ia64/linux-2.6.18-xen.hg

view lib/zlib_deflate/deflate.c @ 740:bf8b1ee634e2

balloon: Fix the build by including <linux/init.h>

Signed-off-by: Keir Fraser <keir.fraser@citrix.com>
author Keir Fraser <keir.fraser@citrix.com>
date Tue Nov 25 11:19:41 2008 +0000 (2008-11-25)
parents 831230e53067
children
line source
1 /* +++ deflate.c */
2 /* deflate.c -- compress data using the deflation algorithm
3 * Copyright (C) 1995-1996 Jean-loup Gailly.
4 * For conditions of distribution and use, see copyright notice in zlib.h
5 */
7 /*
8 * ALGORITHM
9 *
10 * The "deflation" process depends on being able to identify portions
11 * of the input text which are identical to earlier input (within a
12 * sliding window trailing behind the input currently being processed).
13 *
14 * The most straightforward technique turns out to be the fastest for
15 * most input files: try all possible matches and select the longest.
16 * The key feature of this algorithm is that insertions into the string
17 * dictionary are very simple and thus fast, and deletions are avoided
18 * completely. Insertions are performed at each input character, whereas
19 * string matches are performed only when the previous match ends. So it
20 * is preferable to spend more time in matches to allow very fast string
21 * insertions and avoid deletions. The matching algorithm for small
22 * strings is inspired from that of Rabin & Karp. A brute force approach
23 * is used to find longer strings when a small match has been found.
24 * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze
25 * (by Leonid Broukhis).
26 * A previous version of this file used a more sophisticated algorithm
27 * (by Fiala and Greene) which is guaranteed to run in linear amortized
28 * time, but has a larger average cost, uses more memory and is patented.
29 * However the F&G algorithm may be faster for some highly redundant
30 * files if the parameter max_chain_length (described below) is too large.
31 *
32 * ACKNOWLEDGEMENTS
33 *
34 * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and
35 * I found it in 'freeze' written by Leonid Broukhis.
36 * Thanks to many people for bug reports and testing.
37 *
38 * REFERENCES
39 *
40 * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification".
41 * Available in ftp://ds.internic.net/rfc/rfc1951.txt
42 *
43 * A description of the Rabin and Karp algorithm is given in the book
44 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
45 *
46 * Fiala,E.R., and Greene,D.H.
47 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
48 *
49 */
51 #include <linux/module.h>
52 #include <linux/zutil.h>
53 #include "defutil.h"
56 /* ===========================================================================
57 * Function prototypes.
58 */
59 typedef enum {
60 need_more, /* block not completed, need more input or more output */
61 block_done, /* block flush performed */
62 finish_started, /* finish started, need only more output at next deflate */
63 finish_done /* finish done, accept no more input or output */
64 } block_state;
66 typedef block_state (*compress_func) (deflate_state *s, int flush);
67 /* Compression function. Returns the block state after the call. */
69 static void fill_window (deflate_state *s);
70 static block_state deflate_stored (deflate_state *s, int flush);
71 static block_state deflate_fast (deflate_state *s, int flush);
72 static block_state deflate_slow (deflate_state *s, int flush);
73 static void lm_init (deflate_state *s);
74 static void putShortMSB (deflate_state *s, uInt b);
75 static void flush_pending (z_streamp strm);
76 static int read_buf (z_streamp strm, Byte *buf, unsigned size);
77 static uInt longest_match (deflate_state *s, IPos cur_match);
79 #ifdef DEBUG_ZLIB
80 static void check_match (deflate_state *s, IPos start, IPos match,
81 int length);
82 #endif
84 /* ===========================================================================
85 * Local data
86 */
88 #define NIL 0
89 /* Tail of hash chains */
91 #ifndef TOO_FAR
92 # define TOO_FAR 4096
93 #endif
94 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
96 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
97 /* Minimum amount of lookahead, except at the end of the input file.
98 * See deflate.c for comments about the MIN_MATCH+1.
99 */
101 /* Values for max_lazy_match, good_match and max_chain_length, depending on
102 * the desired pack level (0..9). The values given below have been tuned to
103 * exclude worst case performance for pathological files. Better values may be
104 * found for specific files.
105 */
106 typedef struct config_s {
107 ush good_length; /* reduce lazy search above this match length */
108 ush max_lazy; /* do not perform lazy search above this match length */
109 ush nice_length; /* quit search above this match length */
110 ush max_chain;
111 compress_func func;
112 } config;
114 static const config configuration_table[10] = {
115 /* good lazy nice chain */
116 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
117 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* maximum speed, no lazy matches */
118 /* 2 */ {4, 5, 16, 8, deflate_fast},
119 /* 3 */ {4, 6, 32, 32, deflate_fast},
121 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
122 /* 5 */ {8, 16, 32, 32, deflate_slow},
123 /* 6 */ {8, 16, 128, 128, deflate_slow},
124 /* 7 */ {8, 32, 128, 256, deflate_slow},
125 /* 8 */ {32, 128, 258, 1024, deflate_slow},
126 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* maximum compression */
128 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
129 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
130 * meaning.
