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
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://ds.internic.net/rfc/rfc1951.txt
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
54 #define read_buf dread_buf
56 const char deflate_copyright[] =
57 " deflate 1.1.4 Copyright 1995-2002 Jean-loup Gailly ";
59 If you use the zlib library in a product, an acknowledgment is welcome
60 in the documentation of your product. If for some reason you cannot
61 include such an acknowledgment, I would appreciate that you keep this
62 copyright string in the executable of your product.
65 /* ===========================================================================
66 * Function prototypes.
69 need_more, /* block not completed, need more input or more output */
70 block_done, /* block flush performed */
71 finish_started, /* finish started, need only more output at next deflate */
72 finish_done /* finish done, accept no more input or output */
75 typedef block_state (*compress_func) OF((deflate_state *s, int flush));
76 /* Compression function. Returns the block state after the call. */
78 local void fill_window OF((deflate_state *s));
79 local block_state deflate_stored OF((deflate_state *s, int flush));
80 local block_state deflate_fast OF((deflate_state *s, int flush));
81 local block_state deflate_slow OF((deflate_state *s, int flush));
82 local void lm_init OF((deflate_state *s));
83 local void putShortMSB OF((deflate_state *s, uInt b));
84 local void flush_pending OF((z_streamp strm));
85 local int read_buf OF((z_streamp strm, Bytef *buf, unsigned size));
87 void match_init OF((void)); /* asm code initialization */
88 uInt longest_match OF((deflate_state *s, IPos cur_match));
90 local uInt longest_match OF((deflate_state *s, IPos cur_match));
94 local void check_match OF((deflate_state *s, IPos start, IPos match,
98 /* ===========================================================================
103 /* Tail of hash chains */
106 # define TOO_FAR 4096
108 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
110 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
111 /* Minimum amount of lookahead, except at the end of the input file.
112 * See deflate.c for comments about the MIN_MATCH+1.
115 /* Values for max_lazy_match, good_match and max_chain_length, depending on
116 * the desired pack level (0..9). The values given below have been tuned to
117 * exclude worst case performance for pathological files. Better values may be
118 * found for specific files.
120 typedef struct config_s {
121 ush good_length; /* reduce lazy search above this match length */
122 ush max_lazy; /* do not perform lazy search above this match length */
123 ush nice_length; /* quit search above this match length */
128 local const config configuration_table[10] = {
129 /* good lazy nice chain */
130 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
131 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* maximum speed, no lazy matches */
132 /* 2 */ {4, 5, 16, 8, deflate_fast},
133 /* 3 */ {4, 6, 32, 32, deflate_fast},
135 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
136 /* 5 */ {8, 16, 32, 32, deflate_slow},
137 /* 6 */ {8, 16, 128, 128, deflate_slow},
138 /* 7 */ {8, 32, 128, 256, deflate_slow},
139 /* 8 */ {32, 128, 258, 1024, deflate_slow},
140 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* maximum compression */
142 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
143 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
148 /* result of memcmp for equal strings */
150 struct static_tree_desc_s {int dummy;}; /* for buggy compilers */
152 /* ===========================================================================
153 * Update a hash value with the given input byte
154 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
155 * input characters, so that a running hash key can be computed from the
156 * previous key instead of complete recalculation each time.
158 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
161 /* ===========================================================================
162 * Insert string str in the dictionary and set match_head to the previous head
163 * of the hash chain (the most recent string with same hash key). Return
164 * the previous length of the hash chain.
165 * If this file is compiled with -DFASTEST, the compression level is forced
166 * to 1, and no hash chains are maintained.
167 * IN assertion: all calls to to INSERT_STRING are made with consecutive
168 * input characters and the first MIN_MATCH bytes of str are valid
169 * (except for the last MIN_MATCH-1 bytes of the input file).
172 #define INSERT_STRING(s, str, match_head) \
173 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
174 match_head = s->head[s->ins_h], \
175 s->head[s->ins_h] = (Pos)(str))
177 #define INSERT_STRING(s, str, match_head) \
178 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
179 s->prev[(str) & s->w_mask] = match_head = s->head[s->ins_h], \
180 s->head[s->ins_h] = (Pos)(str))
183 /* ===========================================================================
184 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
185 * prev[] will be initialized on the fly.
