4 * A compressor that produces output compatible with the LZMS compression format.
8 * Copyright (C) 2013, 2014 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/.
28 #include "wimlib/compress_common.h"
29 #include "wimlib/compressor_ops.h"
30 #include "wimlib/endianness.h"
31 #include "wimlib/error.h"
32 #include "wimlib/lz_mf.h"
33 #include "wimlib/lz_repsearch.h"
34 #include "wimlib/lzms.h"
35 #include "wimlib/unaligned.h"
36 #include "wimlib/util.h"
42 /* Stucture used for writing raw bits as a series of 16-bit little endian coding
43 * units. This starts at the *end* of the compressed data buffer and proceeds
45 struct lzms_output_bitstream {
47 /* Bits that haven't yet been written to the output buffer. */
50 /* Number of bits currently held in @bitbuf. */
53 /* Pointer to one past the next position in the compressed data buffer
54 * at which to output a 16-bit coding unit. */
57 /* Pointer to the beginning of the output buffer. (The "end" when
58 * writing backwards!) */
62 /* Stucture used for range encoding (raw version). This starts at the
63 * *beginning* of the compressed data buffer and proceeds forward. */
64 struct lzms_range_encoder_raw {
66 /* A 33-bit variable that holds the low boundary of the current range.
67 * The 33rd bit is needed to catch carries. */
70 /* Size of the current range. */
73 /* Next 16-bit coding unit to output. */
76 /* Number of 16-bit coding units whose output has been delayed due to
77 * possible carrying. The first such coding unit is @cache; all
78 * subsequent such coding units are 0xffff. */
81 /* Pointer to the beginning of the output buffer. */
84 /* Pointer to the position in the output buffer at which the next coding
85 * unit must be written. */
88 /* Pointer just past the end of the output buffer. */
92 /* Structure used for range encoding. This wraps around `struct
93 * lzms_range_encoder_raw' to use and maintain probability entries. */
94 struct lzms_range_encoder {
96 /* Pointer to the raw range encoder, which has no persistent knowledge
97 * of probabilities. Multiple lzms_range_encoder's share the same
98 * lzms_range_encoder_raw. */
99 struct lzms_range_encoder_raw *rc;
101 /* Bits recently encoded by this range encoder. This is used as an
102 * index into @prob_entries. */
105 /* Bitmask for @state to prevent its value from exceeding the number of
106 * probability entries. */
109 /* Probability entries being used for this range encoder. */
110 struct lzms_probability_entry prob_entries[LZMS_MAX_NUM_STATES];
113 /* Structure used for Huffman encoding. */
114 struct lzms_huffman_encoder {
116 /* Bitstream to write Huffman-encoded symbols and verbatim bits to.
117 * Multiple lzms_huffman_encoder's share the same lzms_output_bitstream.
119 struct lzms_output_bitstream *os;
121 /* Number of symbols that have been written using this code far. Reset
122 * to 0 whenever the code is rebuilt. */
123 u32 num_syms_written;
125 /* When @num_syms_written reaches this number, the Huffman code must be
129 /* Number of symbols in the represented Huffman code. */
132 /* Running totals of symbol frequencies. These are diluted slightly
133 * whenever the code is rebuilt. */
134 u32 sym_freqs[LZMS_MAX_NUM_SYMS];
136 /* The length, in bits, of each symbol in the Huffman code. */
137 u8 lens[LZMS_MAX_NUM_SYMS];
139 /* The codeword of each symbol in the Huffman code. */
140 u32 codewords[LZMS_MAX_NUM_SYMS];
143 /* Internal compression parameters */
144 struct lzms_compressor_params {
145 u32 min_match_length;
146 u32 nice_match_length;
147 u32 max_search_depth;
148 u32 optim_array_length;
151 /* State of the LZMS compressor */
152 struct lzms_compressor {
154 /* Internal compression parameters */
155 struct lzms_compressor_params params;
157 /* Data currently being compressed */
161 /* Lempel-Ziv match-finder */
164 /* Temporary space to store found matches */
165 struct lz_match *matches;
167 /* Per-position data for near-optimal parsing */
168 struct lzms_mc_pos_data *optimum;
169 struct lzms_mc_pos_data *optimum_end;
171 /* Raw range encoder which outputs to the beginning of the compressed
172 * data buffer, proceeding forwards */
173 struct lzms_range_encoder_raw rc;
175 /* Bitstream which outputs to the end of the compressed data buffer,
176 * proceeding backwards */
177 struct lzms_output_bitstream os;
180 struct lzms_range_encoder main_range_encoder;
181 struct lzms_range_encoder match_range_encoder;
182 struct lzms_range_encoder lz_match_range_encoder;
183 struct lzms_range_encoder lz_repeat_match_range_encoders[LZMS_NUM_RECENT_OFFSETS - 1];
184 struct lzms_range_encoder delta_match_range_encoder;
185 struct lzms_range_encoder delta_repeat_match_range_encoders[LZMS_NUM_RECENT_OFFSETS - 1];
187 /* Huffman encoders */
188 struct lzms_huffman_encoder literal_encoder;
189 struct lzms_huffman_encoder lz_offset_encoder;
190 struct lzms_huffman_encoder length_encoder;
191 struct lzms_huffman_encoder delta_power_encoder;
192 struct lzms_huffman_encoder delta_offset_encoder;
194 /* Used for preprocessing */
195 s32 last_target_usages[65536];
197 #define LZMS_NUM_FAST_LENGTHS 256
198 /* Table: length => length slot for small lengths */
199 u8 length_slot_fast[LZMS_NUM_FAST_LENGTHS];
201 /* Table: length => current cost for small match lengths */
202 u32 length_cost_fast[LZMS_NUM_FAST_LENGTHS];
204 #define LZMS_NUM_FAST_OFFSETS 32768
205 /* Table: offset => offset slot for small offsets */
206 u8 offset_slot_fast[LZMS_NUM_FAST_OFFSETS];
210 * Match chooser position data:
212 * An array of these structures is used during the near-optimal match-choosing
213 * algorithm. They correspond to consecutive positions in the window and are
214 * used to keep track of the cost to reach each position, and the match/literal
215 * choices that need to be chosen to reach that position.
217 struct lzms_mc_pos_data {
219 /* The cost, in bits, of the lowest-cost path that has been found to
220 * reach this position. This can change as progressively lower cost
221 * paths are found to reach this position. */
223 #define MC_INFINITE_COST UINT32_MAX
225 /* The match or literal that was taken to reach this position. This can
226 * change as progressively lower cost paths are found to reach this
229 * This variable is divided into two bitfields.
