aboutsummaryrefslogtreecommitdiffstats
path: root/lib/zlib_deflate/deflate.c
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 18:20:36 -0400
committerLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 18:20:36 -0400
commit1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 (patch)
tree0bba044c4ce775e45a88a51686b5d9f90697ea9d /lib/zlib_deflate/deflate.c
Linux-2.6.12-rc2v2.6.12-rc2
Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
Diffstat (limited to 'lib/zlib_deflate/deflate.c')
-rw-r--r--lib/zlib_deflate/deflate.c1268
1 files changed, 1268 insertions, 0 deletions
diff --git a/lib/zlib_deflate/deflate.c b/lib/zlib_deflate/deflate.c
new file mode 100644
index 000000000000..ad9a1bf4fc63
--- /dev/null
+++ b/lib/zlib_deflate/deflate.c
@@ -0,0 +1,1268 @@
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 */
6
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 */
50
51#include <linux/module.h>
52#include <linux/zutil.h>
53#include "defutil.h"
54
55
56/* ===========================================================================
57 * Function prototypes.
58 */
59typedef 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;
65
66typedef block_state (*compress_func) (deflate_state *s, int flush);
67/* Compression function. Returns the block state after the call. */
68
69static void fill_window (deflate_state *s);
70static block_state deflate_stored (deflate_state *s, int flush);
71static block_state deflate_fast (deflate_state *s, int flush);
72static block_state deflate_slow (deflate_state *s, int flush);
73static void lm_init (deflate_state *s);
74static void putShortMSB (deflate_state *s, uInt b);
75static void flush_pending (z_streamp strm);
76static int read_buf (z_streamp strm, Byte *buf, unsigned size);
77static uInt longest_match (deflate_state *s, IPos cur_match);
78
79#ifdef DEBUG_ZLIB
80static void check_match (deflate_state *s, IPos start, IPos match,
81 int length);
82#endif
83
84/* ===========================================================================
85 * Local data
86 */
87
88#define NIL 0
89/* Tail of hash chains */
90
91#ifndef TOO_FAR
92# define TOO_FAR 4096
93#endif
94/* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
95
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 */
100
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 */
106typedef 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;
113
114static 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},
120
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 */
127
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 */
132
133#define EQUAL 0
134/* result of memcmp for equal strings */
135
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)
143
144
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))
157
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));
165
166/* ========================================================================= */
167int zlib_deflateInit_(
168 z_streamp strm,
169 int level,
170 const char *version,
171 int stream_size
172)
173{
174 return zlib_deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS,
175 DEF_MEM_LEVEL,
176 Z_DEFAULT_STRATEGY, version, stream_size);
177 /* To do: ignore strm->next_in if we use it as window */
178}
179
180/* ========================================================================= */
181int zlib_deflateInit2_(
182 z_streamp strm,
183 int level,
184 int method,
185 int windowBits,
186 int memLevel,
187 int strategy,
188 const char *version,
189 int stream_size
190)
191{
192 deflate_state *s;
193 int noheader = 0;
194 static char* my_version = ZLIB_VERSION;
195 deflate_workspace *mem;
196
197 ush *overlay;
198 /* We overlay pending_buf and d_buf+l_buf. This works since the average
199 * output size for (length,distance) codes is <= 24 bits.
200 */
201
202 if (version == NULL || version[0] != my_version[0] ||
203 stream_size != sizeof(z_stream)) {
204 return Z_VERSION_ERROR;
205 }
206 if (strm == NULL) return Z_STREAM_ERROR;
207
208 strm->msg = NULL;
209
210 if (level == Z_DEFAULT_COMPRESSION) level = 6;
211
212 mem = (deflate_workspace *) strm->workspace;
213
214 if (windowBits < 0) { /* undocumented feature: suppress zlib header */
215 noheader = 1;
216 windowBits = -windowBits;
217 }
218 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
219 windowBits < 9 || windowBits > 15 || level < 0 || level > 9 ||
220 strategy < 0 || strategy > Z_HUFFMAN_ONLY) {
221 return Z_STREAM_ERROR;
222 }
223 s = (deflate_state *) &(mem->deflate_memory);
224 strm->state = (struct internal_state *)s;
225 s->strm = strm;
226
227 s->noheader = noheader;
228 s->w_bits = windowBits;
229 s->w_size = 1 << s->w_bits;
230 s->w_mask = s->w_size - 1;
231
232 s->hash_bits = memLevel + 7;
233 s->hash_size = 1 << s->hash_bits;
234 s->hash_mask = s->hash_size - 1;
235 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
236
237 s->window = (Byte *) mem->window_memory;
238 s->prev = (Pos *) mem->prev_memory;
239 s->head = (Pos *) mem->head_memory;
240
241 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
242
243 overlay = (ush *) mem->overlay_memory;
244 s->pending_buf = (uch *) overlay;
245 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
246
247 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
248 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
249
250 s->level = level;
251 s->strategy = strategy;
252 s->method = (Byte)method;
253
254 return zlib_deflateReset(strm);
255}
256
257/* ========================================================================= */
258int zlib_deflateSetDictionary(
259 z_streamp strm,
260 const Byte *dictionary,
261 uInt dictLength
262)
263{
264 deflate_state *s;
265 uInt length = dictLength;
266 uInt n;
267 IPos hash_head = 0;
268
269 if (strm == NULL || strm->state == NULL || dictionary == NULL)
270 return Z_STREAM_ERROR;
271
272 s = (deflate_state *) strm->state;
273 if (s->status != INIT_STATE) return Z_STREAM_ERROR;
274
275 strm->adler = zlib_adler32(strm->adler, dictionary, dictLength);
276
277 if (length < MIN_MATCH) return Z_OK;
278 if (length > MAX_DIST(s)) {
279 length = MAX_DIST(s);
280#ifndef USE_DICT_HEAD
281 dictionary += dictLength - length; /* use the tail of the dictionary */
282#endif
283 }
284 memcpy((char *)s->window, dictionary, length);
285 s->strstart = length;
286 s->block_start = (long)length;
287
288 /* Insert all strings in the hash table (except for the last two bytes).