131 */
133 #define EQUAL 0
134 /* result of memcmp for equal strings */
136 /* ===========================================================================
137 * Update a hash value with the given input byte
138 * IN assertion: all calls to to UPDATE_HASH are made with consecutive
139 * input characters, so that a running hash key can be computed from the
140 * previous key instead of complete recalculation each time.
141 */
142 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
145 /* ===========================================================================
146 * Insert string str in the dictionary and set match_head to the previous head
147 * of the hash chain (the most recent string with same hash key). Return
148 * the previous length of the hash chain.
149 * IN assertion: all calls to to INSERT_STRING are made with consecutive
150 * input characters and the first MIN_MATCH bytes of str are valid
151 * (except for the last MIN_MATCH-1 bytes of the input file).
152 */
153 #define INSERT_STRING(s, str, match_head) \
154 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
155 s->prev[(str) & s->w_mask] = match_head = s->head[s->ins_h], \
156 s->head[s->ins_h] = (Pos)(str))
158 /* ===========================================================================
159 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
160 * prev[] will be initialized on the fly.
161 */
162 #define CLEAR_HASH(s) \
163 s->head[s->hash_size-1] = NIL; \
164 memset((char *)s->head, 0, (unsigned)(s->hash_size-1)*sizeof(*s->head));
166 /* ========================================================================= */
167 int zlib_deflateInit2(
168 z_streamp strm,
169 int level,
170 int method,
171 int windowBits,
172 int memLevel,
173 int strategy
174 )
175 {
176 deflate_state *s;
177 int noheader = 0;
178 deflate_workspace *mem;
180 ush *overlay;
181 /* We overlay pending_buf and d_buf+l_buf. This works since the average
182 * output size for (length,distance) codes is <= 24 bits.
183 */
185 if (strm == NULL) return Z_STREAM_ERROR;
187 strm->msg = NULL;
189 if (level == Z_DEFAULT_COMPRESSION) level = 6;
191 mem = (deflate_workspace *) strm->workspace;
193 if (windowBits < 0) { /* undocumented feature: suppress zlib header */
194 noheader = 1;
195 windowBits = -windowBits;
196 }
197 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
198 windowBits < 9 || windowBits > 15 || level < 0 || level > 9 ||
199 strategy < 0 || strategy > Z_HUFFMAN_ONLY) {
200 return Z_STREAM_ERROR;
201 }
202 s = (deflate_state *) &(mem->deflate_memory);
203 strm->state = (struct internal_state *)s;
204 s->strm = strm;
206 s->noheader = noheader;
207 s->w_bits = windowBits;
208 s->w_size = 1 << s->w_bits;
209 s->w_mask = s->w_size - 1;
211 s->hash_bits = memLevel + 7;
212 s->hash_size = 1 << s->hash_bits;
213 s->hash_mask = s->hash_size - 1;
214 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
216 s->window = (Byte *) mem->window_memory;
217 s->prev = (Pos *) mem->prev_memory;
218 s->head = (Pos *) mem->head_memory;
220 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
222 overlay = (ush *) mem->overlay_memory;
223 s->pending_buf = (uch *) overlay;
224 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
226 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
227 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
229 s->level = level;
230 s->strategy = strategy;
231 s->method = (Byte)method;
233 return zlib_deflateReset(strm);
234 }
236 /* ========================================================================= */
237 #if 0
238 int zlib_deflateSetDictionary(
239 z_streamp strm,
240 const Byte *dictionary,
241 uInt dictLength
242 )
243 {
244 deflate_state *s;
245 uInt length = dictLength;
246 uInt n;
247 IPos hash_head = 0;
249 if (strm == NULL || strm->state == NULL || dictionary == NULL)
250 return Z_STREAM_ERROR;
252 s = (deflate_state *) strm->state;
253 if (s->status != INIT_STATE) return Z_STREAM_ERROR;
255 strm->adler = zlib_adler32(strm->adler, dictionary, dictLength);
257 if (length < MIN_MATCH) return Z_OK;
258 if (length > MAX_DIST(s)) {
259 length = MAX_DIST(s);
260 #ifndef USE_DICT_HEAD
261 dictionary += dictLength - length; /* use the tail of the dictionary */
262 #endif
263 }
264 memcpy((char *)s->window, dictionary, length);
265 s->strstart = length;
266 s->block_start = (long)length;
268 /* Insert all strings in the hash table (except for the last two bytes).
269 * s->lookahead stays null, so s->ins_h will be recomputed at the next
270 * call of fill_window.