232 * Low bits are 1, high bits are the literal.
234 * Explicit offset matches:
235 * Low bits are the match length, high bits are the offset plus 2.
237 * Repeat offset matches:
238 * Low bits are the match length, high bits are the queue index.
241 #define MC_OFFSET_SHIFT 32
242 #define MC_LEN_MASK (((u64)1 << MC_OFFSET_SHIFT) - 1)
244 /* The LZMS adaptive state that exists at this position. This is filled
245 * in lazily, only after the minimum-cost path to this position is
248 * Note: the way we handle this adaptive state in the "minimum-cost"
249 * parse is actually only an approximation. It's possible for the
250 * globally optimal, minimum cost path to contain a prefix, ending at a
251 * position, where that path prefix is *not* the minimum cost path to
252 * that position. This can happen if such a path prefix results in a
253 * different adaptive state which results in lower costs later. We do
254 * not solve this problem; we only consider the lowest cost to reach
255 * each position, which seems to be an acceptable approximation.
257 * Note: this adaptive state also does not include the probability
258 * entries or current Huffman codewords. Those aren't maintained
259 * per-position and are only updated occassionally. */
260 struct lzms_adaptive_state {
261 struct lzms_lz_lru_queues lru;
265 u8 lz_repeat_match_state[LZMS_NUM_RECENT_OFFSETS - 1];
270 lzms_init_fast_slots(struct lzms_compressor *c)
272 /* Create table mapping small lengths to length slots. */
273 for (unsigned slot = 0, i = 0; i < LZMS_NUM_FAST_LENGTHS; i++) {
274 while (i >= lzms_length_slot_base[slot + 1])
276 c->length_slot_fast[i] = slot;
279 /* Create table mapping small offsets to offset slots. */
280 for (unsigned slot = 0, i = 0; i < LZMS_NUM_FAST_OFFSETS; i++) {
281 while (i >= lzms_offset_slot_base[slot + 1])
283 c->offset_slot_fast[i] = slot;
287 static inline unsigned
288 lzms_get_length_slot_fast(const struct lzms_compressor *c, u32 length)
290 if (likely(length < LZMS_NUM_FAST_LENGTHS))
291 return c->length_slot_fast[length];
293 return lzms_get_length_slot(length);
296 static inline unsigned
297 lzms_get_offset_slot_fast(const struct lzms_compressor *c, u32 offset)
299 if (offset < LZMS_NUM_FAST_OFFSETS)
300 return c->offset_slot_fast[offset];
302 return lzms_get_offset_slot(offset);
305 /* Initialize the output bitstream @os to write backwards to the specified
306 * compressed data buffer @out that is @out_limit 16-bit integers long. */
308 lzms_output_bitstream_init(struct lzms_output_bitstream *os,
309 le16 *out, size_t out_limit)
313 os->next = out + out_limit;
318 * Write some bits, contained in the low @num_bits bits of @bits (ordered from
319 * high-order to low-order), to the output bitstream @os.
321 * @max_num_bits is a compile-time constant that specifies the maximum number of
322 * bits that can ever be written at this call site.
325 lzms_output_bitstream_put_varbits(struct lzms_output_bitstream *os,
326 u32 bits, unsigned num_bits,
327 unsigned max_num_bits)
329 LZMS_ASSERT(num_bits <= 48);
331 /* Add the bits to the bit buffer variable. */
332 os->bitcount += num_bits;
333 os->bitbuf = (os->bitbuf << num_bits) | bits;
335 /* Check whether any coding units need to be written. */
336 while (os->bitcount >= 16) {
340 /* Write a coding unit, unless it would underflow the buffer. */
341 if (os->next != os->begin)
342 put_unaligned_u16_le(os->bitbuf >> os->bitcount, --os->next);
344 /* Optimization for call sites that never write more than 16
346 if (max_num_bits <= 16)
351 /* Flush the output bitstream, ensuring that all bits written to it have been
352 * written to memory. Returns %true if all bits have been output successfully,
353 * or %false if an overrun occurred. */
355 lzms_output_bitstream_flush(struct lzms_output_bitstream *os)
357 if (os->next == os->begin)
360 if (os->bitcount != 0)
361 put_unaligned_u16_le(os->bitbuf << (16 - os->bitcount), --os->next);
366 /* Initialize the range encoder @rc to write forwards to the specified
367 * compressed data buffer @out that is @out_limit 16-bit integers long. */
369 lzms_range_encoder_raw_init(struct lzms_range_encoder_raw *rc,
370 le16 *out, size_t out_limit)
373 rc->range = 0xffffffff;
378 rc->end = out + out_limit;
382 * Attempt to flush bits from the range encoder.
384 * Note: this is based on the public domain code for LZMA written by Igor
385 * Pavlov. The only differences in this function are that in LZMS the bits must
386 * be output in 16-bit coding units instead of 8-bit coding units, and that in
387 * LZMS the first coding unit is not ignored by the decompressor, so the encoder
388 * cannot output a dummy value to that position.
390 * The basic idea is that we're writing bits from @rc->low to the output.
391 * However, due to carrying, the writing of coding units with value 0xffff, as
392 * well as one prior coding unit, must be delayed until it is determined whether
396 lzms_range_encoder_raw_shift_low(struct lzms_range_encoder_raw *rc)
398 if ((u32)(rc->low) < 0xffff0000 ||
399 (u32)(rc->low >> 32) != 0)
401 /* Carry not needed (rc->low < 0xffff0000), or carry occurred
402 * ((rc->low >> 32) != 0, a.k.a. the carry bit is 1). */
404 if (likely(rc->next >= rc->begin)) {
405 if (rc->next != rc->end) {
406 put_unaligned_u16_le(rc->cache +
407 (u16)(rc->low >> 32),
414 } while (--rc->cache_size != 0);
416 rc->cache = (rc->low >> 16) & 0xffff;
419 rc->low = (rc->low & 0xffff) << 16;
423 lzms_range_encoder_raw_normalize(struct lzms_range_encoder_raw *rc)
425 if (rc->range <= 0xffff) {
427 lzms_range_encoder_raw_shift_low(rc);
432 lzms_range_encoder_raw_flush(struct lzms_range_encoder_raw *rc)
434 for (unsigned i = 0; i < 4; i++)
435 lzms_range_encoder_raw_shift_low(rc);
436 return rc->next != rc->end;
439 /* Encode the next bit using the range encoder (raw version).
