1 /* deflate.c -- compress data using the deflation algorithm
2 * Copyright (C) 1995-2002 Jean-loup Gailly.
3 * For conditions of distribution and use, see copyright notice in zlib.h
5 * $FreeBSD: src/lib/libz/deflate.c,v 1.5.2.2 2003/02/01 13:33:12 sobomax Exp $
6 * $DragonFly: src/lib/libz/Attic/deflate.c,v 1.2 2003/06/17 04:26:52 dillon Exp $
12 * The "deflation" process depends on being able to identify portions
13 * of the input text which are identical to earlier input (within a
14 * sliding window trailing behind the input currently being processed).
16 * The most straightforward technique turns out to be the fastest for
17 * most input files: try all possible matches and select the longest.
18 * The key feature of this algorithm is that insertions into the string
19 * dictionary are very simple and thus fast, and deletions are avoided
20 * completely. Insertions are performed at each input character, whereas
21 * string matches are performed only when the previous match ends. So it
22 * is preferable to spend more time in matches to allow very fast string
23 * insertions and avoid deletions. The matching algorithm for small
24 * strings is inspired from that of Rabin & Karp. A brute force approach
25 * is used to find longer strings when a small match has been found.
26 * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze
27 * (by Leonid Broukhis).
28 * A previous version of this file used a more sophisticated algorithm
29 * (by Fiala and Greene) which is guaranteed to run in linear amortized
30 * time, but has a larger average cost, uses more memory and is patented.
31 * However the F&G algorithm may be faster for some highly redundant
32 * files if the parameter max_chain_length (described below) is too large.
36 * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and
37 * I found it in 'freeze' written by Leonid Broukhis.
38 * Thanks to many people for bug reports and testing.
42 * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification".
43 * Available in ftp://ds.internic.net/rfc/rfc1951.txt
45 * A description of the Rabin and Karp algorithm is given in the book
46 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
48 * Fiala,E.R., and Greene,D.H.
49 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
53 /* @(#) $FreeBSD: src/lib/libz/deflate.c,v 1.5.2.2 2003/02/01 13:33:12 sobomax Exp $ */
57 const char deflate_copyright[] =
58 " deflate 1.1.4 Copyright 1995-2002 Jean-loup Gailly ";
60 If you use the zlib library in a product, an acknowledgment is welcome
61 in the documentation of your product. If for some reason you cannot
62 include such an acknowledgment, I would appreciate that you keep this
63 copyright string in the executable of your product.
66 /* ===========================================================================
67 * Function prototypes.
70 need_more, /* block not completed, need more input or more output */
71 block_done, /* block flush performed */
72 finish_started, /* finish started, need only more output at next deflate */
73 finish_done /* finish done, accept no more input or output */
76 typedef block_state (*compress_func) OF((deflate_state *s, int flush));
77 /* Compression function. Returns the block state after the call. */
79 local void fill_window OF((deflate_state *s));
80 local block_state deflate_stored OF((deflate_state *s, int flush));
81 local block_state deflate_fast OF((deflate_state *s, int flush));
82 local block_state deflate_slow OF((deflate_state *s, int flush));
83 local void lm_init OF((deflate_state *s));
84 local void putShortMSB OF((deflate_state *s, uInt b));
85 local void flush_pending OF((z_streamp strm));
86 local int read_buf OF((z_streamp strm, Bytef *buf, unsigned size));
88 void match_init OF((void)); /* asm code initialization */
89 uInt longest_match OF((deflate_state *s, IPos cur_match));
91 local uInt longest_match OF((deflate_state *s, IPos cur_match));
95 local void check_match OF((deflate_state *s, IPos start, IPos match,
99 /* ===========================================================================
104 /* Tail of hash chains */
107 # define TOO_FAR 4096
109 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
111 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
112 /* Minimum amount of lookahead, except at the end of the input file.
113 * See deflate.c for comments about the MIN_MATCH+1.
116 /* Values for max_lazy_match, good_match and max_chain_length, depending on
117 * the desired pack level (0..9). The values given below have been tuned to
118 * exclude worst case performance for pathological files. Better values may be
119 * found for specific files.
