4 * A match-finder for Lempel-Ziv compression based on bottom-up construction and
5 * traversal of the Longest Common Prefix (LCP) interval tree.
7 * The following copying information applies to this specific source code file:
9 * Written in 2014-2015 by Eric Biggers <ebiggers3@gmail.com>
11 * To the extent possible under law, the author(s) have dedicated all copyright
12 * and related and neighboring rights to this software to the public domain
13 * worldwide via the Creative Commons Zero 1.0 Universal Public Domain
14 * Dedication (the "CC0").
16 * This software is distributed in the hope that it will be useful, but WITHOUT
17 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
18 * FOR A PARTICULAR PURPOSE. See the CC0 for more details.
20 * You should have received a copy of the CC0 along with this software; if not
21 * see <http://creativecommons.org/publicdomain/zero/1.0/>.
30 #include "wimlib/divsufsort.h"
31 #include "wimlib/lcpit_matchfinder.h"
32 #include "wimlib/util.h"
35 #define LCP_MAX (((u32)1 << LCP_BITS) - 1)
36 #define LCP_SHIFT (32 - LCP_BITS)
37 #define LCP_MASK (LCP_MAX << LCP_SHIFT)
38 #define POS_MASK (((u32)1 << (32 - LCP_BITS)) - 1)
39 #define MAX_NORMAL_BUFSIZE (POS_MASK + 1)
41 #define HUGE_LCP_BITS 7
42 #define HUGE_LCP_MAX (((u32)1 << HUGE_LCP_BITS) - 1)
43 #define HUGE_LCP_SHIFT (64 - HUGE_LCP_BITS)
44 #define HUGE_LCP_MASK ((u64)HUGE_LCP_MAX << HUGE_LCP_SHIFT)
45 #define HUGE_POS_MASK 0xFFFFFFFF
46 #define MAX_HUGE_BUFSIZE ((u64)HUGE_POS_MASK + 1)
47 #define HUGE_UNVISITED_TAG 0x100000000
49 #define PREFETCH_SAFETY 5
52 * Build the LCP (Longest Common Prefix) array in linear time.
54 * LCP[r] will be the length of the longest common prefix between the suffixes
55 * with positions SA[r - 1] and SA[r]. LCP[0] will be undefined.
57 * Algorithm taken from Kasai et al. (2001), but modified slightly:
59 * - With bytes there is no realistic way to reserve a unique symbol for
60 * end-of-buffer, so use explicit checks for end-of-buffer.
62 * - For decreased memory usage and improved memory locality, pack the two
63 * logically distinct SA and LCP arrays into a single array SA_and_LCP.
65 * - Since SA_and_LCP is accessed randomly, improve the cache behavior by
66 * reading several entries ahead in ISA and prefetching the upcoming
69 * - If an LCP value is less than the minimum match length, then store 0. This
70 * avoids having to do comparisons against the minimum match length later.
72 * - If an LCP value is greater than the "nice match length", then store the
73 * "nice match length". This caps the number of bits needed to store each
74 * LCP value, and this caps the depth of the LCP-interval tree, without
75 * usually hurting the compression ratio too much.
79 * Kasai et al. 2001. Linear-Time Longest-Common-Prefix Computation in
80 * Suffix Arrays and Its Applications. CPM '01 Proceedings of the 12th
81 * Annual Symposium on Combinatorial Pattern Matching pp. 181-192.
84 build_LCP(u32 SA_and_LCP[restrict], const u32 ISA[restrict],
85 const u8 T[restrict], const u32 n,
86 const u32 min_lcp, const u32 max_lcp)
89 for (u32 i = 0; i < n; i++) {
91 prefetchw(&SA_and_LCP[ISA[i + PREFETCH_SAFETY]]);
93 const u32 j = SA_and_LCP[r - 1] & POS_MASK;
94 const u32 lim = min(n - i, n - j);
95 while (h < lim && T[i + h] == T[j + h])
98 if (stored_lcp < min_lcp)
100 else if (stored_lcp > max_lcp)
101 stored_lcp = max_lcp;
102 SA_and_LCP[r] |= stored_lcp << LCP_SHIFT;
110 * Use the suffix array accompanied with the longest-common-prefix array ---
111 * which in combination can be called the "enhanced suffix array" --- to
112 * simulate a bottom-up traversal of the corresponding suffix tree, or
113 * equivalently the lcp-interval tree. Do so in suffix rank order, but save the
114 * superinterval references needed for later bottom-up traversal of the tree in
115 * suffix position order.