271 */
272 s->ins_h = s->window[0];
273 UPDATE_HASH(s, s->ins_h, s->window[1]);
274 for (n = 0; n <= length - MIN_MATCH; n++) {
275 INSERT_STRING(s, n, hash_head);
276 }
277 if (hash_head) hash_head = 0; /* to make compiler happy */
278 return Z_OK;
279 }
280 #endif /* 0 */
282 /* ========================================================================= */
283 int zlib_deflateReset(
284 z_streamp strm
285 )
286 {
287 deflate_state *s;
289 if (strm == NULL || strm->state == NULL)
290 return Z_STREAM_ERROR;
292 strm->total_in = strm->total_out = 0;
293 strm->msg = NULL;
294 strm->data_type = Z_UNKNOWN;
296 s = (deflate_state *)strm->state;
297 s->pending = 0;
298 s->pending_out = s->pending_buf;
300 if (s->noheader < 0) {
301 s->noheader = 0; /* was set to -1 by deflate(..., Z_FINISH); */
302 }
303 s->status = s->noheader ? BUSY_STATE : INIT_STATE;
304 strm->adler = 1;
305 s->last_flush = Z_NO_FLUSH;
307 zlib_tr_init(s);
308 lm_init(s);
310 return Z_OK;
311 }
313 /* ========================================================================= */
314 #if 0
315 int zlib_deflateParams(
316 z_streamp strm,
317 int level,
318 int strategy
319 )
320 {
321 deflate_state *s;
322 compress_func func;
323 int err = Z_OK;
325 if (strm == NULL || strm->state == NULL) return Z_STREAM_ERROR;
326 s = (deflate_state *) strm->state;
328 if (level == Z_DEFAULT_COMPRESSION) {
329 level = 6;
330 }
331 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_HUFFMAN_ONLY) {
332 return Z_STREAM_ERROR;
333 }
334 func = configuration_table[s->level].func;
336 if (func != configuration_table[level].func && strm->total_in != 0) {
337 /* Flush the last buffer: */
338 err = zlib_deflate(strm, Z_PARTIAL_FLUSH);
339 }
340 if (s->level != level) {
341 s->level = level;
342 s->max_lazy_match = configuration_table[level].max_lazy;
343 s->good_match = configuration_table[level].good_length;
344 s->nice_match = configuration_table[level].nice_length;
345 s->max_chain_length = configuration_table[level].max_chain;
346 }
347 s->strategy = strategy;
348 return err;
349 }
350 #endif /* 0 */
352 /* =========================================================================
353 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
354 * IN assertion: the stream state is correct and there is enough room in
355 * pending_buf.
356 */
357 static void putShortMSB(
358 deflate_state *s,
359 uInt b
360 )
361 {
362 put_byte(s, (Byte)(b >> 8));
363 put_byte(s, (Byte)(b & 0xff));
364 }
366 /* =========================================================================
367 * Flush as much pending output as possible. All deflate() output goes
368 * through this function so some applications may wish to modify it
369 * to avoid allocating a large strm->next_out buffer and copying into it.
370 * (See also read_buf()).
371 */
372 static void flush_pending(
373 z_streamp strm
374 )
375 {
376 deflate_state *s = (deflate_state *) strm->state;
377 unsigned len = s->pending;
379 if (len > strm->avail_out) len = strm->avail_out;
380 if (len == 0) return;
382 if (strm->next_out != NULL) {
383 memcpy(strm->next_out, s->pending_out, len);
384 strm->next_out += len;
385 }
386 s->pending_out += len;
387 strm->total_out += len;
388 strm->avail_out -= len;
389 s->pending -= len;
390 if (s->pending == 0) {
391 s->pending_out = s->pending_buf;
392 }
393 }
395 /* ========================================================================= */
396 int zlib_deflate(
397 z_streamp strm,
398 int flush
399 )
400 {
401 int old_flush; /* value of flush param for previous deflate call */
402 deflate_state *s;
404 if (strm == NULL || strm->state == NULL ||
405 flush > Z_FINISH || flush < 0) {
406 return Z_STREAM_ERROR;
407 }
408 s = (deflate_state *) strm->state;
410 if ((strm->next_in == NULL && strm->avail_in != 0) ||
411 (s->status == FINISH_STATE && flush != Z_FINISH)) {
412 return Z_STREAM_ERROR;
413 }
414 if (strm->avail_out == 0) return Z_BUF_ERROR;
416 s->strm = strm; /* just in case */
417 old_flush = s->last_flush;
418 s->last_flush = flush;
420 /* Write the zlib header */
421 if (s->status == INIT_STATE) {
423 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
424 uInt level_flags = (s->level-1) >> 1;
426 if (level_flags > 3) level_flags = 3;
427 header |= (level_flags << 6);
428 if (s->strstart != 0) header |= PRESET_DICT;
429 header += 31 - (header % 31);
431 s->status = BUSY_STATE;
432 putShortMSB(s, header);
434 /* Save the adler32 of the preset dictionary: */
435 if (s->strstart != 0) {
436 putShortMSB(s, (uInt)(strm->adler >> 16));
437 putShortMSB(s, (uInt)(strm->adler & 0xffff));
438 }
439 strm->adler = 1L;
440 }
442 /* Flush as much pending output as possible */
443 if (s->pending != 0) {
444 flush_pending(strm);
445 if (strm->avail_out == 0) {
446 /* Since avail_out is 0, deflate will be called again with
447 * more output space, but possibly with both pending and
448 * avail_in equal to zero. There won't be anything to do,
449 * but this is not an error situation so make sure we
450 * return OK instead of BUF_ERROR at next call of deflate:
451 */
452 s->last_flush = -1;
453 return Z_OK;
454 }
456 /* Make sure there is something to do and avoid duplicate consecutive
457 * flushes. For repeated and useless calls with Z_FINISH, we keep
458 * returning Z_STREAM_END instead of Z_BUFF_ERROR.
