comparison zlib/deflate.c @ 3:5a977ccbc7a9 default tip

Empty changelog
author darius
date Sat, 06 Dec 1997 05:41:29 +0000
parents
children
comparison
equal deleted inserted replaced
2:fba0b6e6cdc7 3:5a977ccbc7a9
1 /* deflate.c -- compress data using the deflation algorithm
2 * Copyright (C) 1995 Jean-loup Gailly.
3 * For conditions of distribution and use, see copyright notice in zlib.h
4 */
5
6 /*
7 * ALGORITHM
8 *
9 * The "deflation" process depends on being able to identify portions
10 * of the input text which are identical to earlier input (within a
11 * sliding window trailing behind the input currently being processed).
12 *
13 * The most straightforward technique turns out to be the fastest for
14 * most input files: try all possible matches and select the longest.
15 * The key feature of this algorithm is that insertions into the string
16 * dictionary are very simple and thus fast, and deletions are avoided
17 * completely. Insertions are performed at each input character, whereas
18 * string matches are performed only when the previous match ends. So it
19 * is preferable to spend more time in matches to allow very fast string
20 * insertions and avoid deletions. The matching algorithm for small
21 * strings is inspired from that of Rabin & Karp. A brute force approach
22 * is used to find longer strings when a small match has been found.
23 * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze
24 * (by Leonid Broukhis).
25 * A previous version of this file used a more sophisticated algorithm
26 * (by Fiala and Greene) which is guaranteed to run in linear amortized
27 * time, but has a larger average cost, uses more memory and is patented.
28 * However the F&G algorithm may be faster for some highly redundant
29 * files if the parameter max_chain_length (described below) is too large.
30 *
31 * ACKNOWLEDGEMENTS
32 *
33 * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and
34 * I found it in 'freeze' written by Leonid Broukhis.
35 * Thanks to many people for bug reports and testing.
36 *
37 * REFERENCES
38 *
39 * Deutsch, L.P.,"'Deflate' Compressed Data Format Specification".
40 * Available in ftp.uu.net:/pub/archiving/zip/doc/deflate-1.1.doc
41 *
42 * A description of the Rabin and Karp algorithm is given in the book
43 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
44 *
45 * Fiala,E.R., and Greene,D.H.
46 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
47 *
48 */
49
50 /* $Id: deflate.c,v 1.1.1.1 1997/12/06 05:41:35 darius Exp $ */
51
52 #include "deflate.h"
53
54 char copyright[] = " deflate Copyright 1995 Jean-loup Gailly ";
55 /*
56 If you use the zlib library in a product, an acknowledgment is welcome
57 in the documentation of your product. If for some reason you cannot
58 include such an acknowledgment, I would appreciate that you keep this
59 copyright string in the executable of your product.
60 */
61
62 #define NIL 0
63 /* Tail of hash chains */
64
65 #ifndef TOO_FAR
66 # define TOO_FAR 4096
67 #endif
68 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
69
70 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
71 /* Minimum amount of lookahead, except at the end of the input file.
72 * See deflate.c for comments about the MIN_MATCH+1.
73 */
74
75 /* Values for max_lazy_match, good_match and max_chain_length, depending on
76 * the desired pack level (0..9). The values given below have been tuned to
77 * exclude worst case performance for pathological files. Better values may be
78 * found for specific files.
79 */
80
81 typedef struct config_s {
82 ush good_length; /* reduce lazy search above this match length */
83 ush max_lazy; /* do not perform lazy search above this match length */
84 ush nice_length; /* quit search above this match length */
85 ush max_chain;
86 } config;
87
88 local config configuration_table[10] = {
89 /* good lazy nice chain */
90 /* 0 */ {0, 0, 0, 0}, /* store only */
91 /* 1 */ {4, 4, 8, 4}, /* maximum speed, no lazy matches */
92 /* 2 */ {4, 5, 16, 8},
93 /* 3 */ {4, 6, 32, 32},
94
95 /* 4 */ {4, 4, 16, 16}, /* lazy matches */
96 /* 5 */ {8, 16, 32, 32},
97 /* 6 */ {8, 16, 128, 128},
98 /* 7 */ {8, 32, 128, 256},
99 /* 8 */ {32, 128, 258, 1024},
100 /* 9 */ {32, 258, 258, 4096}}; /* maximum compression */
101
102 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
103 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
104 * meaning.