441 * @prob is the chance out of LZMS_PROBABILITY_MAX that the next bit is 0. */
443 lzms_range_encoder_raw_encode_bit(struct lzms_range_encoder_raw *rc,
446 lzms_range_encoder_raw_normalize(rc);
448 u32 bound = (rc->range >> LZMS_PROBABILITY_BITS) * prob;
457 /* Encode a bit using the specified range encoder. This wraps around
458 * lzms_range_encoder_raw_encode_bit() to handle using and updating the
459 * appropriate state and probability entry. */
461 lzms_range_encode_bit(struct lzms_range_encoder *enc, int bit)
463 struct lzms_probability_entry *prob_entry;
466 /* Load the probability entry corresponding to the current state. */
467 prob_entry = &enc->prob_entries[enc->state];
469 /* Update the state based on the next bit. */
470 enc->state = ((enc->state << 1) | bit) & enc->mask;
472 /* Get the probability that the bit is 0. */
473 prob = lzms_get_probability(prob_entry);
475 /* Update the probability entry. */
476 lzms_update_probability_entry(prob_entry, bit);
478 /* Encode the bit. */
479 lzms_range_encoder_raw_encode_bit(enc->rc, bit, prob);
482 /* Called when an adaptive Huffman code needs to be rebuilt. */
484 lzms_rebuild_huffman_code(struct lzms_huffman_encoder *enc)
486 make_canonical_huffman_code(enc->num_syms,
487 LZMS_MAX_CODEWORD_LEN,
492 /* Dilute the frequencies. */
493 for (unsigned i = 0; i < enc->num_syms; i++) {
494 enc->sym_freqs[i] >>= 1;
495 enc->sym_freqs[i] += 1;
497 enc->num_syms_written = 0;
500 /* Encode a symbol using the specified Huffman encoder. */
502 lzms_huffman_encode_symbol(struct lzms_huffman_encoder *enc, unsigned sym)
504 lzms_output_bitstream_put_varbits(enc->os,
507 LZMS_MAX_CODEWORD_LEN);
508 ++enc->sym_freqs[sym];
509 if (++enc->num_syms_written == enc->rebuild_freq)
510 lzms_rebuild_huffman_code(enc);
514 lzms_update_fast_length_costs(struct lzms_compressor *c);
516 /* Encode a match length. */
518 lzms_encode_length(struct lzms_compressor *c, u32 length)
521 unsigned num_extra_bits;
524 slot = lzms_get_length_slot_fast(c, length);
526 extra_bits = length - lzms_length_slot_base[slot];
527 num_extra_bits = lzms_extra_length_bits[slot];
529 lzms_huffman_encode_symbol(&c->length_encoder, slot);
530 if (c->length_encoder.num_syms_written == 0)
531 lzms_update_fast_length_costs(c);
533 lzms_output_bitstream_put_varbits(c->length_encoder.os,
534 extra_bits, num_extra_bits, 30);
537 /* Encode an LZ match offset. */
539 lzms_encode_lz_offset(struct lzms_compressor *c, u32 offset)
542 unsigned num_extra_bits;
545 slot = lzms_get_offset_slot_fast(c, offset);
547 extra_bits = offset - lzms_offset_slot_base[slot];
548 num_extra_bits = lzms_extra_offset_bits[slot];
550 lzms_huffman_encode_symbol(&c->lz_offset_encoder, slot);
551 lzms_output_bitstream_put_varbits(c->lz_offset_encoder.os,
552 extra_bits, num_extra_bits, 30);
555 /* Encode a literal byte. */
557 lzms_encode_literal(struct lzms_compressor *c, unsigned literal)
559 /* Main bit: 0 = a literal, not a match. */
560 lzms_range_encode_bit(&c->main_range_encoder, 0);
562 /* Encode the literal using the current literal Huffman code. */
563 lzms_huffman_encode_symbol(&c->literal_encoder, literal);
566 /* Encode an LZ repeat offset match. */
568 lzms_encode_lz_repeat_offset_match(struct lzms_compressor *c,
569 u32 length, unsigned rep_index)
573 /* Main bit: 1 = a match, not a literal. */
574 lzms_range_encode_bit(&c->main_range_encoder, 1);
576 /* Match bit: 0 = an LZ match, not a delta match. */
577 lzms_range_encode_bit(&c->match_range_encoder, 0);
579 /* LZ match bit: 1 = repeat offset, not an explicit offset. */
580 lzms_range_encode_bit(&c->lz_match_range_encoder, 1);
582 /* Encode the repeat offset index. A 1 bit is encoded for each index
583 * passed up. This sequence of 1 bits is terminated by a 0 bit, or
584 * automatically when (LZMS_NUM_RECENT_OFFSETS - 1) 1 bits have been
586 for (i = 0; i < rep_index; i++)
587 lzms_range_encode_bit(&c->lz_repeat_match_range_encoders[i], 1);
589 if (i < LZMS_NUM_RECENT_OFFSETS - 1)
590 lzms_range_encode_bit(&c->lz_repeat_match_range_encoders[i], 0);
592 /* Encode the match length. */
593 lzms_encode_length(c, length);
596 /* Encode an LZ explicit offset match. */
598 lzms_encode_lz_explicit_offset_match(struct lzms_compressor *c,
599 u32 length, u32 offset)
601 /* Main bit: 1 = a match, not a literal. */
602 lzms_range_encode_bit(&c->main_range_encoder, 1);
604 /* Match bit: 0 = an LZ match, not a delta match. */
605 lzms_range_encode_bit(&c->match_range_encoder, 0);
607 /* LZ match bit: 0 = explicit offset, not a repeat offset. */
608 lzms_range_encode_bit(&c->lz_match_range_encoder, 0);
610 /* Encode the match offset. */
611 lzms_encode_lz_offset(c, offset);
613 /* Encode the match length. */
614 lzms_encode_length(c, length);
618 lzms_encode_item(struct lzms_compressor *c, u64 mc_item_data)
620 u32 len = mc_item_data & MC_LEN_MASK;
621 u32 offset_data = mc_item_data >> MC_OFFSET_SHIFT;
624 lzms_encode_literal(c, offset_data);
625 else if (offset_data < LZMS_NUM_RECENT_OFFSETS)
626 lzms_encode_lz_repeat_offset_match(c, len, offset_data);
628 lzms_encode_lz_explicit_offset_match(c, len, offset_data - LZMS_OFFSET_OFFSET);
631 /* Encode a list of matches and literals chosen by the parsing algorithm. */
633 lzms_encode_item_list(struct lzms_compressor *c,
634 struct lzms_mc_pos_data *cur_optimum_ptr)
636 struct lzms_mc_pos_data *end_optimum_ptr;
640 /* The list is currently in reverse order (last item to first item).
