1 /* deflate.c -- compress data using the deflation algorithm
2 * Copyright (C) 1995-1996 Jean-loup Gailly.
3 * For conditions of distribution and use, see copyright notice in zlib.h
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).
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.
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.
39 * Deutsch, L.P.,"'Deflate' Compressed Data Format Specification".
40 * Available in ftp.uu.net:/pub/archiving/zip/doc/deflate-1.1.doc
42 * A description of the Rabin and Karp algorithm is given in the book
43 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
45 * Fiala,E.R., and Greene,D.H.
46 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
50 /* $Id: deflate.c,v 1.15 1996/07/24 13:40:58 me Exp $ */
54 char deflate_copyright[] = " deflate 1.0.4 Copyright 1995-1996 Jean-loup Gailly ";
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.
62 /* ===========================================================================
63 * Function prototypes.
66 need_more, /* block not completed, need more input or more output */
67 block_done, /* block flush performed */
68 finish_started, /* finish started, need only more output at next deflate */
69 finish_done /* finish done, accept no more input or output */
72 typedef block_state (*compress_func) OF((deflate_state *s, int flush));
73 /* Compression function. Returns the block state after the call. */
75 local void fill_window OF((deflate_state *s));
76 local block_state deflate_stored OF((deflate_state *s, int flush));
77 local block_state deflate_fast OF((deflate_state *s, int flush));
78 local block_state deflate_slow OF((deflate_state *s, int flush));
79 local void lm_init OF((deflate_state *s));
80 local uInt longest_match OF((deflate_state *s, IPos cur_match));
81 local void putShortMSB OF((deflate_state *s, uInt b));
82 local void flush_pending OF((z_streamp strm));
83 local int read_buf OF((z_streamp strm, charf *buf, unsigned size));
85 void match_init OF((void)); /* asm code initialization */
89 local void check_match OF((deflate_state *s, IPos start, IPos match,
93 /* ===========================================================================
98 /* Tail of hash chains */
101 # define TOO_FAR 4096
103 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
105 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
106 /* Minimum amount of lookahead, except at the end of the input file.
107 * See deflate.c for comments about the MIN_MATCH+1.
110 /* Values for max_lazy_match, good_match and max_chain_length, depending on
111 * the desired pack level (0..9). The values given below have been tuned to
112 * exclude worst case performance for pathological files. Better values may be
113 * found for specific files.
115 typedef struct config_s {
116 ush good_length; /* reduce lazy search above this match length */
117 ush max_lazy; /* do not perform lazy search above this match length */
118 ush nice_length; /* quit search above this match length */
123 local config configuration_table[10] = {
124 /* good lazy nice chain */
125 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
126 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* maximum speed, no lazy matches */
127 /* 2 */ {4, 5, 16, 8, deflate_fast},
128 /* 3 */ {4, 6, 32, 32, deflate_fast},
130 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
131 /* 5 */ {8, 16, 32, 32, deflate_slow},
132 /* 6 */ {8, 16, 128, 128, deflate_slow},
133 /* 7 */ {8, 32, 128, 256, deflate_slow},
134 /* 8 */ {32, 128, 258, 1024, deflate_slow},
135 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* maximum compression */
137 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
138 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
143 /* result of memcmp for equal strings */
145 struct static_tree_desc_s {int dummy;}; /* for buggy compilers */
147 /* ===========================================================================
148 * Update a hash value with the given input byte
149 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
150 * input characters, so that a running hash key can be computed from the
151 * previous key instead of complete recalculation each time.
153 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
156 /* ===========================================================================
157 * Insert string str in the dictionary and set match_head to the previous head
158 * of the hash chain (the most recent string with same hash key). Return
159 * the previous length of the hash chain.
160 * IN assertion: all calls to to INSERT_STRING are made with consecutive
161 * input characters and the first MIN_MATCH bytes of str are valid
162 * (except for the last MIN_MATCH-1 bytes of the input file).