289 * s->lookahead stays null, so s->ins_h will be recomputed at the next
290 * call of fill_window.
291 */
292 s->ins_h = s->window[0];
293 UPDATE_HASH(s, s->ins_h, s->window[1]);
294 for (n = 0; n <= length - MIN_MATCH; n++) {
295 INSERT_STRING(s, n, hash_head);
296 }
297 if (hash_head) hash_head = 0; /* to make compiler happy */
298 return Z_OK;
299}
300
301/* ========================================================================= */
302int zlib_deflateReset(
303 z_streamp strm
304)
305{
306 deflate_state *s;
307
308 if (strm == NULL || strm->state == NULL)
309 return Z_STREAM_ERROR;
310
311 strm->total_in = strm->total_out = 0;
312 strm->msg = NULL;
313 strm->data_type = Z_UNKNOWN;
314
315 s = (deflate_state *)strm->state;
316 s->pending = 0;
317 s->pending_out = s->pending_buf;
318
319 if (s->noheader < 0) {
320 s->noheader = 0; /* was set to -1 by deflate(..., Z_FINISH); */
321 }
322 s->status = s->noheader ? BUSY_STATE : INIT_STATE;
323 strm->adler = 1;
324 s->last_flush = Z_NO_FLUSH;
325
326 zlib_tr_init(s);
327 lm_init(s);
328
329 return Z_OK;
330}
331
332/* ========================================================================= */
333int zlib_deflateParams(
334 z_streamp strm,
335 int level,
336 int strategy
337)
338{
339 deflate_state *s;
340 compress_func func;
341 int err = Z_OK;
342
343 if (strm == NULL || strm->state == NULL) return Z_STREAM_ERROR;
344 s = (deflate_state *) strm->state;
345
346 if (level == Z_DEFAULT_COMPRESSION) {
347 level = 6;
348 }
349 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_HUFFMAN_ONLY) {
350 return Z_STREAM_ERROR;
351 }
352 func = configuration_table[s->level].func;
353
354 if (func != configuration_table[level].func && strm->total_in != 0) {
355 /* Flush the last buffer: */
356 err = zlib_deflate(strm, Z_PARTIAL_FLUSH);
357 }
358 if (s->level != level) {
359 s->level = level;
360 s->max_lazy_match = configuration_table[level].max_lazy;
361 s->good_match = configuration_table[level].good_length;
362 s->nice_match = configuration_table[level].nice_length;
363 s->max_chain_length = configuration_table[level].max_chain;
364 }
365 s->strategy = strategy;
366 return err;
367}
368
369/* =========================================================================
370 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
371 * IN assertion: the stream state is correct and there is enough room in
372 * pending_buf.
373 */
374static void putShortMSB(
375 deflate_state *s,
376 uInt b
377)
378{
379 put_byte(s, (Byte)(b >> 8));
380 put_byte(s, (Byte)(b & 0xff));
381}
382
383/* =========================================================================
384 * Flush as much pending output as possible. All deflate() output goes
385 * through this function so some applications may wish to modify it
386 * to avoid allocating a large strm->next_out buffer and copying into it.
387 * (See also read_buf()).
