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 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/.
32 #define LZ_MIN_MATCH 3
35 #define HASH_SIZE (1 << HASH_BITS)
36 #define HASH_MASK (HASH_SIZE - 1)
40 #elif LZ_MIN_MATCH == 3
43 #error "Invalid LZ_MIN_MATCH"
46 /* Hash function, based on code from zlib. This function will update and return
47 * the hash value @hash for the string ending on the additional input character
48 * @c. This function must be called for each consecutive character, because it
49 * uses a running hash value rather than computing it separately for each
52 * The AND operation guarantees that only 3 characters will affect the hash
53 * value, so every identical 3-character string will have the same hash value.
55 static inline uint update_hash(uint hash, u8 c)
57 return ((hash << HASH_SHIFT) ^ c) & HASH_MASK;
61 /* Insert a 3-character string at position @str_pos in @window and with hash
62 * code @hash into the hash table described by @hash_tab and @prev_tab. Based
65 * The hash table uses chains (linked lists) for the hash buckets, but there are
66 * no real pointers involved. Indexing `hash_tab' by hash value gives the index
67 * within the window of the last string in the hash bucket. To find the index
68 * of the previous string in the hash chain, the `prev_tab' array is indexed by
69 * the string index. `prev_tab' can be indexed repeatedly by the string index
70 * to walk through the hash chain, until the special index `0' is reached,
71 * indicating the end of the hash chain.
73 static inline uint insert_string(u16 hash_tab[], u16 prev_tab[],
74 const u8 window[], uint str_pos, uint hash)
76 hash = update_hash(hash, window[str_pos + LZ_MIN_MATCH - 1]);
77 prev_tab[str_pos] = hash_tab[hash];
78 hash_tab[hash] = str_pos;
84 * Returns the longest match for a given input position.
86 * @window: The window of uncompressed data.
87 * @bytes_remaining: The number of bytes remaining in the window.
88 * @strstart: The index of the start of the string in the window that
89 * we are trying to find a match for.
90 * @prev_tab: The array of prev pointers for the hash table.
91 * @cur_match: The index of the head of the hash chain for matches
92 * having the hash value of the string beginning
94 * @prev_len: The length of the match that was found for the string
95 * beginning at (@strstart - 1).
96 * @match_start_ret: A location into which the index of the start of the
97 * match will be returned.
98 * @params: Parameters that affect how long the search will proceed
99 * before going with the best that has been found
102 * Returns the length of the match that was found.
104 static uint longest_match(const u8 window[], uint bytes_remaining,
105 uint strstart, const u16 prev_tab[],
106 uint cur_match, uint prev_len,
107 uint *match_start_ret,
108 const struct lz_params *params)
110 uint chain_len = params->max_chain_len;
112 const u8 *scan = window + strstart;
115 uint best_len = prev_len;
116 uint match_start = cur_match;
118 uint nice_match = min(params->nice_match, bytes_remaining);
120 const u8 *strend = scan + min(params->max_match, bytes_remaining);
122 u8 scan_end1 = scan[best_len - 1];
123 u8 scan_end = scan[best_len];
126 /* Do not waste too much time if we already have a good match: */
127 if (best_len >= params->good_match)
131 match = &window[cur_match];
133 /* Skip to next match if the match length cannot increase or if
134 * the match length is less than 2. Note that the checks below
135 * for insufficient lookahead only occur occasionally for
136 * performance reasons. Therefore uninitialized memory will be
137 * accessed, and conditional jumps will be made that depend on
138 * those values. However the length of the match is limited to
139 * the lookahead, so the output of deflate is not affected by
140 * the uninitialized values.
143 if (match[best_len] != scan_end
144 || match[best_len - 1] != scan_end1
146 || *++match != scan[1])
152 } while (scan < strend && *++match == *++scan);
157 *++match == *++scan && *++match == *++scan &&
158 *++match == *++scan && *++match == *++scan &&
159 *++match == *++scan && *++match == *++scan &&
160 *++match == *++scan && *++match == *++scan &&
163 len = match - &window[cur_match];
165 scan = &window[strstart];
167 if (len > best_len) {
168 match_start = cur_match;
170 if (len >= nice_match)
172 scan_end1 = scan[best_len - 1];
173 scan_end = scan[best_len];
175 } while (--chain_len != 0 && (cur_match = prev_tab[cur_match]) != 0);
176 *match_start_ret = match_start;
177 return min(min(best_len, bytes_remaining), params->max_match);
183 * Determines the sequence of matches and literals that a block of data will be
186 * @uncompressed_data: The data that is to be compressed.
187 * @uncompressed_len: The length of @uncompressed_data, in bytes.
188 * @match_tab: An array for the intermediate representation of matches.