164 #define INSERT_STRING(s, str, match_head) \
165 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
166 s->prev[(str) & s->w_mask] = match_head = s->head[s->ins_h], \
167 s->head[s->ins_h] = (Pos)(str))
169 /* ===========================================================================
170 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
171 * prev[] will be initialized on the fly.
173 #define CLEAR_HASH(s) \
174 s->head[s->hash_size-1] = NIL; \
175 zmemzero((charf *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
177 /* ========================================================================= */
178 int deflateInit_(strm, level, version, stream_size)
184 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
185 Z_DEFAULT_STRATEGY, version, stream_size);
186 /* To do: ignore strm->next_in if we use it as window */
189 /* ========================================================================= */
190 int deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
191 version, stream_size)
205 /* We overlay pending_buf and d_buf+l_buf. This works since the average
206 * output size for (length,distance) codes is <= 24 bits.
209 if (version == Z_NULL || version[0] != ZLIB_VERSION[0] ||
210 stream_size != sizeof(z_stream)) {
211 return Z_VERSION_ERROR;
213 if (strm == Z_NULL) return Z_STREAM_ERROR;
216 if (strm->zalloc == Z_NULL) {
217 strm->zalloc = zcalloc;
218 strm->opaque = (voidpf)0;
220 if (strm->zfree == Z_NULL) strm->zfree = zcfree;
222 if (level == Z_DEFAULT_COMPRESSION) level = 6;
224 if (windowBits < 0) { /* undocumented feature: suppress zlib header */
226 windowBits = -windowBits;
228 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
229 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
230 strategy < 0 || strategy > Z_HUFFMAN_ONLY) {
231 return Z_STREAM_ERROR;
233 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
234 if (s == Z_NULL) return Z_MEM_ERROR;
235 strm->state = (struct internal_state FAR *)s;
238 s->noheader = noheader;
239 s->w_bits = windowBits;
240 s->w_size = 1 << s->w_bits;
241 s->w_mask = s->w_size - 1;
243 s->hash_bits = memLevel + 7;
244 s->hash_size = 1 << s->hash_bits;
245 s->hash_mask = s->hash_size - 1;
246 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
248 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
249 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));
250 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
252 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
254 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
255 s->pending_buf = (uchf *) overlay;
257 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
258 s->pending_buf == Z_NULL) {
259 strm->msg = (char*)ERR_MSG(Z_MEM_ERROR);
263 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
264 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
267 s->strategy = strategy;
268 s->method = (Byte)method;
270 return deflateReset(strm);
273 /* ========================================================================= */
274 int deflateSetDictionary (strm, dictionary, dictLength)
276 const Bytef *dictionary;
280 uInt length = dictLength;
284 if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL ||
285 strm->state->status != INIT_STATE) return Z_STREAM_ERROR;
288 strm->adler = adler32(strm->adler, dictionary, dictLength);
290 if (length < MIN_MATCH) return Z_OK;
291 if (length > MAX_DIST(s)) {
292 length = MAX_DIST(s);
293 dictionary += dictLength - length;
295 zmemcpy((charf *)s->window, dictionary, length);
296 s->strstart = length;
297 s->block_start = (long)length;
299 /* Insert all strings in the hash table (except for the last two bytes).
300 * s->lookahead stays null, so s->ins_h will be recomputed at the next
301 * call of fill_window.
