re PR ipa/63566 (i686 bootstrap fails: ICE RTL flag check: INSN_UID used with unexpec...
[official-gcc.git] / zlib / deflate.c
blob99184ff0cef44896d846724d866d7152c341a85f
1 /* deflate.c -- compress data using the deflation algorithm
2 * Copyright (C) 1995-2012 Jean-loup Gailly and Mark Adler
3 * For conditions of distribution and use, see copyright notice in zlib.h
4 */
6 /*
7 * ALGORITHM
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.
31 * ACKNOWLEDGEMENTS
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.
37 * REFERENCES
39 * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification".
40 * Available in http://tools.ietf.org/html/rfc1951
42 * A description of the Rabin and Karp algorithm is given in the book
43 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
45 * Fiala,E.R., and Greene,D.H.
46 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
50 /* @(#) $Id: deflate.c,v 1.1.1.2 2002/03/11 21:53:23 tromey Exp $ */
52 #include "deflate.h"
54 const char deflate_copyright[] =
55 " deflate 1.2.7 Copyright 1995-2012 Jean-loup Gailly and Mark Adler ";
57 If you use the zlib library in a product, an acknowledgment is welcome
58 in the documentation of your product. If for some reason you cannot
59 include such an acknowledgment, I would appreciate that you keep this
60 copyright string in the executable of your product.
63 /* ===========================================================================
64 * Function prototypes.
66 typedef enum {
67 need_more, /* block not completed, need more input or more output */
68 block_done, /* block flush performed */
69 finish_started, /* finish started, need only more output at next deflate */
70 finish_done /* finish done, accept no more input or output */
71 } block_state;
73 typedef block_state (*compress_func) OF((deflate_state *s, int flush));
74 /* Compression function. Returns the block state after the call. */
76 local void fill_window OF((deflate_state *s));
77 local block_state deflate_stored OF((deflate_state *s, int flush));
78 local block_state deflate_fast OF((deflate_state *s, int flush));
79 #ifndef FASTEST
80 local block_state deflate_slow OF((deflate_state *s, int flush));
81 #endif
82 local block_state deflate_rle OF((deflate_state *s, int flush));
83 local block_state deflate_huff OF((deflate_state *s, int flush));
84 local void lm_init OF((deflate_state *s));
85 local void putShortMSB OF((deflate_state *s, uInt b));
86 local void flush_pending OF((z_streamp strm));
87 local int read_buf OF((z_streamp strm, Bytef *buf, unsigned size));
88 #ifdef ASMV
89 void match_init OF((void)); /* asm code initialization */
90 uInt longest_match OF((deflate_state *s, IPos cur_match));
91 #else
92 local uInt longest_match OF((deflate_state *s, IPos cur_match));
93 #endif
95 #ifdef DEBUG
96 local void check_match OF((deflate_state *s, IPos start, IPos match,
97 int length));
98 #endif
100 /* ===========================================================================
101 * Local data
104 #define NIL 0
105 /* Tail of hash chains */
107 #ifndef TOO_FAR
108 # define TOO_FAR 4096
109 #endif
110 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
112 /* Values for max_lazy_match, good_match and max_chain_length, depending on
113 * the desired pack level (0..9). The values given below have been tuned to
114 * exclude worst case performance for pathological files. Better values may be
115 * found for specific files.
117 typedef struct config_s {
118 ush good_length; /* reduce lazy search above this match length */
119 ush max_lazy; /* do not perform lazy search above this match length */
120 ush nice_length; /* quit search above this match length */
121 ush max_chain;
122 compress_func func;
123 } config;
125 #ifdef FASTEST
126 local const config configuration_table[2] = {
127 /* good lazy nice chain */
128 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
129 /* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */
130 #else
131 local const config configuration_table[10] = {
132 /* good lazy nice chain */
133 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
134 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */
135 /* 2 */ {4, 5, 16, 8, deflate_fast},
136 /* 3 */ {4, 6, 32, 32, deflate_fast},
138 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
139 /* 5 */ {8, 16, 32, 32, deflate_slow},
140 /* 6 */ {8, 16, 128, 128, deflate_slow},
141 /* 7 */ {8, 32, 128, 256, deflate_slow},
142 /* 8 */ {32, 128, 258, 1024, deflate_slow},
143 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */
144 #endif
146 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
147 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
148 * meaning.
151 #define EQUAL 0
152 /* result of memcmp for equal strings */
154 #ifndef NO_DUMMY_DECL
155 struct static_tree_desc_s {int dummy;}; /* for buggy compilers */
156 #endif
158 /* rank Z_BLOCK between Z_NO_FLUSH and Z_PARTIAL_FLUSH */
159 #define RANK(f) (((f) << 1) - ((f) > 4 ? 9 : 0))
161 /* ===========================================================================
162 * Update a hash value with the given input byte
163 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
164 * input characters, so that a running hash key can be computed from the
165 * previous key instead of complete recalculation each time.
167 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
170 /* ===========================================================================
171 * Insert string str in the dictionary and set match_head to the previous head
172 * of the hash chain (the most recent string with same hash key). Return
173 * the previous length of the hash chain.
174 * If this file is compiled with -DFASTEST, the compression level is forced
175 * to 1, and no hash chains are maintained.
176 * IN assertion: all calls to to INSERT_STRING are made with consecutive
177 * input characters and the first MIN_MATCH bytes of str are valid
178 * (except for the last MIN_MATCH-1 bytes of the input file).
180 #ifdef FASTEST
181 #define INSERT_STRING(s, str, match_head) \
182 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
183 match_head = s->head[s->ins_h], \
184 s->head[s->ins_h] = (Pos)(str))
185 #else
186 #define INSERT_STRING(s, str, match_head) \
187 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
188 match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \
189 s->head[s->ins_h] = (Pos)(str))
190 #endif
192 /* ===========================================================================
193 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
194 * prev[] will be initialized on the fly.
196 #define CLEAR_HASH(s) \
197 s->head[s->hash_size-1] = NIL; \
198 zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
200 /* ========================================================================= */
201 int ZEXPORT deflateInit_(strm, level, version, stream_size)
202 z_streamp strm;
203 int level;
204 const char *version;
205 int stream_size;
207 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
208 Z_DEFAULT_STRATEGY, version, stream_size);
209 /* To do: ignore strm->next_in if we use it as window */
212 /* ========================================================================= */
213 int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
214 version, stream_size)
215 z_streamp strm;
216 int level;
217 int method;
218 int windowBits;
219 int memLevel;
220 int strategy;
221 const char *version;
222 int stream_size;
224 deflate_state *s;
225 int wrap = 1;
226 static const char my_version[] = ZLIB_VERSION;
228 ushf *overlay;
229 /* We overlay pending_buf and d_buf+l_buf. This works since the average
230 * output size for (length,distance) codes is <= 24 bits.