388 */
389static void flush_pending(
390 z_streamp strm
391)
392{
393 deflate_state *s = (deflate_state *) strm->state;
394 unsigned len = s->pending;
395
396 if (len > strm->avail_out) len = strm->avail_out;
397 if (len == 0) return;
398
399 if (strm->next_out != NULL) {
400 memcpy(strm->next_out, s->pending_out, len);
401 strm->next_out += len;
402 }
403 s->pending_out += len;
404 strm->total_out += len;
405 strm->avail_out -= len;
406 s->pending -= len;
407 if (s->pending == 0) {
408 s->pending_out = s->pending_buf;
409 }
410}
411
412/* ========================================================================= */
413int zlib_deflate(
414 z_streamp strm,
415 int flush
416)
417{
418 int old_flush; /* value of flush param for previous deflate call */
419 deflate_state *s;
420
421 if (strm == NULL || strm->state == NULL ||
422 flush > Z_FINISH || flush < 0) {
423 return Z_STREAM_ERROR;
424 }
425 s = (deflate_state *) strm->state;
426
427 if ((strm->next_in == NULL && strm->avail_in != 0) ||
428 (s->status == FINISH_STATE && flush != Z_FINISH)) {
429 return Z_STREAM_ERROR;
430 }
431 if (strm->avail_out == 0) return Z_BUF_ERROR;
432
433 s->strm = strm; /* just in case */
434 old_flush = s->last_flush;
435 s->last_flush = flush;
436
437 /* Write the zlib header */
438 if (s->status == INIT_STATE) {
439
440 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
441 uInt level_flags = (s->level-1) >> 1;
442
443 if (level_flags > 3) level_flags = 3;
444 header |= (level_flags << 6);
445 if (s->strstart != 0) header |= PRESET_DICT;
446 header += 31 - (header % 31);
447
448 s->status = BUSY_STATE;
449 putShortMSB(s, header);
450
451 /* Save the adler32 of the preset dictionary: */
452 if (s->strstart != 0) {
453 putShortMSB(s, (uInt)(strm->adler >> 16));
454 putShortMSB(s, (uInt)(strm->adler & 0xffff));
455 }
456 strm->adler = 1L;
457 }
458
459 /* Flush as much pending output as possible */
460 if (s->pending != 0) {
461 flush_pending(strm);
462 if (strm->avail_out == 0) {
463 /* Since avail_out is 0, deflate will be called again with
464 * more output space, but possibly with both pending and
465 * avail_in equal to zero. There won't be anything to do,
466 * but this is not an error situation so make sure we
467 * return OK instead of BUF_ERROR at next call of deflate:
468 */
469 s->last_flush = -1;
470 return Z_OK;
471 }
472
473 /* Make sure there is something to do and avoid duplicate consecutive
474 * flushes. For repeated and useless calls with Z_FINISH, we keep
475 * returning Z_STREAM_END instead of Z_BUFF_ERROR.
476 */
477 } else if (strm->avail_in == 0 && flush <= old_flush &&
478 flush != Z_FINISH) {
479 return Z_BUF_ERROR;
480 }
481
482 /* User must not provide more input after the first FINISH: */
483 if (s->status == FINISH_STATE && strm->avail_in != 0) {
484 return Z_BUF_ERROR;
485 }
486
487 /* Start a new block or continue the current one.
488 */
489 if (strm->avail_in != 0 || s->lookahead != 0 ||
490 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
491 block_state bstate;
492
493 bstate = (*(configuration_table[s->level].func))(s, flush);
494
495 if (bstate == finish_started || bstate == finish_done) {
496 s->status = FINISH_STATE;
497 }
498 if (bstate == need_more || bstate == finish_started) {
499 if (strm->avail_out == 0) {
500 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
501 }
502 return Z_OK;
503 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
504 * of deflate should use the same flush parameter to make sure
505 * that the flush is complete. So we don't have to output an
506 * empty block here, this will be done at next call. This also
507 * ensures that for a very small output buffer, we emit at most
508 * one empty block.
509 */
510 }
511 if (bstate == block_done) {
512 if (flush == Z_PARTIAL_FLUSH) {
513 zlib_tr_align(s);
514 } else if (flush == Z_PACKET_FLUSH) {
515 /* Output just the 3-bit `stored' block type value,
516 but not a zero length. */
517 zlib_tr_stored_type_only(s);
518 } else { /* FULL_FLUSH or SYNC_FLUSH */
519 zlib_tr_stored_block(s, (char*)0, 0L, 0);
520 /* For a full flush, this empty block will be recognized
521 * as a special marker by inflate_sync().
522 */
523 if (flush == Z_FULL_FLUSH) {
524 CLEAR_HASH(s); /* forget history */
525 }
526 }
527 flush_pending(strm);
528 if (strm->avail_out == 0) {
529 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
530 return Z_OK;
531 }
532 }
533 }
534 Assert(strm->avail_out > 0, "bug2");
535
536 if (flush != Z_FINISH) return Z_OK;
537 if (s->noheader) return Z_STREAM_END;
538
539 /* Write the zlib trailer (adler32) */
540 putShortMSB(s, (uInt)(strm->adler >> 16));
541 putShortMSB(s, (uInt)(strm->adler & 0xffff));
542 flush_pending(strm);
543 /* If avail_out is zero, the application will call deflate again
544 * to flush the rest.