187 #define CLEAR_HASH(s) \
188 s->head[s->hash_size-1] = NIL; \
189 zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
191 /* ========================================================================= */
192 int ZEXPORT deflateInit_(strm, level, version, stream_size)
198 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
199 Z_DEFAULT_STRATEGY, version, stream_size);
200 /* To do: ignore strm->next_in if we use it as window */
203 /* ========================================================================= */
204 int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
205 version, stream_size)
217 static const char* my_version = ZLIB_VERSION;
220 /* We overlay pending_buf and d_buf+l_buf. This works since the average
221 * output size for (length,distance) codes is <= 24 bits.
224 if (version == Z_NULL || version[0] != my_version[0] ||
225 stream_size != sizeof(z_stream)) {
226 return Z_VERSION_ERROR;
228 if (strm == Z_NULL) return Z_STREAM_ERROR;
231 if (strm->zalloc == Z_NULL) {
232 strm->zalloc = zcalloc;
233 strm->opaque = (voidpf)0;
235 if (strm->zfree == Z_NULL) strm->zfree = zcfree;
237 if (level == Z_DEFAULT_COMPRESSION) level = 6;
242 if (windowBits < 0) { /* undocumented feature: suppress zlib header */
244 windowBits = -windowBits;
246 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
247 windowBits < 9 || windowBits > 15 || level < 0 || level > 9 ||
248 strategy < 0 || strategy > Z_HUFFMAN_ONLY) {
249 return Z_STREAM_ERROR;
251 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
252 if (s == Z_NULL) return Z_MEM_ERROR;
253 strm->state = (struct internal_state FAR *)s;
256 s->noheader = noheader;
257 s->w_bits = windowBits;
258 s->w_size = 1 << s->w_bits;
259 s->w_mask = s->w_size - 1;
261 s->hash_bits = memLevel + 7;
262 s->hash_size = 1 << s->hash_bits;
263 s->hash_mask = s->hash_size - 1;
264 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
266 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
267 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));
268 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
270 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
272 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
273 s->pending_buf = (uchf *) overlay;
274 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
276 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
277 s->pending_buf == Z_NULL) {
278 strm->msg = (char*)ERR_MSG(Z_MEM_ERROR);
282 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
283 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
286 s->strategy = strategy;
287 s->method = (Byte)method;
289 return deflateReset(strm);
292 /* ========================================================================= */
293 int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
295 const Bytef *dictionary;
299 uInt length = dictLength;
303 if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL ||
304 strm->state->status != INIT_STATE) return Z_STREAM_ERROR;
307 strm->adler = adler32(strm->adler, dictionary, dictLength);
309 if (length < MIN_MATCH) return Z_OK;
310 if (length > MAX_DIST(s)) {
311 length = MAX_DIST(s);
312 #ifndef USE_DICT_HEAD
313 dictionary += dictLength - length; /* use the tail of the dictionary */
316 zmemcpy(s->window, dictionary, length);
317 s->strstart = length;
318 s->block_start = (long)length;
320 /* Insert all strings in the hash table (except for the last two bytes).
321 * s->lookahead stays null, so s->ins_h will be recomputed at the next
322 * call of fill_window.
324 s->ins_h = s->window[0];
325 UPDATE_HASH(s, s->ins_h, s->window[1]);
326 for (n = 0; n <= length - MIN_MATCH; n++) {
327 INSERT_STRING(s, n, hash_head);
329 if (hash_head) hash_head = 0; /* to make compiler happy */
333 /* ========================================================================= */
334 int ZEXPORT deflateReset (strm)
339 if (strm == Z_NULL || strm->state == Z_NULL ||
340 strm->zalloc == Z_NULL || strm->zfree == Z_NULL) return Z_STREAM_ERROR;
342 strm->total_in = strm->total_out = 0;
343 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
344 strm->data_type = Z_UNKNOWN;
346 s = (deflate_state *)strm->state;
348 s->pending_out = s->pending_buf;
350 if (s->noheader < 0) {
351 s->noheader = 0; /* was set to -1 by deflate(..., Z_FINISH); */
353 s->status = s->noheader ? BUSY_STATE : INIT_STATE;
355 s->last_flush = Z_NO_FLUSH;
363 /* ========================================================================= */
364 int ZEXPORT deflateParams(strm, level, strategy)
373 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
376 if (level == Z_DEFAULT_COMPRESSION) {
379 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_HUFFMAN_ONLY) {
380 return Z_STREAM_ERROR;
382 func = configuration_table[s->level].func;
384 if (func != configuration_table[level].func && strm->total_in != 0) {
385 /* Flush the last buffer: */
386 err = deflate(strm, Z_PARTIAL_FLUSH);
388 if (s->level != level) {
390 s->max_lazy_match = configuration_table[level].max_lazy;
391 s->good_match = configuration_table[level].good_length;
392 s->nice_match = configuration_table[level].nice_length;
393 s->max_chain_length = configuration_table[level].max_chain;
395 s->strategy = strategy;
399 /* =========================================================================
400 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
401 * IN assertion: the stream state is correct and there is enough room in
404 local void putShortMSB (s, b)
408 put_byte(s, (Byte)(b >> 8));
409 put_byte(s, (Byte)(b & 0xff));
412 /* =========================================================================
413 * Flush as much pending output as possible. All deflate() output goes
414 * through this function so some applications may wish to modify it
415 * to avoid allocating a large strm->next_out buffer and copying into it.
