2 /* deflate.c -- compress data using the deflation algorithm
3 * Copyright (C) 1995-1996 Jean-loup Gailly.
4 * For conditions of distribution and use, see copyright notice in zlib.h
10 * The "deflation" process depends on being able to identify portions
11 * of the input text which are identical to earlier input (within a
12 * sliding window trailing behind the input currently being processed).
14 * The most straightforward technique turns out to be the fastest for
15 * most input files: try all possible matches and select the longest.
16 * The key feature of this algorithm is that insertions into the string
17 * dictionary are very simple and thus fast, and deletions are avoided
18 * completely. Insertions are performed at each input character, whereas
19 * string matches are performed only when the previous match ends. So it
20 * is preferable to spend more time in matches to allow very fast string
21 * insertions and avoid deletions. The matching algorithm for small
22 * strings is inspired from that of Rabin & Karp. A brute force approach
23 * is used to find longer strings when a small match has been found.
24 * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze
25 * (by Leonid Broukhis).
26 * A previous version of this file used a more sophisticated algorithm
27 * (by Fiala and Greene) which is guaranteed to run in linear amortized
28 * time, but has a larger average cost, uses more memory and is patented.
29 * However the F&G algorithm may be faster for some highly redundant
30 * files if the parameter max_chain_length (described below) is too large.
34 * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and
35 * I found it in 'freeze' written by Leonid Broukhis.
36 * Thanks to many people for bug reports and testing.
40 * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification".
41 * Available in ftp://ds.internic.net/rfc/rfc1951.txt
43 * A description of the Rabin and Karp algorithm is given in the book
44 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
46 * Fiala,E.R., and Greene,D.H.
47 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
51 #include <linux/module.h>
52 #include <linux/zutil.h>
56 /* ===========================================================================
57 * Function prototypes.
60 need_more, /* block not completed, need more input or more output */
61 block_done, /* block flush performed */
62 finish_started, /* finish started, need only more output at next deflate */
63 finish_done /* finish done, accept no more input or output */
66 typedef block_state (*compress_func) OF((deflate_state *s, int flush));
67 /* Compression function. Returns the block state after the call. */
69 local void fill_window OF((deflate_state *s));
70 local block_state deflate_stored OF((deflate_state *s, int flush));
71 local block_state deflate_fast OF((deflate_state *s, int flush));
72 local block_state deflate_slow OF((deflate_state *s, int flush));
73 local void lm_init OF((deflate_state *s));
74 local void putShortMSB OF((deflate_state *s, uInt b));
75 local void flush_pending OF((z_streamp strm));
76 local int read_buf OF((z_streamp strm, Bytef *buf, unsigned size));
77 local uInt longest_match OF((deflate_state *s, IPos cur_match));
80 local void check_match OF((deflate_state *s, IPos start, IPos match,
84 /* ===========================================================================
89 /* Tail of hash chains */
94 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
96 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
97 /* Minimum amount of lookahead, except at the end of the input file.
98 * See deflate.c for comments about the MIN_MATCH+1.
101 /* Values for max_lazy_match, good_match and max_chain_length, depending on
102 * the desired pack level (0..9). The values given below have been tuned to
103 * exclude worst case performance for pathological files. Better values may be
104 * found for specific files.
106 typedef struct config_s {
107 ush good_length; /* reduce lazy search above this match length */
108 ush max_lazy; /* do not perform lazy search above this match length */
109 ush nice_length; /* quit search above this match length */
114 local const config configuration_table[10] = {
115 /* good lazy nice chain */
116 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
117 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* maximum speed, no lazy matches */
118 /* 2 */ {4, 5, 16, 8, deflate_fast},
119 /* 3 */ {4, 6, 32, 32, deflate_fast},
121 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
122 /* 5 */ {8, 16, 32, 32, deflate_slow},
123 /* 6 */ {8, 16, 128, 128, deflate_slow},
124 /* 7 */ {8, 32, 128, 256, deflate_slow},
125 /* 8 */ {32, 128, 258, 1024, deflate_slow},
126 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* maximum compression */
128 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
129 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
134 /* result of memcmp for equal strings */
136 /* ===========================================================================
137 * Update a hash value with the given input byte
138 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
139 * input characters, so that a running hash key can be computed from the
140 * previous key instead of complete recalculation each time.