121 typedef struct config_s {
122 ush good_length; /* reduce lazy search above this match length */
123 ush max_lazy; /* do not perform lazy search above this match length */
124 ush nice_length; /* quit search above this match length */
129 local const config configuration_table[10] = {
130 /* good lazy nice chain */
131 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
132 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* maximum speed, no lazy matches */
133 /* 2 */ {4, 5, 16, 8, deflate_fast},
134 /* 3 */ {4, 6, 32, 32, deflate_fast},
136 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
137 /* 5 */ {8, 16, 32, 32, deflate_slow},
138 /* 6 */ {8, 16, 128, 128, deflate_slow},
139 /* 7 */ {8, 32, 128, 256, deflate_slow},
140 /* 8 */ {32, 128, 258, 1024, deflate_slow},
141 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* maximum compression */
143 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
144 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
149 /* result of memcmp for equal strings */
151 struct static_tree_desc_s {int dummy;}; /* for buggy compilers */
153 /* ===========================================================================
154 * Update a hash value with the given input byte
155 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
156 * input characters, so that a running hash key can be computed from the
157 * previous key instead of complete recalculation each time.
159 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
162 /* ===========================================================================
163 * Insert string str in the dictionary and set match_head to the previous head
164 * of the hash chain (the most recent string with same hash key). Return
165 * the previous length of the hash chain.
166 * If this file is compiled with -DFASTEST, the compression level is forced
167 * to 1, and no hash chains are maintained.
168 * IN assertion: all calls to to INSERT_STRING are made with consecutive
169 * input characters and the first MIN_MATCH bytes of str are valid
170 * (except for the last MIN_MATCH-1 bytes of the input file).
173 #define INSERT_STRING(s, str, match_head) \
174 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
175 match_head = s->head[s->ins_h], \
176 s->head[s->ins_h] = (Pos)(str))
178 #define INSERT_STRING(s, str, match_head) \
179 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
180 s->prev[(str) & s->w_mask] = match_head = s->head[s->ins_h], \
181 s->head[s->ins_h] = (Pos)(str))
184 /* ===========================================================================
185 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
186 * prev[] will be initialized on the fly.
188 #define CLEAR_HASH(s) \
189 s->head[s->hash_size-1] = NIL; \
190 zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
192 /* ========================================================================= */
193 int ZEXPORT deflateInit_(strm, level, version, stream_size)
199 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
200 Z_DEFAULT_STRATEGY, version, stream_size);
201 /* To do: ignore strm->next_in if we use it as window */
204 /* ========================================================================= */
205 int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
206 version, stream_size)
218 static const char* my_version = ZLIB_VERSION;
221 /* We overlay pending_buf and d_buf+l_buf. This works since the average
222 * output size for (length,distance) codes is <= 24 bits.
225 if (version == Z_NULL || version[0] != my_version[0] ||
226 stream_size != sizeof(z_stream)) {
227 return Z_VERSION_ERROR;
229 if (strm == Z_NULL) return Z_STREAM_ERROR;
232 if (strm->zalloc == Z_NULL) {
233 strm->zalloc = zcalloc;
234 strm->opaque = (voidpf)0;
236 if (strm->zfree == Z_NULL) strm->zfree = zcfree;
238 if (level == Z_DEFAULT_COMPRESSION) level = 6;
243 if (windowBits < 0) { /* undocumented feature: suppress zlib header */
245 windowBits = -windowBits;
247 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
248 windowBits < 9 || windowBits > 15 || level < 0 || level > 9 ||
249 strategy < 0 || strategy > Z_HUFFMAN_ONLY) {
250 return Z_STREAM_ERROR;
252 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
253 if (s == Z_NULL) return Z_MEM_ERROR;
254 strm->state = (struct internal_state FAR *)s;
257 s->noheader = noheader;
258 s->w_bits = windowBits;
259 s->w_size = 1 << s->w_bits;
260 s->w_mask = s->w_size - 1;
262 s->hash_bits = memLevel + 7;
263 s->hash_size = 1 << s->hash_bits;
264 s->hash_mask = s->hash_size - 1;
265 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
267 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
268 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));
269 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
271 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
273 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
274 s->pending_buf = (uchf *) overlay;
275 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
277 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
278 s->pending_buf == Z_NULL) {
279 strm->msg = (char*)ERR_MSG(Z_MEM_ERROR);
283 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
284 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
287 s->strategy = strategy;
288 s->method = (Byte)method;
290 return deflateReset(strm);
293 /* ========================================================================= */
294 int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
296 const Bytef *dictionary;
300 uInt length = dictLength;
304 if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL ||
305 strm->state->status != INIT_STATE) return Z_STREAM_ERROR;
308 strm->adler = adler32(strm->adler, dictionary, dictLength);
310 if (length < MIN_MATCH) return Z_OK;
311 if (length > MAX_DIST(s)) {
312 length = MAX_DIST(s);
313 #ifndef USE_DICT_HEAD
314 dictionary += dictLength - length; /* use the tail of the dictionary */
317 zmemcpy(s->window, dictionary, length);
318 s->strstart = length;
319 s->block_start = (long)length;
321 /* Insert all strings in the hash table (except for the last two bytes).
322 * s->lookahead stays null, so s->ins_h will be recomputed at the next
323 * call of fill_window.