303 s->ins_h = s->window[0];
304 UPDATE_HASH(s, s->ins_h, s->window[1]);
305 for (n = 0; n <= length - MIN_MATCH; n++) {
306 INSERT_STRING(s, n, hash_head);
308 if (hash_head) hash_head = 0; /* to make compiler happy */
312 /* ========================================================================= */
313 int deflateReset (strm)
318 if (strm == Z_NULL || strm->state == Z_NULL ||
319 strm->zalloc == Z_NULL || strm->zfree == Z_NULL) return Z_STREAM_ERROR;
321 strm->total_in = strm->total_out = 0;
322 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
323 strm->data_type = Z_UNKNOWN;
325 s = (deflate_state *)strm->state;
327 s->pending_out = s->pending_buf;
329 if (s->noheader < 0) {
330 s->noheader = 0; /* was set to -1 by deflate(..., Z_FINISH); */
332 s->status = s->noheader ? BUSY_STATE : INIT_STATE;
334 s->last_flush = Z_NO_FLUSH;
342 /* ========================================================================= */
343 int deflateParams(strm, level, strategy)
352 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
355 if (level == Z_DEFAULT_COMPRESSION) {
358 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_HUFFMAN_ONLY) {
359 return Z_STREAM_ERROR;
361 func = configuration_table[s->level].func;
363 if (func != configuration_table[level].func && strm->total_in != 0) {
364 /* Flush the last buffer: */
365 err = deflate(strm, Z_PARTIAL_FLUSH);
367 if (s->level != level) {
369 s->max_lazy_match = configuration_table[level].max_lazy;
370 s->good_match = configuration_table[level].good_length;
371 s->nice_match = configuration_table[level].nice_length;
372 s->max_chain_length = configuration_table[level].max_chain;
374 s->strategy = strategy;
378 /* =========================================================================
379 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
380 * IN assertion: the stream state is correct and there is enough room in
383 local void putShortMSB (s, b)
387 put_byte(s, (Byte)(b >> 8));
388 put_byte(s, (Byte)(b & 0xff));
391 /* =========================================================================
392 * Flush as much pending output as possible. All deflate() output goes
393 * through this function so some applications may wish to modify it
394 * to avoid allocating a large strm->next_out buffer and copying into it.
395 * (See also read_buf()).
397 local void flush_pending(strm)
400 unsigned len = strm->state->pending;
402 if (len > strm->avail_out) len = strm->avail_out;
403 if (len == 0) return;
405 zmemcpy(strm->next_out, strm->state->pending_out, len);
406 strm->next_out += len;
407 strm->state->pending_out += len;
408 strm->total_out += len;
409 strm->avail_out -= len;
410 strm->state->pending -= len;
411 if (strm->state->pending == 0) {
412 strm->state->pending_out = strm->state->pending_buf;
416 /* ========================================================================= */
417 int deflate (strm, flush)
421 int old_flush; /* value of flush param for previous deflate call */
424 if (strm == Z_NULL || strm->state == Z_NULL ||
425 flush > Z_FINISH || flush < 0) {
426 return Z_STREAM_ERROR;
430 if (strm->next_out == Z_NULL ||
431 (strm->next_in == Z_NULL && strm->avail_in != 0) ||
432 (s->status == FINISH_STATE && flush != Z_FINISH)) {
433 ERR_RETURN(strm, Z_STREAM_ERROR);
435 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
437 s->strm = strm; /* just in case */
438 old_flush = s->last_flush;
439 s->last_flush = flush;
441 /* Write the zlib header */
442 if (s->status == INIT_STATE) {
444 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
445 uInt level_flags = (s->level-1) >> 1;
447 if (level_flags > 3) level_flags = 3;
448 header |= (level_flags << 6);
449 if (s->strstart != 0) header |= PRESET_DICT;
450 header += 31 - (header % 31);
452 s->status = BUSY_STATE;
453 putShortMSB(s, header);
455 /* Save the adler32 of the preset dictionary: */
456 if (s->strstart != 0) {
457 putShortMSB(s, (uInt)(strm->adler >> 16));
458 putShortMSB(s, (uInt)(strm->adler & 0xffff));
463 /* Flush as much pending output as possible */
464 if (s->pending != 0) {
466 if (strm->avail_out == 0) {
467 /* Since avail_out is 0, deflate will be called again with
468 * more output space, but possibly with both pending and
469 * avail_in equal to zero. There won't be anything to do,
470 * but this is not an error situation so make sure we
471 * return OK instead of BUF_ERROR at next call of deflate:
477 /* Make sure there is something to do and avoid duplicate consecutive
478 * flushes. For repeated and useless calls with Z_FINISH, we keep
479 * returning Z_STREAM_END instead of Z_BUFF_ERROR.