142 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
145 /* ===========================================================================
146 * Insert string str in the dictionary and set match_head to the previous head
147 * of the hash chain (the most recent string with same hash key). Return
148 * the previous length of the hash chain.
149 * IN assertion: all calls to to INSERT_STRING are made with consecutive
150 * input characters and the first MIN_MATCH bytes of str are valid
151 * (except for the last MIN_MATCH-1 bytes of the input file).
153 #define INSERT_STRING(s, str, match_head) \
154 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
155 s->prev[(str) & s->w_mask] = match_head = s->head[s->ins_h], \
156 s->head[s->ins_h] = (Pos)(str))
158 /* ===========================================================================
159 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
160 * prev[] will be initialized on the fly.
162 #define CLEAR_HASH(s) \
163 s->head[s->hash_size-1] = NIL; \
164 memset((charf *)s->head, 0, (unsigned)(s->hash_size-1)*sizeof(*s->head));
166 /* ========================================================================= */
167 int zlib_deflateInit_(strm, level, version, stream_size)
173 return zlib_deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS,
175 Z_DEFAULT_STRATEGY, version, stream_size);
176 /* To do: ignore strm->next_in if we use it as window */
179 /* ========================================================================= */
180 int zlib_deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
181 version, stream_size)
193 static char* my_version = ZLIB_VERSION;
194 deflate_workspace *mem;
197 /* We overlay pending_buf and d_buf+l_buf. This works since the average
198 * output size for (length,distance) codes is <= 24 bits.
201 if (version == Z_NULL || version[0] != my_version[0] ||
202 stream_size != sizeof(z_stream)) {
203 return Z_VERSION_ERROR;
205 if (strm == Z_NULL) return Z_STREAM_ERROR;
209 if (level == Z_DEFAULT_COMPRESSION) level = 6;
211 mem = (deflate_workspace *) strm->workspace;
213 if (windowBits < 0) { /* undocumented feature: suppress zlib header */
215 windowBits = -windowBits;
217 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
218 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
219 strategy < 0 || strategy > Z_HUFFMAN_ONLY) {
220 return Z_STREAM_ERROR;
222 s = (deflate_state *) &(mem->deflate_memory);
223 strm->state = (struct internal_state FAR *)s;
226 s->noheader = noheader;
227 s->w_bits = windowBits;
228 s->w_size = 1 << s->w_bits;
229 s->w_mask = s->w_size - 1;
231 s->hash_bits = memLevel + 7;
232 s->hash_size = 1 << s->hash_bits;
233 s->hash_mask = s->hash_size - 1;
234 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
236 s->window = (Bytef *) mem->window_memory;
237 s->prev = (Posf *) mem->prev_memory;
238 s->head = (Posf *) mem->head_memory;
240 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
242 overlay = (ushf *) mem->overlay_memory;
243 s->pending_buf = (uchf *) overlay;
244 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
246 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
247 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
250 s->strategy = strategy;
251 s->method = (Byte)method;
253 return zlib_deflateReset(strm);
256 /* ========================================================================= */
257 int zlib_deflateSetDictionary (strm, dictionary, dictLength)
259 const Bytef *dictionary;
263 uInt length = dictLength;
267 if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL)
268 return Z_STREAM_ERROR;
270 s = (deflate_state *) strm->state;
271 if (s->status != INIT_STATE) return Z_STREAM_ERROR;
273 strm->adler = zlib_adler32(strm->adler, dictionary, dictLength);
275 if (length < MIN_MATCH) return Z_OK;
276 if (length > MAX_DIST(s)) {
277 length = MAX_DIST(s);
278 #ifndef USE_DICT_HEAD
279 dictionary += dictLength - length; /* use the tail of the dictionary */
282 memcpy((charf *)s->window, dictionary, length);
283 s->strstart = length;
284 s->block_start = (long)length;
286 /* Insert all strings in the hash table (except for the last two bytes).
287 * s->lookahead stays null, so s->ins_h will be recomputed at the next
288 * call of fill_window.