325 s->ins_h = s->window[0];
326 UPDATE_HASH(s, s->ins_h, s->window[1]);
327 for (n = 0; n <= length - MIN_MATCH; n++) {
328 INSERT_STRING(s, n, hash_head);
330 if (hash_head) hash_head = 0; /* to make compiler happy */
334 /* ========================================================================= */
335 int ZEXPORT deflateReset (strm)
340 if (strm == Z_NULL || strm->state == Z_NULL ||
341 strm->zalloc == Z_NULL || strm->zfree == Z_NULL) return Z_STREAM_ERROR;
343 strm->total_in = strm->total_out = 0;
344 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
345 strm->data_type = Z_UNKNOWN;
347 s = (deflate_state *)strm->state;
349 s->pending_out = s->pending_buf;
351 if (s->noheader < 0) {
352 s->noheader = 0; /* was set to -1 by deflate(..., Z_FINISH); */
354 s->status = s->noheader ? BUSY_STATE : INIT_STATE;
356 s->last_flush = Z_NO_FLUSH;
364 /* ========================================================================= */
365 int ZEXPORT deflateParams(strm, level, strategy)
374 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
377 if (level == Z_DEFAULT_COMPRESSION) {
380 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_HUFFMAN_ONLY) {
381 return Z_STREAM_ERROR;
383 func = configuration_table[s->level].func;
385 if (func != configuration_table[level].func && strm->total_in != 0) {
386 /* Flush the last buffer: */
387 err = deflate(strm, Z_PARTIAL_FLUSH);
389 if (s->level != level) {
391 s->max_lazy_match = configuration_table[level].max_lazy;
392 s->good_match = configuration_table[level].good_length;
393 s->nice_match = configuration_table[level].nice_length;
394 s->max_chain_length = configuration_table[level].max_chain;
396 s->strategy = strategy;
400 /* =========================================================================
401 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
402 * IN assertion: the stream state is correct and there is enough room in
405 local void putShortMSB (s, b)
409 put_byte(s, (Byte)(b >> 8));
410 put_byte(s, (Byte)(b & 0xff));
413 /* =========================================================================
414 * Flush as much pending output as possible. All deflate() output goes
415 * through this function so some applications may wish to modify it
416 * to avoid allocating a large strm->next_out buffer and copying into it.
417 * (See also read_buf()).
419 local void flush_pending(strm)
422 unsigned len = strm->state->pending;
424 if (len > strm->avail_out) len = strm->avail_out;
425 if (len == 0) return;
427 zmemcpy(strm->next_out, strm->state->pending_out, len);
428 strm->next_out += len;
429 strm->state->pending_out += len;
430 strm->total_out += len;
431 strm->avail_out -= len;
432 strm->state->pending -= len;
433 if (strm->state->pending == 0) {
434 strm->state->pending_out = strm->state->pending_buf;
438 /* ========================================================================= */
439 int ZEXPORT deflate (strm, flush)
443 int old_flush; /* value of flush param for previous deflate call */
446 if (strm == Z_NULL || strm->state == Z_NULL ||
447 flush > Z_FINISH || flush < 0) {
448 return Z_STREAM_ERROR;
452 if (strm->next_out == Z_NULL ||
453 (strm->next_in == Z_NULL && strm->avail_in != 0) ||
454 (s->status == FINISH_STATE && flush != Z_FINISH)) {
455 ERR_RETURN(strm, Z_STREAM_ERROR);
457 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
459 s->strm = strm; /* just in case */
460 old_flush = s->last_flush;
461 s->last_flush = flush;
463 /* Write the zlib header */
464 if (s->status == INIT_STATE) {
466 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
467 uInt level_flags = (s->level-1) >> 1;
469 if (level_flags > 3) level_flags = 3;
470 header |= (level_flags << 6);
471 if (s->strstart != 0) header |= PRESET_DICT;
472 header += 31 - (header % 31);
474 s->status = BUSY_STATE;
475 putShortMSB(s, header);
477 /* Save the adler32 of the preset dictionary: */
478 if (s->strstart != 0) {
479 putShortMSB(s, (uInt)(strm->adler >> 16));
480 putShortMSB(s, (uInt)(strm->adler & 0xffff));
485 /* Flush as much pending output as possible */
486 if (s->pending != 0) {
488 if (strm->avail_out == 0) {
489 /* Since avail_out is 0, deflate will be called again with
490 * more output space, but possibly with both pending and
491 * avail_in equal to zero. There won't be anything to do,
492 * but this is not an error situation so make sure we
493 * return OK instead of BUF_ERROR at next call of deflate:
499 /* Make sure there is something to do and avoid duplicate consecutive
500 * flushes. For repeated and useless calls with Z_FINISH, we keep
501 * returning Z_STREAM_END instead of Z_BUFF_ERROR.