416 * (See also read_buf()).
418 local void flush_pending(strm)
421 unsigned len = strm->state->pending;
423 if (len > strm->avail_out) len = strm->avail_out;
424 if (len == 0) return;
426 zmemcpy(strm->next_out, strm->state->pending_out, len);
427 strm->next_out += len;
428 strm->state->pending_out += len;
429 strm->total_out += len;
430 strm->avail_out -= len;
431 strm->state->pending -= len;
432 if (strm->state->pending == 0) {
433 strm->state->pending_out = strm->state->pending_buf;
437 /* ========================================================================= */
438 int ZEXPORT deflate (strm, flush)
442 int old_flush; /* value of flush param for previous deflate call */
445 if (strm == Z_NULL || strm->state == Z_NULL ||
446 flush > Z_INSERT_ONLY || flush < 0) {
447 return Z_STREAM_ERROR;
451 if (strm->next_out == Z_NULL ||
452 (strm->next_in == Z_NULL && strm->avail_in != 0) ||
453 (s->status == FINISH_STATE && flush != Z_FINISH)) {
454 ERR_RETURN(strm, Z_STREAM_ERROR);
456 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
458 s->strm = strm; /* just in case */
459 old_flush = s->last_flush;
460 s->last_flush = flush;
462 /* Write the zlib header */
463 if (s->status == INIT_STATE) {
465 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
466 uInt level_flags = (s->level-1) >> 1;
468 if (level_flags > 3) level_flags = 3;
469 header |= (level_flags << 6);
470 if (s->strstart != 0) header |= PRESET_DICT;
471 header += 31 - (header % 31);
473 s->status = BUSY_STATE;
474 putShortMSB(s, header);
476 /* Save the adler32 of the preset dictionary: */
477 if (s->strstart != 0) {
478 putShortMSB(s, (uInt)(strm->adler >> 16));
479 putShortMSB(s, (uInt)(strm->adler & 0xffff));
484 /* Flush as much pending output as possible */
485 if (s->pending != 0) {
487 if (strm->avail_out == 0) {
488 /* Since avail_out is 0, deflate will be called again with
489 * more output space, but possibly with both pending and
490 * avail_in equal to zero. There won't be anything to do,
491 * but this is not an error situation so make sure we
492 * return OK instead of BUF_ERROR at next call of deflate:
498 /* Make sure there is something to do and avoid duplicate consecutive
499 * flushes. For repeated and useless calls with Z_FINISH, we keep
500 * returning Z_STREAM_END instead of Z_BUFF_ERROR.
502 } else if (strm->avail_in == 0 && flush <= old_flush &&
504 ERR_RETURN(strm, Z_BUF_ERROR);
507 /* User must not provide more input after the first FINISH: */
508 if (s->status == FINISH_STATE && strm->avail_in != 0) {
509 ERR_RETURN(strm, Z_BUF_ERROR);
512 /* Start a new block or continue the current one.
514 if (strm->avail_in != 0 || s->lookahead != 0 ||
515 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
518 bstate = (*(configuration_table[s->level].func))(s, flush);
520 if (bstate == finish_started || bstate == finish_done) {
521 s->status = FINISH_STATE;
523 if (bstate == need_more || bstate == finish_started) {
524 if (strm->avail_out == 0) {
525 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
528 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
529 * of deflate should use the same flush parameter to make sure
530 * that the flush is complete. So we don't have to output an
531 * empty block here, this will be done at next call. This also
532 * ensures that for a very small output buffer, we emit at most
536 if (bstate == block_done) {
537 if (flush == Z_PARTIAL_FLUSH) {
539 } else { /* FULL_FLUSH or SYNC_FLUSH */
540 _tr_stored_block(s, (char*)0, 0L, 0);
541 /* For a full flush, this empty block will be recognized
542 * as a special marker by inflate_sync().