642 end_optimum_ptr = cur_optimum_ptr;
643 saved_item = cur_optimum_ptr->mc_item_data;
646 cur_optimum_ptr -= item & MC_LEN_MASK;
647 saved_item = cur_optimum_ptr->mc_item_data;
648 cur_optimum_ptr->mc_item_data = item;
649 } while (cur_optimum_ptr != c->optimum);
651 /* Walk the list of items from beginning to end, encoding each item. */
653 lzms_encode_item(c, cur_optimum_ptr->mc_item_data);
654 cur_optimum_ptr += (cur_optimum_ptr->mc_item_data) & MC_LEN_MASK;
655 } while (cur_optimum_ptr != end_optimum_ptr);
658 /* Each bit costs 1 << LZMS_COST_SHIFT units. */
659 #define LZMS_COST_SHIFT 6
661 /*#define LZMS_RC_COSTS_USE_FLOATING_POINT*/
664 lzms_rc_costs[LZMS_PROBABILITY_MAX + 1];
666 #ifdef LZMS_RC_COSTS_USE_FLOATING_POINT
671 lzms_do_init_rc_costs(void)
673 /* Fill in a table that maps range coding probabilities needed to code a
674 * bit X (0 or 1) to the number of bits (scaled by a constant factor, to
675 * handle fractional costs) needed to code that bit X.
677 * Consider the range of the range decoder. To eliminate exactly half
678 * the range (logical probability of 0.5), we need exactly 1 bit. For
679 * lower probabilities we need more bits and for higher probabilities we
680 * need fewer bits. In general, a logical probability of N will
681 * eliminate the proportion 1 - N of the range; this information takes
682 * log2(1 / N) bits to encode.
684 * The below loop is simply calculating this number of bits for each
685 * possible probability allowed by the LZMS compression format, but
686 * without using real numbers. To handle fractional probabilities, each
687 * cost is multiplied by (1 << LZMS_COST_SHIFT). These techniques are
688 * based on those used by LZMA.
690 * Note that in LZMS, a probability x really means x / 64, and 0 / 64 is
691 * really interpreted as 1 / 64 and 64 / 64 is really interpreted as
694 for (u32 i = 0; i <= LZMS_PROBABILITY_MAX; i++) {
699 else if (prob == LZMS_PROBABILITY_MAX)
700 prob = LZMS_PROBABILITY_MAX - 1;
702 #ifdef LZMS_RC_COSTS_USE_FLOATING_POINT
703 lzms_rc_costs[i] = log2((double)LZMS_PROBABILITY_MAX / prob) *
704 (1 << LZMS_COST_SHIFT);
708 for (u32 j = 0; j < LZMS_COST_SHIFT; j++) {
711 while (w >= ((u32)1 << 16)) {
716 lzms_rc_costs[i] = (LZMS_PROBABILITY_BITS << LZMS_COST_SHIFT) -
723 lzms_init_rc_costs(void)
725 static pthread_once_t once = PTHREAD_ONCE_INIT;
727 pthread_once(&once, lzms_do_init_rc_costs);
730 /* Return the cost to range-encode the specified bit from the specified state.*/
732 lzms_rc_bit_cost(const struct lzms_range_encoder *enc, u8 cur_state, int bit)
737 prob_zero = enc->prob_entries[cur_state].num_recent_zero_bits;
740 prob_correct = prob_zero;
742 prob_correct = LZMS_PROBABILITY_MAX - prob_zero;
744 return lzms_rc_costs[prob_correct];
747 /* Return the cost to Huffman-encode the specified symbol. */
749 lzms_huffman_symbol_cost(const struct lzms_huffman_encoder *enc, unsigned sym)
751 return (u32)enc->lens[sym] << LZMS_COST_SHIFT;
754 /* Return the cost to encode the specified literal byte. */
756 lzms_literal_cost(const struct lzms_compressor *c, unsigned literal,
757 const struct lzms_adaptive_state *state)
759 return lzms_rc_bit_cost(&c->main_range_encoder, state->main_state, 0) +
760 lzms_huffman_symbol_cost(&c->literal_encoder, literal);
763 /* Update the table that directly provides the costs for small lengths. */
765 lzms_update_fast_length_costs(struct lzms_compressor *c)
771 for (len = 1; len < LZMS_NUM_FAST_LENGTHS; len++) {
773 while (len >= lzms_length_slot_base[slot + 1]) {
775 cost = (u32)(c->length_encoder.lens[slot] +
776 lzms_extra_length_bits[slot]) << LZMS_COST_SHIFT;
779 c->length_cost_fast[len] = cost;
783 /* Return the cost to encode the specified match length, which must be less than
784 * LZMS_NUM_FAST_LENGTHS. */
786 lzms_fast_length_cost(const struct lzms_compressor *c, u32 length)
788 LZMS_ASSERT(length < LZMS_NUM_FAST_LENGTHS);
789 return c->length_cost_fast[length];
792 /* Return the cost to encode the specified LZ match offset. */
794 lzms_lz_offset_cost(const struct lzms_compressor *c, u32 offset)
796 unsigned slot = lzms_get_offset_slot_fast(c, offset);
798 return (u32)(c->lz_offset_encoder.lens[slot] +
799 lzms_extra_offset_bits[slot]) << LZMS_COST_SHIFT;
803 * Consider coding the match at repeat offset index @rep_idx. Consider each
804 * length from the minimum (2) to the full match length (@rep_len).
807 lzms_consider_lz_repeat_offset_match(const struct lzms_compressor *c,
808 struct lzms_mc_pos_data *cur_optimum_ptr,
809 u32 rep_len, unsigned rep_idx)
816 base_cost = cur_optimum_ptr->cost;
818 base_cost += lzms_rc_bit_cost(&c->main_range_encoder,
819 cur_optimum_ptr->state.main_state, 1);
821 base_cost += lzms_rc_bit_cost(&c->match_range_encoder,
822 cur_optimum_ptr->state.match_state, 0);
824 base_cost += lzms_rc_bit_cost(&c->lz_match_range_encoder,
825 cur_optimum_ptr->state.lz_match_state, 1);
827 for (i = 0; i < rep_idx; i++)
828 base_cost += lzms_rc_bit_cost(&c->lz_repeat_match_range_encoders[i],
829 cur_optimum_ptr->state.lz_repeat_match_state[i], 1);
831 if (i < LZMS_NUM_RECENT_OFFSETS - 1)
832 base_cost += lzms_rc_bit_cost(&c->lz_repeat_match_range_encoders[i],
833 cur_optimum_ptr->state.lz_repeat_match_state[i], 0);
837 cost = base_cost + lzms_fast_length_cost(c, len);
838 if (cost < (cur_optimum_ptr + len)->cost) {
839 (cur_optimum_ptr + len)->mc_item_data =
840 ((u64)rep_idx << MC_OFFSET_SHIFT) | len;
841 (cur_optimum_ptr + len)->cost = cost;
843 } while (++len <= rep_len);
847 * Consider coding each match in @matches as an explicit offset match.