503 } else if (strm->avail_in == 0 && flush <= old_flush &&
505 ERR_RETURN(strm, Z_BUF_ERROR);
508 /* User must not provide more input after the first FINISH: */
509 if (s->status == FINISH_STATE && strm->avail_in != 0) {
510 ERR_RETURN(strm, Z_BUF_ERROR);
513 /* Start a new block or continue the current one.
515 if (strm->avail_in != 0 || s->lookahead != 0 ||
516 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
519 bstate = (*(configuration_table[s->level].func))(s, flush);
521 if (bstate == finish_started || bstate == finish_done) {
522 s->status = FINISH_STATE;
524 if (bstate == need_more || bstate == finish_started) {
525 if (strm->avail_out == 0) {
526 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
529 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
530 * of deflate should use the same flush parameter to make sure
531 * that the flush is complete. So we don't have to output an
532 * empty block here, this will be done at next call. This also
533 * ensures that for a very small output buffer, we emit at most
537 if (bstate == block_done) {
538 if (flush == Z_PARTIAL_FLUSH) {
540 } else { /* FULL_FLUSH or SYNC_FLUSH */
541 _tr_stored_block(s, (char*)0, 0L, 0);
542 /* For a full flush, this empty block will be recognized
543 * as a special marker by inflate_sync().
545 if (flush == Z_FULL_FLUSH) {
546 CLEAR_HASH(s); /* forget history */
550 if (strm->avail_out == 0) {
551 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
556 Assert(strm->avail_out > 0, "bug2");
558 if (flush != Z_FINISH) return Z_OK;
559 if (s->noheader) return Z_STREAM_END;
561 /* Write the zlib trailer (adler32) */
562 putShortMSB(s, (uInt)(strm->adler >> 16));
563 putShortMSB(s, (uInt)(strm->adler & 0xffff));
565 /* If avail_out is zero, the application will call deflate again
568 s->noheader = -1; /* write the trailer only once! */
569 return s->pending != 0 ? Z_OK : Z_STREAM_END;
572 /* ========================================================================= */
573 int ZEXPORT deflateEnd (strm)
578 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
580 status = strm->state->status;
581 if (status != INIT_STATE && status != BUSY_STATE &&
582 status != FINISH_STATE) {
583 return Z_STREAM_ERROR;
586 /* Deallocate in reverse order of allocations: */
587 TRY_FREE(strm, strm->state->pending_buf);
588 TRY_FREE(strm, strm->state->head);
589 TRY_FREE(strm, strm->state->prev);
590 TRY_FREE(strm, strm->state->window);
592 ZFREE(strm, strm->state);
593 strm->state = Z_NULL;
595 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
598 /* =========================================================================
599 * Copy the source state to the destination state.
600 * To simplify the source, this is not supported for 16-bit MSDOS (which
601 * doesn't have enough memory anyway to duplicate compression states).
603 int ZEXPORT deflateCopy (dest, source)
608 return Z_STREAM_ERROR;
615 if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) {
616 return Z_STREAM_ERROR;
623 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
624 if (ds == Z_NULL) return Z_MEM_ERROR;
625 dest->state = (struct internal_state FAR *) ds;
629 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
630 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));
631 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));
632 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
633 ds->pending_buf = (uchf *) overlay;
635 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
636 ds->pending_buf == Z_NULL) {
640 /* following zmemcpy do not work for 16-bit MSDOS */
641 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
642 zmemcpy(ds->prev, ss->prev, ds->w_size * sizeof(Pos));
643 zmemcpy(ds->head, ss->head, ds->hash_size * sizeof(Pos));
644 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
646 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
647 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
648 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
650 ds->l_desc.dyn_tree = ds->dyn_ltree;
651 ds->d_desc.dyn_tree = ds->dyn_dtree;
652 ds->bl_desc.dyn_tree = ds->bl_tree;
658 /* ===========================================================================
659 * Read a new buffer from the current input stream, update the adler32
660 * and total number of bytes read. All deflate() input goes through
661 * this function so some applications may wish to modify it to avoid
662 * allocating a large strm->next_in buffer and copying from it.
663 * (See also flush_pending()).