481 } else if (strm->avail_in == 0 && flush <= old_flush &&
483 ERR_RETURN(strm, Z_BUF_ERROR);
486 /* User must not provide more input after the first FINISH: */
487 if (s->status == FINISH_STATE && strm->avail_in != 0) {
488 ERR_RETURN(strm, Z_BUF_ERROR);
491 /* Start a new block or continue the current one.
493 if (strm->avail_in != 0 || s->lookahead != 0 ||
494 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
497 bstate = (*(configuration_table[s->level].func))(s, flush);
499 if (bstate == finish_started || bstate == finish_done) {
500 s->status = FINISH_STATE;
502 if (bstate == need_more || bstate == finish_started) {
503 if (strm->avail_out == 0) {
504 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
507 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
508 * of deflate should use the same flush parameter to make sure
509 * that the flush is complete. So we don't have to output an
510 * empty block here, this will be done at next call. This also
511 * ensures that for a very small output buffer, we emit at most
515 if (bstate == block_done) {
516 if (flush == Z_PARTIAL_FLUSH) {
518 } else { /* FULL_FLUSH or SYNC_FLUSH */
519 _tr_stored_block(s, (char*)0, 0L, 0);
520 /* For a full flush, this empty block will be recognized
521 * as a special marker by inflate_sync().
523 if (flush == Z_FULL_FLUSH) {
524 CLEAR_HASH(s); /* forget history */
528 if (strm->avail_out == 0) {
529 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
534 Assert(strm->avail_out > 0, "bug2");
536 if (flush != Z_FINISH) return Z_OK;
537 if (s->noheader) return Z_STREAM_END;
539 /* Write the zlib trailer (adler32) */
540 putShortMSB(s, (uInt)(strm->adler >> 16));
541 putShortMSB(s, (uInt)(strm->adler & 0xffff));
543 /* If avail_out is zero, the application will call deflate again
546 s->noheader = -1; /* write the trailer only once! */
547 return s->pending != 0 ? Z_OK : Z_STREAM_END;
550 /* ========================================================================= */
551 int deflateEnd (strm)
556 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
558 /* Deallocate in reverse order of allocations: */
559 TRY_FREE(strm, strm->state->pending_buf);
560 TRY_FREE(strm, strm->state->head);
561 TRY_FREE(strm, strm->state->prev);
562 TRY_FREE(strm, strm->state->window);
564 status = strm->state->status;
565 ZFREE(strm, strm->state);
566 strm->state = Z_NULL;
568 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
571 /* ========================================================================= */
572 int deflateCopy (dest, source)
576 if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) {
577 return Z_STREAM_ERROR;
580 return Z_STREAM_ERROR; /* to be implemented */
582 dest->state = (struct internal_state FAR *)
583 (*dest->zalloc)(1, sizeof(deflate_state));
584 if (dest->state == Z_NULL) return Z_MEM_ERROR;
586 *(dest->state) = *(source->state);
591 /* ===========================================================================
592 * Read a new buffer from the current input stream, update the adler32
593 * and total number of bytes read. All deflate() input goes through
594 * this function so some applications may wish to modify it to avoid
595 * allocating a large strm->next_in buffer and copying from it.
596 * (See also flush_pending()).
598 local int read_buf(strm, buf, size)
603 unsigned len = strm->avail_in;
605 if (len > size) len = size;
606 if (len == 0) return 0;
608 strm->avail_in -= len;
610 if (!strm->state->noheader) {
611 strm->adler = adler32(strm->adler, strm->next_in, len);
613 zmemcpy(buf, strm->next_in, len);
614 strm->next_in += len;
615 strm->total_in += len;
620 /* ===========================================================================
621 * Initialize the "longest match" routines for a new zlib stream
623 local void lm_init (s)
626 s->window_size = (ulg)2L*s->w_size;
630 /* Set the default configuration parameters:
632 s->max_lazy_match = configuration_table[s->level].max_lazy;
633 s->good_match = configuration_table[s->level].good_length;
634 s->nice_match = configuration_table[s->level].nice_length;
635 s->max_chain_length = configuration_table[s->level].max_chain;
640 s->match_length = s->prev_length = MIN_MATCH-1;
641 s->match_available = 0;
644 match_init(); /* initialize the asm code */
648 /* ===========================================================================
649 * Set match_start to the longest match starting at the given string and
650 * return its length. Matches shorter or equal to prev_length are discarded,
651 * in which case the result is equal to prev_length and match_start is
653 * IN assertions: cur_match is the head of the hash chain for the current
654 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
655 * OUT assertion: the match length is not greater than s->lookahead.
