit offset match. */
2402 /* Tally the aligned offset symbol if needed. */
2403 if (adjusted_offset >= LZX_MIN_ALIGNED_OFFSET + LZX_OFFSET_ADJUSTMENT)
2404 c->freqs.aligned[adjusted_offset & LZX_ALIGNED_OFFSET_BITMASK]++;
2406 recent_offsets[2] = recent_offsets[1];
2407 recent_offsets[1] = recent_offsets[0];
2408 recent_offsets[0] = adjusted_offset - LZX_OFFSET_ADJUSTMENT;
2411 /* Reset the literal run length and advance to the next sequence. */
2412 *next_seq_p = next_seq + 1;
2417 * Called when the compressor ends a block. This finshes the last lzx_sequence,
2418 * which is just a literal run with no following match. This literal run might
2421 static forceinline void
2422 lzx_finish_sequence(struct lzx_sequence *last_seq, u32 litrunlen)
2424 last_seq->litrunlen_and_matchlen = litrunlen << SEQ_MATCHLEN_BITS;
2428 * Find the longest repeat offset match with the current position. If a match
2429 * is found, return its length and set *best_rep_idx_ret to the index of its
2430 * offset in @recent_offsets. Otherwise, return 0.
2432 * Don't bother with length 2 matches; consider matches of length >= 3 only.
2433 * Also assume that max_len >= 3.
2436 lzx_find_longest_repeat_offset_match(const u8 * const in_next,
2437 const u32 recent_offsets[],
2438 const unsigned max_len,
2439 unsigned *best_rep_idx_ret)
2441 STATIC_ASSERT(LZX_NUM_RECENT_OFFSETS == 3); /* loop is unrolled */
2443 const u32 seq3 = load_u24_unaligned(in_next);
2445 unsigned best_rep_len = 0;
2446 unsigned best_rep_idx = 0;
2449 /* Check for rep0 match (most recent offset) */
2450 matchptr = in_next - recent_offsets[0];
2451 if (load_u24_unaligned(matchptr) == seq3)
2452 best_rep_len = lz_extend(in_next, matchptr, 3, max_len);
2454 /* Check for rep1 match (second most recent offset) */
2455 matchptr = in_next - recent_offsets[1];
2456 if (load_u24_unaligned(matchptr) == seq3) {
2457 rep_len = lz_extend(in_next, matchptr, 3, max_len);
2458 if (rep_len > best_rep_len) {
2459 best_rep_len = rep_len;
2464 /* Check for rep2 match (third most recent offset) */
2465 matchptr = in_next - recent_offsets[2];
2466 if (load_u24_unaligned(matchptr) == seq3) {
2467 rep_len = lz_extend(in_next, matchptr, 3, max_len);
2468 if (rep_len > best_rep_len) {
2469 best_rep_len = rep_len;
2474 *best_rep_idx_ret = best_rep_idx;
2475 return best_rep_len;
2479 * Fast heuristic scoring for lazy parsing: how "good" is this match?
2480 * This is mainly determined by the length: longer matches are better.
2481 * However, we also give a bonus to close (small offset) matches and to repeat
2482 * offset matches, since those require fewer bits to encode.
2485 static forceinline unsigned
2486 lzx_explicit_offset_match_score(unsigned len, u32 adjusted_offset)
2488 unsigned score = len;
2490 if (adjusted_offset < 4096)
2492 if (adjusted_offset < 256)
2498 static forceinline unsigned
2499 lzx_repeat_offset_match_score(unsigned rep_len, unsigned rep_idx)
2505 * This is the "lazy" LZX compressor. The basic idea is that before it chooses
2506 * a match, it checks to see if there's a longer match at the next position. If
2507 * yes, it chooses a literal and continues to the next position. If no, it
2508 * chooses the match.
2510 * Some additional heuristics are used as well. Repeat offset matches are
2511 * considered favorably and sometimes are chosen immediately. In addition, long
2512 * matches (at least "nice_len" bytes) are chosen immediately as well. Finally,
2513 * when we decide whether a match is "better" than another, we take the offset
2514 * into consideration as well as the length.