459 */
460 } else if (strm->avail_in == 0 && flush <= old_flush &&
461 flush != Z_FINISH) {
462 return Z_BUF_ERROR;
463 }
465 /* User must not provide more input after the first FINISH: */
466 if (s->status == FINISH_STATE && strm->avail_in != 0) {
467 return Z_BUF_ERROR;
468 }
470 /* Start a new block or continue the current one.
471 */
472 if (strm->avail_in != 0 || s->lookahead != 0 ||
473 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
474 block_state bstate;
476 bstate = (*(configuration_table[s->level].func))(s, flush);
478 if (bstate == finish_started || bstate == finish_done) {
479 s->status = FINISH_STATE;
480 }
481 if (bstate == need_more || bstate == finish_started) {
482 if (strm->avail_out == 0) {
483 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
484 }
485 return Z_OK;
486 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
487 * of deflate should use the same flush parameter to make sure
488 * that the flush is complete. So we don't have to output an
489 * empty block here, this will be done at next call. This also
490 * ensures that for a very small output buffer, we emit at most
491 * one empty block.
492 */
493 }
494 if (bstate == block_done) {
495 if (flush == Z_PARTIAL_FLUSH) {
496 zlib_tr_align(s);
497 } else if (flush == Z_PACKET_FLUSH) {
498 /* Output just the 3-bit `stored' block type value,
499 but not a zero length. */
500 zlib_tr_stored_type_only(s);
501 } else { /* FULL_FLUSH or SYNC_FLUSH */
502 zlib_tr_stored_block(s, (char*)0, 0L, 0);
503 /* For a full flush, this empty block will be recognized
504 * as a special marker by inflate_sync().
505 */
506 if (flush == Z_FULL_FLUSH) {
507 CLEAR_HASH(s); /* forget history */
508 }
509 }
510 flush_pending(strm);
511 if (strm->avail_out == 0) {
512 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
513 return Z_OK;
514 }
515 }
516 }
517 Assert(strm->avail_out > 0, "bug2");
519 if (flush != Z_FINISH) return Z_OK;
520 if (s->noheader) return Z_STREAM_END;
522 /* Write the zlib trailer (adler32) */
523 putShortMSB(s, (uInt)(strm->adler >> 16));
524 putShortMSB(s, (uInt)(strm->adler & 0xffff));
525 flush_pending(strm);
526 /* If avail_out is zero, the application will call deflate again
527 * to flush the rest.
528 */
529 s->noheader = -1; /* write the trailer only once! */
530 return s->pending != 0 ? Z_OK : Z_STREAM_END;
531 }
533 /* ========================================================================= */
534 int zlib_deflateEnd(
535 z_streamp strm
536 )
537 {
538 int status;
539 deflate_state *s;
541 if (strm == NULL || strm->state == NULL) return Z_STREAM_ERROR;
542 s = (deflate_state *) strm->state;
544 status = s->status;
545 if (status != INIT_STATE && status != BUSY_STATE &&
546 status != FINISH_STATE) {
547 return Z_STREAM_ERROR;
548 }
550 strm->state = NULL;
552 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
553 }
555 /* =========================================================================
556 * Copy the source state to the destination state.
557 */
558 #if 0
559 int zlib_deflateCopy (
560 z_streamp dest,
561 z_streamp source
562 )
563 {
564 #ifdef MAXSEG_64K
565 return Z_STREAM_ERROR;
566 #else
567 deflate_state *ds;
568 deflate_state *ss;
569 ush *overlay;
570 deflate_workspace *mem;
573 if (source == NULL || dest == NULL || source->state == NULL) {
574 return Z_STREAM_ERROR;
575 }
577 ss = (deflate_state *) source->state;
579 *dest = *source;
581 mem = (deflate_workspace *) dest->workspace;
583 ds = &(mem->deflate_memory);
585 dest->state = (struct internal_state *) ds;
586 *ds = *ss;
587 ds->strm = dest;
589 ds->window = (Byte *) mem->window_memory;
590 ds->prev = (Pos *) mem->prev_memory;
591 ds->head = (Pos *) mem->head_memory;
592 overlay = (ush *) mem->overlay_memory;
593 ds->pending_buf = (uch *) overlay;
595 memcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
596 memcpy(ds->prev, ss->prev, ds->w_size * sizeof(Pos));
597 memcpy(ds->head, ss->head, ds->hash_size * sizeof(Pos));
598 memcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
600 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
601 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
602 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
604 ds->l_desc.dyn_tree = ds->dyn_ltree;
605 ds->d_desc.dyn_tree = ds->dyn_dtree;
606 ds->bl_desc.dyn_tree = ds->bl_tree;
608 return Z_OK;
609 #endif
610 }
611 #endif /* 0 */
613 /* ===========================================================================
614 * Read a new buffer from the current input stream, update the adler32
615 * and total number of bytes read. All deflate() input goes through
616 * this function so some applications may wish to modify it to avoid
617 * allocating a large strm->next_in buffer and copying from it.
618 * (See also flush_pending()).