658 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
659 * match.S. The code will be functionally equivalent.
661 local uInt longest_match(s, cur_match)
663 IPos cur_match; /* current match */
665 unsigned chain_length = s->max_chain_length;/* max hash chain length */
666 register Bytef *scan = s->window + s->strstart; /* current string */
667 register Bytef *match; /* matched string */
668 register int len; /* length of current match */
669 int best_len = s->prev_length; /* best match length so far */
670 int nice_match = s->nice_match; /* stop if match long enough */
671 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
672 s->strstart - (IPos)MAX_DIST(s) : NIL;
673 /* Stop when cur_match becomes <= limit. To simplify the code,
674 * we prevent matches with the string of window index 0.
676 Posf *prev = s->prev;
677 uInt wmask = s->w_mask;
680 /* Compare two bytes at a time. Note: this is not always beneficial.
681 * Try with and without -DUNALIGNED_OK to check.
683 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
684 register ush scan_start = *(ushf*)scan;
685 register ush scan_end = *(ushf*)(scan+best_len-1);
687 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
688 register Byte scan_end1 = scan[best_len-1];
689 register Byte scan_end = scan[best_len];
692 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
693 * It is easy to get rid of this optimization if necessary.
695 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
697 /* Do not waste too much time if we already have a good match: */
698 if (s->prev_length >= s->good_match) {
701 /* Do not look for matches beyond the end of the input. This is necessary
702 * to make deflate deterministic.
704 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
706 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
709 Assert(cur_match < s->strstart, "no future");
710 match = s->window + cur_match;
712 /* Skip to next match if the match length cannot increase
713 * or if the match length is less than 2:
715 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
716 /* This code assumes sizeof(unsigned short) == 2. Do not use
717 * UNALIGNED_OK if your compiler uses a different size.
719 if (*(ushf*)(match+best_len-1) != scan_end ||
720 *(ushf*)match != scan_start) continue;
722 /* It is not necessary to compare scan[2] and match[2] since they are
723 * always equal when the other bytes match, given that the hash keys
724 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
725 * strstart+3, +5, ... up to strstart+257. We check for insufficient
726 * lookahead only every 4th comparison; the 128th check will be made
727 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
728 * necessary to put more guard bytes at the end of the window, or
729 * to check more often for insufficient lookahead.
731 Assert(scan[2] == match[2], "scan[2]?");
734 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
735 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
736 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
737 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
739 /* The funny "do {}" generates better code on most compilers */
741 /* Here, scan <= window+strstart+257 */
742 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
743 if (*scan == *match) scan++;
745 len = (MAX_MATCH - 1) - (int)(strend-scan);
746 scan = strend - (MAX_MATCH-1);
748 #else /* UNALIGNED_OK */
750 if (match[best_len] != scan_end ||
751 match[best_len-1] != scan_end1 ||
753 *++match != scan[1]) continue;
755 /* The check at best_len-1 can be removed because it will be made
756 * again later. (This heuristic is not always a win.)
757 * It is not necessary to compare scan[2] and match[2] since they
758 * are always equal when the other bytes match, given that
759 * the hash keys are equal and that HASH_BITS >= 8.
762 Assert(*scan == *match, "match[2]?");
764 /* We check for insufficient lookahead only every 8th comparison;
765 * the 256th check will be made at strstart+258.