290 s->ins_h = s->window[0];
291 UPDATE_HASH(s, s->ins_h, s->window[1]);
292 for (n = 0; n <= length - MIN_MATCH; n++) {
293 INSERT_STRING(s, n, hash_head);
295 if (hash_head) hash_head = 0; /* to make compiler happy */
299 /* ========================================================================= */
300 int zlib_deflateReset (strm)
305 if (strm == Z_NULL || strm->state == Z_NULL)
306 return Z_STREAM_ERROR;
308 strm->total_in = strm->total_out = 0;
310 strm->data_type = Z_UNKNOWN;
312 s = (deflate_state *)strm->state;
314 s->pending_out = s->pending_buf;
316 if (s->noheader < 0) {
317 s->noheader = 0; /* was set to -1 by deflate(..., Z_FINISH); */
319 s->status = s->noheader ? BUSY_STATE : INIT_STATE;
321 s->last_flush = Z_NO_FLUSH;
329 /* ========================================================================= */
330 int zlib_deflateParams(strm, level, strategy)
339 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
340 s = (deflate_state *) strm->state;
342 if (level == Z_DEFAULT_COMPRESSION) {
345 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_HUFFMAN_ONLY) {
346 return Z_STREAM_ERROR;
348 func = configuration_table[s->level].func;
350 if (func != configuration_table[level].func && strm->total_in != 0) {
351 /* Flush the last buffer: */
352 err = zlib_deflate(strm, Z_PARTIAL_FLUSH);
354 if (s->level != level) {
356 s->max_lazy_match = configuration_table[level].max_lazy;
357 s->good_match = configuration_table[level].good_length;
358 s->nice_match = configuration_table[level].nice_length;
359 s->max_chain_length = configuration_table[level].max_chain;
361 s->strategy = strategy;
365 /* =========================================================================
366 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
367 * IN assertion: the stream state is correct and there is enough room in
370 local void putShortMSB (s, b)
374 put_byte(s, (Byte)(b >> 8));
375 put_byte(s, (Byte)(b & 0xff));
378 /* =========================================================================
379 * Flush as much pending output as possible. All deflate() output goes
380 * through this function so some applications may wish to modify it
381 * to avoid allocating a large strm->next_out buffer and copying into it.
382 * (See also read_buf()).
384 local void flush_pending(strm)
387 deflate_state *s = (deflate_state *) strm->state;
388 unsigned len = s->pending;
390 if (len > strm->avail_out) len = strm->avail_out;
391 if (len == 0) return;
393 if (strm->next_out != Z_NULL) {
394 memcpy(strm->next_out, s->pending_out, len);
395 strm->next_out += len;
397 s->pending_out += len;
398 strm->total_out += len;
399 strm->avail_out -= len;
401 if (s->pending == 0) {
402 s->pending_out = s->pending_buf;
406 /* ========================================================================= */
407 int zlib_deflate (strm, flush)
411 int old_flush; /* value of flush param for previous deflate call */
414 if (strm == Z_NULL || strm->state == Z_NULL ||
415 flush > Z_FINISH || flush < 0) {
416 return Z_STREAM_ERROR;
418 s = (deflate_state *) strm->state;
420 if ((strm->next_in == Z_NULL && strm->avail_in != 0) ||
421 (s->status == FINISH_STATE && flush != Z_FINISH)) {
422 return Z_STREAM_ERROR;
424 if (strm->avail_out == 0) return Z_BUF_ERROR;
426 s->strm = strm; /* just in case */
427 old_flush = s->last_flush;
428 s->last_flush = flush;
430 /* Write the zlib header */
431 if (s->status == INIT_STATE) {
433 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
434 uInt level_flags = (s->level-1) >> 1;
436 if (level_flags > 3) level_flags = 3;
437 header |= (level_flags << 6);
438 if (s->strstart != 0) header |= PRESET_DICT;
439 header += 31 - (header % 31);
441 s->status = BUSY_STATE;
442 putShortMSB(s, header);
444 /* Save the adler32 of the preset dictionary: */
445 if (s->strstart != 0) {
446 putShortMSB(s, (uInt)(strm->adler >> 16));
447 putShortMSB(s, (uInt)(strm->adler & 0xffff));
452 /* Flush as much pending output as possible */
453 if (s->pending != 0) {
455 if (strm->avail_out == 0) {
456 /* Since avail_out is 0, deflate will be called again with
457 * more output space, but possibly with both pending and
458 * avail_in equal to zero. There won't be anything to do,
459 * but this is not an error situation so make sure we
460 * return OK instead of BUF_ERROR at next call of deflate:
466 /* Make sure there is something to do and avoid duplicate consecutive
467 * flushes. For repeated and useless calls with Z_FINISH, we keep
468 * returning Z_STREAM_END instead of Z_BUFF_ERROR.