849 * @matches must be sorted by strictly increasing length and strictly increasing
850 * offset. This is guaranteed by the match-finder.
852 * We consider each length from the minimum (2) to the longest
853 * (matches[num_matches - 1].len). For each length, we consider only the
854 * smallest offset for which that length is available. Although this is not
855 * guaranteed to be optimal due to the possibility of a larger offset costing
856 * less than a smaller offset to code, this is a very useful heuristic.
859 lzms_consider_lz_explicit_offset_matches(const struct lzms_compressor *c,
860 struct lzms_mc_pos_data *cur_optimum_ptr,
861 const struct lz_match matches[],
870 base_cost = cur_optimum_ptr->cost;
872 base_cost += lzms_rc_bit_cost(&c->main_range_encoder,
873 cur_optimum_ptr->state.main_state, 1);
875 base_cost += lzms_rc_bit_cost(&c->match_range_encoder,
876 cur_optimum_ptr->state.match_state, 0);
878 base_cost += lzms_rc_bit_cost(&c->lz_match_range_encoder,
879 cur_optimum_ptr->state.lz_match_state, 0);
883 position_cost = base_cost + lzms_lz_offset_cost(c, matches[i].offset);
885 cost = position_cost + lzms_fast_length_cost(c, len);
886 if (cost < (cur_optimum_ptr + len)->cost) {
887 (cur_optimum_ptr + len)->mc_item_data =
888 ((u64)(matches[i].offset + LZMS_OFFSET_OFFSET)
889 << MC_OFFSET_SHIFT) | len;
890 (cur_optimum_ptr + len)->cost = cost;
892 } while (++len <= matches[i].len);
893 } while (++i != num_matches);
897 lzms_init_adaptive_state(struct lzms_adaptive_state *state)
901 lzms_init_lz_lru_queues(&state->lru);
902 state->main_state = 0;
903 state->match_state = 0;
904 state->lz_match_state = 0;
905 for (i = 0; i < LZMS_NUM_RECENT_OFFSETS - 1; i++)
906 state->lz_repeat_match_state[i] = 0;
910 lzms_update_main_state(struct lzms_adaptive_state *state, int is_match)
912 state->main_state = ((state->main_state << 1) | is_match) % LZMS_NUM_MAIN_STATES;
916 lzms_update_match_state(struct lzms_adaptive_state *state, int is_delta)
918 state->match_state = ((state->match_state << 1) | is_delta) % LZMS_NUM_MATCH_STATES;
922 lzms_update_lz_match_state(struct lzms_adaptive_state *state, int is_repeat_offset)
924 state->lz_match_state = ((state->lz_match_state << 1) | is_repeat_offset) % LZMS_NUM_LZ_MATCH_STATES;
928 lzms_update_lz_repeat_match_state(struct lzms_adaptive_state *state, int rep_idx)
932 for (i = 0; i < rep_idx; i++)
933 state->lz_repeat_match_state[i] =
934 ((state->lz_repeat_match_state[i] << 1) | 1) %
935 LZMS_NUM_LZ_REPEAT_MATCH_STATES;
937 if (i < LZMS_NUM_RECENT_OFFSETS - 1)
938 state->lz_repeat_match_state[i] =
939 ((state->lz_repeat_match_state[i] << 1) | 0) %
940 LZMS_NUM_LZ_REPEAT_MATCH_STATES;
944 * The main near-optimal parsing routine.
946 * Briefly, the algorithm does an approximate minimum-cost path search to find a
947 * "near-optimal" sequence of matches and literals to output, based on the
948 * current cost model. The algorithm steps forward, position by position (byte
949 * by byte), and updates the minimum cost path to reach each later position that
950 * can be reached using a match or literal from the current position. This is
951 * essentially Dijkstra's algorithm in disguise: the graph nodes are positions,
952 * the graph edges are possible matches/literals to code, and the cost of each
953 * edge is the estimated number of bits that will be required to output the
954 * corresponding match or literal. But one difference is that we actually
955 * compute the lowest-cost path in pieces, where each piece is terminated when
956 * there are no choices to be made.
960 * - This does not output any delta matches.
962 * - The costs of literals and matches are estimated using the range encoder
963 * states and the semi-adaptive Huffman codes. Except for range encoding
964 * states, costs are assumed to be constant throughout a single run of the
965 * parsing algorithm, which can parse up to @optim_array_length bytes of data.
966 * This introduces a source of inaccuracy because the probabilities and
967 * Huffman codes can change over this part of the data.
970 lzms_near_optimal_parse(struct lzms_compressor *c)
972 const u8 *window_ptr;
973 const u8 *window_end;
974 struct lzms_mc_pos_data *cur_optimum_ptr;
975 struct lzms_mc_pos_data *end_optimum_ptr;
979 unsigned rep_max_idx;
986 window_ptr = c->cur_window;
987 window_end = window_ptr + c->cur_window_size;
989 lzms_init_adaptive_state(&c->optimum[0].state);
992 /* Start building a new list of items, which will correspond to the next
993 * piece of the overall minimum-cost path. */
995 cur_optimum_ptr = c->optimum;
996 cur_optimum_ptr->cost = 0;
997 end_optimum_ptr = cur_optimum_ptr;
999 /* States should currently be consistent with the encoders. */
1000 LZMS_ASSERT(cur_optimum_ptr->state.main_state == c->main_range_encoder.state);
1001 LZMS_ASSERT(cur_optimum_ptr->state.match_state == c->match_range_encoder.state);
1002 LZMS_ASSERT(cur_optimum_ptr->state.lz_match_state == c->lz_match_range_encoder.state);
1003 for (i = 0; i < LZMS_NUM_RECENT_OFFSETS - 1; i++)
1004 LZMS_ASSERT(cur_optimum_ptr->state.lz_repeat_match_state[i] ==
1005 c->lz_repeat_match_range_encoders[i].state);
1007 if (window_ptr == window_end)
1010 /* The following loop runs once for each per byte in the window, except
1011 * in a couple shortcut cases. */
1014 /* Find explicit offset matches with the current position. */
1015 num_matches = lz_mf_get_matches(c->mf, c->matches);
1019 * Find the longest repeat offset match with the current
1024 * - Only search for repeat offset matches if the
1025 * match-finder already found at least one match.