117 * To enumerate the lcp-intervals, this algorithm scans the suffix array and its
118 * corresponding LCP array linearly. While doing so, it maintains a stack of
119 * lcp-intervals that are currently open, meaning that their left boundaries
120 * have been seen but their right boundaries have not. The bottom of the stack
121 * is the interval which covers the entire suffix array (this has lcp=0), and
122 * the top of the stack is the deepest interval that is currently open (this has
123 * the greatest lcp of any interval on the stack). When this algorithm opens an
124 * lcp-interval, it assigns it a unique index in intervals[] and pushes it onto
125 * the stack. When this algorithm closes an interval, it pops it from the stack
126 * and sets the intervals[] entry of that interval to the index and lcp of that
127 * interval's superinterval, which is the new top of the stack.
129 * This algorithm also set pos_data[pos] for each suffix position 'pos' to the
130 * index and lcp of the deepest lcp-interval containing it. Alternatively, we
131 * can interpret each suffix as being associated with a singleton lcp-interval,
132 * or leaf of the suffix tree. With this interpretation, an entry in pos_data[]
133 * is the superinterval reference for one of these singleton lcp-intervals and
134 * therefore is not fundamentally different from an entry in intervals[].
136 * To reduce memory usage, this algorithm re-uses the suffix array's memory to
137 * store the generated intervals[] array. This is possible because SA and LCP
138 * are accessed linearly, and no more than one interval is generated per suffix.
140 * The techniques used in this algorithm are described in various published
141 * papers. The generation of lcp-intervals from the suffix array (SA) and the
142 * longest-common-prefix array (LCP) is given as Algorithm BottomUpTraverse in
143 * Kasai et al. (2001) and Algorithm 4.1 ("Computation of lcp-intervals") in
144 * Abouelhoda et al. (2004). Both these papers note the equivalence between
145 * lcp-intervals (including the singleton lcp-interval for each suffix) and
146 * nodes of the suffix tree. Abouelhoda et al. (2004) furthermore applies
147 * bottom-up traversal of the lcp-interval tree to Lempel-Ziv factorization, as
148 * does Chen at al. (2008). Algorithm CPS1b of Chen et al. (2008) dynamically
149 * re-uses the suffix array during bottom-up traversal of the lcp-interval tree.
153 * Kasai et al. Linear-Time Longest-Common-Prefix Computation in Suffix
154 * Arrays and Its Applications. 2001. CPM '01 Proceedings of the 12th
155 * Annual Symposium on Combinatorial Pattern Matching pp. 181-192.
157 * M.I. Abouelhoda, S. Kurtz, E. Ohlebusch. 2004. Replacing Suffix Trees
158 * With Enhanced Suffix Arrays. Journal of Discrete Algorithms Volume 2
159 * Issue 1, March 2004, pp. 53-86.
161 * G. Chen, S.J. Puglisi, W.F. Smyth. 2008. Lempel-Ziv Factorization
162 * Using Less Time & Space. Mathematics in Computer Science June 2008,
163 * Volume 1, Issue 4, pp. 605-623.