470 } else if (strm->avail_in == 0 && flush <= old_flush &&
475 /* User must not provide more input after the first FINISH: */
476 if (s->status == FINISH_STATE && strm->avail_in != 0) {
480 /* Start a new block or continue the current one.
482 if (strm->avail_in != 0 || s->lookahead != 0 ||
483 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
486 bstate = (*(configuration_table[s->level].func))(s, flush);
488 if (bstate == finish_started || bstate == finish_done) {
489 s->status = FINISH_STATE;
491 if (bstate == need_more || bstate == finish_started) {
492 if (strm->avail_out == 0) {
493 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
496 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
497 * of deflate should use the same flush parameter to make sure
498 * that the flush is complete. So we don't have to output an
499 * empty block here, this will be done at next call. This also
500 * ensures that for a very small output buffer, we emit at most
504 if (bstate == block_done) {
505 if (flush == Z_PARTIAL_FLUSH) {
507 } else if (flush == Z_PACKET_FLUSH) {
508 /* Output just the 3-bit `stored' block type value,
509 but not a zero length. */
510 zlib_tr_stored_type_only(s);
511 } else { /* FULL_FLUSH or SYNC_FLUSH */
512 zlib_tr_stored_block(s, (char*)0, 0L, 0);
513 /* For a full flush, this empty block will be recognized
514 * as a special marker by inflate_sync().
516 if (flush == Z_FULL_FLUSH) {
517 CLEAR_HASH(s); /* forget history */
521 if (strm->avail_out == 0) {
522 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
527 Assert(strm->avail_out > 0, "bug2");
529 if (flush != Z_FINISH) return Z_OK;
530 if (s->noheader) return Z_STREAM_END;
532 /* Write the zlib trailer (adler32) */
533 putShortMSB(s, (uInt)(strm->adler >> 16));
534 putShortMSB(s, (uInt)(strm->adler & 0xffff));
536 /* If avail_out is zero, the application will call deflate again
539 s->noheader = -1; /* write the trailer only once! */
540 return s->pending != 0 ? Z_OK : Z_STREAM_END;
543 /* ========================================================================= */
544 int zlib_deflateEnd (strm)
550 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
551 s = (deflate_state *) strm->state;
554 if (status != INIT_STATE && status != BUSY_STATE &&
555 status != FINISH_STATE) {
556 return Z_STREAM_ERROR;
559 strm->state = Z_NULL;
561 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
564 /* =========================================================================
565 * Copy the source state to the destination state.
567 int zlib_deflateCopy (dest, source)
572 return Z_STREAM_ERROR;
577 deflate_workspace *mem;
580 if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) {
581 return Z_STREAM_ERROR;
584 ss = (deflate_state *) source->state;
588 mem = (deflate_workspace *) dest->workspace;
590 ds = &(mem->deflate_memory);
592 dest->state = (struct internal_state FAR *) ds;
596 ds->window = (Bytef *) mem->window_memory;
597 ds->prev = (Posf *) mem->prev_memory;
598 ds->head = (Posf *) mem->head_memory;
599 overlay = (ushf *) mem->overlay_memory;
600 ds->pending_buf = (uchf *) overlay;
602 memcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
603 memcpy(ds->prev, ss->prev, ds->w_size * sizeof(Pos));
604 memcpy(ds->head, ss->head, ds->hash_size * sizeof(Pos));
605 memcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
607 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
608 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
609 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
611 ds->l_desc.dyn_tree = ds->dyn_ltree;
612 ds->d_desc.dyn_tree = ds->dyn_dtree;
613 ds->bl_desc.dyn_tree = ds->bl_tree;
619 /* ===========================================================================
620 * Read a new buffer from the current input stream, update the adler32
621 * and total number of bytes read. All deflate() input goes through
622 * this function so some applications may wish to modify it to avoid
623 * allocating a large strm->next_in buffer and copying from it.
624 * (See also flush_pending()).