2516 static forceinline void
2517 lzx_compress_lazy(struct lzx_compressor * restrict c,
2518 const u8 * const restrict in_begin, size_t in_nbytes,
2519 struct lzx_output_bitstream * restrict os, bool is_16_bit)
2521 const u8 * in_next = in_begin;
2522 const u8 * const in_end = in_begin + in_nbytes;
2523 unsigned max_len = LZX_MAX_MATCH_LEN;
2524 unsigned nice_len = min(c->nice_match_length, max_len);
2525 STATIC_ASSERT(LZX_NUM_RECENT_OFFSETS == 3);
2526 u32 recent_offsets[LZX_NUM_RECENT_OFFSETS] = {1, 1, 1};
2527 u32 next_hashes[2] = {0, 0};
2529 /* Initialize the matchfinder. */
2530 CALL_HC_MF(is_16_bit, c, hc_matchfinder_init);
2533 /* Starting a new block */
2535 const u8 * const in_block_begin = in_next;
2536 const u8 * const in_max_block_end =
2537 in_next + min(SOFT_MAX_BLOCK_SIZE, in_end - in_next);
2538 struct lzx_sequence *next_seq = c->chosen_sequences;
2542 u32 cur_adjusted_offset;
2546 u32 next_adjusted_offset;
2547 unsigned next_score;
2548 unsigned best_rep_len;
2549 unsigned best_rep_idx;
2553 lzx_reset_symbol_frequencies(c);
2554 lzx_init_block_split_stats(&c->split_stats);
2557 /* Adjust max_len and nice_len if we're nearing the end
2558 * of the input buffer. */
2559 if (unlikely(max_len > in_end - in_next)) {
2560 max_len = in_end - in_next;
2561 nice_len = min(max_len, nice_len);
2564 /* Find the longest match (subject to the
2565 * max_search_depth cutoff parameter) with the current
2566 * position. Don't bother with length 2 matches; only
2567 * look for matches of length >= 3. */
2568 cur_len = CALL_HC_MF(is_16_bit, c,
2569 hc_matchfinder_longest_match,
2575 c->max_search_depth,
2579 /* If there was no match found, or the only match found
2580 * was a distant short match, then choose a literal. */
2583 cur_offset >= 8192 - LZX_OFFSET_ADJUSTMENT &&
2584 cur_offset != recent_offsets[0] &&
2585 cur_offset != recent_offsets[1] &&
2586 cur_offset != recent_offsets[2]))
2588 lzx_choose_literal(c, *in_next, &litrunlen);
2593 /* Heuristic: if this match has the most recent offset,
2594 * then go ahead and choose it as a rep0 match. */
2595 if (cur_offset == recent_offsets[0]) {
2597 skip_len = cur_len - 1;
2598 cur_adjusted_offset = 0;
2599 goto choose_cur_match;
2602 /* Compute the longest match's score as an explicit
2604 cur_adjusted_offset = cur_offset + LZX_OFFSET_ADJUSTMENT;
2605 cur_score = lzx_explicit_offset_match_score(cur_len, cur_adjusted_offset);
2607 /* Find the longest repeat offset match at this
2608 * position. If we find one and it's "better" than the
2609 * explicit offset match we found, then go ahead and
2610 * choose the repeat offset match immediately. */
2611 best_rep_len = lzx_find_longest_repeat_offset_match(in_next,
2617 if (best_rep_len != 0 &&
2618 (rep_score = lzx_repeat_offset_match_score(best_rep_len,
2619 best_rep_idx)) >= cur_score)
2621 cur_len = best_rep_len;
2622 cur_adjusted_offset = best_rep_idx;
2623 skip_len = best_rep_len - 1;
2624 goto choose_cur_match;
2629 * We have a match at the current position. If the
2630 * match is very long, then choose it immediately.
2631 * Otherwise, see if there's a better match at the next
2635 if (cur_len >= nice_len) {
2636 skip_len = cur_len - 1;
2637 goto choose_cur_match;
2640 if (unlikely(max_len > in_end - in_next)) {
2641 max_len = in_end - in_next;
2642 nice_len = min(max_len, nice_len);
2645 next_len = CALL_HC_MF(is_16_bit, c,
2646 hc_matchfinder_longest_match,
2652 c->max_search_depth / 2,
2656 if (next_len <= cur_len - 2) {
2657 /* No potentially better match was found. */
2659 skip_len = cur_len - 2;
2660 goto choose_cur_match;
2663 next_adjusted_offset = next_offset + LZX_OFFSET_ADJUSTMENT;
2664 next_score = lzx_explicit_offset_match_score(next_len, next_adjusted_offset);
2666 best_rep_len = lzx_find_longest_repeat_offset_match(in_next,
2672 if (best_rep_len != 0 &&
2673 (rep_score = lzx_repeat_offset_match_score(best_rep_len,
2674 best_rep_idx)) >= next_score)
2677 if (rep_score > cur_score) {