105 */
106
107 #define EQUAL 0
108 /* result of memcmp for equal strings */
109
110 struct static_tree_desc_s {int dummy;}; /* for buggy compilers */
111
112 /* ===========================================================================
113 * Prototypes for local functions.
114 */
115
116 local void fill_window __P((deflate_state *s));
117 local int deflate_fast __P((deflate_state *s, int flush));
118 local int deflate_slow __P((deflate_state *s, int flush));
119 local void lm_init __P((deflate_state *s));
120 local int longest_match __P((deflate_state *s, IPos cur_match));
121 local void putShortMSB __P((deflate_state *s, uInt b));
122 local void flush_pending __P((z_stream *strm));
123 local int read_buf __P((z_stream *strm, char *buf, unsigned size));
124 #ifdef ASMV
125 void match_init __P((void)); /* asm code initialization */
126 #endif
127
128 #ifdef DEBUG
129 local void check_match __P((deflate_state *s, IPos start, IPos match,
130 int length));
131 #endif
132
133
134 /* ===========================================================================
135 * Update a hash value with the given input byte
136 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
137 * input characters, so that a running hash key can be computed from the
138 * previous key instead of complete recalculation each time.
139 */
140 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
141
142 /* ===========================================================================
143 * Insert string str in the dictionary and set match_head to the previous head
144 * of the hash chain (the most recent string with same hash key). Return
145 * the previous length of the hash chain.
146 * IN assertion: all calls to to INSERT_STRING are made with consecutive
147 * input characters and the first MIN_MATCH bytes of str are valid
148 * (except for the last MIN_MATCH-1 bytes of the input file).
149 */
150 #define INSERT_STRING(s, str, match_head) \
151 (UPDATE_HASH(s, s->ins_h, s->window[(str) + MIN_MATCH-1]), \
152 s->prev[(str) & s->w_mask] = match_head = s->head[s->ins_h], \
153 s->head[s->ins_h] = (str))
154
155 /* ===========================================================================
156 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
157 * prev[] will be initialized on the fly.
158 */
159 #define CLEAR_HASH(s) \
160 s->head[s->hash_size-1] = NIL; \
161 zmemzero((char*)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
162
163 /* ========================================================================= */
164 int deflateInit (strm, level)
165 z_stream *strm;
166 int level;
167 {
168 return deflateInit2 (strm, level, DEFLATED, MAX_WBITS, DEF_MEM_LEVEL, 0);
169 /* To do: ignore strm->next_in if we use it as window */
170 }
171
172 /* ========================================================================= */
173 int deflateInit2 (strm, level, method, windowBits, memLevel, strategy)
174 z_stream *strm;
175 int level;
176 int method;
177 int windowBits;
178 int memLevel;
179 int strategy;
180 {
181 deflate_state *s;
182 int noheader = 0;
183
184 if (strm == Z_NULL) return Z_STREAM_ERROR;
185
186 strm->msg = Z_NULL;
187 if (strm->zalloc == Z_NULL) strm->zalloc = zcalloc;
188 if (strm->zfree == Z_NULL) strm->zfree = zcfree;
189
190 if (level == Z_DEFAULT_COMPRESSION) level = 6;
191
192 if (windowBits < 0) { /* undocumented feature: suppress zlib header */
193 noheader = 1;
194 windowBits = -windowBits;
195 }
196 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != DEFLATED ||
197 windowBits < 8 || windowBits > 15 || level < 1 || level > 9) {
198 return Z_STREAM_ERROR;
199 }
200 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
201 if (s == Z_NULL) return Z_MEM_ERROR;
202 strm->state = (struct internal_state *)s;
203 s->strm = strm;
204
205 s->noheader = noheader;
206 s->w_bits = windowBits;
207 s->w_size = 1 << s->w_bits;
208 s->w_mask = s->w_size - 1;
209
210 s->hash_bits = memLevel + 7;
211 s->hash_size = 1 << s->hash_bits;
212 s->hash_mask = s->hash_size - 1;
213 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
214
215 s->window = (Byte*) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
216 s->prev = (Pos*) ZALLOC(strm, s->w_size, sizeof(Pos));
217 s->head = (Pos*) ZALLOC(strm, s->hash_size, sizeof(Pos));
218
219 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
220
221 s->pending_buf = (uch*) ZALLOC(strm, s->lit_bufsize, 2*sizeof(ush));
222
223 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
224 s->pending_buf == Z_NULL) {
225 strm->msg = z_errmsg[1-Z_MEM_ERROR];
226 deflateEnd (strm);
227 return Z_MEM_ERROR;
228 }
229 s->d_buf = (ush*) &(s->pending_buf[s->lit_bufsize]);
230 s->l_buf = (uch*) &(s->pending_buf[3*s->lit_bufsize]);
231 /* We overlay pending_buf and d_buf+l_buf. This works since the average
232 * output size for (length,distance) codes is <= 32 bits (worst case
233 * is 15+15+13=33).