626 local int read_buf(strm, buf, size)
631 unsigned len = strm->avail_in;
633 if (len > size) len = size;
634 if (len == 0) return 0;
636 strm->avail_in -= len;
638 if (!((deflate_state *)(strm->state))->noheader) {
639 strm->adler = zlib_adler32(strm->adler, strm->next_in, len);
641 memcpy(buf, strm->next_in, len);
642 strm->next_in += len;
643 strm->total_in += len;
648 /* ===========================================================================
649 * Initialize the "longest match" routines for a new zlib stream
651 local void lm_init (s)
654 s->window_size = (ulg)2L*s->w_size;
658 /* Set the default configuration parameters:
660 s->max_lazy_match = configuration_table[s->level].max_lazy;
661 s->good_match = configuration_table[s->level].good_length;
662 s->nice_match = configuration_table[s->level].nice_length;
663 s->max_chain_length = configuration_table[s->level].max_chain;
668 s->match_length = s->prev_length = MIN_MATCH-1;
669 s->match_available = 0;
673 /* ===========================================================================
674 * Set match_start to the longest match starting at the given string and
675 * return its length. Matches shorter or equal to prev_length are discarded,
676 * in which case the result is equal to prev_length and match_start is
678 * IN assertions: cur_match is the head of the hash chain for the current
679 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
680 * OUT assertion: the match length is not greater than s->lookahead.
682 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
683 * match.S. The code will be functionally equivalent.
685 local uInt longest_match(s, cur_match)
687 IPos cur_match; /* current match */
689 unsigned chain_length = s->max_chain_length;/* max hash chain length */
690 register Bytef *scan = s->window + s->strstart; /* current string */
691 register Bytef *match; /* matched string */
692 register int len; /* length of current match */
693 int best_len = s->prev_length; /* best match length so far */
694 int nice_match = s->nice_match; /* stop if match long enough */
695 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
696 s->strstart - (IPos)MAX_DIST(s) : NIL;
697 /* Stop when cur_match becomes <= limit. To simplify the code,
698 * we prevent matches with the string of window index 0.
700 Posf *prev = s->prev;
701 uInt wmask = s->w_mask;
704 /* Compare two bytes at a time. Note: this is not always beneficial.
705 * Try with and without -DUNALIGNED_OK to check.
707 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
708 register ush scan_start = *(ushf*)scan;
709 register ush scan_end = *(ushf*)(scan+best_len-1);
711 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
712 register Byte scan_end1 = scan[best_len-1];
713 register Byte scan_end = scan[best_len];
716 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
717 * It is easy to get rid of this optimization if necessary.
719 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
721 /* Do not waste too much time if we already have a good match: */
722 if (s->prev_length >= s->good_match) {
725 /* Do not look for matches beyond the end of the input. This is necessary
726 * to make deflate deterministic.
728 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
730 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
733 Assert(cur_match < s->strstart, "no future");
734 match = s->window + cur_match;
736 /* Skip to next match if the match length cannot increase
737 * or if the match length is less than 2:
739 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
740 /* This code assumes sizeof(unsigned short) == 2. Do not use
741 * UNALIGNED_OK if your compiler uses a different size.
743 if (*(ushf*)(match+best_len-1) != scan_end ||
744 *(ushf*)match != scan_start) continue;
746 /* It is not necessary to compare scan[2] and match[2] since they are
747 * always equal when the other bytes match, given that the hash keys
748 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
749 * strstart+3, +5, ... up to strstart+257. We check for insufficient
750 * lookahead only every 4th comparison; the 128th check will be made
751 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
752 * necessary to put more guard bytes at the end of the window, or
753 * to check more often for insufficient lookahead.
755 Assert(scan[2] == match[2], "scan[2]?");
758 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
759 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
760 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
761 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
763 /* The funny "do {}" generates better code on most compilers */
765 /* Here, scan <= window+strstart+257 */
766 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
767 if (*scan == *match) scan++;
769 len = (MAX_MATCH - 1) - (int)(strend-scan);
770 scan = strend - (MAX_MATCH-1);
772 #else /* UNALIGNED_OK */
774 if (match[best_len] != scan_end ||
775 match[best_len-1] != scan_end1 ||
777 *++match != scan[1]) continue;
779 /* The check at best_len-1 can be removed because it will be made
780 * again later. (This heuristic is not always a win.)
781 * It is not necessary to compare scan[2] and match[2] since they
782 * are always equal when the other bytes match, given that
783 * the hash keys are equal and that HASH_BITS >= 8.
786 Assert(*scan == *match, "match[2]?");
788 /* We check for insufficient lookahead only every 8th comparison;
789 * the 256th check will be made at strstart+258.