619 */
620 static int read_buf(
621 z_streamp strm,
622 Byte *buf,
623 unsigned size
624 )
625 {
626 unsigned len = strm->avail_in;
628 if (len > size) len = size;
629 if (len == 0) return 0;
631 strm->avail_in -= len;
633 if (!((deflate_state *)(strm->state))->noheader) {
634 strm->adler = zlib_adler32(strm->adler, strm->next_in, len);
635 }
636 memcpy(buf, strm->next_in, len);
637 strm->next_in += len;
638 strm->total_in += len;
640 return (int)len;
641 }
643 /* ===========================================================================
644 * Initialize the "longest match" routines for a new zlib stream
645 */
646 static void lm_init(
647 deflate_state *s
648 )
649 {
650 s->window_size = (ulg)2L*s->w_size;
652 CLEAR_HASH(s);
654 /* Set the default configuration parameters:
655 */
656 s->max_lazy_match = configuration_table[s->level].max_lazy;
657 s->good_match = configuration_table[s->level].good_length;
658 s->nice_match = configuration_table[s->level].nice_length;
659 s->max_chain_length = configuration_table[s->level].max_chain;
661 s->strstart = 0;
662 s->block_start = 0L;
663 s->lookahead = 0;
664 s->match_length = s->prev_length = MIN_MATCH-1;
665 s->match_available = 0;
666 s->ins_h = 0;
667 }
669 /* ===========================================================================
670 * Set match_start to the longest match starting at the given string and
671 * return its length. Matches shorter or equal to prev_length are discarded,
672 * in which case the result is equal to prev_length and match_start is
673 * garbage.
674 * IN assertions: cur_match is the head of the hash chain for the current
675 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
676 * OUT assertion: the match length is not greater than s->lookahead.
677 */
678 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
679 * match.S. The code will be functionally equivalent.
680 */
681 static uInt longest_match(
682 deflate_state *s,
683 IPos cur_match /* current match */
684 )
685 {
686 unsigned chain_length = s->max_chain_length;/* max hash chain length */
687 register Byte *scan = s->window + s->strstart; /* current string */
688 register Byte *match; /* matched string */
689 register int len; /* length of current match */
690 int best_len = s->prev_length; /* best match length so far */
691 int nice_match = s->nice_match; /* stop if match long enough */
692 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
693 s->strstart - (IPos)MAX_DIST(s) : NIL;
694 /* Stop when cur_match becomes <= limit. To simplify the code,
695 * we prevent matches with the string of window index 0.
696 */
697 Pos *prev = s->prev;
698 uInt wmask = s->w_mask;
700 #ifdef UNALIGNED_OK
701 /* Compare two bytes at a time. Note: this is not always beneficial.
702 * Try with and without -DUNALIGNED_OK to check.
703 */
704 register Byte *strend = s->window + s->strstart + MAX_MATCH - 1;
705 register ush scan_start = *(ush*)scan;
706 register ush scan_end = *(ush*)(scan+best_len-1);
707 #else
708 register Byte *strend = s->window + s->strstart + MAX_MATCH;
709 register Byte scan_end1 = scan[best_len-1];
710 register Byte scan_end = scan[best_len];
711 #endif
713 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
714 * It is easy to get rid of this optimization if necessary.
715 */
716 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
718 /* Do not waste too much time if we already have a good match: */
719 if (s->prev_length >= s->good_match) {
720 chain_length >>= 2;
721 }
722 /* Do not look for matches beyond the end of the input. This is necessary
723 * to make deflate deterministic.
724 */
725 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
727 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
729 do {
730 Assert(cur_match < s->strstart, "no future");
731 match = s->window + cur_match;
733 /* Skip to next match if the match length cannot increase
734 * or if the match length is less than 2:
735 */
736 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
737 /* This code assumes sizeof(unsigned short) == 2. Do not use
738 * UNALIGNED_OK if your compiler uses a different size.
739 */
740 if (*(ush*)(match+best_len-1) != scan_end ||
741 *(ush*)match != scan_start) continue;
743 /* It is not necessary to compare scan[2] and match[2] since they are
744 * always equal when the other bytes match, given that the hash keys
745 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
746 * strstart+3, +5, ... up to strstart+257. We check for insufficient
747 * lookahead only every 4th comparison; the 128th check will be made
748 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
749 * necessary to put more guard bytes at the end of the window, or
750 * to check more often for insufficient lookahead.
751 */
752 Assert(scan[2] == match[2], "scan[2]?");
753 scan++, match++;
754 do {
755 } while (*(ush*)(scan+=2) == *(ush*)(match+=2) &&
756 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
757 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
758 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
759 scan < strend);
760 /* The funny "do {}" generates better code on most compilers */
762 /* Here, scan <= window+strstart+257 */
763 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
764 if (*scan == *match) scan++;
766 len = (MAX_MATCH - 1) - (int)(strend-scan);
767 scan = strend - (MAX_MATCH-1);
769 #else /* UNALIGNED_OK */
771 if (match[best_len] != scan_end ||
772 match[best_len-1] != scan_end1 ||
773 *match != *scan ||
774 *++match != scan[1]) continue;
776 /* The check at best_len-1 can be removed because it will be made
777 * again later. (This heuristic is not always a win.)
778 * It is not necessary to compare scan[2] and match[2] since they
779 * are always equal when the other bytes match, given that
780 * the hash keys are equal and that HASH_BITS >= 8.