2678 /* The next match is better, and it's a
2679 * repeat offset match. */
2680 lzx_choose_literal(c, *(in_next - 2),
2682 cur_len = best_rep_len;
2683 cur_adjusted_offset = best_rep_idx;
2684 skip_len = cur_len - 1;
2685 goto choose_cur_match;
2688 if (next_score > cur_score) {
2689 /* The next match is better, and it's an
2690 * explicit offset match. */
2691 lzx_choose_literal(c, *(in_next - 2),
2694 cur_adjusted_offset = next_adjusted_offset;
2695 cur_score = next_score;
2696 goto have_cur_match;
2700 /* The original match was better; choose it. */
2701 skip_len = cur_len - 2;
2704 /* Choose a match and have the matchfinder skip over its
2705 * remaining bytes. */
2706 lzx_choose_match(c, cur_len, cur_adjusted_offset,
2707 recent_offsets, is_16_bit,
2708 &litrunlen, &next_seq);
2709 CALL_HC_MF(is_16_bit, c,
2710 hc_matchfinder_skip_bytes,
2716 in_next += skip_len;
2718 /* Keep going until it's time to end the block. */
2719 } while (in_next < in_max_block_end &&
2720 !(c->split_stats.num_new_observations >=
2721 NUM_OBSERVATIONS_PER_BLOCK_CHECK &&
2722 in_next - in_block_begin >= MIN_BLOCK_SIZE &&
2723 in_end - in_next >= MIN_BLOCK_SIZE &&
2724 lzx_should_end_block(&c->split_stats)));
2726 /* Flush the block. */
2727 lzx_finish_sequence(next_seq, litrunlen);
2728 lzx_flush_block(c, os, in_block_begin, in_next - in_block_begin, 0);
2730 /* Keep going until we've reached the end of the input buffer. */
2731 } while (in_next != in_end);
2735 lzx_compress_lazy_16(struct lzx_compressor *c, const u8 *in, size_t in_nbytes,
2736 struct lzx_output_bitstream *os)
2738 lzx_compress_lazy(c, in, in_nbytes, os, true);
2742 lzx_compress_lazy_32(struct lzx_compressor *c, const u8 *in, size_t in_nbytes,
2743 struct lzx_output_bitstream *os)
2745 lzx_compress_lazy(c, in, in_nbytes, os, false);
2748 /******************************************************************************/
2749 /* Compressor operations */
2750 /*----------------------------------------------------------------------------*/
2753 * Generate tables for mapping match offsets (actually, "adjusted" match
2754 * offsets) to offset slots.
2757 lzx_init_offset_slot_tabs(struct lzx_compressor *c)
2759 u32 adjusted_offset = 0;
2763 for (; adjusted_offset < ARRAY_LEN(c->offset_slot_tab_1);
2766 if (adjusted_offset >= lzx_offset_slot_base[slot + 1] +
2767 LZX_OFFSET_ADJUSTMENT)
2769 c->offset_slot_tab_1[adjusted_offset] = slot;
2772 /* slots [30, 49] */
2773 for (; adjusted_offset < LZX_MAX_WINDOW_SIZE;
2774 adjusted_offset += (u32)1 << 14)
2776 if (adjusted_offset >= lzx_offset_slot_base[slot + 1] +
2777 LZX_OFFSET_ADJUSTMENT)
2779 c->offset_slot_tab_2[adjusted_offset >> 14] = slot;
2784 lzx_get_compressor_size(size_t max_bufsize, unsigned compression_level)
2786 if (compression_level <= MAX_FAST_LEVEL) {
2787 if (lzx_is_16_bit(max_bufsize))
2788 return offsetof(struct lzx_compressor, hc_mf_16) +
2789 hc_matchfinder_size_16(max_bufsize);
2791 return offsetof(struct lzx_compressor, hc_mf_32) +
2792 hc_matchfinder_size_32(max_bufsize);
2794 if (lzx_is_16_bit(max_bufsize))
2795 return offsetof(struct lzx_compressor, bt_mf_16) +
2796 bt_matchfinder_size_16(max_bufsize);
2798 return offsetof(struct lzx_compressor, bt_mf_32) +
2799 bt_matchfinder_size_32(max_bufsize);
2803 /* Compute the amount of memory needed to allocate an LZX compressor. */
2805 lzx_get_needed_memory(size_t max_bufsize, unsigned compression_level,
2810 if (max_bufsize > LZX_MAX_WINDOW_SIZE)
2813 size += lzx_get_compressor_size(max_bufsize, compression_level);
2815 size += max_bufsize; /* account for in_buffer */
2819 /* Allocate an LZX compressor. */
2821 lzx_create_compressor(size_t max_bufsize, unsigned compression_level,
2822 bool destructive, void **c_ret)
2824 unsigned window_order;
2825 struct lzx_compressor *c;
2827 /* Validate the maximum buffer size and get the window order from it. */
2828 window_order = lzx_get_window_order(max_bufsize);
2829 if (window_order == 0)
2830 return WIMLIB_ERR_INVALID_PARAM;