544 if (flush == Z_FULL_FLUSH) {
545 CLEAR_HASH(s); /* forget history */
549 if (strm->avail_out == 0) {
550 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
555 Assert(strm->avail_out > 0, "bug2");
557 if (flush != Z_FINISH) return Z_OK;
558 if (s->noheader) return Z_STREAM_END;
560 /* Write the zlib trailer (adler32) */
561 putShortMSB(s, (uInt)(strm->adler >> 16));
562 putShortMSB(s, (uInt)(strm->adler & 0xffff));
564 /* If avail_out is zero, the application will call deflate again
567 s->noheader = -1; /* write the trailer only once! */
568 return s->pending != 0 ? Z_OK : Z_STREAM_END;
571 /* ========================================================================= */
572 int ZEXPORT deflateEnd (strm)
577 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
579 status = strm->state->status;
580 if (status != INIT_STATE && status != BUSY_STATE &&
581 status != FINISH_STATE) {
582 return Z_STREAM_ERROR;
585 /* Deallocate in reverse order of allocations: */
586 TRY_FREE(strm, strm->state->pending_buf);
587 TRY_FREE(strm, strm->state->head);
588 TRY_FREE(strm, strm->state->prev);
589 TRY_FREE(strm, strm->state->window);
591 ZFREE(strm, strm->state);
592 strm->state = Z_NULL;
594 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
597 /* =========================================================================
598 * Copy the source state to the destination state.
599 * To simplify the source, this is not supported for 16-bit MSDOS (which
600 * doesn't have enough memory anyway to duplicate compression states).
602 int ZEXPORT deflateCopy (dest, source)
607 return Z_STREAM_ERROR;
614 if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) {
615 return Z_STREAM_ERROR;
622 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
623 if (ds == Z_NULL) return Z_MEM_ERROR;
624 dest->state = (struct internal_state FAR *) ds;
628 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
629 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));
630 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));
631 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
632 ds->pending_buf = (uchf *) overlay;
634 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
635 ds->pending_buf == Z_NULL) {
639 /* following zmemcpy do not work for 16-bit MSDOS */
640 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
641 zmemcpy(ds->prev, ss->prev, ds->w_size * sizeof(Pos));
642 zmemcpy(ds->head, ss->head, ds->hash_size * sizeof(Pos));
643 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
645 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
646 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
647 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
649 ds->l_desc.dyn_tree = ds->dyn_ltree;
650 ds->d_desc.dyn_tree = ds->dyn_dtree;
651 ds->bl_desc.dyn_tree = ds->bl_tree;
657 /* ===========================================================================
658 * Read a new buffer from the current input stream, update the adler32
659 * and total number of bytes read. All deflate() input goes through
660 * this function so some applications may wish to modify it to avoid
661 * allocating a large strm->next_in buffer and copying from it.
662 * (See also flush_pending()).
664 local int read_buf(strm, buf, size)
669 unsigned len = strm->avail_in;
671 if (len > size) len = size;
672 if (len == 0) return 0;
674 strm->avail_in -= len;
676 if (!strm->state->noheader) {
677 strm->adler = adler32(strm->adler, strm->next_in, len);
679 zmemcpy(buf, strm->next_in, len);
680 strm->next_in += len;
681 strm->total_in += len;
686 /* ===========================================================================
687 * Initialize the "longest match" routines for a new zlib stream
689 local void lm_init (s)
692 s->window_size = (ulg)2L*s->w_size;
696 /* Set the default configuration parameters:
698 s->max_lazy_match = configuration_table[s->level].max_lazy;
699 s->good_match = configuration_table[s->level].good_length;
700 s->nice_match = configuration_table[s->level].nice_length;
701 s->max_chain_length = configuration_table[s->level].max_chain;
706 s->match_length = s->prev_length = MIN_MATCH-1;
707 s->match_available = 0;
710 match_init(); /* initialize the asm code */
714 /* ===========================================================================
715 * Set match_start to the longest match starting at the given string and
716 * return its length. Matches shorter or equal to prev_length are discarded,
717 * in which case the result is equal to prev_length and match_start is
719 * IN assertions: cur_match is the head of the hash chain for the current
720 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
721 * OUT assertion: the match length is not greater than s->lookahead.
724 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
725 * match.S. The code will be functionally equivalent.