1027 * - Only consider the longest repeat offset match. It
1028 * seems to be rare for the optimal parse to include a
1029 * repeat offset match that doesn't have the longest
1030 * length (allowing for the possibility that not all
1031 * of that length is actually used).
1033 if (likely(window_ptr - c->cur_window >= LZMS_MAX_INIT_RECENT_OFFSET)) {
1034 BUILD_BUG_ON(LZMS_NUM_RECENT_OFFSETS != 3);
1035 rep_max_len = lz_repsearch3(window_ptr,
1036 window_end - window_ptr,
1037 cur_optimum_ptr->state.lru.recent_offsets,
1044 /* If there's a very long repeat offset match,
1045 * choose it immediately. */
1046 if (rep_max_len >= c->params.nice_match_length) {
1048 lz_mf_skip_positions(c->mf, rep_max_len - 1);
1049 window_ptr += rep_max_len;
1051 if (cur_optimum_ptr != c->optimum)
1052 lzms_encode_item_list(c, cur_optimum_ptr);
1054 lzms_encode_lz_repeat_offset_match(c, rep_max_len,
1057 c->optimum[0].state = cur_optimum_ptr->state;
1059 lzms_update_main_state(&c->optimum[0].state, 1);
1060 lzms_update_match_state(&c->optimum[0].state, 0);
1061 lzms_update_lz_match_state(&c->optimum[0].state, 1);
1062 lzms_update_lz_repeat_match_state(&c->optimum[0].state,
1065 c->optimum[0].state.lru.upcoming_offset =
1066 c->optimum[0].state.lru.recent_offsets[rep_max_idx];
1068 for (i = rep_max_idx; i < LZMS_NUM_RECENT_OFFSETS; i++)
1069 c->optimum[0].state.lru.recent_offsets[i] =
1070 c->optimum[0].state.lru.recent_offsets[i + 1];
1072 lzms_update_lz_lru_queue(&c->optimum[0].state.lru);
1076 /* If reaching any positions for the first time,
1077 * initialize their costs to "infinity". */
1078 while (end_optimum_ptr < cur_optimum_ptr + rep_max_len)
1079 (++end_optimum_ptr)->cost = MC_INFINITE_COST;
1081 /* Consider coding a repeat offset match. */
1082 lzms_consider_lz_repeat_offset_match(c, cur_optimum_ptr,
1083 rep_max_len, rep_max_idx);
1086 longest_len = c->matches[num_matches - 1].len;
1088 /* If there's a very long explicit offset match, choose
1089 * it immediately. */
1090 if (longest_len >= c->params.nice_match_length) {
1092 lz_mf_skip_positions(c->mf, longest_len - 1);
1093 window_ptr += longest_len;
1095 if (cur_optimum_ptr != c->optimum)
1096 lzms_encode_item_list(c, cur_optimum_ptr);
1098 lzms_encode_lz_explicit_offset_match(c, longest_len,
1099 c->matches[num_matches - 1].offset);
1101 c->optimum[0].state = cur_optimum_ptr->state;
1103 lzms_update_main_state(&c->optimum[0].state, 1);
1104 lzms_update_match_state(&c->optimum[0].state, 0);
1105 lzms_update_lz_match_state(&c->optimum[0].state, 0);
1107 c->optimum[0].state.lru.upcoming_offset =
1108 c->matches[num_matches - 1].offset;
1110 lzms_update_lz_lru_queue(&c->optimum[0].state.lru);
1114 /* If reaching any positions for the first time,
1115 * initialize their costs to "infinity". */
1116 while (end_optimum_ptr < cur_optimum_ptr + longest_len)
1117 (++end_optimum_ptr)->cost = MC_INFINITE_COST;
1119 /* Consider coding an explicit offset match. */
1120 lzms_consider_lz_explicit_offset_matches(c, cur_optimum_ptr,
1121 c->matches, num_matches);
1123 /* No matches found. The only choice at this position
1124 * is to code a literal. */
1126 if (end_optimum_ptr == cur_optimum_ptr)
1127 (++end_optimum_ptr)->cost = MC_INFINITE_COST;
1130 /* Consider coding a literal.
1132 * To avoid an extra unpredictable brench, actually checking the
1133 * preferability of coding a literal is integrated into the
1134 * adaptive state update code below. */
1135 literal = *window_ptr++;
1136 cost = cur_optimum_ptr->cost +
1137 lzms_literal_cost(c, literal, &cur_optimum_ptr->state);
1139 /* Advance to the next position. */
1142 /* The lowest-cost path to the current position is now known.
1143 * Finalize the adaptive state that results from taking this
1144 * lowest-cost path. */
1146 if (cost < cur_optimum_ptr->cost) {
1148 cur_optimum_ptr->cost = cost;
1149 cur_optimum_ptr->mc_item_data = ((u64)literal << MC_OFFSET_SHIFT) | 1;
1151 cur_optimum_ptr->state = (cur_optimum_ptr - 1)->state;
1153 lzms_update_main_state(&cur_optimum_ptr->state, 0);
1155 cur_optimum_ptr->state.lru.upcoming_offset = 0;
1158 len = cur_optimum_ptr->mc_item_data & MC_LEN_MASK;
1159 offset_data = cur_optimum_ptr->mc_item_data >> MC_OFFSET_SHIFT;
1161 cur_optimum_ptr->state = (cur_optimum_ptr - len)->state;
1163 lzms_update_main_state(&cur_optimum_ptr->state, 1);
1164 lzms_update_match_state(&cur_optimum_ptr->state, 0);
1166 if (offset_data >= LZMS_NUM_RECENT_OFFSETS) {
1168 /* Explicit offset LZ match */
1170 lzms_update_lz_match_state(&cur_optimum_ptr->state, 0);
1172 cur_optimum_ptr->state.lru.upcoming_offset =
1173 offset_data - LZMS_OFFSET_OFFSET;
1175 /* Repeat offset LZ match */
1177 lzms_update_lz_match_state(&cur_optimum_ptr->state, 1);
1178 lzms_update_lz_repeat_match_state(&cur_optimum_ptr->state,
1181 cur_optimum_ptr->state.lru.upcoming_offset =
1182 cur_optimum_ptr->state.lru.recent_offsets[offset_data];
1184 for (i = offset_data; i < LZMS_NUM_RECENT_OFFSETS; i++)
1185 cur_optimum_ptr->state.lru.recent_offsets[i] =
1186 cur_optimum_ptr->state.lru.recent_offsets[i + 1];
1190 lzms_update_lz_lru_queue(&cur_optimum_ptr->state.lru);
1193 * This loop will terminate when either of the following
1194 * conditions is true:
1196 * (1) cur_optimum_ptr == end_optimum_ptr
1198 * There are no paths that extend beyond the current
1199 * position. In this case, any path to a later position
1200 * must pass through the current position, so we can go
1201 * ahead and choose the list of items that led to this
1204 * (2) cur_optimum_ptr == c->optimum_end
1206 * This bounds the number of times the algorithm can step
1207 * forward before it is guaranteed to start choosing items.