2832 /* Allocate the compressor. */
2833 c = MALLOC(lzx_get_compressor_size(max_bufsize, compression_level));
2837 c->window_order = window_order;
2838 c->num_main_syms = lzx_get_num_main_syms(window_order);
2839 c->destructive = destructive;
2841 /* Allocate the buffer for preprocessed data if needed. */
2842 if (!c->destructive) {
2843 c->in_buffer = MALLOC(max_bufsize);
2848 if (compression_level <= MAX_FAST_LEVEL) {
2850 /* Fast compression: Use lazy parsing. */
2851 if (lzx_is_16_bit(max_bufsize))
2852 c->impl = lzx_compress_lazy_16;
2854 c->impl = lzx_compress_lazy_32;
2856 /* Scale max_search_depth and nice_match_length with the
2857 * compression level. */
2858 c->max_search_depth = (60 * compression_level) / 20;
2859 c->nice_match_length = (80 * compression_level) / 20;
2861 /* lzx_compress_lazy() needs max_search_depth >= 2 because it
2862 * halves the max_search_depth when attempting a lazy match, and
2863 * max_search_depth must be at least 1. */
2864 c->max_search_depth = max(c->max_search_depth, 2);
2867 /* Normal / high compression: Use near-optimal parsing. */
2868 if (lzx_is_16_bit(max_bufsize))
2869 c->impl = lzx_compress_near_optimal_16;
2871 c->impl = lzx_compress_near_optimal_32;
2873 /* Scale max_search_depth and nice_match_length with the
2874 * compression level. */
2875 c->max_search_depth = (24 * compression_level) / 50;
2876 c->nice_match_length = (48 * compression_level) / 50;
2878 /* Also scale num_optim_passes with the compression level. But
2879 * the more passes there are, the less they help --- so don't
2880 * add them linearly. */
2881 c->num_optim_passes = 1;
2882 c->num_optim_passes += (compression_level >= 45);
2883 c->num_optim_passes += (compression_level >= 70);
2884 c->num_optim_passes += (compression_level >= 100);
2885 c->num_optim_passes += (compression_level >= 150);
2886 c->num_optim_passes += (compression_level >= 200);
2887 c->num_optim_passes += (compression_level >= 300);
2889 /* max_search_depth must be at least 1. */
2890 c->max_search_depth = max(c->max_search_depth, 1);
2893 /* Prepare the offset => offset slot mapping. */
2894 lzx_init_offset_slot_tabs(c);
2902 return WIMLIB_ERR_NOMEM;
2905 /* Compress a buffer of data. */
2907 lzx_compress(const void *restrict in, size_t in_nbytes,
2908 void *restrict out, size_t out_nbytes_avail, void *restrict _c)
2910 struct lzx_compressor *c = _c;
2911 struct lzx_output_bitstream os;
2914 /* Don't bother trying to compress very small inputs. */
2918 /* If the compressor is in "destructive" mode, then we can directly
2919 * preprocess the input data. Otherwise, we need to copy it into an
2920 * internal buffer first. */
2921 if (!c->destructive) {
2922 memcpy(c->in_buffer, in, in_nbytes);
2926 /* Preprocess the input data. */
2927 lzx_preprocess((void *)in, in_nbytes);
2929 /* Initially, the previous Huffman codeword lengths are all zeroes. */
2931 memset(&c->codes[1].lens, 0, sizeof(struct lzx_lens));
2933 /* Initialize the output bitstream. */
2934 lzx_init_output(&os, out, out_nbytes_avail);
2936 /* Call the compression level-specific compress() function. */
2937 (*c->impl)(c, in, in_nbytes, &os);
2939 /* Flush the output bitstream. */
2940 result = lzx_flush_output(&os);
2942 /* If the data did not compress to less than its original size and we
2943 * preprocessed the original buffer, then postprocess it to restore it
2944 * to its original state. */
2945 if (result == 0 && c->destructive)
2946 lzx_postprocess((void *)in, in_nbytes);
2948 /* Return the number of compressed bytes, or 0 if the input did not
2949 * compress to less than its original size. */
2953 /* Free an LZX compressor. */
2955 lzx_free_compressor(void *_c)
2957 struct lzx_compressor *c = _c;
2959 if (!c->destructive)
2964 const struct compressor_ops lzx_compressor_ops = {
2965 .get_needed_memory = lzx_get_needed_memory,
2966 .create_compressor = lzx_create_compressor,
2967 .compress = lzx_compress,
2968 .free_compressor = lzx_free_compressor,