728 local uInt longest_match(s, cur_match)
730 IPos cur_match; /* current match */
732 unsigned chain_length = s->max_chain_length;/* max hash chain length */
733 register Bytef *scan = s->window + s->strstart; /* current string */
734 register Bytef *match; /* matched string */
735 register int len; /* length of current match */
736 int best_len = s->prev_length; /* best match length so far */
737 int nice_match = s->nice_match; /* stop if match long enough */
738 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
739 s->strstart - (IPos)MAX_DIST(s) : NIL;
740 /* Stop when cur_match becomes <= limit. To simplify the code,
741 * we prevent matches with the string of window index 0.
743 Posf *prev = s->prev;
744 uInt wmask = s->w_mask;
747 /* Compare two bytes at a time. Note: this is not always beneficial.
748 * Try with and without -DUNALIGNED_OK to check.
750 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
751 register ush scan_start = *(ushf*)scan;
752 register ush scan_end = *(ushf*)(scan+best_len-1);
754 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
755 register Byte scan_end1 = scan[best_len-1];
756 register Byte scan_end = scan[best_len];
759 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
760 * It is easy to get rid of this optimization if necessary.
762 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
764 /* Do not waste too much time if we already have a good match: */
765 if (s->prev_length >= s->good_match) {
768 /* Do not look for matches beyond the end of the input. This is necessary
769 * to make deflate deterministic.
771 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
773 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
776 Assert(cur_match < s->strstart, "no future");
777 match = s->window + cur_match;
779 /* Skip to next match if the match length cannot increase
780 * or if the match length is less than 2:
782 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
783 /* This code assumes sizeof(unsigned short) == 2. Do not use
784 * UNALIGNED_OK if your compiler uses a different size.
786 if (*(ushf*)(match+best_len-1) != scan_end ||
787 *(ushf*)match != scan_start) continue;
789 /* It is not necessary to compare scan[2] and match[2] since they are
790 * always equal when the other bytes match, given that the hash keys
791 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
792 * strstart+3, +5, ... up to strstart+257. We check for insufficient
793 * lookahead only every 4th comparison; the 128th check will be made
794 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
795 * necessary to put more guard bytes at the end of the window, or
796 * to check more often for insufficient lookahead.
798 Assert(scan[2] == match[2], "scan[2]?");
801 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
802 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
803 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
804 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
806 /* The funny "do {}" generates better code on most compilers */
808 /* Here, scan <= window+strstart+257 */
809 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
810 if (*scan == *match) scan++;
812 len = (MAX_MATCH - 1) - (int)(strend-scan);
813 scan = strend - (MAX_MATCH-1);
815 #else /* UNALIGNED_OK */
817 if (match[best_len] != scan_end ||
818 match[best_len-1] != scan_end1 ||
820 *++match != scan[1]) continue;
822 /* The check at best_len-1 can be removed because it will be made
823 * again later. (This heuristic is not always a win.)
824 * It is not necessary to compare scan[2] and match[2] since they
825 * are always equal when the other bytes match, given that
826 * the hash keys are equal and that HASH_BITS >= 8.
829 Assert(*scan == *match, "match[2]?");
831 /* We check for insufficient lookahead only every 8th comparison;
832 * the 256th check will be made at strstart+258.
835 } while (*++scan == *++match && *++scan == *++match &&
836 *++scan == *++match && *++scan == *++match &&
837 *++scan == *++match && *++scan == *++match &&
838 *++scan == *++match && *++scan == *++match &&
841 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
843 len = MAX_MATCH - (int)(strend - scan);
844 scan = strend - MAX_MATCH;
846 #endif /* UNALIGNED_OK */
848 if (len > best_len) {
849 s->match_start = cur_match;
851 if (len >= nice_match) break;
853 scan_end = *(ushf*)(scan+best_len-1);
855 scan_end1 = scan[best_len-1];
856 scan_end = scan[best_len];
859 } while ((cur_match = prev[cur_match & wmask]) > limit
860 && --chain_length != 0);
862 if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
867 /* ---------------------------------------------------------------------------
868 * Optimized version for level == 1 only
870 local uInt longest_match(s, cur_match)
872 IPos cur_match; /* current match */
874 register Bytef *scan = s->window + s->strstart; /* current string */
875 register Bytef *match; /* matched string */
876 register int len; /* length of current match */
877 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
879 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
880 * It is easy to get rid of this optimization if necessary.
882 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
884 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
886 Assert(cur_match < s->strstart, "no future");
888 match = s->window + cur_match;
890 /* Return failure if the match length is less than 2:
892 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
894 /* The check at best_len-1 can be removed because it will be made
895 * again later. (This heuristic is not always a win.)