768 } while (*++scan == *++match && *++scan == *++match &&
769 *++scan == *++match && *++scan == *++match &&
770 *++scan == *++match && *++scan == *++match &&
771 *++scan == *++match && *++scan == *++match &&
774 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
776 len = MAX_MATCH - (int)(strend - scan);
777 scan = strend - MAX_MATCH;
779 #endif /* UNALIGNED_OK */
781 if (len > best_len) {
782 s->match_start = cur_match;
784 if (len >= nice_match) break;
786 scan_end = *(ushf*)(scan+best_len-1);
788 scan_end1 = scan[best_len-1];
789 scan_end = scan[best_len];
792 } while ((cur_match = prev[cur_match & wmask]) > limit
793 && --chain_length != 0);
795 if ((uInt)best_len <= s->lookahead) return best_len;
801 /* ===========================================================================
802 * Check that the match at match_start is indeed a match.
804 local void check_match(s, start, match, length)
809 /* check that the match is indeed a match */
810 if (zmemcmp((charf *)s->window + match,
811 (charf *)s->window + start, length) != EQUAL) {
812 fprintf(stderr, " start %u, match %u, length %d\n",
813 start, match, length);
815 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
816 } while (--length != 0);
817 z_error("invalid match");
820 fprintf(stderr,"\\[%d,%d]", start-match, length);
821 do { putc(s->window[start++], stderr); } while (--length != 0);
825 # define check_match(s, start, match, length)
828 /* ===========================================================================
829 * Fill the window when the lookahead becomes insufficient.
830 * Updates strstart and lookahead.
832 * IN assertion: lookahead < MIN_LOOKAHEAD
833 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
834 * At least one byte has been read, or avail_in == 0; reads are
835 * performed for at least two bytes (required for the zip translate_eol
836 * option -- not supported here).
838 local void fill_window(s)
841 register unsigned n, m;
843 unsigned more; /* Amount of free space at the end of the window. */
844 uInt wsize = s->w_size;
847 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
849 /* Deal with !@#$% 64K limit: */
850 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
853 } else if (more == (unsigned)(-1)) {
854 /* Very unlikely, but possible on 16 bit machine if strstart == 0
855 * and lookahead == 1 (input done one byte at time)
859 /* If the window is almost full and there is insufficient lookahead,
860 * move the upper half to the lower one to make room in the upper half.
862 } else if (s->strstart >= wsize+MAX_DIST(s)) {
864 zmemcpy((charf *)s->window, (charf *)s->window+wsize,
866 s->match_start -= wsize;
867 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
869 s->block_start -= (long) wsize;
871 /* Slide the hash table (could be avoided with 32 bit values
872 at the expense of memory usage):
878 *p = (Pos)(m >= wsize ? m-wsize : NIL);
885 *p = (Pos)(m >= wsize ? m-wsize : NIL);
886 /* If n is not on any hash chain, prev[n] is garbage but
887 * its value will never be used.
893 if (s->strm->avail_in == 0) return;
895 /* If there was no sliding:
896 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
897 * more == window_size - lookahead - strstart
898 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
899 * => more >= window_size - 2*WSIZE + 2
900 * In the BIG_MEM or MMAP case (not yet supported),
901 * window_size == input_size + MIN_LOOKAHEAD &&
902 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
903 * Otherwise, window_size == 2*WSIZE so more >= 2.
904 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
906 Assert(more >= 2, "more < 2");
908 n = read_buf(s->strm, (charf *)s->window + s->strstart + s->lookahead,
912 /* Initialize the hash value now that we have some input: */
913 if (s->lookahead >= MIN_MATCH) {
914 s->ins_h = s->window[s->strstart];
915 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
917 Call UPDATE_HASH() MIN_MATCH-3 more times
920 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
921 * but this is not important since only literal bytes will be emitted.
924 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
927 /* ===========================================================================
928 * Flush the current block, with given end-of-file flag.
929 * IN assertion: strstart is set to the end of the current match.
931 #define FLUSH_BLOCK_ONLY(s, eof) { \
932 _tr_flush_block(s, (s->block_start >= 0L ? \
933 (charf *)&s->window[(unsigned)s->block_start] : \
935 (ulg)((long)s->strstart - s->block_start), \
937 s->block_start = s->strstart; \
938 flush_pending(s->strm); \
939 Tracev((stderr,"[FLUSH]")); \
942 /* Same but force premature exit if necessary. */
943 #define FLUSH_BLOCK(s, eof) { \
944 FLUSH_BLOCK_ONLY(s, eof); \
945 if (s->strm->avail_out == 0) return (eof) ? finish_started : need_more; \
948 /* ===========================================================================
949 * Copy without compression as much as possible from the input stream, return
950 * the current block state.