233 if (version == Z_NULL || version[0] != my_version[0] ||
234 stream_size != sizeof(z_stream)) {
235 return Z_VERSION_ERROR;
237 if (strm == Z_NULL) return Z_STREAM_ERROR;
239 strm->msg = Z_NULL;
240 if (strm->zalloc == (alloc_func)0) {
241 #ifdef Z_SOLO
242 return Z_STREAM_ERROR;
243 #else
244 strm->zalloc = zcalloc;
245 strm->opaque = (voidpf)0;
246 #endif
248 if (strm->zfree == (free_func)0)
249 #ifdef Z_SOLO
250 return Z_STREAM_ERROR;
251 #else
252 strm->zfree = zcfree;
253 #endif
255 #ifdef FASTEST
256 if (level != 0) level = 1;
257 #else
258 if (level == Z_DEFAULT_COMPRESSION) level = 6;
259 #endif
261 if (windowBits < 0) { /* suppress zlib wrapper */
262 wrap = 0;
263 windowBits = -windowBits;
265 #ifdef GZIP
266 else if (windowBits > 15) {
267 wrap = 2; /* write gzip wrapper instead */
268 windowBits -= 16;
270 #endif
271 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
272 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
273 strategy < 0 || strategy > Z_FIXED) {
274 return Z_STREAM_ERROR;
276 if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */
277 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
278 if (s == Z_NULL) return Z_MEM_ERROR;
279 strm->state = (struct internal_state FAR *)s;
280 s->strm = strm;
282 s->wrap = wrap;
283 s->gzhead = Z_NULL;
284 s->w_bits = windowBits;
285 s->w_size = 1 << s->w_bits;
286 s->w_mask = s->w_size - 1;
288 s->hash_bits = memLevel + 7;
289 s->hash_size = 1 << s->hash_bits;
290 s->hash_mask = s->hash_size - 1;
291 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
293 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
294 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));
295 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
297 s->high_water = 0; /* nothing written to s->window yet */
299 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
301 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
302 s->pending_buf = (uchf *) overlay;
303 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
305 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
306 s->pending_buf == Z_NULL) {
307 s->status = FINISH_STATE;
308 strm->msg = (char*)ERR_MSG(Z_MEM_ERROR);
309 deflateEnd (strm);
310 return Z_MEM_ERROR;
312 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
313 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
315 s->level = level;
316 s->strategy = strategy;
317 s->method = (Byte)method;
319 return deflateReset(strm);
322 /* ========================================================================= */
323 int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
324 z_streamp strm;
325 const Bytef *dictionary;
326 uInt dictLength;
328 deflate_state *s;
329 uInt str, n;
330 int wrap;
331 unsigned avail;
332 unsigned char *next;
334 if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL)
335 return Z_STREAM_ERROR;
336 s = strm->state;
337 wrap = s->wrap;
338 if (wrap == 2 || (wrap == 1 && s->status != INIT_STATE) || s->lookahead)
339 return Z_STREAM_ERROR;
341 /* when using zlib wrappers, compute Adler-32 for provided dictionary */
342 if (wrap == 1)
343 strm->adler = adler32(strm->adler, dictionary, dictLength);
344 s->wrap = 0; /* avoid computing Adler-32 in read_buf */
346 /* if dictionary would fill window, just replace the history */
347 if (dictLength >= s->w_size) {
348 if (wrap == 0) { /* already empty otherwise */
349 CLEAR_HASH(s);
350 s->strstart = 0;
351 s->block_start = 0L;
352 s->insert = 0;
354 dictionary += dictLength - s->w_size; /* use the tail */
355 dictLength = s->w_size;
358 /* insert dictionary into window and hash */
359 avail = strm->avail_in;
360 next = strm->next_in;
361 strm->avail_in = dictLength;
362 strm->next_in = (Bytef *)dictionary;
363 fill_window(s);
364 while (s->lookahead >= MIN_MATCH) {
365 str = s->strstart;
366 n = s->lookahead - (MIN_MATCH-1);
367 do {
368 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
369 #ifndef FASTEST
370 s->prev[str & s->w_mask] = s->head[s->ins_h];
371 #endif
372 s->head[s->ins_h] = (Pos)str;
373 str++;
374 } while (--n);
375 s->strstart = str;
376 s->lookahead = MIN_MATCH-1;
377 fill_window(s);
379 s->strstart += s->lookahead;
380 s->block_start = (long)s->strstart;
381 s->insert = s->lookahead;
382 s->lookahead = 0;
383 s->match_length = s->prev_length = MIN_MATCH-1;
384 s->match_available = 0;
385 strm->next_in = next;
386 strm->avail_in = avail;
387 s->wrap = wrap;
388 return Z_OK;
391 /* ========================================================================= */
392 int ZEXPORT deflateResetKeep (strm)
393 z_streamp strm;
395 deflate_state *s;
397 if (strm == Z_NULL || strm->state == Z_NULL ||
398 strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) {
399 return Z_STREAM_ERROR;
402 strm->total_in = strm->total_out = 0;
403 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
404 strm->data_type = Z_UNKNOWN;
406 s = (deflate_state *)strm->state;
407 s->pending = 0;
408 s->pending_out = s->pending_buf;
410 if (s->wrap < 0) {
411 s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */
413 s->status = s->wrap ? INIT_STATE : BUSY_STATE;
414 strm->adler =
415 #ifdef GZIP
416 s->wrap == 2 ? crc32(0L, Z_NULL, 0) :
417 #endif
418 adler32(0L, Z_NULL, 0);
419 s->last_flush = Z_NO_FLUSH;
421 _tr_init(s);
423 return Z_OK;
426 /* ========================================================================= */
427 int ZEXPORT deflateReset (strm)
428 z_streamp strm;
430 int ret;
432 ret = deflateResetKeep(strm);
433 if (ret == Z_OK)
434 lm_init(strm->state);
435 return ret;
438 /* ========================================================================= */
439 int ZEXPORT deflateSetHeader (strm, head)
440 z_streamp strm;
441 gz_headerp head;
443 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
444 if (strm->state->wrap != 2) return Z_STREAM_ERROR;
445 strm->state->gzhead = head;
446 return Z_OK;
449 /* ========================================================================= */
450 int ZEXPORT deflatePending (strm, pending, bits)
451 unsigned *pending;
452 int *bits;
453 z_streamp strm;
455 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
456 if (pending != Z_NULL)
457 *pending = strm->state->pending;
458 if (bits != Z_NULL)
459 *bits = strm->state->bi_valid;
460 return Z_OK;
463 /* ========================================================================= */
464 int ZEXPORT deflatePrime (strm, bits, value)
465 z_streamp strm;
466 int bits;
467 int value;
469 deflate_state *s;
470 int put;
472 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
473 s = strm->state;
474 if ((Bytef *)(s->d_buf) < s->pending_out + ((Buf_size + 7) >> 3))
475 return Z_BUF_ERROR;
476 do {
477 put = Buf_size - s->bi_valid;
478 if (put > bits)
479 put = bits;
480 s->bi_buf |= (ush)((value & ((1 << put) - 1)) << s->bi_valid);
481 s->bi_valid += put;
482 _tr_flush_bits(s);
483 value >>= put;
484 bits -= put;
485 } while (bits);
486 return Z_OK;
489 /* ========================================================================= */
490 int ZEXPORT deflateParams(strm, level, strategy)
491 z_streamp strm;
492 int level;
493 int strategy;
495 deflate_state *s;
496 compress_func func;
497 int err = Z_OK;
499 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
500 s = strm->state;
502 #ifdef FASTEST
503 if (level != 0) level = 1;
504 #else
505 if (level == Z_DEFAULT_COMPRESSION) level = 6;
506 #endif
507 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
508 return Z_STREAM_ERROR;
510 func = configuration_table[s->level].func;
512 if ((strategy != s->strategy || func != configuration_table[level].func) &&
513 strm->total_in != 0) {
514 /* Flush the last buffer: */
515 err = deflate(strm, Z_BLOCK);
517 if (s->level != level) {
518 s->level = level;
519 s->max_lazy_match = configuration_table[level].max_lazy;
520 s->good_match = configuration_table[level].good_length;
521 s->nice_match = configuration_table[level].nice_length;
522 s->max_chain_length = configuration_table[level].max_chain;
524 s->strategy = strategy;
525 return err;
528 /* ========================================================================= */
529 int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain)
530 z_streamp strm;
531 int good_length;
532 int max_lazy;
533 int nice_length;
534 int max_chain;
536 deflate_state *s;
538 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
539 s = strm->state;
540 s->good_match = good_length;
541 s->max_lazy_match = max_lazy;
542 s->nice_match = nice_length;
543 s->max_chain_length = max_chain;
544 return Z_OK;
547 /* =========================================================================
548 * For the default windowBits of 15 and memLevel of 8, this function returns
549 * a close to exact, as well as small, upper bound on the compressed size.