896 * It is not necessary to compare scan[2] and match[2] since they
897 * are always equal when the other bytes match, given that
898 * the hash keys are equal and that HASH_BITS >= 8.
900 scan += 2, match += 2;
901 Assert(*scan == *match, "match[2]?");
903 /* We check for insufficient lookahead only every 8th comparison;
904 * the 256th check will be made at strstart+258.
907 } while (*++scan == *++match && *++scan == *++match &&
908 *++scan == *++match && *++scan == *++match &&
909 *++scan == *++match && *++scan == *++match &&
910 *++scan == *++match && *++scan == *++match &&
913 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
915 len = MAX_MATCH - (int)(strend - scan);
917 if (len < MIN_MATCH) return MIN_MATCH - 1;
919 s->match_start = cur_match;
920 return len <= s->lookahead ? len : s->lookahead;
926 /* ===========================================================================
927 * Check that the match at match_start is indeed a match.
929 local void check_match(s, start, match, length)
934 /* check that the match is indeed a match */
935 if (zmemcmp(s->window + match,
936 s->window + start, length) != EQUAL) {
937 fprintf(stderr, " start %u, match %u, length %d\n",
938 start, match, length);
940 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
941 } while (--length != 0);
942 z_error("invalid match");
945 fprintf(stderr,"\\[%d,%d]", start-match, length);
946 do { putc(s->window[start++], stderr); } while (--length != 0);
950 # define check_match(s, start, match, length)
953 /* ===========================================================================
954 * Fill the window when the lookahead becomes insufficient.
955 * Updates strstart and lookahead.
957 * IN assertion: lookahead < MIN_LOOKAHEAD
958 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
959 * At least one byte has been read, or avail_in == 0; reads are
960 * performed for at least two bytes (required for the zip translate_eol
961 * option -- not supported here).
963 local void fill_window(s)
966 register unsigned n, m;
968 unsigned more; /* Amount of free space at the end of the window. */
969 uInt wsize = s->w_size;
972 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
974 /* Deal with !@#$% 64K limit: */
975 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
978 } else if (more == (unsigned)(-1)) {
979 /* Very unlikely, but possible on 16 bit machine if strstart == 0
980 * and lookahead == 1 (input done one byte at time)
984 /* If the window is almost full and there is insufficient lookahead,
985 * move the upper half to the lower one to make room in the upper half.
987 } else if (s->strstart >= wsize+MAX_DIST(s)) {
989 zmemcpy(s->window, s->window+wsize, (unsigned)wsize);
990 s->match_start -= wsize;
991 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
992 s->block_start -= (long) wsize;
994 /* Slide the hash table (could be avoided with 32 bit values
995 at the expense of memory usage). We slide even when level == 0
996 to keep the hash table consistent if we switch back to level > 0
997 later. (Using level 0 permanently is not an optimal usage of
998 zlib, so we don't care about this pathological case.)
1004 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1012 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1013 /* If n is not on any hash chain, prev[n] is garbage but
1014 * its value will never be used.
1020 if (s->strm->avail_in == 0) return;
1022 /* If there was no sliding:
1023 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1024 * more == window_size - lookahead - strstart
1025 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1026 * => more >= window_size - 2*WSIZE + 2
1027 * In the BIG_MEM or MMAP case (not yet supported),
1028 * window_size == input_size + MIN_LOOKAHEAD &&
1029 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1030 * Otherwise, window_size == 2*WSIZE so more >= 2.
1031 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1033 Assert(more >= 2, "more < 2");
1035 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1038 /* Initialize the hash value now that we have some input: */
1039 if (s->lookahead >= MIN_MATCH) {
1040 s->ins_h = s->window[s->strstart];
1041 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1043 Call UPDATE_HASH() MIN_MATCH-3 more times
1046 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1047 * but this is not important since only literal bytes will be emitted.
1050 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1053 /* ===========================================================================
1054 * Flush the current block, with given end-of-file flag.
1055 * IN assertion: strstart is set to the end of the current match.
1057 #define FLUSH_BLOCK_ONLY(s, eof) { \
1058 _tr_flush_block(s, (s->block_start >= 0L ? \
1059 (charf *)&s->window[(unsigned)s->block_start] : \
1061 (ulg)((long)s->strstart - s->block_start), \
1063 s->block_start = s->strstart; \
1064 flush_pending(s->strm); \
1065 Tracev((stderr,"[FLUSH]")); \
1068 /* Same but force premature exit if necessary. */
1069 #define FLUSH_BLOCK(s, eof) { \
1070 FLUSH_BLOCK_ONLY(s, eof); \
1071 if (s->strm->avail_out == 0) return (eof) ? finish_started : need_more; \
1074 /* ===========================================================================
1075 * Copy without compression as much as possible from the input stream, return
1076 * the current block state.