166 build_LCPIT(u32 intervals[restrict], u32 pos_data[restrict], const u32 n)
168 u32 * const SA_and_LCP = intervals;
169 u32 next_interval_idx;
170 u32 open_intervals[LCP_MAX + 1];
171 u32 *top = open_intervals;
172 u32 prev_pos = SA_and_LCP[0] & POS_MASK;
176 next_interval_idx = 1;
178 for (u32 r = 1; r < n; r++) {
179 const u32 next_pos = SA_and_LCP[r] & POS_MASK;
180 const u32 next_lcp = SA_and_LCP[r] & LCP_MASK;
181 const u32 top_lcp = *top & LCP_MASK;
183 prefetchw(&pos_data[SA_and_LCP[r + PREFETCH_SAFETY] & POS_MASK]);
185 if (next_lcp == top_lcp) {
186 /* Continuing the deepest open interval */
187 pos_data[prev_pos] = *top;
188 } else if (next_lcp > top_lcp) {
189 /* Opening a new interval */
190 *++top = next_lcp | next_interval_idx++;
191 pos_data[prev_pos] = *top;
193 /* Closing the deepest open interval */
194 pos_data[prev_pos] = *top;
196 const u32 closed_interval_idx = *top-- & POS_MASK;
197 const u32 superinterval_lcp = *top & LCP_MASK;
199 if (next_lcp == superinterval_lcp) {
200 /* Continuing the superinterval */
201 intervals[closed_interval_idx] = *top;
203 } else if (next_lcp > superinterval_lcp) {
204 /* Creating a new interval that is a
205 * superinterval of the one being
206 * closed, but still a subinterval of
207 * its superinterval */
208 *++top = next_lcp | next_interval_idx++;
209 intervals[closed_interval_idx] = *top;
212 /* Also closing the superinterval */
213 intervals[closed_interval_idx] = *top;
220 /* Close any still-open intervals. */
221 pos_data[prev_pos] = *top;
222 for (; top > open_intervals; top--)
223 intervals[*top & POS_MASK] = *(top - 1);
227 * Advance the LCP-interval tree matchfinder by one byte.
229 * If @record_matches is true, then matches are written to the @matches array
230 * sorted by strictly decreasing length and strictly decreasing offset, and the
231 * return value is the number of matches found. Otherwise, @matches is ignored
232 * and the return value is always 0.
234 * How this algorithm works:
236 * 'cur_pos' is the position of the current suffix, which is the suffix being
237 * matched against. 'cur_pos' starts at 0 and is incremented each time this
238 * function is called. This function finds each suffix with position less than
239 * 'cur_pos' that shares a prefix with the current suffix, but for each distinct
240 * prefix length it finds only the suffix with the greatest position (i.e. the
241 * most recently seen in the linear traversal by position). This function
242 * accomplishes this using the lcp-interval tree data structure that was built
243 * by build_LCPIT() and is updated dynamically by this function.
245 * The first step is to read 'pos_data[cur_pos]', which gives the index and lcp
246 * value of the deepest lcp-interval containing the current suffix --- or,
247 * equivalently, the parent of the conceptual singleton lcp-interval that
248 * contains the current suffix.
250 * The second step is to ascend the lcp-interval tree until reaching an interval
251 * that has not yet been visited, and link the intervals to the current suffix
252 * along the way. An lcp-interval has been visited if and only if it has been
253 * linked to a suffix. Initially, no lcp-intervals are linked to suffixes.
255 * The third step is to continue ascending the lcp-interval tree, but indirectly
256 * through suffix links rather than through the original superinterval
257 * references, and continuing to form links with the current suffix along the
258 * way. Each suffix visited during this step, except in a special case to
259 * handle outdated suffixes, is a match which can be written to matches[]. Each
260 * intervals[] entry contains the position of the next suffix to visit, which we
261 * shall call 'match_pos'; this is the most recently seen suffix that belongs to
262 * that lcp-interval. 'pos_data[match_pos]' then contains the lcp and interval
263 * index of the next lcp-interval that should be visited.
265 * We can view these arrays as describing a new set of links that gets overlaid
266 * on top of the original superinterval references of the lcp-interval tree.
267 * Each such link can connect a node of the lcp-interval tree to an ancestor
268 * more than one generation removed.