951 * This function does not insert new strings in the dictionary since
952 * uncompressible data is probably not useful. This function is used
953 * only for the level=0 compression option.
954 * NOTE: this function should be optimized to avoid extra copying.
956 local block_state deflate_stored(s, flush)
961 /* Fill the window as much as possible: */
962 if (s->lookahead <= 1) {
964 Assert(s->strstart < s->w_size+MAX_DIST(s) ||
965 s->block_start >= (long)s->w_size, "slide too late");
968 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
970 if (s->lookahead == 0) break; /* flush the current block */
972 Assert(s->block_start >= 0L, "block gone");
974 s->strstart += s->lookahead;
977 /* Stored blocks are limited to 0xffff bytes: */
978 if (s->strstart == 0 || s->strstart > 0xfffe) {
979 /* strstart == 0 is possible when wraparound on 16-bit machine */
980 s->lookahead = s->strstart - 0xffff;
981 s->strstart = 0xffff;
984 /* Emit a stored block if it is large enough: */
985 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
989 FLUSH_BLOCK(s, flush == Z_FINISH);
990 return flush == Z_FINISH ? finish_done : block_done;
993 /* ===========================================================================
994 * Compress as much as possible from the input stream, return the current
996 * This function does not perform lazy evaluation of matches and inserts
997 * new strings in the dictionary only for unmatched strings or for short
998 * matches. It is used only for the fast compression options.
1000 local block_state deflate_fast(s, flush)
1004 IPos hash_head = NIL; /* head of the hash chain */
1005 int bflush; /* set if current block must be flushed */
1008 /* Make sure that we always have enough lookahead, except
1009 * at the end of the input file. We need MAX_MATCH bytes
1010 * for the next match, plus MIN_MATCH bytes to insert the
1011 * string following the next match.
1013 if (s->lookahead < MIN_LOOKAHEAD) {
1015 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1018 if (s->lookahead == 0) break; /* flush the current block */
1021 /* Insert the string window[strstart .. strstart+2] in the
1022 * dictionary, and set hash_head to the head of the hash chain:
1024 if (s->lookahead >= MIN_MATCH) {
1025 INSERT_STRING(s, s->strstart, hash_head);
1028 /* Find the longest match, discarding those <= prev_length.
1029 * At this point we have always match_length < MIN_MATCH
1031 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1032 /* To simplify the code, we prevent matches with the string
1033 * of window index 0 (in particular we have to avoid a match
1034 * of the string with itself at the start of the input file).
1036 if (s->strategy != Z_HUFFMAN_ONLY) {
1037 s->match_length = longest_match (s, hash_head);
1039 /* longest_match() sets match_start */
1041 if (s->match_length >= MIN_MATCH) {
1042 check_match(s, s->strstart, s->match_start, s->match_length);
1044 bflush = _tr_tally(s, s->strstart - s->match_start,
1045 s->match_length - MIN_MATCH);
1047 s->lookahead -= s->match_length;
1049 /* Insert new strings in the hash table only if the match length
1050 * is not too large. This saves time but degrades compression.
1052 if (s->match_length <= s->max_insert_length &&
1053 s->lookahead >= MIN_MATCH) {
1054 s->match_length--; /* string at strstart already in hash table */
1057 INSERT_STRING(s, s->strstart, hash_head);
1058 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1059 * always MIN_MATCH bytes ahead.
1061 } while (--s->match_length != 0);
1064 s->strstart += s->match_length;
1065 s->match_length = 0;
1066 s->ins_h = s->window[s->strstart];
1067 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1069 Call UPDATE_HASH() MIN_MATCH-3 more times
1071 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1072 * matter since it will be recomputed at next deflate call.