234 */
235
236 s->level = level;
237 s->strategy = strategy;
238 s->method = (Byte)method;
239
240 return deflateReset(strm);
241 }
242
243 /* ========================================================================= */
244 int deflateReset (strm)
245 z_stream *strm;
246 {
247 deflate_state *s;
248
249 if (strm == Z_NULL || strm->state == Z_NULL ||
250 strm->zalloc == Z_NULL || strm->zfree == Z_NULL) return Z_STREAM_ERROR;
251
252 strm->total_in = strm->total_out = 0;
253 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
254 strm->data_type = Z_UNKNOWN;
255
256 s = (deflate_state *)strm->state;
257 s->pending = 0;
258 s->pending_out = s->pending_buf;
259
260 s->status = s->noheader ? BUSY_STATE : INIT_STATE;
261 s->adler = 1;
262
263 ct_init(s);
264 lm_init(s);
265
266 return Z_OK;
267 }
268
269 /* =========================================================================
270 * Put a short the pending_out buffer. The 16-bit value is put in MSB order.
271 * IN assertion: the stream state is correct and there is enough room in
272 * the pending_out buffer.
273 */
274 local void putShortMSB (s, b)
275 deflate_state *s;
276 uInt b;
277 {
278 put_byte(s, (Byte)(b >> 8));
279 put_byte(s, (Byte)(b & 0xff));
280 }
281
282 /* =========================================================================
283 * Flush as much pending output as possible.
284 */
285 local void flush_pending(strm)
286 z_stream *strm;
287 {
288 unsigned len = strm->state->pending;
289
290 if (len > strm->avail_out) len = strm->avail_out;
291 if (len == 0) return;
292
293 zmemcpy(strm->next_out, strm->state->pending_out, len);
294 strm->next_out += len;
295 strm->state->pending_out += len;
296 strm->total_out += len;
297 strm->avail_out -= len;
298 strm->state->pending -= len;
299 if (strm->state->pending == 0) {
300 strm->state->pending_out = strm->state->pending_buf;
301 }
302 }
303
304 /* ========================================================================= */
305 int deflate (strm, flush)
306 z_stream *strm;
307 int flush;
308 {
309 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
310
311 if (strm->next_out == Z_NULL || strm->next_in == Z_NULL) {
312 ERR_RETURN(strm, Z_STREAM_ERROR);
313 }
314 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
315
316 strm->state->strm = strm; /* just in case */
317
318 /* Write the zlib header */
319 if (strm->state->status == INIT_STATE) {
320
321 uInt header = (DEFLATED + ((strm->state->w_bits-8)<<4)) << 8;
322 uInt level_flags = (strm->state->level-1) >> 1;
323
324 if (level_flags > 3) level_flags = 3;
325 header |= (level_flags << 6);
326 header += 31 - (header % 31);
327
328 strm->state->status = BUSY_STATE;
329 putShortMSB(strm->state, header);
330 }
331
332 /* Flush as much pending output as possible */
333 if (strm->state->pending != 0) {
334 flush_pending(strm);
335 if (strm->avail_out == 0) return Z_OK;
336 }
337
338 /* User must not provide more input after the first FINISH: */
339 if (strm->state->status == FINISH_STATE && strm->avail_in != 0) {
340 ERR_RETURN(strm, Z_BUF_ERROR);
341 }
342
343 /* Start a new block or continue the current one.