270 * For each one-byte advance, the current position becomes the most recently
271 * seen suffix for a continuous sequence of lcp-intervals from a leaf interval
272 * to the root interval. Conceptually, this algorithm needs to update all these
273 * nodes to link to 'cur_pos', and then update 'pos_data[cur_pos]' to a "null"
274 * link. But actually, if some of these nodes have the same most recently seen
275 * suffix, then this algorithm just visits the pos_data[] entry for that suffix
276 * and skips over all these nodes in one step. Updating the extra nodes is
277 * accomplished by creating a redirection from the previous suffix to the
280 * Using this shortcutting scheme, it's possible for a suffix to become out of
281 * date, which means that it is no longer the most recently seen suffix for the
282 * lcp-interval under consideration. This case is detected by noticing when the
283 * next lcp-interval link actually points deeper in the tree, and it is worked
284 * around by just continuing until we get to a link that actually takes us
285 * higher in the tree. This can be described as a lazy-update scheme.
287 static forceinline u32
288 lcpit_advance_one_byte(const u32 cur_pos,
289 u32 pos_data[restrict],
290 u32 intervals[restrict],
292 struct lz_match matches[restrict],
293 const bool record_matches)
298 struct lz_match *matchptr;
300 /* Get the deepest lcp-interval containing the current suffix. */
301 ref = pos_data[cur_pos];
303 /* Prefetch upcoming data, up to 3 positions ahead. Assume the
304 * intervals are already visited. */
306 /* Prefetch the superinterval via a suffix link for the deepest
307 * lcp-interval containing the suffix starting 1 position from now. */
308 prefetchw(&intervals[pos_data[next[0]] & POS_MASK]);
310 /* Prefetch suffix link for the deepest lcp-interval containing the
311 * suffix starting 2 positions from now. */
312 next[0] = intervals[next[1]] & POS_MASK;
313 prefetchw(&pos_data[next[0]]);
315 /* Prefetch the deepest lcp-interval containing the suffix starting 3
316 * positions from now. */
317 next[1] = pos_data[cur_pos + 3] & POS_MASK;
318 prefetchw(&intervals[next[1]]);
320 /* There is no "next suffix" after the current one. */
321 pos_data[cur_pos] = 0;
323 /* Ascend until we reach a visited interval, the root, or a child of the
324 * root. Link unvisited intervals to the current suffix as we go. */
325 while ((super_ref = intervals[ref & POS_MASK]) & LCP_MASK) {
326 intervals[ref & POS_MASK] = cur_pos;
330 if (super_ref == 0) {
331 /* In this case, the current interval may be any of:
333 * (2) an unvisited child of the root;
334 * (3) an interval last visited by suffix 0
336 * We could avoid the ambiguity with (3) by using an lcp
337 * placeholder value other than 0 to represent "visited", but
338 * it's fastest to use 0. So we just don't allow matches with
341 if (ref != 0) /* Not the root? */
342 intervals[ref & POS_MASK] = cur_pos;
346 /* Ascend indirectly via pos_data[] links. */
347 match_pos = super_ref;
350 while ((super_ref = pos_data[match_pos]) > ref)
351 match_pos = intervals[super_ref & POS_MASK];
352 intervals[ref & POS_MASK] = cur_pos;
353 pos_data[match_pos] = ref;
354 if (record_matches) {
355 matchptr->length = ref >> LCP_SHIFT;
356 matchptr->offset = cur_pos - match_pos;
362 match_pos = intervals[ref & POS_MASK];
364 return matchptr - matches;
367 /* Expand SA from 32 bits to 64 bits. */
369 expand_SA(void *p, u32 n)
371 typedef u32 _may_alias_attribute aliased_u32_t;
372 typedef u64 _may_alias_attribute aliased_u64_t;
374 aliased_u32_t *SA = p;
375 aliased_u64_t *SA64 = p;
383 /* Like build_LCP(), but for buffers larger than MAX_NORMAL_BUFSIZE. */
385 build_LCP_huge(u64 SA_and_LCP64[restrict], const u32 ISA[restrict],
386 const u8 T[restrict], const u32 n,
387 const u32 min_lcp, const u32 max_lcp)
390 for (u32 i = 0; i < n; i++) {
391 const u32 r = ISA[i];
392 prefetchw(&SA_and_LCP64[ISA[i + PREFETCH_SAFETY]]);
394 const u32 j = SA_and_LCP64[r - 1] & HUGE_POS_MASK;
395 const u32 lim = min(n - i, n - j);
396 while (h < lim && T[i + h] == T[j + h])
399 if (stored_lcp < min_lcp)
401 else if (stored_lcp > max_lcp)
402 stored_lcp = max_lcp;
403 SA_and_LCP64[r] |= (u64)stored_lcp << HUGE_LCP_SHIFT;
411 * Like build_LCPIT(), but for buffers larger than MAX_NORMAL_BUFSIZE.