550 * They are coded as constants here for a reason--if the #define's are
551 * changed, then this function needs to be changed as well. The return
552 * value for 15 and 8 only works for those exact settings.
554 * For any setting other than those defaults for windowBits and memLevel,
555 * the value returned is a conservative worst case for the maximum expansion
556 * resulting from using fixed blocks instead of stored blocks, which deflate
557 * can emit on compressed data for some combinations of the parameters.
559 * This function could be more sophisticated to provide closer upper bounds for
560 * every combination of windowBits and memLevel. But even the conservative
561 * upper bound of about 14% expansion does not seem onerous for output buffer
562 * allocation.
564 uLong ZEXPORT deflateBound(strm, sourceLen)
565 z_streamp strm;
566 uLong sourceLen;
568 deflate_state *s;
569 uLong complen, wraplen;
570 Bytef *str;
572 /* conservative upper bound for compressed data */
573 complen = sourceLen +
574 ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5;
576 /* if can't get parameters, return conservative bound plus zlib wrapper */
577 if (strm == Z_NULL || strm->state == Z_NULL)
578 return complen + 6;
580 /* compute wrapper length */
581 s = strm->state;
582 switch (s->wrap) {
583 case 0: /* raw deflate */
584 wraplen = 0;
585 break;
586 case 1: /* zlib wrapper */
587 wraplen = 6 + (s->strstart ? 4 : 0);
588 break;
589 case 2: /* gzip wrapper */
590 wraplen = 18;
591 if (s->gzhead != Z_NULL) { /* user-supplied gzip header */
592 if (s->gzhead->extra != Z_NULL)
593 wraplen += 2 + s->gzhead->extra_len;
594 str = s->gzhead->name;
595 if (str != Z_NULL)
596 do {
597 wraplen++;
598 } while (*str++);
599 str = s->gzhead->comment;
600 if (str != Z_NULL)
601 do {
602 wraplen++;
603 } while (*str++);
604 if (s->gzhead->hcrc)
605 wraplen += 2;
607 break;
608 default: /* for compiler happiness */
609 wraplen = 6;
612 /* if not default parameters, return conservative bound */
613 if (s->w_bits != 15 || s->hash_bits != 8 + 7)
614 return complen + wraplen;
616 /* default settings: return tight bound for that case */
617 return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) +
618 (sourceLen >> 25) + 13 - 6 + wraplen;
621 /* =========================================================================
622 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
623 * IN assertion: the stream state is correct and there is enough room in
624 * pending_buf.
626 local void putShortMSB (s, b)
627 deflate_state *s;
628 uInt b;
630 put_byte(s, (Byte)(b >> 8));
631 put_byte(s, (Byte)(b & 0xff));
634 /* =========================================================================
635 * Flush as much pending output as possible. All deflate() output goes
636 * through this function so some applications may wish to modify it
637 * to avoid allocating a large strm->next_out buffer and copying into it.
638 * (See also read_buf()).
640 local void flush_pending(strm)
641 z_streamp strm;
643 unsigned len;
644 deflate_state *s = strm->state;
646 _tr_flush_bits(s);
647 len = s->pending;
648 if (len > strm->avail_out) len = strm->avail_out;
649 if (len == 0) return;
651 zmemcpy(strm->next_out, s->pending_out, len);
652 strm->next_out += len;
653 s->pending_out += len;
654 strm->total_out += len;
655 strm->avail_out -= len;
656 s->pending -= len;
657 if (s->pending == 0) {
658 s->pending_out = s->pending_buf;
662 /* ========================================================================= */
663 int ZEXPORT deflate (strm, flush)
664 z_streamp strm;
665 int flush;
667 int old_flush; /* value of flush param for previous deflate call */
668 deflate_state *s;
670 if (strm == Z_NULL || strm->state == Z_NULL ||
671 flush > Z_BLOCK || flush < 0) {
672 return Z_STREAM_ERROR;
674 s = strm->state;
676 if (strm->next_out == Z_NULL ||
677 (strm->next_in == Z_NULL && strm->avail_in != 0) ||
678 (s->status == FINISH_STATE && flush != Z_FINISH)) {
679 ERR_RETURN(strm, Z_STREAM_ERROR);
681 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
683 s->strm = strm; /* just in case */
684 old_flush = s->last_flush;
685 s->last_flush = flush;
687 /* Write the header */
688 if (s->status == INIT_STATE) {
689 #ifdef GZIP
690 if (s->wrap == 2) {
691 strm->adler = crc32(0L, Z_NULL, 0);
692 put_byte(s, 31);
693 put_byte(s, 139);
694 put_byte(s, 8);
695 if (s->gzhead == Z_NULL) {
696 put_byte(s, 0);
697 put_byte(s, 0);
698 put_byte(s, 0);
699 put_byte(s, 0);
700 put_byte(s, 0);
701 put_byte(s, s->level == 9 ? 2 :
702 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
703 4 : 0));
704 put_byte(s, OS_CODE);
705 s->status = BUSY_STATE;
707 else {
708 put_byte(s, (s->gzhead->text ? 1 : 0) +
709 (s->gzhead->hcrc ? 2 : 0) +
710 (s->gzhead->extra == Z_NULL ? 0 : 4) +
711 (s->gzhead->name == Z_NULL ? 0 : 8) +
712 (s->gzhead->comment == Z_NULL ? 0 : 16)
714 put_byte(s, (Byte)(s->gzhead->time & 0xff));
715 put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));
716 put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));
717 put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));
718 put_byte(s, s->level == 9 ? 2 :
719 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
720 4 : 0));
721 put_byte(s, s->gzhead->os & 0xff);
722 if (s->gzhead->extra != Z_NULL) {
723 put_byte(s, s->gzhead->extra_len & 0xff);
724 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);
726 if (s->gzhead->hcrc)
727 strm->adler = crc32(strm->adler, s->pending_buf,
728 s->pending);
729 s->gzindex = 0;
730 s->status = EXTRA_STATE;
733 else
734 #endif
736 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
737 uInt level_flags;
739 if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)
740 level_flags = 0;
741 else if (s->level < 6)
742 level_flags = 1;
743 else if (s->level == 6)
744 level_flags = 2;
745 else
746 level_flags = 3;
747 header |= (level_flags << 6);
748 if (s->strstart != 0) header |= PRESET_DICT;
749 header += 31 - (header % 31);
751 s->status = BUSY_STATE;
752 putShortMSB(s, header);
754 /* Save the adler32 of the preset dictionary: */
755 if (s->strstart != 0) {
756 putShortMSB(s, (uInt)(strm->adler >> 16));
757 putShortMSB(s, (uInt)(strm->adler & 0xffff));
759 strm->adler = adler32(0L, Z_NULL, 0);
762 #ifdef GZIP
763 if (s->status == EXTRA_STATE) {
764 if (s->gzhead->extra != Z_NULL) {
765 uInt beg = s->pending; /* start of bytes to update crc */
767 while (s->gzindex < (s->gzhead->extra_len & 0xffff)) {
768 if (s->pending == s->pending_buf_size) {
769 if (s->gzhead->hcrc && s->pending > beg)
770 strm->adler = crc32(strm->adler, s->pending_buf + beg,
771 s->pending - beg);
772 flush_pending(strm);
773 beg = s->pending;
774 if (s->pending == s->pending_buf_size)
775 break;
777 put_byte(s, s->gzhead->extra[s->gzindex]);
778 s->gzindex++;