545 */
546 s->noheader = -1; /* write the trailer only once! */
547 return s->pending != 0 ? Z_OK : Z_STREAM_END;
548}
549
550/* ========================================================================= */
551int zlib_deflateEnd(
552 z_streamp strm
553)
554{
555 int status;
556 deflate_state *s;
557
558 if (strm == NULL || strm->state == NULL) return Z_STREAM_ERROR;
559 s = (deflate_state *) strm->state;
560
561 status = s->status;
562 if (status != INIT_STATE && status != BUSY_STATE &&
563 status != FINISH_STATE) {
564 return Z_STREAM_ERROR;
565 }
566
567 strm->state = NULL;
568
569 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
570}
571
572/* =========================================================================
573 * Copy the source state to the destination state.
574 */
575int zlib_deflateCopy (
576 z_streamp dest,
577 z_streamp source
578)
579{
580#ifdef MAXSEG_64K
581 return Z_STREAM_ERROR;
582#else
583 deflate_state *ds;
584 deflate_state *ss;
585 ush *overlay;
586 deflate_workspace *mem;
587
588
589 if (source == NULL || dest == NULL || source->state == NULL) {
590 return Z_STREAM_ERROR;
591 }
592
593 ss = (deflate_state *) source->state;
594
595 *dest = *source;
596
597 mem = (deflate_workspace *) dest->workspace;
598
599 ds = &(mem->deflate_memory);
600
601 dest->state = (struct internal_state *) ds;
602 *ds = *ss;
603 ds->strm = dest;
604
605 ds->window = (Byte *) mem->window_memory;
606 ds->prev = (Pos *) mem->prev_memory;
607 ds->head = (Pos *) mem->head_memory;
608 overlay = (ush *) mem->overlay_memory;
609 ds->pending_buf = (uch *) overlay;
610
611 memcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
612 memcpy(ds->prev, ss->prev, ds->w_size * sizeof(Pos));
613 memcpy(ds->head, ss->head, ds->hash_size * sizeof(Pos));
614 memcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
615
616 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
617 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
618 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
619
620 ds->l_desc.dyn_tree = ds->dyn_ltree;
621 ds->d_desc.dyn_tree = ds->dyn_dtree;
622 ds->bl_desc.dyn_tree = ds->bl_tree;
623
624 return Z_OK;
625#endif
626}
627
628/* ===========================================================================
629 * Read a new buffer from the current input stream, update the adler32
630 * and total number of bytes read. All deflate() input goes through
631 * this function so some applications may wish to modify it to avoid
632 * allocating a large strm->next_in buffer and copying from it.
633 * (See also flush_pending()).
634 */
635static int read_buf(
636 z_streamp strm,
637 Byte *buf,
638 unsigned size
639)
640{
641 unsigned len = strm->avail_in;
642
643 if (len > size) len = size;
644 if (len == 0) return 0;
645
646 strm->avail_in -= len;
647
648 if (!((deflate_state *)(strm->state))->noheader) {
649 strm->adler = zlib_adler32(strm->adler, strm->next_in, len);
650 }
651 memcpy(buf, strm->next_in, len);
652 strm->next_in += len;
653 strm->total_in += len;
654
655 return (int)len;
656}
657
658/* ===========================================================================
659 * Initialize the "longest match" routines for a new zlib stream
660 */
661static void lm_init(
662 deflate_state *s
663)
664{
665 s->window_size = (ulg)2L*s->w_size;
666
667 CLEAR_HASH(s);
668
669 /* Set the default configuration parameters:
670 */
671 s->max_lazy_match = configuration_table[s->level].max_lazy;
672 s->good_match = configuration_table[s->level].good_length;
673 s->nice_match = configuration_table[s->level].nice_length;
674 s->max_chain_length = configuration_table[s->level].max_chain;
675
676 s->strstart = 0;
677 s->block_start = 0L;
678 s->lookahead = 0;
679 s->match_length = s->prev_length = MIN_MATCH-1;
680 s->match_available = 0;
681 s->ins_h = 0;
682}
683
684/* ===========================================================================
685 * Set match_start to the longest match starting at the given string and
686 * return its length. Matches shorter or equal to prev_length are discarded,
687 * in which case the result is equal to prev_length and match_start is
688 * garbage.
689 * IN assertions: cur_match is the head of the hash chain for the current
690 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
691 * OUT assertion: the match length is not greater than s->lookahead.
692 */
693/* For 80x86 and 680x0, an optimized version will be provided in match.asm or
694 * match.S. The code will be functionally equivalent.
695 */
696static uInt longest_match(
697 deflate_state *s,
698 IPos cur_match /* current match */
699)
700{
701 unsigned chain_length = s->max_chain_length;/* max hash chain length */
702 register Byte *scan = s->window + s->strstart; /* current string */
703 register Byte *match; /* matched string */
704 register int len; /* length of current match */
705 int best_len = s->prev_length; /* best match length so far */
706 int nice_match = s->nice_match; /* stop if match long enough */
707 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
708 s->strstart - (IPos)MAX_DIST(s) : NIL;
709 /* Stop when cur_match becomes <= limit. To simplify the code,
710 * we prevent matches with the string of window index 0.