1208 * This limits the memory usage. It also guarantees that
1209 * the parser will not go too long without updating the
1210 * probability tables.
1212 * Note: no check for end-of-window is needed because
1213 * end-of-window will trigger condition (1).
1215 if (cur_optimum_ptr == end_optimum_ptr ||
1216 cur_optimum_ptr == c->optimum_end)
1218 c->optimum[0].state = cur_optimum_ptr->state;
1223 /* Output the current list of items that constitute the minimum-cost
1224 * path to the current position. */
1225 lzms_encode_item_list(c, cur_optimum_ptr);
1230 lzms_init_range_encoder(struct lzms_range_encoder *enc,
1231 struct lzms_range_encoder_raw *rc, u32 num_states)
1235 LZMS_ASSERT(is_power_of_2(num_states));
1236 enc->mask = num_states - 1;
1237 for (u32 i = 0; i < num_states; i++) {
1238 enc->prob_entries[i].num_recent_zero_bits = LZMS_INITIAL_PROBABILITY;
1239 enc->prob_entries[i].recent_bits = LZMS_INITIAL_RECENT_BITS;
1244 lzms_init_huffman_encoder(struct lzms_huffman_encoder *enc,
1245 struct lzms_output_bitstream *os,
1247 unsigned rebuild_freq)
1250 enc->num_syms_written = 0;
1251 enc->rebuild_freq = rebuild_freq;
1252 enc->num_syms = num_syms;
1253 for (unsigned i = 0; i < num_syms; i++)
1254 enc->sym_freqs[i] = 1;
1256 make_canonical_huffman_code(enc->num_syms,
1257 LZMS_MAX_CODEWORD_LEN,
1263 /* Prepare the LZMS compressor for compressing a block of data. */
1265 lzms_prepare_compressor(struct lzms_compressor *c, const u8 *udata, u32 ulen,
1266 le16 *cdata, u32 clen16)
1268 unsigned num_offset_slots;
1270 /* Copy the uncompressed data into the @c->cur_window buffer. */
1271 memcpy(c->cur_window, udata, ulen);
1272 c->cur_window_size = ulen;
1274 /* Initialize the raw range encoder (writing forwards). */
1275 lzms_range_encoder_raw_init(&c->rc, cdata, clen16);
1277 /* Initialize the output bitstream for Huffman symbols and verbatim bits
1278 * (writing backwards). */
1279 lzms_output_bitstream_init(&c->os, cdata, clen16);
1281 /* Calculate the number of offset slots required. */
1282 num_offset_slots = lzms_get_offset_slot(ulen - 1) + 1;
1284 /* Initialize a Huffman encoder for each alphabet. */
1285 lzms_init_huffman_encoder(&c->literal_encoder, &c->os,
1286 LZMS_NUM_LITERAL_SYMS,
1287 LZMS_LITERAL_CODE_REBUILD_FREQ);
1289 lzms_init_huffman_encoder(&c->lz_offset_encoder, &c->os,
1291 LZMS_LZ_OFFSET_CODE_REBUILD_FREQ);
1293 lzms_init_huffman_encoder(&c->length_encoder, &c->os,
1295 LZMS_LENGTH_CODE_REBUILD_FREQ);
1297 lzms_init_huffman_encoder(&c->delta_offset_encoder, &c->os,
1299 LZMS_DELTA_OFFSET_CODE_REBUILD_FREQ);
1301 lzms_init_huffman_encoder(&c->delta_power_encoder, &c->os,
1302 LZMS_NUM_DELTA_POWER_SYMS,
1303 LZMS_DELTA_POWER_CODE_REBUILD_FREQ);
1305 /* Initialize range encoders, all of which wrap around the same
1306 * lzms_range_encoder_raw. */
1307 lzms_init_range_encoder(&c->main_range_encoder,
1308 &c->rc, LZMS_NUM_MAIN_STATES);
1310 lzms_init_range_encoder(&c->match_range_encoder,
1311 &c->rc, LZMS_NUM_MATCH_STATES);
1313 lzms_init_range_encoder(&c->lz_match_range_encoder,
1314 &c->rc, LZMS_NUM_LZ_MATCH_STATES);
1316 for (unsigned i = 0; i < ARRAY_LEN(c->lz_repeat_match_range_encoders); i++)
1317 lzms_init_range_encoder(&c->lz_repeat_match_range_encoders[i],
1318 &c->rc, LZMS_NUM_LZ_REPEAT_MATCH_STATES);
1320 lzms_init_range_encoder(&c->delta_match_range_encoder,
1321 &c->rc, LZMS_NUM_DELTA_MATCH_STATES);
1323 for (unsigned i = 0; i < ARRAY_LEN(c->delta_repeat_match_range_encoders); i++)
1324 lzms_init_range_encoder(&c->delta_repeat_match_range_encoders[i],
1325 &c->rc, LZMS_NUM_DELTA_REPEAT_MATCH_STATES);
1327 /* Set initial length costs for lengths < LZMS_NUM_FAST_LENGTHS. */
1328 lzms_update_fast_length_costs(c);
1331 /* Flush the output streams, prepare the final compressed data, and return its
1334 * A return value of 0 indicates that the data could not be compressed to fit in
1335 * the available space. */
1337 lzms_finalize(struct lzms_compressor *c, u8 *cdata, size_t csize_avail)
1339 size_t num_forwards_bytes;
1340 size_t num_backwards_bytes;
1342 /* Flush both the forwards and backwards streams, and make sure they
1343 * didn't cross each other and start overwriting each other's data. */
1344 if (!lzms_output_bitstream_flush(&c->os))
1347 if (!lzms_range_encoder_raw_flush(&c->rc))
1350 if (c->rc.next > c->os.next)
1353 /* Now the compressed buffer contains the data output by the forwards
1354 * bitstream, then empty space, then data output by the backwards
1355 * bitstream. Move the data output by the backwards bitstream to be
1356 * adjacent to the data output by the forward bitstream, and calculate
1357 * the compressed size that this results in. */
1358 num_forwards_bytes = (u8*)c->rc.next - (u8*)cdata;
1359 num_backwards_bytes = ((u8*)cdata + csize_avail) - (u8*)c->os.next;
1361 memmove(cdata + num_forwards_bytes, c->os.next, num_backwards_bytes);
1363 return num_forwards_bytes + num_backwards_bytes;
1366 /* Set internal compression parameters for the specified compression level and
1367 * maximum window size. */
1369 lzms_build_params(unsigned int compression_level,
1370 struct lzms_compressor_params *params)
1372 /* Allow length 2 matches if the compression level is sufficiently high.