780 if (s->gzhead->hcrc && s->pending > beg)
781 strm->adler = crc32(strm->adler, s->pending_buf + beg,
782 s->pending - beg);
783 if (s->gzindex == s->gzhead->extra_len) {
784 s->gzindex = 0;
785 s->status = NAME_STATE;
788 else
789 s->status = NAME_STATE;
791 if (s->status == NAME_STATE) {
792 if (s->gzhead->name != Z_NULL) {
793 uInt beg = s->pending; /* start of bytes to update crc */
794 int val;
796 do {
797 if (s->pending == s->pending_buf_size) {
798 if (s->gzhead->hcrc && s->pending > beg)
799 strm->adler = crc32(strm->adler, s->pending_buf + beg,
800 s->pending - beg);
801 flush_pending(strm);
802 beg = s->pending;
803 if (s->pending == s->pending_buf_size) {
804 val = 1;
805 break;
808 val = s->gzhead->name[s->gzindex++];
809 put_byte(s, val);
810 } while (val != 0);
811 if (s->gzhead->hcrc && s->pending > beg)
812 strm->adler = crc32(strm->adler, s->pending_buf + beg,
813 s->pending - beg);
814 if (val == 0) {
815 s->gzindex = 0;
816 s->status = COMMENT_STATE;
819 else
820 s->status = COMMENT_STATE;
822 if (s->status == COMMENT_STATE) {
823 if (s->gzhead->comment != Z_NULL) {
824 uInt beg = s->pending; /* start of bytes to update crc */
825 int val;
827 do {
828 if (s->pending == s->pending_buf_size) {
829 if (s->gzhead->hcrc && s->pending > beg)
830 strm->adler = crc32(strm->adler, s->pending_buf + beg,
831 s->pending - beg);
832 flush_pending(strm);
833 beg = s->pending;
834 if (s->pending == s->pending_buf_size) {
835 val = 1;
836 break;
839 val = s->gzhead->comment[s->gzindex++];
840 put_byte(s, val);
841 } while (val != 0);
842 if (s->gzhead->hcrc && s->pending > beg)
843 strm->adler = crc32(strm->adler, s->pending_buf + beg,
844 s->pending - beg);
845 if (val == 0)
846 s->status = HCRC_STATE;
848 else
849 s->status = HCRC_STATE;
851 if (s->status == HCRC_STATE) {
852 if (s->gzhead->hcrc) {
853 if (s->pending + 2 > s->pending_buf_size)
854 flush_pending(strm);
855 if (s->pending + 2 <= s->pending_buf_size) {
856 put_byte(s, (Byte)(strm->adler & 0xff));
857 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
858 strm->adler = crc32(0L, Z_NULL, 0);
859 s->status = BUSY_STATE;
862 else
863 s->status = BUSY_STATE;
865 #endif
867 /* Flush as much pending output as possible */
868 if (s->pending != 0) {
869 flush_pending(strm);
870 if (strm->avail_out == 0) {
871 /* Since avail_out is 0, deflate will be called again with
872 * more output space, but possibly with both pending and
873 * avail_in equal to zero. There won't be anything to do,
874 * but this is not an error situation so make sure we
875 * return OK instead of BUF_ERROR at next call of deflate:
877 s->last_flush = -1;
878 return Z_OK;
881 /* Make sure there is something to do and avoid duplicate consecutive
882 * flushes. For repeated and useless calls with Z_FINISH, we keep
883 * returning Z_STREAM_END instead of Z_BUF_ERROR.
885 } else if (strm->avail_in == 0 && RANK(flush) <= RANK(old_flush) &&
886 flush != Z_FINISH) {
887 ERR_RETURN(strm, Z_BUF_ERROR);
890 /* User must not provide more input after the first FINISH: */
891 if (s->status == FINISH_STATE && strm->avail_in != 0) {
892 ERR_RETURN(strm, Z_BUF_ERROR);
895 /* Start a new block or continue the current one.
897 if (strm->avail_in != 0 || s->lookahead != 0 ||
898 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
899 block_state bstate;
901 bstate = s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) :
902 (s->strategy == Z_RLE ? deflate_rle(s, flush) :
903 (*(configuration_table[s->level].func))(s, flush));
905 if (bstate == finish_started || bstate == finish_done) {
906 s->status = FINISH_STATE;
908 if (bstate == need_more || bstate == finish_started) {
909 if (strm->avail_out == 0) {
910 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
912 return Z_OK;
913 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
914 * of deflate should use the same flush parameter to make sure
915 * that the flush is complete. So we don't have to output an
916 * empty block here, this will be done at next call. This also
917 * ensures that for a very small output buffer, we emit at most
918 * one empty block.
921 if (bstate == block_done) {
922 if (flush == Z_PARTIAL_FLUSH) {
923 _tr_align(s);
924 } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */
925 _tr_stored_block(s, (char*)0, 0L, 0);
926 /* For a full flush, this empty block will be recognized
927 * as a special marker by inflate_sync().
929 if (flush == Z_FULL_FLUSH) {
930 CLEAR_HASH(s); /* forget history */
931 if (s->lookahead == 0) {
932 s->strstart = 0;
933 s->block_start = 0L;
934 s->insert = 0;
938 flush_pending(strm);
939 if (strm->avail_out == 0) {
940 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
941 return Z_OK;
945 Assert(strm->avail_out > 0, "bug2");
947 if (flush != Z_FINISH) return Z_OK;
948 if (s->wrap <= 0) return Z_STREAM_END;
950 /* Write the trailer */
951 #ifdef GZIP
952 if (s->wrap == 2) {
953 put_byte(s, (Byte)(strm->adler & 0xff));
954 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
955 put_byte(s, (Byte)((strm->adler >> 16) & 0xff));
956 put_byte(s, (Byte)((strm->adler >> 24) & 0xff));
957 put_byte(s, (Byte)(strm->total_in & 0xff));
958 put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));
959 put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));
960 put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));
962 else
963 #endif
965 putShortMSB(s, (uInt)(strm->adler >> 16));
966 putShortMSB(s, (uInt)(strm->adler & 0xffff));
968 flush_pending(strm);
969 /* If avail_out is zero, the application will call deflate again
970 * to flush the rest.
972 if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
973 return s->pending != 0 ? Z_OK : Z_STREAM_END;
976 /* ========================================================================= */
977 int ZEXPORT deflateEnd (strm)
978 z_streamp strm;
980 int status;
982 if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR;
984 status = strm->state->status;
985 if (status != INIT_STATE &&
986 status != EXTRA_STATE &&
987 status != NAME_STATE &&
988 status != COMMENT_STATE &&
989 status != HCRC_STATE &&
990 status != BUSY_STATE &&
991 status != FINISH_STATE) {
992 return Z_STREAM_ERROR;
995 /* Deallocate in reverse order of allocations: */
996 TRY_FREE(strm, strm->state->pending_buf);
997 TRY_FREE(strm, strm->state->head);
998 TRY_FREE(strm, strm->state->prev);
999 TRY_FREE(strm, strm->state->window);
1001 ZFREE(strm, strm->state);
1002 strm->state = Z_NULL;
1004 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
1007 /* =========================================================================
1008 * Copy the source state to the destination state.
1009 * To simplify the source, this is not supported for 16-bit MSDOS (which
1010 * doesn't have enough memory anyway to duplicate compression states).