781 */
782 scan += 2, match++;
783 Assert(*scan == *match, "match[2]?");
785 /* We check for insufficient lookahead only every 8th comparison;
786 * the 256th check will be made at strstart+258.
787 */
788 do {
789 } while (*++scan == *++match && *++scan == *++match &&
790 *++scan == *++match && *++scan == *++match &&
791 *++scan == *++match && *++scan == *++match &&
792 *++scan == *++match && *++scan == *++match &&
793 scan < strend);
795 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
797 len = MAX_MATCH - (int)(strend - scan);
798 scan = strend - MAX_MATCH;
800 #endif /* UNALIGNED_OK */
802 if (len > best_len) {
803 s->match_start = cur_match;
804 best_len = len;
805 if (len >= nice_match) break;
806 #ifdef UNALIGNED_OK
807 scan_end = *(ush*)(scan+best_len-1);
808 #else
809 scan_end1 = scan[best_len-1];
810 scan_end = scan[best_len];
811 #endif
812 }
813 } while ((cur_match = prev[cur_match & wmask]) > limit
814 && --chain_length != 0);
816 if ((uInt)best_len <= s->lookahead) return best_len;
817 return s->lookahead;
818 }
820 #ifdef DEBUG_ZLIB
821 /* ===========================================================================
822 * Check that the match at match_start is indeed a match.
823 */
824 static void check_match(
825 deflate_state *s,
826 IPos start,
827 IPos match,
828 int length
829 )
830 {
831 /* check that the match is indeed a match */
832 if (memcmp((char *)s->window + match,
833 (char *)s->window + start, length) != EQUAL) {
834 fprintf(stderr, " start %u, match %u, length %d\n",
835 start, match, length);
836 do {
837 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
838 } while (--length != 0);
839 z_error("invalid match");
840 }
841 if (z_verbose > 1) {
842 fprintf(stderr,"\\[%d,%d]", start-match, length);
843 do { putc(s->window[start++], stderr); } while (--length != 0);
844 }
845 }
846 #else
847 # define check_match(s, start, match, length)
848 #endif
850 /* ===========================================================================
851 * Fill the window when the lookahead becomes insufficient.
852 * Updates strstart and lookahead.
853 *
854 * IN assertion: lookahead < MIN_LOOKAHEAD
855 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
856 * At least one byte has been read, or avail_in == 0; reads are
857 * performed for at least two bytes (required for the zip translate_eol
858 * option -- not supported here).
859 */
860 static void fill_window(
861 deflate_state *s
862 )
863 {
864 register unsigned n, m;
865 register Pos *p;
866 unsigned more; /* Amount of free space at the end of the window. */
867 uInt wsize = s->w_size;
869 do {
870 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
872 /* Deal with !@#$% 64K limit: */
873 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
874 more = wsize;
876 } else if (more == (unsigned)(-1)) {
877 /* Very unlikely, but possible on 16 bit machine if strstart == 0
878 * and lookahead == 1 (input done one byte at time)
879 */
880 more--;
882 /* If the window is almost full and there is insufficient lookahead,
883 * move the upper half to the lower one to make room in the upper half.
884 */
885 } else if (s->strstart >= wsize+MAX_DIST(s)) {
887 memcpy((char *)s->window, (char *)s->window+wsize,
888 (unsigned)wsize);
889 s->match_start -= wsize;
890 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
891 s->block_start -= (long) wsize;
893 /* Slide the hash table (could be avoided with 32 bit values
894 at the expense of memory usage). We slide even when level == 0
895 to keep the hash table consistent if we switch back to level > 0
896 later. (Using level 0 permanently is not an optimal usage of
897 zlib, so we don't care about this pathological case.)
898 */
899 n = s->hash_size;
900 p = &s->head[n];
901 do {
902 m = *--p;
903 *p = (Pos)(m >= wsize ? m-wsize : NIL);
904 } while (--n);
906 n = wsize;
907 p = &s->prev[n];
908 do {
909 m = *--p;
910 *p = (Pos)(m >= wsize ? m-wsize : NIL);
911 /* If n is not on any hash chain, prev[n] is garbage but
912 * its value will never be used.
913 */
914 } while (--n);
915 more += wsize;
916 }
917 if (s->strm->avail_in == 0) return;
919 /* If there was no sliding:
920 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
921 * more == window_size - lookahead - strstart
922 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
923 * => more >= window_size - 2*WSIZE + 2
924 * In the BIG_MEM or MMAP case (not yet supported),
925 * window_size == input_size + MIN_LOOKAHEAD &&
926 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
927 * Otherwise, window_size == 2*WSIZE so more >= 2.
928 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
929 */
930 Assert(more >= 2, "more < 2");
932 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
933 s->lookahead += n;
935 /* Initialize the hash value now that we have some input: */
936 if (s->lookahead >= MIN_MATCH) {
937 s->ins_h = s->window[s->strstart];
938 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
939 #if MIN_MATCH != 3
940 Call UPDATE_HASH() MIN_MATCH-3 more times
941 #endif
942 }
943 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
944 * but this is not important since only literal bytes will be emitted.
945 */
947 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
948 }
950 /* ===========================================================================
951 * Flush the current block, with given end-of-file flag.
952 * IN assertion: strstart is set to the end of the current match.