413 * This "huge" version is also slightly different in that the lcp value stored
414 * in each intervals[] entry is the lcp value for that interval, not its
415 * superinterval. This lcp value stays put in intervals[] and doesn't get moved
416 * to pos_data[] during lcpit_advance_one_byte_huge(). One consequence of this
417 * is that we have to use a special flag to distinguish visited from unvisited
418 * intervals. But overall, this scheme keeps the memory usage at 12n instead of
419 * 16n. (The non-huge version is 8n.)
422 build_LCPIT_huge(u64 intervals64[restrict], u32 pos_data[restrict], const u32 n)
424 u64 * const SA_and_LCP64 = intervals64;
425 u32 next_interval_idx;
426 u32 open_intervals[HUGE_LCP_MAX + 1];
427 u32 *top = open_intervals;
428 u32 prev_pos = SA_and_LCP64[0] & HUGE_POS_MASK;
432 next_interval_idx = 1;
434 for (u32 r = 1; r < n; r++) {
435 const u32 next_pos = SA_and_LCP64[r] & HUGE_POS_MASK;
436 const u64 next_lcp = SA_and_LCP64[r] & HUGE_LCP_MASK;
437 const u64 top_lcp = intervals64[*top];
439 prefetchw(&pos_data[SA_and_LCP64[r + PREFETCH_SAFETY] & HUGE_POS_MASK]);
441 if (next_lcp == top_lcp) {
442 /* Continuing the deepest open interval */
443 pos_data[prev_pos] = *top;
444 } else if (next_lcp > top_lcp) {
445 /* Opening a new interval */
446 intervals64[next_interval_idx] = next_lcp;
447 pos_data[prev_pos] = next_interval_idx;
448 *++top = next_interval_idx++;
450 /* Closing the deepest open interval */
451 pos_data[prev_pos] = *top;
453 const u32 closed_interval_idx = *top--;
454 const u64 superinterval_lcp = intervals64[*top];
456 if (next_lcp == superinterval_lcp) {
457 /* Continuing the superinterval */
458 intervals64[closed_interval_idx] |=
459 HUGE_UNVISITED_TAG | *top;
461 } else if (next_lcp > superinterval_lcp) {
462 /* Creating a new interval that is a
463 * superinterval of the one being
464 * closed, but still a subinterval of
465 * its superinterval */
466 intervals64[next_interval_idx] = next_lcp;
467 intervals64[closed_interval_idx] |=
468 HUGE_UNVISITED_TAG | next_interval_idx;
469 *++top = next_interval_idx++;
472 /* Also closing the superinterval */
473 intervals64[closed_interval_idx] |=
474 HUGE_UNVISITED_TAG | *top;
481 /* Close any still-open intervals. */
482 pos_data[prev_pos] = *top;
483 for (; top > open_intervals; top--)
484 intervals64[*top] |= HUGE_UNVISITED_TAG | *(top - 1);
487 /* Like lcpit_advance_one_byte(), but for buffers larger than
488 * MAX_NORMAL_BUFSIZE. */
489 static forceinline u32
490 lcpit_advance_one_byte_huge(const u32 cur_pos,
491 u32 pos_data[restrict],
492 u64 intervals64[restrict],
493 u32 prefetch_next[restrict],
494 struct lz_match matches[restrict],
495 const bool record_matches)
498 u32 next_interval_idx;
502 struct lz_match *matchptr;
504 interval_idx = pos_data[cur_pos];
506 prefetchw(&intervals64[pos_data[prefetch_next[0]] & HUGE_POS_MASK]);
508 prefetch_next[0] = intervals64[prefetch_next[1]] & HUGE_POS_MASK;
509 prefetchw(&pos_data[prefetch_next[0]]);
511 prefetch_next[1] = pos_data[cur_pos + 3] & HUGE_POS_MASK;
512 prefetchw(&intervals64[prefetch_next[1]]);