344 */
345 if (strm->avail_in != 0 ||
346 (flush == Z_FINISH && strm->state->status != FINISH_STATE)) {
347 int quit;
348
349 if (flush == Z_FINISH) {
350 strm->state->status = FINISH_STATE;
351 }
352 if (strm->state->level <= 3) {
353 quit = deflate_fast(strm->state, flush);
354 } else {
355 quit = deflate_slow(strm->state, flush);
356 }
357 if (flush == Z_FULL_FLUSH) {
358 ct_stored_block(strm->state, (char*)0, 0L, 0); /* special marker */
359 flush_pending(strm);
360 CLEAR_HASH(strm->state); /* forget history */
361 if (strm->avail_out == 0) return Z_OK;
362 }
363 if (quit) return Z_OK;
364 }
365 Assert(strm->avail_out > 0, "bug2");
366
367 if (flush != Z_FINISH) return Z_OK;
368 if (strm->state->noheader) return Z_STREAM_END;
369
370 /* Write the zlib trailer (adler32) */
371 putShortMSB(strm->state, (uInt)(strm->state->adler >> 16));
372 putShortMSB(strm->state, (uInt)(strm->state->adler & 0xffff));
373 flush_pending(strm);
374 /* If avail_out is zero, the application will call deflate again
375 * to flush the rest.
376 */
377 strm->state->noheader = 1; /* write the trailer only once! */
378 return strm->state->pending != 0 ? Z_OK : Z_STREAM_END;
379 }
380
381 /* ========================================================================= */
382 int deflateEnd (strm)
383 z_stream *strm;
384 {
385 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
386
387 TRY_FREE(strm, strm->state->window);
388 TRY_FREE(strm, strm->state->prev);
389 TRY_FREE(strm, strm->state->head);
390 TRY_FREE(strm, strm->state->pending_buf);
391
392 ZFREE(strm, strm->state);
393 strm->state = Z_NULL;
394
395 return Z_OK;
396 }
397
398 /* ========================================================================= */
399 int deflateCopy (dest, source)
400 z_stream *dest;
401 z_stream *source;
402 {
403 if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) {
404 return Z_STREAM_ERROR;
405 }
406 *dest = *source;
407 return Z_STREAM_ERROR; /* to be implemented */
408 #if 0
409 dest->state = (struct internal_state *)
410 (*dest->zalloc)(1, sizeof(deflate_state));
411 if (dest->state == Z_NULL) return Z_MEM_ERROR;
412
413 *(dest->state) = *(source->state);
414 return Z_OK;
415 #endif
416 }
417
418 /* ===========================================================================
419 * Read a new buffer from the current input stream, update the adler32
420 * and total number of bytes read.
421 */
422 local int read_buf(strm, buf, size)
423 z_stream *strm;
424 char *buf;
425 unsigned size;
426 {
427 unsigned len = strm->avail_in;
428
429 if (len > size) len = size;
430 if (len == 0) return 0;
431
432 strm->avail_in -= len;
433
434 if (!strm->state->noheader) {
435 strm->state->adler = adler32(strm->state->adler, strm->next_in, len);
436 }
437 zmemcpy(buf, strm->next_in, len);
438 strm->next_in += len;
439 strm->total_in += len;
440
441 return (int)len;
442 }
443
444 /* ===========================================================================
445 * Initialize the "longest match" routines for a new zlib stream
446 */
447 local void lm_init (s)
448 deflate_state *s;
449 {
450 register unsigned j;
451
452 s->window_size = (ulg)2L*s->w_size;
453
454 CLEAR_HASH(s);
455
456 /* Set the default configuration parameters:
457 */
458 s->max_lazy_match = configuration_table[s->level].max_lazy;
459 s->good_match = configuration_table[s->level].good_length;
460 s->nice_match = configuration_table[s->level].nice_length;
461 s->max_chain_length = configuration_table[s->level].max_chain;
462
463 s->strstart = 0;
464 s->block_start = 0L;
465 s->lookahead = 0;
466 s->match_length = MIN_MATCH-1;
467 s->match_available = 0;
468 #ifdef ASMV
469 match_init(); /* initialize the asm code */
470 #endif
471
472 s->ins_h = 0;
473 for (j=0; j<MIN_MATCH-1; j++) UPDATE_HASH(s, s->ins_h, s->window[j]);
474 /* If lookahead < MIN_MATCH, ins_h is garbage, but this is
475 * not important since only literal bytes will be emitted.