665 local int read_buf(strm, buf, size)
670 unsigned len = strm->avail_in;
672 if (len > size) len = size;
673 if (len == 0) return 0;
675 strm->avail_in -= len;
677 if (!strm->state->noheader) {
678 strm->adler = adler32(strm->adler, strm->next_in, len);
680 zmemcpy(buf, strm->next_in, len);
681 strm->next_in += len;
682 strm->total_in += len;
687 /* ===========================================================================
688 * Initialize the "longest match" routines for a new zlib stream
690 local void lm_init (s)
693 s->window_size = (ulg)2L*s->w_size;
697 /* Set the default configuration parameters:
699 s->max_lazy_match = configuration_table[s->level].max_lazy;
700 s->good_match = configuration_table[s->level].good_length;
701 s->nice_match = configuration_table[s->level].nice_length;
702 s->max_chain_length = configuration_table[s->level].max_chain;
707 s->match_length = s->prev_length = MIN_MATCH-1;
708 s->match_available = 0;
711 match_init(); /* initialize the asm code */
715 /* ===========================================================================
716 * Set match_start to the longest match starting at the given string and
717 * return its length. Matches shorter or equal to prev_length are discarded,
718 * in which case the result is equal to prev_length and match_start is
720 * IN assertions: cur_match is the head of the hash chain for the current
721 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
722 * OUT assertion: the match length is not greater than s->lookahead.
725 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
726 * match.S. The code will be functionally equivalent.
729 local uInt longest_match(s, cur_match)
731 IPos cur_match; /* current match */
733 unsigned chain_length = s->max_chain_length;/* max hash chain length */
734 register Bytef *scan = s->window + s->strstart; /* current string */
735 register Bytef *match; /* matched string */
736 register int len; /* length of current match */
737 int best_len = s->prev_length; /* best match length so far */
738 int nice_match = s->nice_match; /* stop if match long enough */
739 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
740 s->strstart - (IPos)MAX_DIST(s) : NIL;
741 /* Stop when cur_match becomes <= limit. To simplify the code,
742 * we prevent matches with the string of window index 0.
744 Posf *prev = s->prev;
745 uInt wmask = s->w_mask;
748 /* Compare two bytes at a time. Note: this is not always beneficial.
749 * Try with and without -DUNALIGNED_OK to check.
751 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
752 register ush scan_start = *(ushf*)scan;
753 register ush scan_end = *(ushf*)(scan+best_len-1);
755 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
756 register Byte scan_end1 = scan[best_len-1];
757 register Byte scan_end = scan[best_len];
760 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
761 * It is easy to get rid of this optimization if necessary.
763 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
765 /* Do not waste too much time if we already have a good match: */
766 if (s->prev_length >= s->good_match) {
769 /* Do not look for matches beyond the end of the input. This is necessary
770 * to make deflate deterministic.
772 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
774 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
777 Assert(cur_match < s->strstart, "no future");
778 match = s->window + cur_match;
780 /* Skip to next match if the match length cannot increase
781 * or if the match length is less than 2:
783 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
784 /* This code assumes sizeof(unsigned short) == 2. Do not use
785 * UNALIGNED_OK if your compiler uses a different size.
787 if (*(ushf*)(match+best_len-1) != scan_end ||
788 *(ushf*)match != scan_start) continue;
790 /* It is not necessary to compare scan[2] and match[2] since they are
791 * always equal when the other bytes match, given that the hash keys
792 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
793 * strstart+3, +5, ... up to strstart+257. We check for insufficient
794 * lookahead only every 4th comparison; the 128th check will be made
795 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
796 * necessary to put more guard bytes at the end of the window, or
797 * to check more often for insufficient lookahead.
799 Assert(scan[2] == match[2], "scan[2]?");
802 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
803 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
804 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
805 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
807 /* The funny "do {}" generates better code on most compilers */
809 /* Here, scan <= window+strstart+257 */
810 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
811 if (*scan == *match) scan++;
813 len = (MAX_MATCH - 1) - (int)(strend-scan);
814 scan = strend - (MAX_MATCH-1);
816 #else /* UNALIGNED_OK */
818 if (match[best_len] != scan_end ||
819 match[best_len-1] != scan_end1 ||
821 *++match != scan[1]) continue;
823 /* The check at best_len-1 can be removed because it will be made
824 * again later. (This heuristic is not always a win.)
825 * It is not necessary to compare scan[2] and match[2] since they
826 * are always equal when the other bytes match, given that
827 * the hash keys are equal and that HASH_BITS >= 8.
830 Assert(*scan == *match, "match[2]?");
832 /* We check for insufficient lookahead only every 8th comparison;
833 * the 256th check will be made at strstart+258.