1076 /* No match, output a literal byte */
1077 Tracevv((stderr,"%c", s->window[s->strstart]));
1078 bflush = _tr_tally (s, 0, s->window[s->strstart]);
1082 if (bflush) FLUSH_BLOCK(s, 0);
1084 FLUSH_BLOCK(s, flush == Z_FINISH);
1085 return flush == Z_FINISH ? finish_done : block_done;
1088 /* ===========================================================================
1089 * Same as above, but achieves better compression. We use a lazy
1090 * evaluation for matches: a match is finally adopted only if there is
1091 * no better match at the next window position.
1093 local block_state deflate_slow(s, flush)
1097 IPos hash_head = NIL; /* head of hash chain */
1098 int bflush; /* set if current block must be flushed */
1100 /* Process the input block. */
1102 /* Make sure that we always have enough lookahead, except
1103 * at the end of the input file. We need MAX_MATCH bytes
1104 * for the next match, plus MIN_MATCH bytes to insert the
1105 * string following the next match.
1107 if (s->lookahead < MIN_LOOKAHEAD) {
1109 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1112 if (s->lookahead == 0) break; /* flush the current block */
1115 /* Insert the string window[strstart .. strstart+2] in the
1116 * dictionary, and set hash_head to the head of the hash chain:
1118 if (s->lookahead >= MIN_MATCH) {
1119 INSERT_STRING(s, s->strstart, hash_head);
1122 /* Find the longest match, discarding those <= prev_length.
1124 s->prev_length = s->match_length, s->prev_match = s->match_start;
1125 s->match_length = MIN_MATCH-1;
1127 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1128 s->strstart - hash_head <= MAX_DIST(s)) {
1129 /* To simplify the code, we prevent matches with the string
1130 * of window index 0 (in particular we have to avoid a match
1131 * of the string with itself at the start of the input file).
1133 if (s->strategy != Z_HUFFMAN_ONLY) {
1134 s->match_length = longest_match (s, hash_head);
1136 /* longest_match() sets match_start */
1138 if (s->match_length <= 5 && (s->strategy == Z_FILTERED ||
1139 (s->match_length == MIN_MATCH &&
1140 s->strstart - s->match_start > TOO_FAR))) {
1142 /* If prev_match is also MIN_MATCH, match_start is garbage
1143 * but we will ignore the current match anyway.
1145 s->match_length = MIN_MATCH-1;
1148 /* If there was a match at the previous step and the current
1149 * match is not better, output the previous match:
1151 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1152 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1153 /* Do not insert strings in hash table beyond this. */
1155 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1157 bflush = _tr_tally(s, s->strstart -1 - s->prev_match,
1158 s->prev_length - MIN_MATCH);
1160 /* Insert in hash table all strings up to the end of the match.
1161 * strstart-1 and strstart are already inserted. If there is not
1162 * enough lookahead, the last two strings are not inserted in
1165 s->lookahead -= s->prev_length-1;
1166 s->prev_length -= 2;
1168 if (++s->strstart <= max_insert) {
1169 INSERT_STRING(s, s->strstart, hash_head);
1171 } while (--s->prev_length != 0);
1172 s->match_available = 0;
1173 s->match_length = MIN_MATCH-1;
1176 if (bflush) FLUSH_BLOCK(s, 0);
1178 } else if (s->match_available) {
1179 /* If there was no match at the previous position, output a
1180 * single literal. If there was a match but the current match
1181 * is longer, truncate the previous match to a single literal.
1183 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1184 if (_tr_tally (s, 0, s->window[s->strstart-1])) {
1185 FLUSH_BLOCK_ONLY(s, 0);
1189 if (s->strm->avail_out == 0) return need_more;
1191 /* There is no previous match to compare with, wait for
1192 * the next step to decide.
1194 s->match_available = 1;
1199 Assert (flush != Z_NO_FLUSH, "no flush?");
1200 if (s->match_available) {
1201 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1202 _tr_tally (s, 0, s->window[s->strstart-1]);
1203 s->match_available = 0;
1205 FLUSH_BLOCK(s, flush == Z_FINISH);
1206 return flush == Z_FINISH ? finish_done : block_done;