476 */
477 }
478
479 /* ===========================================================================
480 * Set match_start to the longest match starting at the given string and
481 * return its length. Matches shorter or equal to prev_length are discarded,
482 * in which case the result is equal to prev_length and match_start is
483 * garbage.
484 * IN assertions: cur_match is the head of the hash chain for the current
485 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
486 */
487 #ifndef ASMV
488 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
489 * match.S. The code will be functionally equivalent.
490 */
491 local INLINE int longest_match(s, cur_match)
492 deflate_state *s;
493 IPos cur_match; /* current match */
494 {
495 unsigned chain_length = s->max_chain_length;/* max hash chain length */
496 register Byte *scan = s->window + s->strstart; /* current string */
497 register Byte *match; /* matched string */
498 register int len; /* length of current match */
499 int best_len = s->prev_length; /* best match length so far */
500 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
501 s->strstart - (IPos)MAX_DIST(s) : NIL;
502 /* Stop when cur_match becomes <= limit. To simplify the code,
503 * we prevent matches with the string of window index 0.
504 */
505 Pos *prev = s->prev;
506 uInt wmask = s->w_mask;
507
508 #ifdef UNALIGNED_OK
509 /* Compare two bytes at a time. Note: this is not always beneficial.
510 * Try with and without -DUNALIGNED_OK to check.
511 */
512 register Byte *strend = s->window + s->strstart + MAX_MATCH - 1;
513 register ush scan_start = *(ush*)scan;
514 register ush scan_end = *(ush*)(scan+best_len-1);
515 #else
516 register Byte *strend = s->window + s->strstart + MAX_MATCH;
517 register Byte scan_end1 = scan[best_len-1];
518 register Byte scan_end = scan[best_len];
519 #endif
520
521 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
522 * It is easy to get rid of this optimization if necessary.
523 */
524 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
525
526 /* Do not waste too much time if we already have a good match: */
527 if (s->prev_length >= s->good_match) {
528 chain_length >>= 2;
529 }
530 Assert(s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
531
532 do {
533 Assert(cur_match < s->strstart, "no future");
534 match = s->window + cur_match;
535
536 /* Skip to next match if the match length cannot increase
537 * or if the match length is less than 2:
538 */
539 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
540 /* This code assumes sizeof(unsigned short) == 2. Do not use
541 * UNALIGNED_OK if your compiler uses a different size.
542 */
543 if (*(ush*)(match+best_len-1) != scan_end ||
544 *(ush*)match != scan_start) continue;
545
546 /* It is not necessary to compare scan[2] and match[2] since they are
547 * always equal when the other bytes match, given that the hash keys
548 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
549 * strstart+3, +5, ... up to strstart+257. We check for insufficient
550 * lookahead only every 4th comparison; the 128th check will be made
551 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
552 * necessary to put more guard bytes at the end of the window, or
553 * to check more often for insufficient lookahead.
554 */
555 scan++, match++;
556 do {
557 } while (*(ush*)(scan+=2) == *(ush*)(match+=2) &&
558 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
559 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
560 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
561 scan < strend);
562 /* The funny "do {}" generates better code on most compilers */
563
564 /* Here, scan <= window+strstart+257 */
565 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
566 if (*scan == *match) scan++;
567
568 len = (MAX_MATCH - 1) - (int)(strend-scan);
569 scan = strend - (MAX_MATCH-1);
570
571 #else /* UNALIGNED_OK */
572
573 if (match[best_len] != scan_end ||
574 match[best_len-1] != scan_end1 ||
575 *match != *scan ||
576 *++match != scan[1]) continue;
577
578 /* The check at best_len-1 can be removed because it will be made
579 * again later. (This heuristic is not always a win.)
580 * It is not necessary to compare scan[2] and match[2] since they
581 * are always equal when the other bytes match, given that
582 * the hash keys are equal and that HASH_BITS >= 8.
583 */
584 scan += 2, match++;
585
586 /* We check for insufficient lookahead only every 8th comparison;
587 * the 256th check will be made at strstart+258.
