4 * This file provides the code to analyze a buffer of uncompressed data for
5 * matches, as per the LZ77 algorithm. It uses a hash table to accelerate the
6 * process. This is based on code from zlib (v. 1.2.5).
10 * Copyright (C) 2012, 2013 Eric Biggers
11 * Copyright (C) 1995-2010 Jean-loup Gailly and Mark Adler
13 * This file is part of wimlib, a library for working with WIM files.
15 * wimlib is free software; you can redistribute it and/or modify it under the
16 * terms of the GNU General Public License as published by the Free
17 * Software Foundation; either version 3 of the License, or (at your option)
20 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
21 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
22 * A PARTICULAR PURPOSE. See the GNU General Public License for more
25 * You should have received a copy of the GNU General Public License
26 * along with wimlib; if not, see http://www.gnu.org/licenses/.
33 #include "wimlib/compress.h"
34 #include "wimlib/util.h"
38 #define LZ_MIN_MATCH 3
41 #define HASH_SIZE (1 << HASH_BITS)
42 #define HASH_MASK (HASH_SIZE - 1)
46 #elif LZ_MIN_MATCH == 3
49 #error "Invalid LZ_MIN_MATCH"
52 /* Hash function, based on code from zlib. This function will update and return
53 * the hash value @hash for the string ending on the additional input character
54 * @c. This function must be called for each consecutive character, because it
55 * uses a running hash value rather than computing it separately for each
58 * The AND operation guarantees that only 3 characters will affect the hash
59 * value, so every identical 3-character string will have the same hash value.
61 static inline unsigned
62 update_hash(unsigned hash, u8 c)
64 return ((hash << HASH_SHIFT) ^ c) & HASH_MASK;
68 /* Insert a 3-character string at position @str_pos in @window and with hash
69 * code @hash into the hash table described by @hash_tab and @prev_tab. Based
72 * The hash table uses chains (linked lists) for the hash buckets, but there are
73 * no real pointers involved. Indexing `hash_tab' by hash value gives the index
74 * within the window of the last string in the hash bucket. To find the index
75 * of the previous string in the hash chain, the `prev_tab' array is indexed by
76 * the string index. `prev_tab' can be indexed repeatedly by the string index
77 * to walk through the hash chain, until the special index `0' is reached,
78 * indicating the end of the hash chain.
80 static inline unsigned
81 insert_string(u16 hash_tab[], u16 prev_tab[],
82 const u8 window[], unsigned str_pos,
85 hash = update_hash(hash, window[str_pos + LZ_MIN_MATCH - 1]);
86 prev_tab[str_pos] = hash_tab[hash];
87 hash_tab[hash] = str_pos;
93 * Returns the longest match for a given input position.
95 * @window: The window of uncompressed data.
96 * @bytes_remaining: The number of bytes remaining in the window.
97 * @strstart: The index of the start of the string in the window that
98 * we are trying to find a match for.
99 * @prev_tab: The array of prev pointers for the hash table.
100 * @cur_match: The index of the head of the hash chain for matches
101 * having the hash value of the string beginning
102 * at index @strstart.
103 * @prev_len: The length of the match that was found for the string
104 * beginning at (@strstart - 1).
105 * @match_start_ret: A location into which the index of the start of the
106 * match will be returned.
107 * @params: Parameters that affect how long the search will proceed
108 * before going with the best that has been found
111 * Returns the length of the match that was found.
114 longest_match(const u8 window[], unsigned bytes_remaining,
115 unsigned strstart, const u16 prev_tab[],
116 unsigned cur_match, unsigned prev_len,
117 unsigned *match_start_ret,
118 const struct lz_params *params)
120 unsigned chain_len = params->max_chain_len;
122 const u8 *scan = window + strstart;
125 unsigned best_len = prev_len;
126 unsigned match_start = cur_match;
128 unsigned nice_match = min(params->nice_match, bytes_remaining);
130 const u8 *strend = scan + min(params->max_match, bytes_remaining);
132 u8 scan_end1 = scan[best_len - 1];
133 u8 scan_end = scan[best_len];
136 /* Do not waste too much time if we already have a good match: */
137 if (best_len >= params->good_match)
141 match = &window[cur_match];
143 /* Skip to next match if the match length cannot increase or if
144 * the match length is less than 2. Note that the checks below
145 * for insufficient lookahead only occur occasionally for
146 * performance reasons. Therefore uninitialized memory will be
147 * accessed, and conditional jumps will be made that depend on
148 * those values. However the length of the match is limited to
149 * the lookahead, so the output of deflate is not affected by
150 * the uninitialized values.
153 if (match[best_len] != scan_end
154 || match[best_len - 1] != scan_end1
156 || *++match != scan[1])
162 } while (scan < strend && *++match == *++scan);
167 *++match == *++scan && *++match == *++scan &&
168 *++match == *++scan && *++match == *++scan &&
169 *++match == *++scan && *++match == *++scan &&
170 *++match == *++scan && *++match == *++scan &&
173 len = match - &window[cur_match];
175 scan = &window[strstart];
177 if (len > best_len) {
178 match_start = cur_match;
180 if (len >= nice_match)
182 scan_end1 = scan[best_len - 1];
183 scan_end = scan[best_len];
185 } while (--chain_len != 0 && (cur_match = prev_tab[cur_match]) != 0);
186 *match_start_ret = match_start;
187 return min(min(best_len, bytes_remaining), params->max_match);
193 * Determines the sequence of matches and literals that a block of data will be
196 * @uncompressed_data: The data that is to be compressed.