189 * @record_match: A function that will be called to produce the
190 * intermediate representation of a match, given
191 * the offset and length. This function should also
192 * update the appropriate symbol frequency counts
193 * so that any needed Huffman codes can be made
195 * @record_literal: A function that will be called to produce the
196 * intermediate representation of a literal, given
197 * the character of the literal. This function
198 * should also update the appropriate symbol
199 * frequency counts so that any needed Huffman
200 * codes can be made later.
201 * @record_match_arg_1:
202 * @record_match_arg_2: Extra arguments to be passed to @record_match.
203 * @record_literal_arg: Extra arguments to be passed to @record_literal.
204 * @params: Structure that contains parameters that affect how the
205 * analysis proceeds (mainly how good the matches
208 * Returns the total number of matches and literal bytes that were found; this
209 * is the number of slots in @match_tab that have been filled with the
210 * intermediate representation of a match or literal byte.
212 uint lz_analyze_block(const u8 uncompressed_data[], uint uncompressed_len,
213 u32 match_tab[], lz_record_match_t record_match,
214 lz_record_literal_t record_literal, void *record_match_arg1,
215 void *record_match_arg2, void *record_literal_arg,
216 const struct lz_params *params)
218 uint cur_match_pos = 0;
219 uint cur_input_pos = 0;
222 uint prev_len = params->min_match - 1;
224 uint match_len = params->min_match - 1;
225 uint match_start = 0;
226 bool match_available = false;
227 u16 hash_tab[HASH_SIZE];
229 u16 prev_tab[uncompressed_len];
231 ZERO_ARRAY(hash_tab);
232 ZERO_ARRAY(prev_tab);
235 /* If there are at least 3 characters remaining in the input,
236 * insert the 3-character string beginning at
237 * uncompressed_data[cur_input_pos] into the hash table.
239 * hash_head is set to the index of the previous string in the
240 * hash bucket, or 0 if there is no such string */
241 if (uncompressed_len - cur_input_pos >= params->min_match) {
242 hash = insert_string(hash_tab, prev_tab,
244 cur_input_pos, hash);
245 hash_head = prev_tab[cur_input_pos];
251 /* Find the longest match, discarding those <= prev_len. */
252 prev_len = match_len;
253 prev_start = match_start;
254 match_len = params->min_match - 1;
256 if (hash_head != 0 && prev_len < params->max_lazy_match) {
257 /* To simplify the code, we prevent matches with the
258 * string of window index 0 (in particular we have to
259 * avoid a match of the string with itself at the start
260 * of the input file). */
261 match_len = longest_match(uncompressed_data,
262 uncompressed_len - cur_input_pos,
263 cur_input_pos, prev_tab,
265 &match_start, params);
267 if (match_len == params->min_match &&
268 cur_input_pos - match_start > params->too_far)
269 match_len = params->min_match - 1;
272 /* If there was a match at the previous step and the current
273 * match is not better, output the previous match:
275 if (prev_len >= params->min_match && match_len <= prev_len) {
277 /* Do not insert strings in hash table beyond this. */
278 uint max_insert = uncompressed_len - params->min_match;
280 /*DEBUG("Recording match (pos = %u, offset = %u, len = %u)\n",*/
281 /*cur_input_pos - 1, */
282 /*cur_input_pos - 1 - prev_start,*/
285 match = (*record_match)(cur_input_pos - 1 - prev_start,
290 match_tab[cur_match_pos++] = match;
292 /* Insert in hash table all strings up to the end of the match.
293 * strstart-1 and strstart are already inserted. If there is not
294 * enough lookahead, the last two strings are not inserted in
297 #if LZ_MIN_MATCH == 2
304 if (++cur_input_pos <= max_insert) {
305 hash = insert_string(hash_tab, prev_tab,
310 } while (--prev_len != 0);
312 match_available = false;
313 match_len = params->min_match - 1;
314 } else if (match_available) {
315 /* If there was no match at the previous position, output a
316 * single literal. If there was a match but the current match
317 * is longer, truncate the previous match to a single literal.
320 /*DEBUG("Recording litrl (pos = %u, value = %u)\n",*/
321 /*cur_input_pos - 1, */
322 /*uncompressed_data[cur_input_pos - 1]);*/
324 match = (*record_literal)(
325 uncompressed_data[cur_input_pos - 1],
327 match_tab[cur_match_pos++] = match;
329 /* There is no previous match to compare with, wait for
330 * the next step to decide. */
331 match_available = true;
333 } while (++cur_input_pos < uncompressed_len);
335 if (match_available) {
336 match = (*record_literal)(uncompressed_data[cur_input_pos - 1],
338 match_tab[cur_match_pos++] = match;
340 return cur_match_pos;