588 */
589 do {
590 } while (*++scan == *++match && *++scan == *++match &&
591 *++scan == *++match && *++scan == *++match &&
592 *++scan == *++match && *++scan == *++match &&
593 *++scan == *++match && *++scan == *++match &&
594 scan < strend);
595
596 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
597
598 len = MAX_MATCH - (int)(strend - scan);
599 scan = strend - MAX_MATCH;
600
601 #endif /* UNALIGNED_OK */
602
603 if (len > best_len) {
604 s->match_start = cur_match;
605 best_len = len;
606 if (len >= s->nice_match) break;
607 #ifdef UNALIGNED_OK
608 scan_end = *(ush*)(scan+best_len-1);
609 #else
610 scan_end1 = scan[best_len-1];
611 scan_end = scan[best_len];
612 #endif
613 }
614 } while ((cur_match = prev[cur_match & wmask]) > limit
615 && --chain_length != 0);
616
617 return best_len;
618 }
619 #endif /* ASMV */
620
621 #ifdef DEBUG
622 /* ===========================================================================
623 * Check that the match at match_start is indeed a match.
624 */
625 local void check_match(s, start, match, length)
626 deflate_state *s;
627 IPos start, match;
628 int length;
629 {
630 /* check that the match is indeed a match */
631 if (memcmp((char*)s->window + match,
632 (char*)s->window + start, length) != EQUAL) {
633 fprintf(stderr,
634 " start %d, match %d, length %d\n",
635 start, match, length);
636 z_error("invalid match");
637 }
638 if (verbose > 1) {
639 fprintf(stderr,"\\[%d,%d]", start-match, length);
640 do { putc(s->window[start++], stderr); } while (--length != 0);
641 }
642 }
643 #else
644 # define check_match(s, start, match, length)
645 #endif
646
647 /* ===========================================================================
648 * Fill the window when the lookahead becomes insufficient.
649 * Updates strstart and lookahead.
650 *
651 * IN assertion: lookahead < MIN_LOOKAHEAD
652 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
653 * At least one byte has been read, or avail_in == 0; reads are
654 * performed for at least two bytes (required for the zip translate_eol
655 * option -- not supported here).
656 */
657 local void fill_window(s)
658 deflate_state *s;
659 {
660 register unsigned n, m;
661 register Pos *p;
662 unsigned more; /* Amount of free space at the end of the window. */
663 uInt wsize = s->w_size;
664
665 do {
666 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
667
668 /* Deal with !@#$% 64K limit: */
669 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
670 more = wsize;
671 } else if (more == (unsigned)(-1)) {
672 /* Very unlikely, but possible on 16 bit machine if strstart == 0
673 * and lookahead == 1 (input done one byte at time)
674 */
675 more--;
676
677 /* If the window is almost full and there is insufficient lookahead,
678 * move the upper half to the lower one to make room in the upper half.
679 */
680 } else if (s->strstart >= wsize+MAX_DIST(s)) {
681
682 /* By the IN assertion, the window is not empty so we can't confuse
683 * more == 0 with more == 64K on a 16 bit machine.
684 */
685 zmemcpy((char*)s->window, (char*)s->window+wsize,
686 (unsigned)wsize);
687 s->match_start -= wsize;
688 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
689
690 s->block_start -= (long) wsize;
691
692 /* Slide the hash table (could be avoided with 32 bit values
693 at the expense of memory usage):
694 */
695 n = s->hash_size;
696 p = &s->head[n-1];
697 do {
698 m = *p;
699 *p-- = (Pos)(m >= wsize ? m-wsize : NIL);
700 } while (--n);
701
702 n = wsize;
703 p = &s->prev[n-1];
704 do {
705 m = *p;
706 *p-- = (Pos)(m >= wsize ? m-wsize : NIL);
707 /* If n is not on any hash chain, prev[n] is garbage but
708 * its value will never be used.
709 */
710 } while (--n);
711
712 more += wsize;
713 }
714 if (s->strm->avail_in == 0) return;
715
716 /* If there was no sliding:
717 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
718 * more == window_size - lookahead - strstart
719 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
720 * => more >= window_size - 2*WSIZE + 2
721 * In the BIG_MEM or MMAP case (not yet supported),
722 * window_size == input_size + MIN_LOOKAHEAD &&
723 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
724 * Otherwise, window_size == 2*WSIZE so more >= 2.