953 */
954 #define FLUSH_BLOCK_ONLY(s, eof) { \
955 zlib_tr_flush_block(s, (s->block_start >= 0L ? \
956 (char *)&s->window[(unsigned)s->block_start] : \
957 NULL), \
958 (ulg)((long)s->strstart - s->block_start), \
959 (eof)); \
960 s->block_start = s->strstart; \
961 flush_pending(s->strm); \
962 Tracev((stderr,"[FLUSH]")); \
963 }
965 /* Same but force premature exit if necessary. */
966 #define FLUSH_BLOCK(s, eof) { \
967 FLUSH_BLOCK_ONLY(s, eof); \
968 if (s->strm->avail_out == 0) return (eof) ? finish_started : need_more; \
969 }
971 /* ===========================================================================
972 * Copy without compression as much as possible from the input stream, return
973 * the current block state.
974 * This function does not insert new strings in the dictionary since
975 * uncompressible data is probably not useful. This function is used
976 * only for the level=0 compression option.
977 * NOTE: this function should be optimized to avoid extra copying from
978 * window to pending_buf.
979 */
980 static block_state deflate_stored(
981 deflate_state *s,
982 int flush
983 )
984 {
985 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited
986 * to pending_buf_size, and each stored block has a 5 byte header:
987 */
988 ulg max_block_size = 0xffff;
989 ulg max_start;
991 if (max_block_size > s->pending_buf_size - 5) {
992 max_block_size = s->pending_buf_size - 5;
993 }
995 /* Copy as much as possible from input to output: */
996 for (;;) {
997 /* Fill the window as much as possible: */
998 if (s->lookahead <= 1) {
1000 Assert(s->strstart < s->w_size+MAX_DIST(s) ||
1001 s->block_start >= (long)s->w_size, "slide too late");
1003 fill_window(s);
1004 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
1006 if (s->lookahead == 0) break; /* flush the current block */
1008 Assert(s->block_start >= 0L, "block gone");
1010 s->strstart += s->lookahead;
1011 s->lookahead = 0;
1013 /* Emit a stored block if pending_buf will be full: */
1014 max_start = s->block_start + max_block_size;
1015 if (s->strstart == 0 || (ulg)s->strstart >= max_start) {
1016 /* strstart == 0 is possible when wraparound on 16-bit machine */
1017 s->lookahead = (uInt)(s->strstart - max_start);
1018 s->strstart = (uInt)max_start;
1019 FLUSH_BLOCK(s, 0);
1021 /* Flush if we may have to slide, otherwise block_start may become
1022 * negative and the data will be gone:
1023 */
1024 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
1025 FLUSH_BLOCK(s, 0);
1028 FLUSH_BLOCK(s, flush == Z_FINISH);
1029 return flush == Z_FINISH ? finish_done : block_done;
1032 /* ===========================================================================
1033 * Compress as much as possible from the input stream, return the current
1034 * block state.
1035 * This function does not perform lazy evaluation of matches and inserts
1036 * new strings in the dictionary only for unmatched strings or for short
1037 * matches. It is used only for the fast compression options.
1038 */
1039 static block_state deflate_fast(
1040 deflate_state *s,
1041 int flush
1044 IPos hash_head = NIL; /* head of the hash chain */
1045 int bflush; /* set if current block must be flushed */
1047 for (;;) {
1048 /* Make sure that we always have enough lookahead, except
1049 * at the end of the input file. We need MAX_MATCH bytes
1050 * for the next match, plus MIN_MATCH bytes to insert the
1051 * string following the next match.
1052 */
1053 if (s->lookahead < MIN_LOOKAHEAD) {
1054 fill_window(s);
1055 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1056 return need_more;
1058 if (s->lookahead == 0) break; /* flush the current block */
1061 /* Insert the string window[strstart .. strstart+2] in the
1062 * dictionary, and set hash_head to the head of the hash chain:
1063 */
1064 if (s->lookahead >= MIN_MATCH) {
1065 INSERT_STRING(s, s->strstart, hash_head);
1068 /* Find the longest match, discarding those <= prev_length.
1069 * At this point we have always match_length < MIN_MATCH
1070 */
1071 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1072 /* To simplify the code, we prevent matches with the string
1073 * of window index 0 (in particular we have to avoid a match
1074 * of the string with itself at the start of the input file).
1075 */
1076 if (s->strategy != Z_HUFFMAN_ONLY) {
1077 s->match_length = longest_match (s, hash_head);
1079 /* longest_match() sets match_start */
1081 if (s->match_length >= MIN_MATCH) {
1082 check_match(s, s->strstart, s->match_start, s->match_length);
1084 bflush = zlib_tr_tally(s, s->strstart - s->match_start,
1085 s->match_length - MIN_MATCH);
1087 s->lookahead -= s->match_length;
1089 /* Insert new strings in the hash table only if the match length
1090 * is not too large. This saves time but degrades compression.
1091 */
1092 if (s->match_length <= s->max_insert_length &&
1093 s->lookahead >= MIN_MATCH) {
1094 s->match_length--; /* string at strstart already in hash table */
1095 do {
1096 s->strstart++;
1097 INSERT_STRING(s, s->strstart, hash_head);
1098 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1099 * always MIN_MATCH bytes ahead.