1012 int ZEXPORT deflateCopy (dest, source)
1013 z_streamp dest;
1014 z_streamp source;
1016 #ifdef MAXSEG_64K
1017 return Z_STREAM_ERROR;
1018 #else
1019 deflate_state *ds;
1020 deflate_state *ss;
1021 ushf *overlay;
1024 if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) {
1025 return Z_STREAM_ERROR;
1028 ss = source->state;
1030 zmemcpy((voidpf)dest, (voidpf)source, sizeof(z_stream));
1032 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
1033 if (ds == Z_NULL) return Z_MEM_ERROR;
1034 dest->state = (struct internal_state FAR *) ds;
1035 zmemcpy((voidpf)ds, (voidpf)ss, sizeof(deflate_state));
1036 ds->strm = dest;
1038 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
1039 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));
1040 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));
1041 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
1042 ds->pending_buf = (uchf *) overlay;
1044 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
1045 ds->pending_buf == Z_NULL) {
1046 deflateEnd (dest);
1047 return Z_MEM_ERROR;
1049 /* following zmemcpy do not work for 16-bit MSDOS */
1050 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
1051 zmemcpy((voidpf)ds->prev, (voidpf)ss->prev, ds->w_size * sizeof(Pos));
1052 zmemcpy((voidpf)ds->head, (voidpf)ss->head, ds->hash_size * sizeof(Pos));
1053 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
1055 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
1056 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
1057 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
1059 ds->l_desc.dyn_tree = ds->dyn_ltree;
1060 ds->d_desc.dyn_tree = ds->dyn_dtree;
1061 ds->bl_desc.dyn_tree = ds->bl_tree;
1063 return Z_OK;
1064 #endif /* MAXSEG_64K */
1067 /* ===========================================================================
1068 * Read a new buffer from the current input stream, update the adler32
1069 * and total number of bytes read. All deflate() input goes through
1070 * this function so some applications may wish to modify it to avoid
1071 * allocating a large strm->next_in buffer and copying from it.
1072 * (See also flush_pending()).
1074 local int read_buf(strm, buf, size)
1075 z_streamp strm;
1076 Bytef *buf;
1077 unsigned size;
1079 unsigned len = strm->avail_in;
1081 if (len > size) len = size;
1082 if (len == 0) return 0;
1084 strm->avail_in -= len;
1086 zmemcpy(buf, strm->next_in, len);
1087 if (strm->state->wrap == 1) {
1088 strm->adler = adler32(strm->adler, buf, len);
1090 #ifdef GZIP
1091 else if (strm->state->wrap == 2) {
1092 strm->adler = crc32(strm->adler, buf, len);
1094 #endif
1095 strm->next_in += len;
1096 strm->total_in += len;
1098 return (int)len;
1101 /* ===========================================================================
1102 * Initialize the "longest match" routines for a new zlib stream
1104 local void lm_init (s)
1105 deflate_state *s;
1107 s->window_size = (ulg)2L*s->w_size;
1109 CLEAR_HASH(s);
1111 /* Set the default configuration parameters:
1113 s->max_lazy_match = configuration_table[s->level].max_lazy;
1114 s->good_match = configuration_table[s->level].good_length;
1115 s->nice_match = configuration_table[s->level].nice_length;
1116 s->max_chain_length = configuration_table[s->level].max_chain;
1118 s->strstart = 0;
1119 s->block_start = 0L;
1120 s->lookahead = 0;
1121 s->insert = 0;
1122 s->match_length = s->prev_length = MIN_MATCH-1;
1123 s->match_available = 0;
1124 s->ins_h = 0;
1125 #ifndef FASTEST
1126 #ifdef ASMV
1127 match_init(); /* initialize the asm code */
1128 #endif
1129 #endif
1132 #ifndef FASTEST
1133 /* ===========================================================================
1134 * Set match_start to the longest match starting at the given string and
1135 * return its length. Matches shorter or equal to prev_length are discarded,
1136 * in which case the result is equal to prev_length and match_start is
1137 * garbage.
1138 * IN assertions: cur_match is the head of the hash chain for the current
1139 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
1140 * OUT assertion: the match length is not greater than s->lookahead.
1142 #ifndef ASMV
1143 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
1144 * match.S. The code will be functionally equivalent.
1146 local uInt longest_match(s, cur_match)
1147 deflate_state *s;
1148 IPos cur_match; /* current match */
1150 unsigned chain_length = s->max_chain_length;/* max hash chain length */
1151 register Bytef *scan = s->window + s->strstart; /* current string */
1152 register Bytef *match; /* matched string */
1153 register int len; /* length of current match */
1154 int best_len = s->prev_length; /* best match length so far */
1155 int nice_match = s->nice_match; /* stop if match long enough */
1156 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
1157 s->strstart - (IPos)MAX_DIST(s) : NIL;
1158 /* Stop when cur_match becomes <= limit. To simplify the code,
1159 * we prevent matches with the string of window index 0.
1161 Posf *prev = s->prev;
1162 uInt wmask = s->w_mask;
1164 #ifdef UNALIGNED_OK
1165 /* Compare two bytes at a time. Note: this is not always beneficial.
1166 * Try with and without -DUNALIGNED_OK to check.
1168 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
1169 register ush scan_start = *(ushf*)scan;
1170 register ush scan_end = *(ushf*)(scan+best_len-1);
1171 #else
1172 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1173 register Byte scan_end1 = scan[best_len-1];
1174 register Byte scan_end = scan[best_len];
1175 #endif
1177 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1178 * It is easy to get rid of this optimization if necessary.
1180 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1182 /* Do not waste too much time if we already have a good match: */
1183 if (s->prev_length >= s->good_match) {
1184 chain_length >>= 2;
1186 /* Do not look for matches beyond the end of the input. This is necessary
1187 * to make deflate deterministic.
1189 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
1191 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1193 do {
1194 Assert(cur_match < s->strstart, "no future");
1195 match = s->window + cur_match;
1197 /* Skip to next match if the match length cannot increase
1198 * or if the match length is less than 2. Note that the checks below
1199 * for insufficient lookahead only occur occasionally for performance
1200 * reasons. Therefore uninitialized memory will be accessed, and
1201 * conditional jumps will be made that depend on those values.
1202 * However the length of the match is limited to the lookahead, so
1203 * the output of deflate is not affected by the uninitialized values.
1205 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
1206 /* This code assumes sizeof(unsigned short) == 2. Do not use
1207 * UNALIGNED_OK if your compiler uses a different size.
1209 if (*(ushf*)(match+best_len-1) != scan_end ||
1210 *(ushf*)match != scan_start) continue;
1212 /* It is not necessary to compare scan[2] and match[2] since they are
1213 * always equal when the other bytes match, given that the hash keys
1214 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
1215 * strstart+3, +5, ... up to strstart+257. We check for insufficient
1216 * lookahead only every 4th comparison; the 128th check will be made
1217 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
1218 * necessary to put more guard bytes at the end of the window, or
1219 * to check more often for insufficient lookahead.
1221 Assert(scan[2] == match[2], "scan[2]?");
1222 scan++, match++;
1223 do {
1224 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1225 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1226 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1227 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1228 scan < strend);
1229 /* The funny "do {}" generates better code on most compilers */
1231 /* Here, scan <= window+strstart+257 */
1232 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1233 if (*scan == *match) scan++;
1235 len = (MAX_MATCH - 1) - (int)(strend-scan);
1236 scan = strend - (MAX_MATCH-1);
1238 #else /* UNALIGNED_OK */
1240 if (match[best_len] != scan_end ||
1241 match[best_len-1] != scan_end1 ||
1242 *match != *scan ||
1243 *++match != scan[1]) continue;
1245 /* The check at best_len-1 can be removed because it will be made
1246 * again later. (This heuristic is not always a win.)
1247 * It is not necessary to compare scan[2] and match[2] since they
1248 * are always equal when the other bytes match, given that
1249 * the hash keys are equal and that HASH_BITS >= 8.
1251 scan += 2, match++;
1252 Assert(*scan == *match, "match[2]?");
1254 /* We check for insufficient lookahead only every 8th comparison;
1255 * the 256th check will be made at strstart+258.