711 */
712 Pos *prev = s->prev;
713 uInt wmask = s->w_mask;
714
715#ifdef UNALIGNED_OK
716 /* Compare two bytes at a time. Note: this is not always beneficial.
717 * Try with and without -DUNALIGNED_OK to check.
718 */
719 register Byte *strend = s->window + s->strstart + MAX_MATCH - 1;
720 register ush scan_start = *(ush*)scan;
721 register ush scan_end = *(ush*)(scan+best_len-1);
722#else
723 register Byte *strend = s->window + s->strstart + MAX_MATCH;
724 register Byte scan_end1 = scan[best_len-1];
725 register Byte scan_end = scan[best_len];
726#endif
727
728 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
729 * It is easy to get rid of this optimization if necessary.
730 */
731 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
732
733 /* Do not waste too much time if we already have a good match: */
734 if (s->prev_length >= s->good_match) {
735 chain_length >>= 2;
736 }
737 /* Do not look for matches beyond the end of the input. This is necessary
738 * to make deflate deterministic.
739 */
740 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
741
742 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
743
744 do {
745 Assert(cur_match < s->strstart, "no future");
746 match = s->window + cur_match;
747
748 /* Skip to next match if the match length cannot increase
749 * or if the match length is less than 2:
750 */
751#if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
752 /* This code assumes sizeof(unsigned short) == 2. Do not use
753 * UNALIGNED_OK if your compiler uses a different size.
754 */
755 if (*(ush*)(match+best_len-1) != scan_end ||
756 *(ush*)match != scan_start) continue;
757
758 /* It is not necessary to compare scan[2] and match[2] since they are
759 * always equal when the other bytes match, given that the hash keys
760 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
761 * strstart+3, +5, ... up to strstart+257. We check for insufficient
762 * lookahead only every 4th comparison; the 128th check will be made
763 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
764 * necessary to put more guard bytes at the end of the window, or
765 * to check more often for insufficient lookahead.
766 */
767 Assert(scan[2] == match[2], "scan[2]?");
768 scan++, match++;
769 do {
770 } while (*(ush*)(scan+=2) == *(ush*)(match+=2) &&
771 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
772 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
773 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
774 scan < strend);
775 /* The funny "do {}" generates better code on most compilers */
776
777 /* Here, scan <= window+strstart+257 */
778 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
779 if (*scan == *match) scan++;
780
781 len = (MAX_MATCH - 1) - (int)(strend-scan);
782 scan = strend - (MAX_MATCH-1);
783
784#else /* UNALIGNED_OK */
785
786 if (match[best_len] != scan_end ||
787 match[best_len-1] != scan_end1 ||
788 *match != *scan ||
789 *++match != scan[1]) continue;
790
791 /* The check at best_len-1 can be removed because it will be made
792 * again later. (This heuristic is not always a win.)
793 * It is not necessary to compare scan[2] and match[2] since they
794 * are always equal when the other bytes match, given that
795 * the hash keys are equal and that HASH_BITS >= 8.
796 */
797 scan += 2, match++;
798 Assert(*scan == *match, "match[2]?");
799
800 /* We check for insufficient lookahead only every 8th comparison;
801 * the 256th check will be made at strstart+258.
802 */
803 do {
804 } while (*++scan == *++match && *++scan == *++match &&
805 *++scan == *++match && *++scan == *++match &&
806 *++scan == *++match && *++scan == *++match &&
807 *++scan == *++match && *++scan == *++match &&
808 scan < strend);
809
810 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
811
812 len = MAX_MATCH - (int)(strend - scan);
813 scan = strend - MAX_MATCH;
814
815#endif /* UNALIGNED_OK */
816
817 if (len > best_len) {
818 s->match_start = cur_match;
819 best_len = len;
820 if (len >= nice_match) break;
821#ifdef UNALIGNED_OK
822 scan_end = *(ush*)(scan+best_len-1);
823#else
824 scan_end1 = scan[best_len-1];
825 scan_end = scan[best_len];
826#endif
827 }
828 } while ((cur_match = prev[cur_match & wmask]) > limit
829 && --chain_length != 0);
830
831 if ((uInt)best_len <= s->lookahead) return best_len;
832 return s->lookahead;
833}
834
835#ifdef DEBUG_ZLIB
836/* ===========================================================================
837 * Check that the match at match_start is indeed a match.