836 } while (*++scan == *++match && *++scan == *++match &&
837 *++scan == *++match && *++scan == *++match &&
838 *++scan == *++match && *++scan == *++match &&
839 *++scan == *++match && *++scan == *++match &&
842 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
844 len = MAX_MATCH - (int)(strend - scan);
845 scan = strend - MAX_MATCH;
847 #endif /* UNALIGNED_OK */
849 if (len > best_len) {
850 s->match_start = cur_match;
852 if (len >= nice_match) break;
854 scan_end = *(ushf*)(scan+best_len-1);
856 scan_end1 = scan[best_len-1];
857 scan_end = scan[best_len];
860 } while ((cur_match = prev[cur_match & wmask]) > limit
861 && --chain_length != 0);
863 if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
868 /* ---------------------------------------------------------------------------
869 * Optimized version for level == 1 only
871 local uInt longest_match(s, cur_match)
873 IPos cur_match; /* current match */
875 register Bytef *scan = s->window + s->strstart; /* current string */
876 register Bytef *match; /* matched string */
877 register int len; /* length of current match */
878 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
880 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
881 * It is easy to get rid of this optimization if necessary.
883 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
885 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
887 Assert(cur_match < s->strstart, "no future");
889 match = s->window + cur_match;
891 /* Return failure if the match length is less than 2:
893 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
895 /* The check at best_len-1 can be removed because it will be made
896 * again later. (This heuristic is not always a win.)
897 * It is not necessary to compare scan[2] and match[2] since they
898 * are always equal when the other bytes match, given that
899 * the hash keys are equal and that HASH_BITS >= 8.
901 scan += 2, match += 2;
902 Assert(*scan == *match, "match[2]?");
904 /* We check for insufficient lookahead only every 8th comparison;
905 * the 256th check will be made at strstart+258.
908 } while (*++scan == *++match && *++scan == *++match &&
909 *++scan == *++match && *++scan == *++match &&
910 *++scan == *++match && *++scan == *++match &&
911 *++scan == *++match && *++scan == *++match &&
914 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
916 len = MAX_MATCH - (int)(strend - scan);
918 if (len < MIN_MATCH) return MIN_MATCH - 1;
920 s->match_start = cur_match;
921 return len <= s->lookahead ? len : s->lookahead;
927 /* ===========================================================================
928 * Check that the match at match_start is indeed a match.
930 local void check_match(s, start, match, length)
935 /* check that the match is indeed a match */
936 if (zmemcmp(s->window + match,
937 s->window + start, length) != EQUAL) {
938 fprintf(stderr, " start %u, match %u, length %d\n",
939 start, match, length);
941 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
942 } while (--length != 0);
943 z_error("invalid match");
946 fprintf(stderr,"\\[%d,%d]", start-match, length);
947 do { putc(s->window[start++], stderr); } while (--length != 0);
951 # define check_match(s, start, match, length)
954 /* ===========================================================================
955 * Fill the window when the lookahead becomes insufficient.
956 * Updates strstart and lookahead.
958 * IN assertion: lookahead < MIN_LOOKAHEAD
959 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
960 * At least one byte has been read, or avail_in == 0; reads are
961 * performed for at least two bytes (required for the zip translate_eol
962 * option -- not supported here).
964 local void fill_window(s)
967 register unsigned n, m;
969 unsigned more; /* Amount of free space at the end of the window. */
970 uInt wsize = s->w_size;
973 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
975 /* Deal with !@#$% 64K limit: */
976 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
979 } else if (more == (unsigned)(-1)) {
980 /* Very unlikely, but possible on 16 bit machine if strstart == 0
981 * and lookahead == 1 (input done one byte at time)
985 /* If the window is almost full and there is insufficient lookahead,
986 * move the upper half to the lower one to make room in the upper half.
988 } else if (s->strstart >= wsize+MAX_DIST(s)) {
990 zmemcpy(s->window, s->window+wsize, (unsigned)wsize);
991 s->match_start -= wsize;
992 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
993 s->block_start -= (long) wsize;
995 /* Slide the hash table (could be avoided with 32 bit values
996 at the expense of memory usage). We slide even when level == 0
997 to keep the hash table consistent if we switch back to level > 0
998 later. (Using level 0 permanently is not an optimal usage of
999 zlib, so we don't care about this pathological case.)
1005 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1013 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1014 /* If n is not on any hash chain, prev[n] is garbage but
1015 * its value will never be used.
1021 if (s->strm->avail_in == 0) return;
1023 /* If there was no sliding:
1024 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1025 * more == window_size - lookahead - strstart
1026 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1027 * => more >= window_size - 2*WSIZE + 2
1028 * In the BIG_MEM or MMAP case (not yet supported),
1029 * window_size == input_size + MIN_LOOKAHEAD &&
1030 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1031 * Otherwise, window_size == 2*WSIZE so more >= 2.
1032 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1034 Assert(more >= 2, "more < 2");
1036 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1039 /* Initialize the hash value now that we have some input: */
1040 if (s->lookahead >= MIN_MATCH) {
1041 s->ins_h = s->window[s->strstart];
1042 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1044 Call UPDATE_HASH() MIN_MATCH-3 more times
1047 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1048 * but this is not important since only literal bytes will be emitted.