1100 */
1101 } while (--s->match_length != 0);
1102 s->strstart++;
1103 } else {
1104 s->strstart += s->match_length;
1105 s->match_length = 0;
1106 s->ins_h = s->window[s->strstart];
1107 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1108 #if MIN_MATCH != 3
1109 Call UPDATE_HASH() MIN_MATCH-3 more times
1110 #endif
1111 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1112 * matter since it will be recomputed at next deflate call.
1113 */
1115 } else {
1116 /* No match, output a literal byte */
1117 Tracevv((stderr,"%c", s->window[s->strstart]));
1118 bflush = zlib_tr_tally (s, 0, s->window[s->strstart]);
1119 s->lookahead--;
1120 s->strstart++;
1122 if (bflush) FLUSH_BLOCK(s, 0);
1124 FLUSH_BLOCK(s, flush == Z_FINISH);
1125 return flush == Z_FINISH ? finish_done : block_done;
1128 /* ===========================================================================
1129 * Same as above, but achieves better compression. We use a lazy
1130 * evaluation for matches: a match is finally adopted only if there is
1131 * no better match at the next window position.
1132 */
1133 static block_state deflate_slow(
1134 deflate_state *s,
1135 int flush
1138 IPos hash_head = NIL; /* head of hash chain */
1139 int bflush; /* set if current block must be flushed */
1141 /* Process the input block. */
1142 for (;;) {
1143 /* Make sure that we always have enough lookahead, except
1144 * at the end of the input file. We need MAX_MATCH bytes
1145 * for the next match, plus MIN_MATCH bytes to insert the
1146 * string following the next match.
1147 */
1148 if (s->lookahead < MIN_LOOKAHEAD) {
1149 fill_window(s);
1150 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1151 return need_more;
1153 if (s->lookahead == 0) break; /* flush the current block */
1156 /* Insert the string window[strstart .. strstart+2] in the
1157 * dictionary, and set hash_head to the head of the hash chain:
1158 */
1159 if (s->lookahead >= MIN_MATCH) {
1160 INSERT_STRING(s, s->strstart, hash_head);
1163 /* Find the longest match, discarding those <= prev_length.
1164 */
1165 s->prev_length = s->match_length, s->prev_match = s->match_start;
1166 s->match_length = MIN_MATCH-1;
1168 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1169 s->strstart - hash_head <= MAX_DIST(s)) {
1170 /* To simplify the code, we prevent matches with the string
1171 * of window index 0 (in particular we have to avoid a match
1172 * of the string with itself at the start of the input file).
1173 */
1174 if (s->strategy != Z_HUFFMAN_ONLY) {
1175 s->match_length = longest_match (s, hash_head);
1177 /* longest_match() sets match_start */
1179 if (s->match_length <= 5 && (s->strategy == Z_FILTERED ||
1180 (s->match_length == MIN_MATCH &&
1181 s->strstart - s->match_start > TOO_FAR))) {
1183 /* If prev_match is also MIN_MATCH, match_start is garbage
1184 * but we will ignore the current match anyway.
1185 */
1186 s->match_length = MIN_MATCH-1;
1189 /* If there was a match at the previous step and the current
1190 * match is not better, output the previous match:
1191 */
1192 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1193 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1194 /* Do not insert strings in hash table beyond this. */
1196 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1198 bflush = zlib_tr_tally(s, s->strstart -1 - s->prev_match,
1199 s->prev_length - MIN_MATCH);
1201 /* Insert in hash table all strings up to the end of the match.
1202 * strstart-1 and strstart are already inserted. If there is not
1203 * enough lookahead, the last two strings are not inserted in
1204 * the hash table.
1205 */
1206 s->lookahead -= s->prev_length-1;
1207 s->prev_length -= 2;
1208 do {
1209 if (++s->strstart <= max_insert) {
1210 INSERT_STRING(s, s->strstart, hash_head);
1212 } while (--s->prev_length != 0);
1213 s->match_available = 0;
1214 s->match_length = MIN_MATCH-1;
1215 s->strstart++;
1217 if (bflush) FLUSH_BLOCK(s, 0);
1219 } else if (s->match_available) {
1220 /* If there was no match at the previous position, output a
1221 * single literal. If there was a match but the current match
1222 * is longer, truncate the previous match to a single literal.
1223 */
1224 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1225 if (zlib_tr_tally (s, 0, s->window[s->strstart-1])) {
1226 FLUSH_BLOCK_ONLY(s, 0);
1228 s->strstart++;
1229 s->lookahead--;
1230 if (s->strm->avail_out == 0) return need_more;
1231 } else {
1232 /* There is no previous match to compare with, wait for
1233 * the next step to decide.
1234 */
1235 s->match_available = 1;
1236 s->strstart++;
1237 s->lookahead--;
1240 Assert (flush != Z_NO_FLUSH, "no flush?");
1241 if (s->match_available) {
1242 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1243 zlib_tr_tally (s, 0, s->window[s->strstart-1]);
1244 s->match_available = 0;
1246 FLUSH_BLOCK(s, flush == Z_FINISH);
1247 return flush == Z_FINISH ? finish_done : block_done;
1250 int zlib_deflate_workspacesize(void)
1252 return sizeof(deflate_workspace);