838 */
839static void check_match(
840 deflate_state *s,
841 IPos start,
842 IPos match,
843 int length
844)
845{
846 /* check that the match is indeed a match */
847 if (memcmp((char *)s->window + match,
848 (char *)s->window + start, length) != EQUAL) {
849 fprintf(stderr, " start %u, match %u, length %d\n",
850 start, match, length);
851 do {
852 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
853 } while (--length != 0);
854 z_error("invalid match");
855 }
856 if (z_verbose > 1) {
857 fprintf(stderr,"\\[%d,%d]", start-match, length);
858 do { putc(s->window[start++], stderr); } while (--length != 0);
859 }
860}
861#else
862# define check_match(s, start, match, length)
863#endif
864
865/* ===========================================================================
866 * Fill the window when the lookahead becomes insufficient.
867 * Updates strstart and lookahead.
868 *
869 * IN assertion: lookahead < MIN_LOOKAHEAD
870 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
871 * At least one byte has been read, or avail_in == 0; reads are
872 * performed for at least two bytes (required for the zip translate_eol
873 * option -- not supported here).
874 */
875static void fill_window(
876 deflate_state *s
877)
878{
879 register unsigned n, m;
880 register Pos *p;
881 unsigned more; /* Amount of free space at the end of the window. */
882 uInt wsize = s->w_size;
883
884 do {
885 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
886
887 /* Deal with !@#$% 64K limit: */
888 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
889 more = wsize;
890
891 } else if (more == (unsigned)(-1)) {
892 /* Very unlikely, but possible on 16 bit machine if strstart == 0
893 * and lookahead == 1 (input done one byte at time)
894 */
895 more--;
896
897 /* If the window is almost full and there is insufficient lookahead,
898 * move the upper half to the lower one to make room in the upper half.
899 */
900 } else if (s->strstart >= wsize+MAX_DIST(s)) {
901
902 memcpy((char *)s->window, (char *)s->window+wsize,
903 (unsigned)wsize);
904 s->match_start -= wsize;
905 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
906 s->block_start -= (long) wsize;
907
908 /* Slide the hash table (could be avoided with 32 bit values
909 at the expense of memory usage). We slide even when level == 0
910 to keep the hash table consistent if we switch back to level > 0
911 later. (Using level 0 permanently is not an optimal usage of
912 zlib, so we don't care about this pathological case.)
913 */
914 n = s->hash_size;
915 p = &s->head[n];
916 do {
917 m = *--p;
918 *p = (Pos)(m >= wsize ? m-wsize : NIL);
919 } while (--n);
920
921 n = wsize;
922 p = &s->prev[n];
923 do {
924 m = *--p;
925 *p = (Pos)(m >= wsize ? m-wsize : NIL);
926 /* If n is not on any hash chain, prev[n] is garbage but
927 * its value will never be used.
928 */
929 } while (--n);
930 more += wsize;
931 }
932 if (s->strm->avail_in == 0) return;
933
934 /* If there was no sliding:
935 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
936 * more == window_size - lookahead - strstart
937 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
938 * => more >= window_size - 2*WSIZE + 2
939 * In the BIG_MEM or MMAP case (not yet supported),
940 * window_size == input_size + MIN_LOOKAHEAD &&
941 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
942 * Otherwise, window_size == 2*WSIZE so more >= 2.
943 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
944 */
945 Assert(more >= 2, "more < 2");
946
947 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
948 s->lookahead += n;
949
950 /* Initialize the hash value now that we have some input: */
951 if (s->lookahead >= MIN_MATCH) {
952 s->ins_h = s->window[s->strstart];
953 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
954#if MIN_MATCH != 3
955 Call UPDATE_HASH() MIN_MATCH-3 more times
956#endif
957 }
958 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
959 * but this is not important since only literal bytes will be emitted.
960 */
961
962 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
963}
964
965/* ===========================================================================
966 * Flush the current block, with given end-of-file flag.
967 * IN assertion: strstart is set to the end of the current match.
968 */
969#define FLUSH_BLOCK_ONLY(s, eof) { \
970 zlib_tr_flush_block(s, (s->block_start >= 0L ? \
971 (char *)&s->window[(unsigned)s->block_start] : \
972 NULL), \
973 (ulg)((long)s->strstart - s->block_start), \
974 (eof)); \
975 s->block_start = s->strstart; \
976 flush_pending(s->strm); \
977 Tracev((stderr,"[FLUSH]")); \
978}
979
980/* Same but force premature exit if necessary. */
981#define FLUSH_BLOCK(s, eof) { \
982 FLUSH_BLOCK_ONLY(s, eof); \
983 if (s->strm->avail_out == 0) return (eof) ? finish_started : need_more; \
984}
985
986/* ===========================================================================
987 * Copy without compression as much as possible from the input stream, return
988 * the current block state.
989 * This function does not insert new strings in the dictionary since
990 * uncompressible data is probably not useful. This function is used
991 * only for the level=0 compression option.
992 * NOTE: this function should be optimized to avoid extra copying from
993 * window to pending_buf.