1077 * This function does not insert new strings in the dictionary since
1078 * uncompressible data is probably not useful. This function is used
1079 * only for the level=0 compression option.
1080 * NOTE: this function should be optimized to avoid extra copying from
1081 * window to pending_buf.
1083 local block_state deflate_stored(s, flush)
1087 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited
1088 * to pending_buf_size, and each stored block has a 5 byte header:
1090 ulg max_block_size = 0xffff;
1093 if (max_block_size > s->pending_buf_size - 5) {
1094 max_block_size = s->pending_buf_size - 5;
1097 /* Copy as much as possible from input to output: */
1099 /* Fill the window as much as possible: */
1100 if (s->lookahead <= 1) {
1102 Assert(s->strstart < s->w_size+MAX_DIST(s) ||
1103 s->block_start >= (long)s->w_size, "slide too late");
1106 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
1108 if (s->lookahead == 0) break; /* flush the current block */
1110 Assert(s->block_start >= 0L, "block gone");
1112 s->strstart += s->lookahead;
1115 if (flush == Z_INSERT_ONLY) {
1116 s->block_start = s->strstart;
1120 /* Emit a stored block if pending_buf will be full: */
1121 max_start = s->block_start + max_block_size;
1122 if (s->strstart == 0 || (ulg)s->strstart >= max_start) {
1123 /* strstart == 0 is possible when wraparound on 16-bit machine */
1124 s->lookahead = (uInt)(s->strstart - max_start);
1125 s->strstart = (uInt)max_start;
1128 /* Flush if we may have to slide, otherwise block_start may become
1129 * negative and the data will be gone:
1131 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
1135 if (flush == Z_INSERT_ONLY) {
1136 s->block_start = s->strstart;
1140 FLUSH_BLOCK(s, flush == Z_FINISH);
1141 return flush == Z_FINISH ? finish_done : block_done;
1144 /* ===========================================================================
1145 * Compress as much as possible from the input stream, return the current
1147 * This function does not perform lazy evaluation of matches and inserts
1148 * new strings in the dictionary only for unmatched strings or for short
1149 * matches. It is used only for the fast compression options.
1151 local block_state deflate_fast(s, flush)
1155 IPos hash_head = NIL; /* head of the hash chain */
1156 int bflush; /* set if current block must be flushed */
1159 /* Make sure that we always have enough lookahead, except
1160 * at the end of the input file. We need MAX_MATCH bytes
1161 * for the next match, plus MIN_MATCH bytes to insert the
1162 * string following the next match.
1164 if (s->lookahead < MIN_LOOKAHEAD) {
1166 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1169 if (s->lookahead == 0) break; /* flush the current block */
1172 /* Insert the string window[strstart .. strstart+2] in the
1173 * dictionary, and set hash_head to the head of the hash chain:
1175 if (s->lookahead >= MIN_MATCH) {
1176 INSERT_STRING(s, s->strstart, hash_head);
1179 if (flush == Z_INSERT_ONLY) {
1185 /* Find the longest match, discarding those <= prev_length.
1186 * At this point we have always match_length < MIN_MATCH
1188 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1189 /* To simplify the code, we prevent matches with the string
1190 * of window index 0 (in particular we have to avoid a match
1191 * of the string with itself at the start of the input file).
1193 if (s->strategy != Z_HUFFMAN_ONLY) {
1194 s->match_length = longest_match (s, hash_head);
1196 /* longest_match() sets match_start */
1198 if (s->match_length >= MIN_MATCH) {
1199 check_match(s, s->strstart, s->match_start, s->match_length);
1201 _tr_tally_dist(s, s->strstart - s->match_start,
1202 s->match_length - MIN_MATCH, bflush);
1204 s->lookahead -= s->match_length;
1206 /* Insert new strings in the hash table only if the match length
1207 * is not too large. This saves time but degrades compression.
1210 if (s->match_length <= s->max_insert_length &&
1211 s->lookahead >= MIN_MATCH) {
1212 s->match_length--; /* string at strstart already in hash table */
1215 INSERT_STRING(s, s->strstart, hash_head);
1216 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1217 * always MIN_MATCH bytes ahead.