1051 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1054 /* ===========================================================================
1055 * Flush the current block, with given end-of-file flag.
1056 * IN assertion: strstart is set to the end of the current match.
1058 #define FLUSH_BLOCK_ONLY(s, eof) { \
1059 _tr_flush_block(s, (s->block_start >= 0L ? \
1060 (charf *)&s->window[(unsigned)s->block_start] : \
1062 (ulg)((long)s->strstart - s->block_start), \
1064 s->block_start = s->strstart; \
1065 flush_pending(s->strm); \
1066 Tracev((stderr,"[FLUSH]")); \
1069 /* Same but force premature exit if necessary. */
1070 #define FLUSH_BLOCK(s, eof) { \
1071 FLUSH_BLOCK_ONLY(s, eof); \
1072 if (s->strm->avail_out == 0) return (eof) ? finish_started : need_more; \
1075 /* ===========================================================================
1076 * Copy without compression as much as possible from the input stream, return
1077 * the current block state.
1078 * This function does not insert new strings in the dictionary since
1079 * uncompressible data is probably not useful. This function is used
1080 * only for the level=0 compression option.
1081 * NOTE: this function should be optimized to avoid extra copying from
1082 * window to pending_buf.
1084 local block_state deflate_stored(s, flush)
1088 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited
1089 * to pending_buf_size, and each stored block has a 5 byte header:
1091 ulg max_block_size = 0xffff;
1094 if (max_block_size > s->pending_buf_size - 5) {
1095 max_block_size = s->pending_buf_size - 5;
1098 /* Copy as much as possible from input to output: */
1100 /* Fill the window as much as possible: */
1101 if (s->lookahead <= 1) {
1103 Assert(s->strstart < s->w_size+MAX_DIST(s) ||
1104 s->block_start >= (long)s->w_size, "slide too late");
1107 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
1109 if (s->lookahead == 0) break; /* flush the current block */
1111 Assert(s->block_start >= 0L, "block gone");
1113 s->strstart += s->lookahead;
1116 /* Emit a stored block if pending_buf will be full: */
1117 max_start = s->block_start + max_block_size;
1118 if (s->strstart == 0 || (ulg)s->strstart >= max_start) {
1119 /* strstart == 0 is possible when wraparound on 16-bit machine */
1120 s->lookahead = (uInt)(s->strstart - max_start);
1121 s->strstart = (uInt)max_start;
1124 /* Flush if we may have to slide, otherwise block_start may become
1125 * negative and the data will be gone:
1127 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
1131 FLUSH_BLOCK(s, flush == Z_FINISH);
1132 return flush == Z_FINISH ? finish_done : block_done;
1135 /* ===========================================================================
1136 * Compress as much as possible from the input stream, return the current
1138 * This function does not perform lazy evaluation of matches and inserts
1139 * new strings in the dictionary only for unmatched strings or for short
1140 * matches. It is used only for the fast compression options.
1142 local block_state deflate_fast(s, flush)
1146 IPos hash_head = NIL; /* head of the hash chain */
1147 int bflush; /* set if current block must be flushed */
1150 /* Make sure that we always have enough lookahead, except
1151 * at the end of the input file. We need MAX_MATCH bytes
1152 * for the next match, plus MIN_MATCH bytes to insert the
1153 * string following the next match.
1155 if (s->lookahead < MIN_LOOKAHEAD) {
1157 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1160 if (s->lookahead == 0) break; /* flush the current block */
1163 /* Insert the string window[strstart .. strstart+2] in the
1164 * dictionary, and set hash_head to the head of the hash chain:
1166 if (s->lookahead >= MIN_MATCH) {
1167 INSERT_STRING(s, s->strstart, hash_head);
1170 /* Find the longest match, discarding those <= prev_length.
1171 * At this point we have always match_length < MIN_MATCH
1173 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1174 /* To simplify the code, we prevent matches with the string
1175 * of window index 0 (in particular we have to avoid a match
1176 * of the string with itself at the start of the input file).
1178 if (s->strategy != Z_HUFFMAN_ONLY) {
1179 s->match_length = longest_match (s, hash_head);
1181 /* longest_match() sets match_start */
1183 if (s->match_length >= MIN_MATCH) {
1184 check_match(s, s->strstart, s->match_start, s->match_length);
1186 _tr_tally_dist(s, s->strstart - s->match_start,
1187 s->match_length - MIN_MATCH, bflush);
1189 s->lookahead -= s->match_length;
1191 /* Insert new strings in the hash table only if the match length
1192 * is not too large. This saves time but degrades compression.