792 } while (*++scan == *++match && *++scan == *++match &&
793 *++scan == *++match && *++scan == *++match &&
794 *++scan == *++match && *++scan == *++match &&
795 *++scan == *++match && *++scan == *++match &&
798 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
800 len = MAX_MATCH - (int)(strend - scan);
801 scan = strend - MAX_MATCH;
803 #endif /* UNALIGNED_OK */
805 if (len > best_len) {
806 s->match_start = cur_match;
808 if (len >= nice_match) break;
810 scan_end = *(ushf*)(scan+best_len-1);
812 scan_end1 = scan[best_len-1];
813 scan_end = scan[best_len];
816 } while ((cur_match = prev[cur_match & wmask]) > limit
817 && --chain_length != 0);
819 if ((uInt)best_len <= s->lookahead) return best_len;
824 /* ===========================================================================
825 * Check that the match at match_start is indeed a match.
827 local void check_match(s, start, match, length)
832 /* check that the match is indeed a match */
833 if (memcmp((charf *)s->window + match,
834 (charf *)s->window + start, length) != EQUAL) {
835 fprintf(stderr, " start %u, match %u, length %d\n",
836 start, match, length);
838 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
839 } while (--length != 0);
840 z_error("invalid match");
843 fprintf(stderr,"\\[%d,%d]", start-match, length);
844 do { putc(s->window[start++], stderr); } while (--length != 0);
848 # define check_match(s, start, match, length)
851 /* ===========================================================================
852 * Fill the window when the lookahead becomes insufficient.
853 * Updates strstart and lookahead.
855 * IN assertion: lookahead < MIN_LOOKAHEAD
856 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
857 * At least one byte has been read, or avail_in == 0; reads are
858 * performed for at least two bytes (required for the zip translate_eol
859 * option -- not supported here).
861 local void fill_window(s)
864 register unsigned n, m;
866 unsigned more; /* Amount of free space at the end of the window. */
867 uInt wsize = s->w_size;
870 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
872 /* Deal with !@#$% 64K limit: */
873 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
876 } else if (more == (unsigned)(-1)) {
877 /* Very unlikely, but possible on 16 bit machine if strstart == 0
878 * and lookahead == 1 (input done one byte at time)
882 /* If the window is almost full and there is insufficient lookahead,
883 * move the upper half to the lower one to make room in the upper half.
885 } else if (s->strstart >= wsize+MAX_DIST(s)) {
887 memcpy((charf *)s->window, (charf *)s->window+wsize,
889 s->match_start -= wsize;
890 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
891 s->block_start -= (long) wsize;
893 /* Slide the hash table (could be avoided with 32 bit values
894 at the expense of memory usage). We slide even when level == 0
895 to keep the hash table consistent if we switch back to level > 0
896 later. (Using level 0 permanently is not an optimal usage of
897 zlib, so we don't care about this pathological case.)
903 *p = (Pos)(m >= wsize ? m-wsize : NIL);
910 *p = (Pos)(m >= wsize ? m-wsize : NIL);
911 /* If n is not on any hash chain, prev[n] is garbage but
912 * its value will never be used.
917 if (s->strm->avail_in == 0) return;
919 /* If there was no sliding:
920 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
921 * more == window_size - lookahead - strstart
922 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
923 * => more >= window_size - 2*WSIZE + 2
924 * In the BIG_MEM or MMAP case (not yet supported),
925 * window_size == input_size + MIN_LOOKAHEAD &&
926 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
927 * Otherwise, window_size == 2*WSIZE so more >= 2.
928 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
930 Assert(more >= 2, "more < 2");
932 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
935 /* Initialize the hash value now that we have some input: */
936 if (s->lookahead >= MIN_MATCH) {
937 s->ins_h = s->window[s->strstart];
938 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
940 Call UPDATE_HASH() MIN_MATCH-3 more times
943 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
944 * but this is not important since only literal bytes will be emitted.
947 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
950 /* ===========================================================================
951 * Flush the current block, with given end-of-file flag.
952 * IN assertion: strstart is set to the end of the current match.
954 #define FLUSH_BLOCK_ONLY(s, eof) { \
955 zlib_tr_flush_block(s, (s->block_start >= 0L ? \
956 (charf *)&s->window[(unsigned)s->block_start] : \
958 (ulg)((long)s->strstart - s->block_start), \
960 s->block_start = s->strstart; \
961 flush_pending(s->strm); \
962 Tracev((stderr,"[FLUSH]")); \
965 /* Same but force premature exit if necessary. */
966 #define FLUSH_BLOCK(s, eof) { \
967 FLUSH_BLOCK_ONLY(s, eof); \
968 if (s->strm->avail_out == 0) return (eof) ? finish_started : need_more; \
971 /* ===========================================================================
972 * Copy without compression as much as possible from the input stream, return
973 * the current block state.