197 * @uncompressed_len: The length of @uncompressed_data, in bytes.
198 * @match_tab: An array for the intermediate representation of matches.
199 * @record_match: A function that will be called to produce the
200 * intermediate representation of a match, given
201 * the offset and length. This function should also
202 * update the appropriate symbol frequency counts
203 * so that any needed Huffman codes can be made
205 * @record_literal: A function that will be called to produce the
206 * intermediate representation of a literal, given
207 * the character of the literal. This function
208 * should also update the appropriate symbol
209 * frequency counts so that any needed Huffman
210 * codes can be made later.
211 * @record_match_arg_1:
212 * @record_match_arg_2: Extra arguments to be passed to @record_match.
213 * @record_literal_arg: Extra arguments to be passed to @record_literal.
214 * @params: Structure that contains parameters that affect how the
215 * analysis proceeds (mainly how good the matches
218 * Returns the total number of matches and literal bytes that were found; this
219 * is the number of slots in @match_tab that have been filled with the
220 * intermediate representation of a match or literal byte.
223 lz_analyze_block(const u8 uncompressed_data[],
224 unsigned uncompressed_len,
226 lz_record_match_t record_match,
227 lz_record_literal_t record_literal,
228 void *record_match_arg1,
229 void *record_match_arg2,
230 void *record_literal_arg,
231 const struct lz_params *params)
233 unsigned cur_match_pos = 0;
234 unsigned cur_input_pos = 0;
236 unsigned hash_head = 0;
237 unsigned prev_len = params->min_match - 1;
239 unsigned match_len = params->min_match - 1;
240 unsigned match_start = 0;
241 bool match_available = false;
242 u16 hash_tab[HASH_SIZE];
244 u16 prev_tab[uncompressed_len];
246 ZERO_ARRAY(hash_tab);
247 ZERO_ARRAY(prev_tab);
250 /* If there are at least 3 characters remaining in the input,
251 * insert the 3-character string beginning at
252 * uncompressed_data[cur_input_pos] into the hash table.
254 * hash_head is set to the index of the previous string in the
255 * hash bucket, or 0 if there is no such string */
256 if (uncompressed_len - cur_input_pos >= params->min_match) {
257 hash = insert_string(hash_tab, prev_tab,
259 cur_input_pos, hash);
260 hash_head = prev_tab[cur_input_pos];
266 /* Find the longest match, discarding those <= prev_len. */
267 prev_len = match_len;
268 prev_start = match_start;
269 match_len = params->min_match - 1;
271 if (hash_head != 0 && prev_len < params->max_lazy_match) {
272 /* To simplify the code, we prevent matches with the
273 * string of window index 0 (in particular we have to
274 * avoid a match of the string with itself at the start
275 * of the input file). */
276 match_len = longest_match(uncompressed_data,
277 uncompressed_len - cur_input_pos,
278 cur_input_pos, prev_tab,
280 &match_start, params);
282 if (match_len == params->min_match &&
283 cur_input_pos - match_start > params->too_far)
284 match_len = params->min_match - 1;
287 /* If there was a match at the previous step and the current
288 * match is not better, output the previous match:
290 if (prev_len >= params->min_match && match_len <= prev_len) {
292 /* Do not insert strings in hash table beyond this. */
293 unsigned max_insert = uncompressed_len - params->min_match;
295 /*DEBUG("Recording match (pos = %u, offset = %u, len = %u)\n",*/
296 /*cur_input_pos - 1, */
297 /*cur_input_pos - 1 - prev_start,*/
300 match = (*record_match)(cur_input_pos - 1 - prev_start,
305 match_tab[cur_match_pos++] = match;
307 /* Insert in hash table all strings up to the end of the match.
308 * strstart-1 and strstart are already inserted. If there is not
309 * enough lookahead, the last two strings are not inserted in
312 #if LZ_MIN_MATCH == 2
319 if (++cur_input_pos <= max_insert) {
320 hash = insert_string(hash_tab, prev_tab,
325 } while (--prev_len != 0);
327 match_available = false;
328 match_len = params->min_match - 1;
329 } else if (match_available) {
330 /* If there was no match at the previous position, output a
331 * single literal. If there was a match but the current match
332 * is longer, truncate the previous match to a single literal.
335 /*DEBUG("Recording litrl (pos = %u, value = %u)\n",*/
336 /*cur_input_pos - 1, */
337 /*uncompressed_data[cur_input_pos - 1]);*/
339 match = (*record_literal)(
340 uncompressed_data[cur_input_pos - 1],
342 match_tab[cur_match_pos++] = match;
344 /* There is no previous match to compare with, wait for
345 * the next step to decide. */
346 match_available = true;
348 } while (++cur_input_pos < uncompressed_len);
350 if (match_available) {
351 match = (*record_literal)(uncompressed_data[cur_input_pos - 1],
353 match_tab[cur_match_pos++] = match;
355 return cur_match_pos;