1195 if (s->match_length <= s->max_insert_length &&
1196 s->lookahead >= MIN_MATCH) {
1197 s->match_length--; /* string at strstart already in hash table */
1200 INSERT_STRING(s, s->strstart, hash_head);
1201 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1202 * always MIN_MATCH bytes ahead.
1204 } while (--s->match_length != 0);
1209 s->strstart += s->match_length;
1210 s->match_length = 0;
1211 s->ins_h = s->window[s->strstart];
1212 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1214 Call UPDATE_HASH() MIN_MATCH-3 more times
1216 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1217 * matter since it will be recomputed at next deflate call.
1221 /* No match, output a literal byte */
1222 Tracevv((stderr,"%c", s->window[s->strstart]));
1223 _tr_tally_lit (s, s->window[s->strstart], bflush);
1227 if (bflush) FLUSH_BLOCK(s, 0);
1229 FLUSH_BLOCK(s, flush == Z_FINISH);
1230 return flush == Z_FINISH ? finish_done : block_done;
1233 /* ===========================================================================
1234 * Same as above, but achieves better compression. We use a lazy
1235 * evaluation for matches: a match is finally adopted only if there is
1236 * no better match at the next window position.
1238 local block_state deflate_slow(s, flush)
1242 IPos hash_head = NIL; /* head of hash chain */
1243 int bflush; /* set if current block must be flushed */
1245 /* Process the input block. */
1247 /* Make sure that we always have enough lookahead, except
1248 * at the end of the input file. We need MAX_MATCH bytes
1249 * for the next match, plus MIN_MATCH bytes to insert the
1250 * string following the next match.
1252 if (s->lookahead < MIN_LOOKAHEAD) {
1254 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1257 if (s->lookahead == 0) break; /* flush the current block */
1260 /* Insert the string window[strstart .. strstart+2] in the
1261 * dictionary, and set hash_head to the head of the hash chain:
1263 if (s->lookahead >= MIN_MATCH) {
1264 INSERT_STRING(s, s->strstart, hash_head);
1267 /* Find the longest match, discarding those <= prev_length.
1269 s->prev_length = s->match_length, s->prev_match = s->match_start;
1270 s->match_length = MIN_MATCH-1;
1272 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1273 s->strstart - hash_head <= MAX_DIST(s)) {
1274 /* To simplify the code, we prevent matches with the string
1275 * of window index 0 (in particular we have to avoid a match
1276 * of the string with itself at the start of the input file).
1278 if (s->strategy != Z_HUFFMAN_ONLY) {
1279 s->match_length = longest_match (s, hash_head);
1281 /* longest_match() sets match_start */
1283 if (s->match_length <= 5 && (s->strategy == Z_FILTERED ||
1284 (s->match_length == MIN_MATCH &&
1285 s->strstart - s->match_start > TOO_FAR))) {
1287 /* If prev_match is also MIN_MATCH, match_start is garbage
1288 * but we will ignore the current match anyway.
1290 s->match_length = MIN_MATCH-1;
1293 /* If there was a match at the previous step and the current
1294 * match is not better, output the previous match:
1296 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1297 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1298 /* Do not insert strings in hash table beyond this. */
1300 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1302 _tr_tally_dist(s, s->strstart -1 - s->prev_match,
1303 s->prev_length - MIN_MATCH, bflush);
1305 /* Insert in hash table all strings up to the end of the match.
1306 * strstart-1 and strstart are already inserted. If there is not
1307 * enough lookahead, the last two strings are not inserted in
1310 s->lookahead -= s->prev_length-1;
1311 s->prev_length -= 2;
1313 if (++s->strstart <= max_insert) {
1314 INSERT_STRING(s, s->strstart, hash_head);
1316 } while (--s->prev_length != 0);
1317 s->match_available = 0;
1318 s->match_length = MIN_MATCH-1;
1321 if (bflush) FLUSH_BLOCK(s, 0);
1323 } else if (s->match_available) {
1324 /* If there was no match at the previous position, output a
1325 * single literal. If there was a match but the current match
1326 * is longer, truncate the previous match to a single literal.
1328 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1329 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1331 FLUSH_BLOCK_ONLY(s, 0);
1335 if (s->strm->avail_out == 0) return need_more;
1337 /* There is no previous match to compare with, wait for
1338 * the next step to decide.
1340 s->match_available = 1;
1345 Assert (flush != Z_NO_FLUSH, "no flush?");
1346 if (s->match_available) {
1347 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1348 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1349 s->match_available = 0;
1351 FLUSH_BLOCK(s, flush == Z_FINISH);
1352 return flush == Z_FINISH ? finish_done : block_done;