1257 do {
1258 } while (*++scan == *++match && *++scan == *++match &&
1259 *++scan == *++match && *++scan == *++match &&
1260 *++scan == *++match && *++scan == *++match &&
1261 *++scan == *++match && *++scan == *++match &&
1262 scan < strend);
1264 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1266 len = MAX_MATCH - (int)(strend - scan);
1267 scan = strend - MAX_MATCH;
1269 #endif /* UNALIGNED_OK */
1271 if (len > best_len) {
1272 s->match_start = cur_match;
1273 best_len = len;
1274 if (len >= nice_match) break;
1275 #ifdef UNALIGNED_OK
1276 scan_end = *(ushf*)(scan+best_len-1);
1277 #else
1278 scan_end1 = scan[best_len-1];
1279 scan_end = scan[best_len];
1280 #endif
1282 } while ((cur_match = prev[cur_match & wmask]) > limit
1283 && --chain_length != 0);
1285 if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
1286 return s->lookahead;
1288 #endif /* ASMV */
1290 #else /* FASTEST */
1292 /* ---------------------------------------------------------------------------
1293 * Optimized version for FASTEST only
1295 local uInt longest_match(s, cur_match)
1296 deflate_state *s;
1297 IPos cur_match; /* current match */
1299 register Bytef *scan = s->window + s->strstart; /* current string */
1300 register Bytef *match; /* matched string */
1301 register int len; /* length of current match */
1302 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1304 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1305 * It is easy to get rid of this optimization if necessary.
1307 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1309 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1311 Assert(cur_match < s->strstart, "no future");
1313 match = s->window + cur_match;
1315 /* Return failure if the match length is less than 2:
1317 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
1319 /* The check at best_len-1 can be removed because it will be made
1320 * again later. (This heuristic is not always a win.)
1321 * It is not necessary to compare scan[2] and match[2] since they
1322 * are always equal when the other bytes match, given that
1323 * the hash keys are equal and that HASH_BITS >= 8.
1325 scan += 2, match += 2;
1326 Assert(*scan == *match, "match[2]?");
1328 /* We check for insufficient lookahead only every 8th comparison;
1329 * the 256th check will be made at strstart+258.
1331 do {
1332 } while (*++scan == *++match && *++scan == *++match &&
1333 *++scan == *++match && *++scan == *++match &&
1334 *++scan == *++match && *++scan == *++match &&
1335 *++scan == *++match && *++scan == *++match &&
1336 scan < strend);
1338 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1340 len = MAX_MATCH - (int)(strend - scan);
1342 if (len < MIN_MATCH) return MIN_MATCH - 1;
1344 s->match_start = cur_match;
1345 return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;
1348 #endif /* FASTEST */
1350 #ifdef DEBUG
1351 /* ===========================================================================
1352 * Check that the match at match_start is indeed a match.
1354 local void check_match(s, start, match, length)
1355 deflate_state *s;
1356 IPos start, match;
1357 int length;
1359 /* check that the match is indeed a match */
1360 if (zmemcmp(s->window + match,
1361 s->window + start, length) != EQUAL) {
1362 fprintf(stderr, " start %u, match %u, length %d\n",
1363 start, match, length);
1364 do {
1365 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
1366 } while (--length != 0);
1367 z_error("invalid match");
1369 if (z_verbose > 1) {
1370 fprintf(stderr,"\\[%d,%d]", start-match, length);
1371 do { putc(s->window[start++], stderr); } while (--length != 0);
1374 #else
1375 # define check_match(s, start, match, length)
1376 #endif /* DEBUG */
1378 /* ===========================================================================
1379 * Fill the window when the lookahead becomes insufficient.
1380 * Updates strstart and lookahead.
1382 * IN assertion: lookahead < MIN_LOOKAHEAD
1383 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
1384 * At least one byte has been read, or avail_in == 0; reads are
1385 * performed for at least two bytes (required for the zip translate_eol
1386 * option -- not supported here).
1388 local void fill_window(s)
1389 deflate_state *s;
1391 register unsigned n, m;
1392 register Posf *p;
1393 unsigned more; /* Amount of free space at the end of the window. */
1394 uInt wsize = s->w_size;
1396 Assert(s->lookahead < MIN_LOOKAHEAD, "already enough lookahead");
1398 do {
1399 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
1401 /* Deal with !@#$% 64K limit: */
1402 if (sizeof(int) <= 2) {
1403 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
1404 more = wsize;
1406 } else if (more == (unsigned)(-1)) {
1407 /* Very unlikely, but possible on 16 bit machine if
1408 * strstart == 0 && lookahead == 1 (input done a byte at time)
1410 more--;
1414 /* If the window is almost full and there is insufficient lookahead,
1415 * move the upper half to the lower one to make room in the upper half.
1417 if (s->strstart >= wsize+MAX_DIST(s)) {
1419 zmemcpy(s->window, s->window+wsize, (unsigned)wsize);
1420 s->match_start -= wsize;
1421 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
1422 s->block_start -= (long) wsize;
1424 /* Slide the hash table (could be avoided with 32 bit values
1425 at the expense of memory usage). We slide even when level == 0
1426 to keep the hash table consistent if we switch back to level > 0
1427 later. (Using level 0 permanently is not an optimal usage of
1428 zlib, so we don't care about this pathological case.)
1430 n = s->hash_size;
1431 p = &s->head[n];
1432 do {
1433 m = *--p;
1434 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1435 } while (--n);
1437 n = wsize;
1438 #ifndef FASTEST
1439 p = &s->prev[n];
1440 do {
1441 m = *--p;
1442 *p = (Pos)(m >= wsize ? m-wsize : NIL);
1443 /* If n is not on any hash chain, prev[n] is garbage but
1444 * its value will never be used.
1446 } while (--n);
1447 #endif
1448 more += wsize;
1450 if (s->strm->avail_in == 0) break;
1452 /* If there was no sliding:
1453 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1454 * more == window_size - lookahead - strstart
1455 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1456 * => more >= window_size - 2*WSIZE + 2
1457 * In the BIG_MEM or MMAP case (not yet supported),
1458 * window_size == input_size + MIN_LOOKAHEAD &&
1459 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1460 * Otherwise, window_size == 2*WSIZE so more >= 2.
1461 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1463 Assert(more >= 2, "more < 2");
1465 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1466 s->lookahead += n;
1468 /* Initialize the hash value now that we have some input: */
1469 if (s->lookahead + s->insert >= MIN_MATCH) {
1470 uInt str = s->strstart - s->insert;
1471 s->ins_h = s->window[str];
1472 UPDATE_HASH(s, s->ins_h, s->window[str + 1]);
1473 #if MIN_MATCH != 3
1474 Call UPDATE_HASH() MIN_MATCH-3 more times
1475 #endif
1476 while (s->insert) {
1477 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
1478 #ifndef FASTEST
1479 s->prev[str & s->w_mask] = s->head[s->ins_h];
1480 #endif
1481 s->head[s->ins_h] = (Pos)str;
1482 str++;
1483 s->insert--;
1484 if (s->lookahead + s->insert < MIN_MATCH)
1485 break;
1488 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1489 * but this is not important since only literal bytes will be emitted.
1492 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1494 /* If the WIN_INIT bytes after the end of the current data have never been
1495 * written, then zero those bytes in order to avoid memory check reports of
1496 * the use of uninitialized (or uninitialised as Julian writes) bytes by
1497 * the longest match routines. Update the high water mark for the next
1498 * time through here. WIN_INIT is set to MAX_MATCH since the longest match
1499 * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead.