725 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
726 */
727 Assert(more >= 2, "more < 2");
728
729 n = read_buf(s->strm, (char*)s->window + s->strstart + s->lookahead,
730 more);
731 s->lookahead += n;
732
733 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
734 }
735
736 /* ===========================================================================
737 * Flush the current block, with given end-of-file flag.
738 * IN assertion: strstart is set to the end of the current match.
739 */
740 #define FLUSH_BLOCK_ONLY(s, eof) { \
741 ct_flush_block(s, (s->block_start >= 0L ? \
742 (char*)&s->window[(unsigned)s->block_start] : \
743 (char*)Z_NULL), (long)s->strstart - s->block_start, (eof)); \
744 s->block_start = s->strstart; \
745 flush_pending(s->strm); \
746 }
747
748 /* Same but force premature exit if necessary. */
749 #define FLUSH_BLOCK(s, eof) { \
750 FLUSH_BLOCK_ONLY(s, eof); \
751 if (s->strm->avail_out == 0) return 1; \
752 }
753
754 /* ===========================================================================
755 * Compress as much as possible from the input stream, return true if
756 * processing was terminated prematurely (no more input or output space).
757 * This function does not perform lazy evaluationof matches and inserts
758 * new strings in the dictionary only for unmatched strings or for short
759 * matches. It is used only for the fast compression options.
760 */
761 local int deflate_fast(s, flush)
762 deflate_state *s;
763 int flush;
764 {
765 IPos hash_head; /* head of the hash chain */
766 int bflush; /* set if current block must be flushed */
767
768 s->prev_length = MIN_MATCH-1;
769
770 for (;;) {
771 /* Make sure that we always have enough lookahead, except
772 * at the end of the input file. We need MAX_MATCH bytes
773 * for the next match, plus MIN_MATCH bytes to insert the
774 * string following the next match.
775 */
776 if (s->lookahead < MIN_LOOKAHEAD) {
777 fill_window(s);
778 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) return 1;
779
780 if (s->lookahead == 0) break; /* flush the current block */
781 }
782
783 /* Insert the string window[strstart .. strstart+2] in the
784 * dictionary, and set hash_head to the head of the hash chain:
785 */
786 INSERT_STRING(s, s->strstart, hash_head);
787
788 /* Find the longest match, discarding those <= prev_length.
789 * At this point we have always match_length < MIN_MATCH
790 */
791 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
792 /* To simplify the code, we prevent matches with the string
793 * of window index 0 (in particular we have to avoid a match
794 * of the string with itself at the start of the input file).
795 */
796 if (s->strategy != Z_HUFFMAN_ONLY) {
797 s->match_length = longest_match (s, hash_head);
798 }
799 /* longest_match() sets match_start */
800
801 if (s->match_length > s->lookahead) s->match_length = s->lookahead;
802 }
803 if (s->match_length >= MIN_MATCH) {
804 check_match(s, s->strstart, s->match_start, s->match_length);
805
806 bflush = ct_tally(s, s->strstart - s->match_start,
807 s->match_length - MIN_MATCH);
808
809 s->lookahead -= s->match_length;
810
811 /* Insert new strings in the hash table only if the match length
812 * is not too large. This saves time but degrades compression.
813 */
814 if (s->match_length <= s->max_insert_length) {
815 s->match_length--; /* string at strstart already in hash table */
816 do {
817 s->strstart++;
818 INSERT_STRING(s, s->strstart, hash_head);
819 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
820 * always MIN_MATCH bytes ahead. If lookahead < MIN_MATCH
821 * these bytes are garbage, but it does not matter since
822 * the next lookahead bytes will be emitted as literals.
823 */
824 } while (--s->match_length != 0);
825 s->strstart++;
826 } else {
827 s->strstart += s->match_length;
828 s->match_length = 0;
829 s->ins_h = s->window[s->strstart];
830 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
831 #if MIN_MATCH != 3
832 Call UPDATE_HASH() MIN_MATCH-3 more times
833 #endif
834 }
835 } else {
836 /* No match, output a literal byte */
837 Tracevv((stderr,"%c", s->window[s->strstart]));
838 bflush = ct_tally (s, 0, s->window[s->strstart]);
839 s->lookahead--;
840 s->strstart++;
841 }
842 if (bflush) FLUSH_BLOCK(s, 0);
843 }
844 FLUSH_BLOCK(s, flush == Z_FINISH);
845 return 0; /* normal exit */
846 }
847
848 /* ===========================================================================
849 * Same as above, but achieves better compression. We use a lazy
850 * evaluation for matches: a match is finally adopted only if there is
851 * no better match at the next window position.