994 */
995static block_state deflate_stored(
996 deflate_state *s,
997 int flush
998)
999{
1000 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited
1001 * to pending_buf_size, and each stored block has a 5 byte header:
1002 */
1003 ulg max_block_size = 0xffff;
1004 ulg max_start;
1005
1006 if (max_block_size > s->pending_buf_size - 5) {
1007 max_block_size = s->pending_buf_size - 5;
1008 }
1009
1010 /* Copy as much as possible from input to output: */
1011 for (;;) {
1012 /* Fill the window as much as possible: */
1013 if (s->lookahead <= 1) {
1014
1015 Assert(s->strstart < s->w_size+MAX_DIST(s) ||
1016 s->block_start >= (long)s->w_size, "slide too late");
1017
1018 fill_window(s);
1019 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
1020
1021 if (s->lookahead == 0) break; /* flush the current block */
1022 }
1023 Assert(s->block_start >= 0L, "block gone");
1024
1025 s->strstart += s->lookahead;
1026 s->lookahead = 0;
1027
1028 /* Emit a stored block if pending_buf will be full: */
1029 max_start = s->block_start + max_block_size;
1030 if (s->strstart == 0 || (ulg)s->strstart >= max_start) {
1031 /* strstart == 0 is possible when wraparound on 16-bit machine */
1032 s->lookahead = (uInt)(s->strstart - max_start);
1033 s->strstart = (uInt)max_start;
1034 FLUSH_BLOCK(s, 0);
1035 }
1036 /* Flush if we may have to slide, otherwise block_start may become
1037 * negative and the data will be gone:
1038 */
1039 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
1040 FLUSH_BLOCK(s, 0);
1041 }
1042 }
1043 FLUSH_BLOCK(s, flush == Z_FINISH);
1044 return flush == Z_FINISH ? finish_done : block_done;
1045}
1046
1047/* ===========================================================================
1048 * Compress as much as possible from the input stream, return the current
1049 * block state.
1050 * This function does not perform lazy evaluation of matches and inserts
1051 * new strings in the dictionary only for unmatched strings or for short
1052 * matches. It is used only for the fast compression options.
1053 */
1054static block_state deflate_fast(
1055 deflate_state *s,
1056 int flush
1057)
1058{
1059 IPos hash_head = NIL; /* head of the hash chain */
1060 int bflush; /* set if current block must be flushed */
1061
1062 for (;;) {
1063 /* Make sure that we always have enough lookahead, except
1064 * at the end of the input file. We need MAX_MATCH bytes
1065 * for the next match, plus MIN_MATCH bytes to insert the
1066 * string following the next match.
1067 */
1068 if (s->lookahead < MIN_LOOKAHEAD) {
1069 fill_window(s);
1070 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1071 return need_more;
1072 }
1073 if (s->lookahead == 0) break; /* flush the current block */
1074 }
1075
1076 /* Insert the string window[strstart .. strstart+2] in the
1077 * dictionary, and set hash_head to the head of the hash chain:
1078 */
1079 if (s->lookahead >= MIN_MATCH) {
1080 INSERT_STRING(s, s->strstart, hash_head);
1081 }
1082
1083 /* Find the longest match, discarding those <= prev_length.
1084 * At this point we have always match_length < MIN_MATCH
1085 */
1086 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1087 /* To simplify the code, we prevent matches with the string
1088 * of window index 0 (in particular we have to avoid a match
1089 * of the string with itself at the start of the input file).
1090 */
1091 if (s->strategy != Z_HUFFMAN_ONLY) {
1092 s->match_length = longest_match (s, hash_head);
1093 }
1094 /* longest_match() sets match_start */
1095 }
1096 if (s->match_length >= MIN_MATCH) {
1097 check_match(s, s->strstart, s->match_start, s->match_length);
1098
1099 bflush = zlib_tr_tally(s, s->strstart - s->match_start,
1100 s->match_length - MIN_MATCH);
1101
1102 s->lookahead -= s->match_length;
1103
1104 /* Insert new strings in the hash table only if the match length
1105 * is not too large. This saves time but degrades compression.
1106 */
1107 if (s->match_length <= s->max_insert_length &&
1108 s->lookahead >= MIN_MATCH) {
1109 s->match_length--; /* string at strstart already in hash table */
1110 do {
1111 s->strstart++;
1112 INSERT_STRING(s, s->strstart, hash_head);
1113 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1114 * always MIN_MATCH bytes ahead.
1115 */
1116 } while (--s->match_length != 0);
1117 s->strstart++;
1118 } else {
1119 s->strstart += s->match_length;
1120 s->match_length = 0;
1121 s->ins_h = s->window[s->strstart];
1122 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1123#if MIN_MATCH != 3
1124 Call UPDATE_HASH() MIN_MATCH-3 more times
1125#endif
1126 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1127 * matter since it will be recomputed at next deflate call.