514 pos_data[cur_pos] = 0;
516 while ((next = intervals64[interval_idx]) & HUGE_UNVISITED_TAG) {
517 intervals64[interval_idx] = (next & HUGE_LCP_MASK) | cur_pos;
518 interval_idx = next & HUGE_POS_MASK;
522 while (next & HUGE_LCP_MASK) {
525 match_pos = next & HUGE_POS_MASK;
526 next_interval_idx = pos_data[match_pos];
527 next = intervals64[next_interval_idx];
528 } while (next > cur);
529 intervals64[interval_idx] = (cur & HUGE_LCP_MASK) | cur_pos;
530 pos_data[match_pos] = interval_idx;
531 if (record_matches) {
532 matchptr->length = cur >> HUGE_LCP_SHIFT;
533 matchptr->offset = cur_pos - match_pos;
536 interval_idx = next_interval_idx;
538 return matchptr - matches;
541 static forceinline u64
542 get_pos_data_size(size_t max_bufsize)
544 return (u64)max((u64)max_bufsize + PREFETCH_SAFETY,
545 DIVSUFSORT_TMP_LEN) * sizeof(u32);
548 static forceinline u64
549 get_intervals_size(size_t max_bufsize)
551 return ((u64)max_bufsize + PREFETCH_SAFETY) *
552 (max_bufsize <= MAX_NORMAL_BUFSIZE ? sizeof(u32) : sizeof(u64));
556 * Calculate the number of bytes of memory needed for the LCP-interval tree
559 * @max_bufsize - maximum buffer size to support
561 * Returns the number of bytes required.
564 lcpit_matchfinder_get_needed_memory(size_t max_bufsize)
566 return get_pos_data_size(max_bufsize) + get_intervals_size(max_bufsize);
570 * Initialize the LCP-interval tree matchfinder.
572 * @mf - the matchfinder structure to initialize
573 * @max_bufsize - maximum buffer size to support
574 * @min_match_len - minimum match length in bytes
575 * @nice_match_len - only consider this many bytes of each match
577 * Returns true if successfully initialized; false if out of memory.
580 lcpit_matchfinder_init(struct lcpit_matchfinder *mf, size_t max_bufsize,
581 u32 min_match_len, u32 nice_match_len)
583 if (lcpit_matchfinder_get_needed_memory(max_bufsize) > SIZE_MAX)
585 if (max_bufsize > MAX_HUGE_BUFSIZE - PREFETCH_SAFETY)
588 mf->pos_data = MALLOC(get_pos_data_size(max_bufsize));
589 mf->intervals = MALLOC(get_intervals_size(max_bufsize));
590 if (!mf->pos_data || !mf->intervals) {
591 lcpit_matchfinder_destroy(mf);
595 mf->min_match_len = min_match_len;
596 mf->nice_match_len = min(nice_match_len,
597 (max_bufsize <= MAX_NORMAL_BUFSIZE) ?
598 LCP_MAX : HUGE_LCP_MAX);
603 * Build the suffix array SA for the specified byte array T of length n.
605 * The suffix array is a sorted array of the byte array's suffixes, represented
606 * by indices into the byte array. It can equivalently be viewed as a mapping
607 * from suffix rank to suffix position.
609 * To build the suffix array, we use libdivsufsort, which uses an
610 * induced-sorting-based algorithm. In practice, this seems to be the fastest
611 * suffix array construction algorithm currently available.
615 * Y. Mori. libdivsufsort, a lightweight suffix-sorting library.
616 * https://github.com/y-256/libdivsufsort
618 * G. Nong, S. Zhang, and W.H. Chan. 2009. Linear Suffix Array
619 * Construction by Almost Pure Induced-Sorting. Data Compression
620 * Conference, 2009. DCC '09. pp. 193 - 202.