852 */
853 local int deflate_slow(s, flush)
854 deflate_state *s;
855 int flush;
856 {
857 IPos hash_head; /* head of hash chain */
858 int bflush; /* set if current block must be flushed */
859
860 /* Process the input block. */
861 for (;;) {
862 /* Make sure that we always have enough lookahead, except
863 * at the end of the input file. We need MAX_MATCH bytes
864 * for the next match, plus MIN_MATCH bytes to insert the
865 * string following the next match.
866 */
867 if (s->lookahead < MIN_LOOKAHEAD) {
868 fill_window(s);
869 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) return 1;
870
871 if (s->lookahead == 0) break; /* flush the current block */
872 }
873
874 /* Insert the string window[strstart .. strstart+2] in the
875 * dictionary, and set hash_head to the head of the hash chain:
876 */
877 INSERT_STRING(s, s->strstart, hash_head);
878
879 /* Find the longest match, discarding those <= prev_length.
880 */
881 s->prev_length = s->match_length, s->prev_match = s->match_start;
882 s->match_length = MIN_MATCH-1;
883
884 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
885 s->strstart - hash_head <= MAX_DIST(s)) {
886 /* To simplify the code, we prevent matches with the string
887 * of window index 0 (in particular we have to avoid a match
888 * of the string with itself at the start of the input file).
889 */
890 if (s->strategy != Z_HUFFMAN_ONLY) {
891 s->match_length = longest_match (s, hash_head);
892 }
893 /* longest_match() sets match_start */
894 if (s->match_length > s->lookahead) s->match_length = s->lookahead;
895
896 if (s->match_length <= 5 && (s->strategy == Z_FILTERED ||
897 (s->match_length == MIN_MATCH &&
898 s->strstart - s->match_start > TOO_FAR))) {
899
900 /* If prev_match is also MIN_MATCH, match_start is garbage
901 * but we will ignore the current match anyway.
902 */
903 s->match_length = MIN_MATCH-1;
904 }
905 }
906 /* If there was a match at the previous step and the current
907 * match is not better, output the previous match:
908 */
909 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
910
911 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
912
913 bflush = ct_tally(s, s->strstart -1 - s->prev_match,
914 s->prev_length - MIN_MATCH);
915
916 /* Insert in hash table all strings up to the end of the match.
917 * strstart-1 and strstart are already inserted.
918 */
919 s->lookahead -= s->prev_length-1;
920 s->prev_length -= 2;
921 do {
922 s->strstart++;
923 INSERT_STRING(s, s->strstart, hash_head);
924 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
925 * always MIN_MATCH bytes ahead. If lookahead < MIN_MATCH
926 * these bytes are garbage, but it does not matter since the
927 * next lookahead bytes will always be emitted as literals.
928 */
929 } while (--s->prev_length != 0);
930 s->match_available = 0;
931 s->match_length = MIN_MATCH-1;
932 s->strstart++;
933
934 if (bflush) FLUSH_BLOCK(s, 0);
935
936 } else if (s->match_available) {
937 /* If there was no match at the previous position, output a
938 * single literal. If there was a match but the current match
939 * is longer, truncate the previous match to a single literal.
940 */
941 Tracevv((stderr,"%c", s->window[s->strstart-1]));
942 if (ct_tally (s, 0, s->window[s->strstart-1])) {
943 FLUSH_BLOCK_ONLY(s, 0);
944 }
945 s->strstart++;
946 s->lookahead--;
947 if (s->strm->avail_out == 0) return 1;
948 } else {
949 /* There is no previous match to compare with, wait for
950 * the next step to decide.
951 */
952 s->match_available = 1;
953 s->strstart++;
954 s->lookahead--;
955 }
956 }
957 if (s->match_available) {
958 ct_tally (s, 0, s->window[s->strstart-1]);
959 s->match_available = 0;
960 }
961 FLUSH_BLOCK(s, flush == Z_FINISH);
962 return 0;
963 }