1501 if (s->high_water < s->window_size) {
1502 ulg curr = s->strstart + (ulg)(s->lookahead);
1503 ulg init;
1505 if (s->high_water < curr) {
1506 /* Previous high water mark below current data -- zero WIN_INIT
1507 * bytes or up to end of window, whichever is less.
1509 init = s->window_size - curr;
1510 if (init > WIN_INIT)
1511 init = WIN_INIT;
1512 zmemzero(s->window + curr, (unsigned)init);
1513 s->high_water = curr + init;
1515 else if (s->high_water < (ulg)curr + WIN_INIT) {
1516 /* High water mark at or above current data, but below current data
1517 * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up
1518 * to end of window, whichever is less.
1520 init = (ulg)curr + WIN_INIT - s->high_water;
1521 if (init > s->window_size - s->high_water)
1522 init = s->window_size - s->high_water;
1523 zmemzero(s->window + s->high_water, (unsigned)init);
1524 s->high_water += init;
1528 Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD,
1529 "not enough room for search");
1532 /* ===========================================================================
1533 * Flush the current block, with given end-of-file flag.
1534 * IN assertion: strstart is set to the end of the current match.
1536 #define FLUSH_BLOCK_ONLY(s, last) { \
1537 _tr_flush_block(s, (s->block_start >= 0L ? \
1538 (charf *)&s->window[(unsigned)s->block_start] : \
1539 (charf *)Z_NULL), \
1540 (ulg)((long)s->strstart - s->block_start), \
1541 (last)); \
1542 s->block_start = s->strstart; \
1543 flush_pending(s->strm); \
1544 Tracev((stderr,"[FLUSH]")); \
1547 /* Same but force premature exit if necessary. */
1548 #define FLUSH_BLOCK(s, last) { \
1549 FLUSH_BLOCK_ONLY(s, last); \
1550 if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \
1553 /* ===========================================================================
1554 * Copy without compression as much as possible from the input stream, return
1555 * the current block state.
1556 * This function does not insert new strings in the dictionary since
1557 * uncompressible data is probably not useful. This function is used
1558 * only for the level=0 compression option.
1559 * NOTE: this function should be optimized to avoid extra copying from
1560 * window to pending_buf.
1562 local block_state deflate_stored(s, flush)
1563 deflate_state *s;
1564 int flush;
1566 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited
1567 * to pending_buf_size, and each stored block has a 5 byte header:
1569 ulg max_block_size = 0xffff;
1570 ulg max_start;
1572 if (max_block_size > s->pending_buf_size - 5) {
1573 max_block_size = s->pending_buf_size - 5;
1576 /* Copy as much as possible from input to output: */
1577 for (;;) {
1578 /* Fill the window as much as possible: */
1579 if (s->lookahead <= 1) {
1581 Assert(s->strstart < s->w_size+MAX_DIST(s) ||
1582 s->block_start >= (long)s->w_size, "slide too late");
1584 fill_window(s);
1585 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
1587 if (s->lookahead == 0) break; /* flush the current block */
1589 Assert(s->block_start >= 0L, "block gone");
1591 s->strstart += s->lookahead;
1592 s->lookahead = 0;
1594 /* Emit a stored block if pending_buf will be full: */
1595 max_start = s->block_start + max_block_size;
1596 if (s->strstart == 0 || (ulg)s->strstart >= max_start) {
1597 /* strstart == 0 is possible when wraparound on 16-bit machine */
1598 s->lookahead = (uInt)(s->strstart - max_start);
1599 s->strstart = (uInt)max_start;
1600 FLUSH_BLOCK(s, 0);
1602 /* Flush if we may have to slide, otherwise block_start may become
1603 * negative and the data will be gone:
1605 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
1606 FLUSH_BLOCK(s, 0);
1609 s->insert = 0;
1610 if (flush == Z_FINISH) {
1611 FLUSH_BLOCK(s, 1);
1612 return finish_done;
1614 if ((long)s->strstart > s->block_start)
1615 FLUSH_BLOCK(s, 0);
1616 return block_done;
1619 /* ===========================================================================
1620 * Compress as much as possible from the input stream, return the current
1621 * block state.
1622 * This function does not perform lazy evaluation of matches and inserts
1623 * new strings in the dictionary only for unmatched strings or for short
1624 * matches. It is used only for the fast compression options.
1626 local block_state deflate_fast(s, flush)
1627 deflate_state *s;
1628 int flush;
1630 IPos hash_head; /* head of the hash chain */
1631 int bflush; /* set if current block must be flushed */
1633 for (;;) {
1634 /* Make sure that we always have enough lookahead, except
1635 * at the end of the input file. We need MAX_MATCH bytes
1636 * for the next match, plus MIN_MATCH bytes to insert the
1637 * string following the next match.
1639 if (s->lookahead < MIN_LOOKAHEAD) {
1640 fill_window(s);
1641 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1642 return need_more;
1644 if (s->lookahead == 0) break; /* flush the current block */
1647 /* Insert the string window[strstart .. strstart+2] in the
1648 * dictionary, and set hash_head to the head of the hash chain:
1650 hash_head = NIL;
1651 if (s->lookahead >= MIN_MATCH) {
1652 INSERT_STRING(s, s->strstart, hash_head);
1655 /* Find the longest match, discarding those <= prev_length.
1656 * At this point we have always match_length < MIN_MATCH
1658 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1659 /* To simplify the code, we prevent matches with the string
1660 * of window index 0 (in particular we have to avoid a match
1661 * of the string with itself at the start of the input file).
1663 s->match_length = longest_match (s, hash_head);
1664 /* longest_match() sets match_start */
1666 if (s->match_length >= MIN_MATCH) {
1667 check_match(s, s->strstart, s->match_start, s->match_length);
1669 _tr_tally_dist(s, s->strstart - s->match_start,
1670 s->match_length - MIN_MATCH, bflush);
1672 s->lookahead -= s->match_length;
1674 /* Insert new strings in the hash table only if the match length
1675 * is not too large. This saves time but degrades compression.
1677 #ifndef FASTEST
1678 if (s->match_length <= s->max_insert_length &&
1679 s->lookahead >= MIN_MATCH) {
1680 s->match_length--; /* string at strstart already in table */
1681 do {
1682 s->strstart++;
1683 INSERT_STRING(s, s->strstart, hash_head);
1684 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1685 * always MIN_MATCH bytes ahead.
1687 } while (--s->match_length != 0);
1688 s->strstart++;
1689 } else
1690 #endif
1692 s->strstart += s->match_length;
1693 s->match_length = 0;
1694 s->ins_h = s->window[s->strstart];
1695 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1696 #if MIN_MATCH != 3
1697 Call UPDATE_HASH() MIN_MATCH-3 more times
1698 #endif
1699 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1700 * matter since it will be recomputed at next deflate call.
1703 } else {
1704 /* No match, output a literal byte */
1705 Tracevv((stderr,"%c", s->window[s->strstart]));
1706 _tr_tally_lit (s, s->window[s->strstart], bflush);
1707 s->lookahead--;
1708 s->strstart++;
1710 if (bflush) FLUSH_BLOCK(s, 0);
1712 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
1713 if (flush == Z_FINISH) {
1714 FLUSH_BLOCK(s, 1);
1715 return finish_done;
1717 if (s->last_lit)
1718 FLUSH_BLOCK(s, 0);
1719 return block_done;
1722 #ifndef FASTEST
1723 /* ===========================================================================
1724 * Same as above, but achieves better compression. We use a lazy
1725 * evaluation for matches: a match is finally adopted only if there is
1726 * no better match at the next window position.
1728 local block_state deflate_slow(s, flush)
1729 deflate_state *s;
1730 int flush;
1732 IPos hash_head; /* head of hash chain */
1733 int bflush; /* set if current block must be flushed */
1735 /* Process the input block. */
1736 for (;;) {
1737 /* Make sure that we always have enough lookahead, except
1738 * at the end of the input file. We need MAX_MATCH bytes
1739 * for the next match, plus MIN_MATCH bytes to insert the
1740 * string following the next match.