1219 } while (--s->match_length != 0);
1224 s->strstart += s->match_length;
1225 s->match_length = 0;
1226 s->ins_h = s->window[s->strstart];
1227 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1229 Call UPDATE_HASH() MIN_MATCH-3 more times
1231 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1232 * matter since it will be recomputed at next deflate call.
1236 /* No match, output a literal byte */
1237 Tracevv((stderr,"%c", s->window[s->strstart]));
1238 _tr_tally_lit (s, s->window[s->strstart], bflush);
1242 if (bflush) FLUSH_BLOCK(s, 0);
1244 if (flush == Z_INSERT_ONLY) {
1245 s->block_start = s->strstart;
1248 FLUSH_BLOCK(s, flush == Z_FINISH);
1249 return flush == Z_FINISH ? finish_done : block_done;
1252 /* ===========================================================================
1253 * Same as above, but achieves better compression. We use a lazy
1254 * evaluation for matches: a match is finally adopted only if there is
1255 * no better match at the next window position.
1257 local block_state deflate_slow(s, flush)
1261 IPos hash_head = NIL; /* head of hash chain */
1262 int bflush; /* set if current block must be flushed */
1264 /* Process the input block. */
1266 /* Make sure that we always have enough lookahead, except
1267 * at the end of the input file. We need MAX_MATCH bytes
1268 * for the next match, plus MIN_MATCH bytes to insert the
1269 * string following the next match.
1271 if (s->lookahead < MIN_LOOKAHEAD) {
1273 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1276 if (s->lookahead == 0) break; /* flush the current block */
1279 /* Insert the string window[strstart .. strstart+2] in the
1280 * dictionary, and set hash_head to the head of the hash chain:
1282 if (s->lookahead >= MIN_MATCH) {
1283 INSERT_STRING(s, s->strstart, hash_head);
1286 if (flush == Z_INSERT_ONLY) {
1292 /* Find the longest match, discarding those <= prev_length.
1294 s->prev_length = s->match_length, s->prev_match = s->match_start;
1295 s->match_length = MIN_MATCH-1;
1297 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1298 s->strstart - hash_head <= MAX_DIST(s)) {
1299 /* To simplify the code, we prevent matches with the string
1300 * of window index 0 (in particular we have to avoid a match
1301 * of the string with itself at the start of the input file).
1303 if (s->strategy != Z_HUFFMAN_ONLY) {
1304 s->match_length = longest_match (s, hash_head);
1306 /* longest_match() sets match_start */
1308 if (s->match_length <= 5 && (s->strategy == Z_FILTERED ||
1309 (s->match_length == MIN_MATCH &&
1310 s->strstart - s->match_start > TOO_FAR))) {
1312 /* If prev_match is also MIN_MATCH, match_start is garbage
1313 * but we will ignore the current match anyway.
1315 s->match_length = MIN_MATCH-1;
1318 /* If there was a match at the previous step and the current
1319 * match is not better, output the previous match:
1321 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1322 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1323 /* Do not insert strings in hash table beyond this. */
1325 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1327 _tr_tally_dist(s, s->strstart -1 - s->prev_match,
1328 s->prev_length - MIN_MATCH, bflush);
1330 /* Insert in hash table all strings up to the end of the match.
1331 * strstart-1 and strstart are already inserted. If there is not
1332 * enough lookahead, the last two strings are not inserted in
1335 s->lookahead -= s->prev_length-1;
1336 s->prev_length -= 2;
1338 if (++s->strstart <= max_insert) {
1339 INSERT_STRING(s, s->strstart, hash_head);
1341 } while (--s->prev_length != 0);
1342 s->match_available = 0;
1343 s->match_length = MIN_MATCH-1;
1346 if (bflush) FLUSH_BLOCK(s, 0);
1348 } else if (s->match_available) {
1349 /* If there was no match at the previous position, output a
1350 * single literal. If there was a match but the current match
1351 * is longer, truncate the previous match to a single literal.
1353 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1354 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1356 FLUSH_BLOCK_ONLY(s, 0);
1360 if (s->strm->avail_out == 0) return need_more;
1362 /* There is no previous match to compare with, wait for
1363 * the next step to decide.
1365 s->match_available = 1;
1370 if (flush == Z_INSERT_ONLY) {
1371 s->block_start = s->strstart;
1374 Assert (flush != Z_NO_FLUSH, "no flush?");
1375 if (s->match_available) {
1376 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1377 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1378 s->match_available = 0;
1380 FLUSH_BLOCK(s, flush == Z_FINISH);
1381 return flush == Z_FINISH ? finish_done : block_done;