1128 */
1129 }
1130 } else {
1131 /* No match, output a literal byte */
1132 Tracevv((stderr,"%c", s->window[s->strstart]));
1133 bflush = zlib_tr_tally (s, 0, s->window[s->strstart]);
1134 s->lookahead--;
1135 s->strstart++;
1136 }
1137 if (bflush) FLUSH_BLOCK(s, 0);
1138 }
1139 FLUSH_BLOCK(s, flush == Z_FINISH);
1140 return flush == Z_FINISH ? finish_done : block_done;
1141}
1142
1143/* ===========================================================================
1144 * Same as above, but achieves better compression. We use a lazy
1145 * evaluation for matches: a match is finally adopted only if there is
1146 * no better match at the next window position.
1147 */
1148static block_state deflate_slow(
1149 deflate_state *s,
1150 int flush
1151)
1152{
1153 IPos hash_head = NIL; /* head of hash chain */
1154 int bflush; /* set if current block must be flushed */
1155
1156 /* Process the input block. */
1157 for (;;) {
1158 /* Make sure that we always have enough lookahead, except
1159 * at the end of the input file. We need MAX_MATCH bytes
1160 * for the next match, plus MIN_MATCH bytes to insert the
1161 * string following the next match.
1162 */
1163 if (s->lookahead < MIN_LOOKAHEAD) {
1164 fill_window(s);
1165 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1166 return need_more;
1167 }
1168 if (s->lookahead == 0) break; /* flush the current block */
1169 }
1170
1171 /* Insert the string window[strstart .. strstart+2] in the
1172 * dictionary, and set hash_head to the head of the hash chain:
1173 */
1174 if (s->lookahead >= MIN_MATCH) {
1175 INSERT_STRING(s, s->strstart, hash_head);
1176 }
1177
1178 /* Find the longest match, discarding those <= prev_length.
1179 */
1180 s->prev_length = s->match_length, s->prev_match = s->match_start;
1181 s->match_length = MIN_MATCH-1;
1182
1183 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1184 s->strstart - hash_head <= MAX_DIST(s)) {
1185 /* To simplify the code, we prevent matches with the string
1186 * of window index 0 (in particular we have to avoid a match
1187 * of the string with itself at the start of the input file).
1188 */
1189 if (s->strategy != Z_HUFFMAN_ONLY) {
1190 s->match_length = longest_match (s, hash_head);
1191 }
1192 /* longest_match() sets match_start */
1193
1194 if (s->match_length <= 5 && (s->strategy == Z_FILTERED ||
1195 (s->match_length == MIN_MATCH &&
1196 s->strstart - s->match_start > TOO_FAR))) {
1197
1198 /* If prev_match is also MIN_MATCH, match_start is garbage
1199 * but we will ignore the current match anyway.
1200 */
1201 s->match_length = MIN_MATCH-1;
1202 }
1203 }
1204 /* If there was a match at the previous step and the current
1205 * match is not better, output the previous match:
1206 */
1207 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1208 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1209 /* Do not insert strings in hash table beyond this. */
1210
1211 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1212
1213 bflush = zlib_tr_tally(s, s->strstart -1 - s->prev_match,
1214 s->prev_length - MIN_MATCH);
1215
1216 /* Insert in hash table all strings up to the end of the match.
1217 * strstart-1 and strstart are already inserted. If there is not
1218 * enough lookahead, the last two strings are not inserted in
1219 * the hash table.
1220 */
1221 s->lookahead -= s->prev_length-1;
1222 s->prev_length -= 2;
1223 do {
1224 if (++s->strstart <= max_insert) {
1225 INSERT_STRING(s, s->strstart, hash_head);
1226 }
1227 } while (--s->prev_length != 0);
1228 s->match_available = 0;
1229 s->match_length = MIN_MATCH-1;
1230 s->strstart++;
1231
1232 if (bflush) FLUSH_BLOCK(s, 0);
1233
1234 } else if (s->match_available) {
1235 /* If there was no match at the previous position, output a
1236 * single literal. If there was a match but the current match
1237 * is longer, truncate the previous match to a single literal.
1238 */
1239 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1240 if (zlib_tr_tally (s, 0, s->window[s->strstart-1])) {
1241 FLUSH_BLOCK_ONLY(s, 0);
1242 }
1243 s->strstart++;
1244 s->lookahead--;
1245 if (s->strm->avail_out == 0) return need_more;
1246 } else {
1247 /* There is no previous match to compare with, wait for
1248 * the next step to decide.
1249 */
1250 s->match_available = 1;
1251 s->strstart++;
1252 s->lookahead--;
1253 }
1254 }
1255 Assert (flush != Z_NO_FLUSH, "no flush?");
1256 if (s->match_available) {
1257 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1258 zlib_tr_tally (s, 0, s->window[s->strstart-1]);
1259 s->match_available = 0;
1260 }
1261 FLUSH_BLOCK(s, flush == Z_FINISH);
1262 return flush == Z_FINISH ? finish_done : block_done;
1263}
1264
1265int zlib_deflate_workspacesize(void)
1266{
1267 return sizeof(deflate_workspace);
1268}