1742 if (s->lookahead < MIN_LOOKAHEAD) {
1743 fill_window(s);
1744 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1745 return need_more;
1747 if (s->lookahead == 0) break; /* flush the current block */
1750 /* Insert the string window[strstart .. strstart+2] in the
1751 * dictionary, and set hash_head to the head of the hash chain:
1753 hash_head = NIL;
1754 if (s->lookahead >= MIN_MATCH) {
1755 INSERT_STRING(s, s->strstart, hash_head);
1758 /* Find the longest match, discarding those <= prev_length.
1760 s->prev_length = s->match_length, s->prev_match = s->match_start;
1761 s->match_length = MIN_MATCH-1;
1763 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1764 s->strstart - hash_head <= MAX_DIST(s)) {
1765 /* To simplify the code, we prevent matches with the string
1766 * of window index 0 (in particular we have to avoid a match
1767 * of the string with itself at the start of the input file).
1769 s->match_length = longest_match (s, hash_head);
1770 /* longest_match() sets match_start */
1772 if (s->match_length <= 5 && (s->strategy == Z_FILTERED
1773 #if TOO_FAR <= 32767
1774 || (s->match_length == MIN_MATCH &&
1775 s->strstart - s->match_start > TOO_FAR)
1776 #endif
1777 )) {
1779 /* If prev_match is also MIN_MATCH, match_start is garbage
1780 * but we will ignore the current match anyway.
1782 s->match_length = MIN_MATCH-1;
1785 /* If there was a match at the previous step and the current
1786 * match is not better, output the previous match:
1788 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1789 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1790 /* Do not insert strings in hash table beyond this. */
1792 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1794 _tr_tally_dist(s, s->strstart -1 - s->prev_match,
1795 s->prev_length - MIN_MATCH, bflush);
1797 /* Insert in hash table all strings up to the end of the match.
1798 * strstart-1 and strstart are already inserted. If there is not
1799 * enough lookahead, the last two strings are not inserted in
1800 * the hash table.
1802 s->lookahead -= s->prev_length-1;
1803 s->prev_length -= 2;
1804 do {
1805 if (++s->strstart <= max_insert) {
1806 INSERT_STRING(s, s->strstart, hash_head);
1808 } while (--s->prev_length != 0);
1809 s->match_available = 0;
1810 s->match_length = MIN_MATCH-1;
1811 s->strstart++;
1813 if (bflush) FLUSH_BLOCK(s, 0);
1815 } else if (s->match_available) {
1816 /* If there was no match at the previous position, output a
1817 * single literal. If there was a match but the current match
1818 * is longer, truncate the previous match to a single literal.
1820 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1821 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1822 if (bflush) {
1823 FLUSH_BLOCK_ONLY(s, 0);
1825 s->strstart++;
1826 s->lookahead--;
1827 if (s->strm->avail_out == 0) return need_more;
1828 } else {
1829 /* There is no previous match to compare with, wait for
1830 * the next step to decide.
1832 s->match_available = 1;
1833 s->strstart++;
1834 s->lookahead--;
1837 Assert (flush != Z_NO_FLUSH, "no flush?");
1838 if (s->match_available) {
1839 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1840 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
1841 s->match_available = 0;
1843 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
1844 if (flush == Z_FINISH) {
1845 FLUSH_BLOCK(s, 1);
1846 return finish_done;
1848 if (s->last_lit)
1849 FLUSH_BLOCK(s, 0);
1850 return block_done;
1852 #endif /* FASTEST */
1854 /* ===========================================================================
1855 * For Z_RLE, simply look for runs of bytes, generate matches only of distance
1856 * one. Do not maintain a hash table. (It will be regenerated if this run of
1857 * deflate switches away from Z_RLE.)
1859 local block_state deflate_rle(s, flush)
1860 deflate_state *s;
1861 int flush;
1863 int bflush; /* set if current block must be flushed */
1864 uInt prev; /* byte at distance one to match */
1865 Bytef *scan, *strend; /* scan goes up to strend for length of run */
1867 for (;;) {
1868 /* Make sure that we always have enough lookahead, except
1869 * at the end of the input file. We need MAX_MATCH bytes
1870 * for the longest run, plus one for the unrolled loop.
1872 if (s->lookahead <= MAX_MATCH) {
1873 fill_window(s);
1874 if (s->lookahead <= MAX_MATCH && flush == Z_NO_FLUSH) {
1875 return need_more;
1877 if (s->lookahead == 0) break; /* flush the current block */
1880 /* See how many times the previous byte repeats */
1881 s->match_length = 0;
1882 if (s->lookahead >= MIN_MATCH && s->strstart > 0) {
1883 scan = s->window + s->strstart - 1;
1884 prev = *scan;
1885 if (prev == *++scan && prev == *++scan && prev == *++scan) {
1886 strend = s->window + s->strstart + MAX_MATCH;
1887 do {
1888 } while (prev == *++scan && prev == *++scan &&
1889 prev == *++scan && prev == *++scan &&
1890 prev == *++scan && prev == *++scan &&
1891 prev == *++scan && prev == *++scan &&
1892 scan < strend);
1893 s->match_length = MAX_MATCH - (int)(strend - scan);
1894 if (s->match_length > s->lookahead)
1895 s->match_length = s->lookahead;
1897 Assert(scan <= s->window+(uInt)(s->window_size-1), "wild scan");
1900 /* Emit match if have run of MIN_MATCH or longer, else emit literal */
1901 if (s->match_length >= MIN_MATCH) {
1902 check_match(s, s->strstart, s->strstart - 1, s->match_length);
1904 _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush);
1906 s->lookahead -= s->match_length;
1907 s->strstart += s->match_length;
1908 s->match_length = 0;
1909 } else {
1910 /* No match, output a literal byte */
1911 Tracevv((stderr,"%c", s->window[s->strstart]));
1912 _tr_tally_lit (s, s->window[s->strstart], bflush);
1913 s->lookahead--;
1914 s->strstart++;
1916 if (bflush) FLUSH_BLOCK(s, 0);
1918 s->insert = 0;
1919 if (flush == Z_FINISH) {
1920 FLUSH_BLOCK(s, 1);
1921 return finish_done;
1923 if (s->last_lit)
1924 FLUSH_BLOCK(s, 0);
1925 return block_done;
1928 /* ===========================================================================
1929 * For Z_HUFFMAN_ONLY, do not look for matches. Do not maintain a hash table.
1930 * (It will be regenerated if this run of deflate switches away from Huffman.)
1932 local block_state deflate_huff(s, flush)
1933 deflate_state *s;
1934 int flush;
1936 int bflush; /* set if current block must be flushed */
1938 for (;;) {
1939 /* Make sure that we have a literal to write. */
1940 if (s->lookahead == 0) {
1941 fill_window(s);
1942 if (s->lookahead == 0) {
1943 if (flush == Z_NO_FLUSH)
1944 return need_more;
1945 break; /* flush the current block */
1949 /* Output a literal byte */
1950 s->match_length = 0;
1951 Tracevv((stderr,"%c", s->window[s->strstart]));
1952 _tr_tally_lit (s, s->window[s->strstart], bflush);
1953 s->lookahead--;
1954 s->strstart++;
1955 if (bflush) FLUSH_BLOCK(s, 0);
1957 s->insert = 0;
1958 if (flush == Z_FINISH) {
1959 FLUSH_BLOCK(s, 1);
1960 return finish_done;
1962 if (s->last_lit)
1963 FLUSH_BLOCK(s, 0);
1964 return block_done;