974 * This function does not insert new strings in the dictionary since
975 * uncompressible data is probably not useful. This function is used
976 * only for the level=0 compression option.
977 * NOTE: this function should be optimized to avoid extra copying from
978 * window to pending_buf.
980 local block_state deflate_stored(s, flush)
984 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited
985 * to pending_buf_size, and each stored block has a 5 byte header:
987 ulg max_block_size = 0xffff;
990 if (max_block_size > s->pending_buf_size - 5) {
991 max_block_size = s->pending_buf_size - 5;
994 /* Copy as much as possible from input to output: */
996 /* Fill the window as much as possible: */
997 if (s->lookahead <= 1) {
999 Assert(s->strstart < s->w_size+MAX_DIST(s) ||
1000 s->block_start >= (long)s->w_size, "slide too late");
1003 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
1005 if (s->lookahead == 0) break; /* flush the current block */
1007 Assert(s->block_start >= 0L, "block gone");
1009 s->strstart += s->lookahead;
1012 /* Emit a stored block if pending_buf will be full: */
1013 max_start = s->block_start + max_block_size;
1014 if (s->strstart == 0 || (ulg)s->strstart >= max_start) {
1015 /* strstart == 0 is possible when wraparound on 16-bit machine */
1016 s->lookahead = (uInt)(s->strstart - max_start);
1017 s->strstart = (uInt)max_start;
1020 /* Flush if we may have to slide, otherwise block_start may become
1021 * negative and the data will be gone:
1023 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
1027 FLUSH_BLOCK(s, flush == Z_FINISH);
1028 return flush == Z_FINISH ? finish_done : block_done;
1031 /* ===========================================================================
1032 * Compress as much as possible from the input stream, return the current
1034 * This function does not perform lazy evaluation of matches and inserts
1035 * new strings in the dictionary only for unmatched strings or for short
1036 * matches. It is used only for the fast compression options.
1038 local block_state deflate_fast(s, flush)
1042 IPos hash_head = NIL; /* head of the hash chain */
1043 int bflush; /* set if current block must be flushed */
1046 /* Make sure that we always have enough lookahead, except
1047 * at the end of the input file. We need MAX_MATCH bytes
1048 * for the next match, plus MIN_MATCH bytes to insert the
1049 * string following the next match.
1051 if (s->lookahead < MIN_LOOKAHEAD) {
1053 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1056 if (s->lookahead == 0) break; /* flush the current block */
1059 /* Insert the string window[strstart .. strstart+2] in the
1060 * dictionary, and set hash_head to the head of the hash chain:
1062 if (s->lookahead >= MIN_MATCH) {
1063 INSERT_STRING(s, s->strstart, hash_head);
1066 /* Find the longest match, discarding those <= prev_length.
1067 * At this point we have always match_length < MIN_MATCH
1069 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1070 /* To simplify the code, we prevent matches with the string
1071 * of window index 0 (in particular we have to avoid a match
1072 * of the string with itself at the start of the input file).
1074 if (s->strategy != Z_HUFFMAN_ONLY) {
1075 s->match_length = longest_match (s, hash_head);
1077 /* longest_match() sets match_start */
1079 if (s->match_length >= MIN_MATCH) {
1080 check_match(s, s->strstart, s->match_start, s->match_length);
1082 bflush = zlib_tr_tally(s, s->strstart - s->match_start,
1083 s->match_length - MIN_MATCH);
1085 s->lookahead -= s->match_length;
1087 /* Insert new strings in the hash table only if the match length
1088 * is not too large. This saves time but degrades compression.
1090 if (s->match_length <= s->max_insert_length &&
1091 s->lookahead >= MIN_MATCH) {
1092 s->match_length--; /* string at strstart already in hash table */
1095 INSERT_STRING(s, s->strstart, hash_head);
1096 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1097 * always MIN_MATCH bytes ahead.
1099 } while (--s->match_length != 0);
1102 s->strstart += s->match_length;
1103 s->match_length = 0;
1104 s->ins_h = s->window[s->strstart];
1105 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1107 Call UPDATE_HASH() MIN_MATCH-3 more times
1109 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1110 * matter since it will be recomputed at next deflate call.