1374 if (compression_level >= 45)
1375 params->min_match_length = 2;
1377 params->min_match_length = 3;
1379 /* Scale nice_match_length and max_search_depth with the compression
1380 * level. But to allow an optimization on length cost calculations,
1381 * don't allow nice_match_length to exceed LZMS_NUM_FAST_LENGTH. */
1382 params->nice_match_length = ((u64)compression_level * 32) / 50;
1383 if (params->nice_match_length < params->min_match_length)
1384 params->nice_match_length = params->min_match_length;
1385 if (params->nice_match_length > LZMS_NUM_FAST_LENGTHS)
1386 params->nice_match_length = LZMS_NUM_FAST_LENGTHS;
1387 params->max_search_depth = compression_level;
1389 params->optim_array_length = 1024;
1392 /* Given the internal compression parameters and maximum window size, build the
1393 * Lempel-Ziv match-finder parameters. */
1395 lzms_build_mf_params(const struct lzms_compressor_params *lzms_params,
1396 u32 max_window_size, struct lz_mf_params *mf_params)
1398 memset(mf_params, 0, sizeof(*mf_params));
1400 /* Choose an appropriate match-finding algorithm. */
1401 if (max_window_size <= 2097152)
1402 mf_params->algorithm = LZ_MF_BINARY_TREES;
1403 else if (max_window_size <= 33554432)
1404 mf_params->algorithm = LZ_MF_LCP_INTERVAL_TREE;
1406 mf_params->algorithm = LZ_MF_LINKED_SUFFIX_ARRAY;
1408 mf_params->max_window_size = max_window_size;
1409 mf_params->min_match_len = lzms_params->min_match_length;
1410 mf_params->max_search_depth = lzms_params->max_search_depth;
1411 mf_params->nice_match_len = lzms_params->nice_match_length;
1415 lzms_free_compressor(void *_c);
1418 lzms_get_needed_memory(size_t max_block_size, unsigned int compression_level)
1420 struct lzms_compressor_params params;
1421 struct lz_mf_params mf_params;
1424 if (max_block_size >= INT32_MAX)
1427 lzms_build_params(compression_level, ¶ms);
1428 lzms_build_mf_params(¶ms, max_block_size, &mf_params);
1430 size += sizeof(struct lzms_compressor);
1433 size += max_block_size;
1436 size += lz_mf_get_needed_memory(mf_params.algorithm, max_block_size);
1439 size += min(params.max_search_depth, params.nice_match_length) *
1440 sizeof(struct lz_match);
1443 size += (params.optim_array_length + params.nice_match_length) *
1444 sizeof(struct lzms_mc_pos_data);
1450 lzms_create_compressor(size_t max_block_size, unsigned int compression_level,
1453 struct lzms_compressor *c;
1454 struct lzms_compressor_params params;
1455 struct lz_mf_params mf_params;
1457 if (max_block_size >= INT32_MAX)
1458 return WIMLIB_ERR_INVALID_PARAM;
1460 lzms_build_params(compression_level, ¶ms);
1461 lzms_build_mf_params(¶ms, max_block_size, &mf_params);
1462 if (!lz_mf_params_valid(&mf_params))
1463 return WIMLIB_ERR_INVALID_PARAM;
1465 c = CALLOC(1, sizeof(struct lzms_compressor));
1471 c->cur_window = MALLOC(max_block_size);
1475 c->mf = lz_mf_alloc(&mf_params);
1479 c->matches = MALLOC(min(params.max_search_depth,
1480 params.nice_match_length) *
1481 sizeof(struct lz_match));
1485 c->optimum = MALLOC((params.optim_array_length +
1486 params.nice_match_length) *
1487 sizeof(struct lzms_mc_pos_data));
1490 c->optimum_end = &c->optimum[params.optim_array_length];
1494 lzms_init_rc_costs();
1496 lzms_init_fast_slots(c);
1502 lzms_free_compressor(c);
1503 return WIMLIB_ERR_NOMEM;
1507 lzms_compress(const void *uncompressed_data, size_t uncompressed_size,
1508 void *compressed_data, size_t compressed_size_avail, void *_c)
1510 struct lzms_compressor *c = _c;
1512 /* Don't bother compressing extremely small inputs. */
1513 if (uncompressed_size < 4)
1516 /* Cap the available compressed size to a 32-bit integer and round it
1517 * down to the nearest multiple of 2. */
1518 if (compressed_size_avail > UINT32_MAX)
1519 compressed_size_avail = UINT32_MAX;
1520 if (compressed_size_avail & 1)
1521 compressed_size_avail--;
1523 /* Initialize the compressor structures. */
1524 lzms_prepare_compressor(c, uncompressed_data, uncompressed_size,
1525 compressed_data, compressed_size_avail / 2);
1527 /* Preprocess the uncompressed data. */
1528 lzms_x86_filter(c->cur_window, c->cur_window_size,
1529 c->last_target_usages, false);
1531 /* Load the window into the match-finder. */
1532 lz_mf_load_window(c->mf, c->cur_window, c->cur_window_size);
1534 /* Compute and encode a literal/match sequence that decompresses to the
1535 * preprocessed data. */
1536 lzms_near_optimal_parse(c);
1538 /* Return the compressed data size or 0. */
1539 return lzms_finalize(c, compressed_data, compressed_size_avail);
1543 lzms_free_compressor(void *_c)
1545 struct lzms_compressor *c = _c;
1548 FREE(c->cur_window);
1556 const struct compressor_ops lzms_compressor_ops = {
1557 .get_needed_memory = lzms_get_needed_memory,
1558 .create_compressor = lzms_create_compressor,
1559 .compress = lzms_compress,
1560 .free_compressor = lzms_free_compressor,