622 * S.J. Puglisi, W.F. Smyth, and A. Turpin. 2007. A Taxonomy of Suffix
623 * Array Construction Algorithms. ACM Computing Surveys (CSUR) Volume 39
624 * Issue 2, 2007 Article No. 4.
627 build_SA(u32 SA[], const u8 T[], u32 n, u32 *tmp)
629 /* Note: divsufsort() requires a fixed amount of temporary space. The
630 * implementation of divsufsort() has been modified from the original to
631 * use the provided temporary space instead of allocating its own, since
632 * we don't want to have to deal with malloc() failures here. */
633 divsufsort(T, SA, n, tmp);
637 * Build the inverse suffix array ISA from the suffix array SA.
639 * Whereas the suffix array is a mapping from suffix rank to suffix position,
640 * the inverse suffix array is a mapping from suffix position to suffix rank.
643 build_ISA(u32 ISA[restrict], const u32 SA[restrict], u32 n)
645 for (u32 r = 0; r < n; r++)
650 * Prepare the LCP-interval tree matchfinder for a new input buffer.
652 * @mf - the initialized matchfinder structure
653 * @T - the input buffer
654 * @n - size of the input buffer in bytes. This must be nonzero and can be at
655 * most the max_bufsize with which lcpit_matchfinder_init() was called.
658 lcpit_matchfinder_load_buffer(struct lcpit_matchfinder *mf, const u8 *T, u32 n)
660 /* intervals[] temporarily stores SA and LCP packed together.
661 * pos_data[] temporarily stores ISA.
662 * pos_data[] is also used as the temporary space for divsufsort(). */
664 build_SA(mf->intervals, T, n, mf->pos_data);
665 build_ISA(mf->pos_data, mf->intervals, n);
666 if (n <= MAX_NORMAL_BUFSIZE) {
667 for (u32 i = 0; i < PREFETCH_SAFETY; i++) {
668 mf->intervals[n + i] = 0;
669 mf->pos_data[n + i] = 0;
671 build_LCP(mf->intervals, mf->pos_data, T, n,
672 mf->min_match_len, mf->nice_match_len);
673 build_LCPIT(mf->intervals, mf->pos_data, n);
674 mf->huge_mode = false;
676 for (u32 i = 0; i < PREFETCH_SAFETY; i++) {
677 mf->intervals64[n + i] = 0;
678 mf->pos_data[n + i] = 0;
680 expand_SA(mf->intervals, n);
681 build_LCP_huge(mf->intervals64, mf->pos_data, T, n,
682 mf->min_match_len, mf->nice_match_len);
683 build_LCPIT_huge(mf->intervals64, mf->pos_data, n);
684 mf->huge_mode = true;
686 mf->cur_pos = 0; /* starting at beginning of input buffer */
687 for (u32 i = 0; i < ARRAY_LEN(mf->next); i++)
692 * Retrieve a list of matches with the next position.
694 * The matches will be recorded in the @matches array, ordered by strictly
695 * decreasing length and strictly decreasing offset.
697 * The return value is the number of matches found and written to @matches.
698 * This can be any value in [0, nice_match_len - min_match_len + 1].
701 lcpit_matchfinder_get_matches(struct lcpit_matchfinder *mf,
702 struct lz_match *matches)
705 return lcpit_advance_one_byte_huge(mf->cur_pos++, mf->pos_data,
706 mf->intervals64, mf->next,
709 return lcpit_advance_one_byte(mf->cur_pos++, mf->pos_data,
710 mf->intervals, mf->next,
715 * Skip the next @count bytes (don't search for matches at them). @count is
719 lcpit_matchfinder_skip_bytes(struct lcpit_matchfinder *mf, u32 count)
723 lcpit_advance_one_byte_huge(mf->cur_pos++, mf->pos_data,
724 mf->intervals64, mf->next,
729 lcpit_advance_one_byte(mf->cur_pos++, mf->pos_data,
730 mf->intervals, mf->next,
737 * Destroy an LCP-interval tree matchfinder that was previously initialized with
738 * lcpit_matchfinder_init().
741 lcpit_matchfinder_destroy(struct lcpit_matchfinder *mf)