1114 /* No match, output a literal byte */
1115 Tracevv((stderr,"%c", s->window[s->strstart]));
1116 bflush = zlib_tr_tally (s, 0, s->window[s->strstart]);
1120 if (bflush) FLUSH_BLOCK(s, 0);
1122 FLUSH_BLOCK(s, flush == Z_FINISH);
1123 return flush == Z_FINISH ? finish_done : block_done;
1126 /* ===========================================================================
1127 * Same as above, but achieves better compression. We use a lazy
1128 * evaluation for matches: a match is finally adopted only if there is
1129 * no better match at the next window position.
1131 local block_state deflate_slow(s, flush)
1135 IPos hash_head = NIL; /* head of hash chain */
1136 int bflush; /* set if current block must be flushed */
1138 /* Process the input block. */
1140 /* Make sure that we always have enough lookahead, except
1141 * at the end of the input file. We need MAX_MATCH bytes
1142 * for the next match, plus MIN_MATCH bytes to insert the
1143 * string following the next match.
1145 if (s->lookahead < MIN_LOOKAHEAD) {
1147 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1150 if (s->lookahead == 0) break; /* flush the current block */
1153 /* Insert the string window[strstart .. strstart+2] in the
1154 * dictionary, and set hash_head to the head of the hash chain:
1156 if (s->lookahead >= MIN_MATCH) {
1157 INSERT_STRING(s, s->strstart, hash_head);
1160 /* Find the longest match, discarding those <= prev_length.
1162 s->prev_length = s->match_length, s->prev_match = s->match_start;
1163 s->match_length = MIN_MATCH-1;
1165 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1166 s->strstart - hash_head <= MAX_DIST(s)) {
1167 /* To simplify the code, we prevent matches with the string
1168 * of window index 0 (in particular we have to avoid a match
1169 * of the string with itself at the start of the input file).
1171 if (s->strategy != Z_HUFFMAN_ONLY) {
1172 s->match_length = longest_match (s, hash_head);
1174 /* longest_match() sets match_start */
1176 if (s->match_length <= 5 && (s->strategy == Z_FILTERED ||
1177 (s->match_length == MIN_MATCH &&
1178 s->strstart - s->match_start > TOO_FAR))) {
1180 /* If prev_match is also MIN_MATCH, match_start is garbage
1181 * but we will ignore the current match anyway.
1183 s->match_length = MIN_MATCH-1;
1186 /* If there was a match at the previous step and the current
1187 * match is not better, output the previous match:
1189 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1190 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1191 /* Do not insert strings in hash table beyond this. */
1193 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1195 bflush = zlib_tr_tally(s, s->strstart -1 - s->prev_match,
1196 s->prev_length - MIN_MATCH);
1198 /* Insert in hash table all strings up to the end of the match.
1199 * strstart-1 and strstart are already inserted. If there is not
1200 * enough lookahead, the last two strings are not inserted in
1203 s->lookahead -= s->prev_length-1;
1204 s->prev_length -= 2;
1206 if (++s->strstart <= max_insert) {
1207 INSERT_STRING(s, s->strstart, hash_head);
1209 } while (--s->prev_length != 0);
1210 s->match_available = 0;
1211 s->match_length = MIN_MATCH-1;
1214 if (bflush) FLUSH_BLOCK(s, 0);
1216 } else if (s->match_available) {
1217 /* If there was no match at the previous position, output a
1218 * single literal. If there was a match but the current match
1219 * is longer, truncate the previous match to a single literal.
1221 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1222 if (zlib_tr_tally (s, 0, s->window[s->strstart-1])) {
1223 FLUSH_BLOCK_ONLY(s, 0);
1227 if (s->strm->avail_out == 0) return need_more;
1229 /* There is no previous match to compare with, wait for
1230 * the next step to decide.
1232 s->match_available = 1;
1237 Assert (flush != Z_NO_FLUSH, "no flush?");
1238 if (s->match_available) {
1239 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1240 zlib_tr_tally (s, 0, s->window[s->strstart-1]);
1241 s->match_available = 0;
1243 FLUSH_BLOCK(s, flush == Z_FINISH);
1244 return flush == Z_FINISH ? finish_done : block_done;
1247 ZEXTERN int ZEXPORT zlib_deflate_workspacesize ()
1249 return sizeof(deflate_workspace);