ASoC: Mark WM5100 register map cache only when going into BIAS_OFF
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / lib / zlib_deflate / deflate.c
blobd63381e8e3331064463283420d872842e27c8a9a
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
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).
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.
32 * ACKNOWLEDGEMENTS
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.
38 * REFERENCES
40 * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification".
41 * Available in ftp://ds.internic.net/rfc/rfc1951.txt
43 * A description of the Rabin and Karp algorithm is given in the book
44 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
46 * Fiala,E.R., and Greene,D.H.
47 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
51 #include <linux/module.h>
52 #include <linux/zutil.h>
53 #include "defutil.h"
56 /* ===========================================================================
57 * Function prototypes.
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
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.
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.
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.
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 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.
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 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).
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.
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
176 deflate_state *s;
177 int noheader = 0;
178 deflate_workspace *mem;
179 char *next;
181 ush *overlay;
182 /* We overlay pending_buf and d_buf+l_buf. This works since the average
183 * output size for (length,distance) codes is <= 24 bits.
186 if (strm == NULL) return Z_STREAM_ERROR;
188 strm->msg = NULL;
190 if (level == Z_DEFAULT_COMPRESSION) level = 6;
192 mem = (deflate_workspace *) strm->workspace;
194 if (windowBits < 0) { /* undocumented feature: suppress zlib header */
195 noheader = 1;
196 windowBits = -windowBits;
198 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
199 windowBits < 9 || windowBits > 15 || level < 0 || level > 9 ||
200 strategy < 0 || strategy > Z_HUFFMAN_ONLY) {
201 return Z_STREAM_ERROR;
205 * Direct the workspace's pointers to the chunks that were allocated
206 * along with the deflate_workspace struct.
208 next = (char *) mem;
209 next += sizeof(*mem);
210 mem->window_memory = (Byte *) next;
211 next += zlib_deflate_window_memsize(windowBits);
212 mem->prev_memory = (Pos *) next;
213 next += zlib_deflate_prev_memsize(windowBits);
214 mem->head_memory = (Pos *) next;
215 next += zlib_deflate_head_memsize(memLevel);
216 mem->overlay_memory = next;
218 s = (deflate_state *) &(mem->deflate_memory);
219 strm->state = (struct internal_state *)s;
220 s->strm = strm;
222 s->noheader = noheader;
223 s->w_bits = windowBits;
224 s->w_size = 1 << s->w_bits;
225 s->w_mask = s->w_size - 1;
227 s->hash_bits = memLevel + 7;
228 s->hash_size = 1 << s->hash_bits;
229 s->hash_mask = s->hash_size - 1;
230 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
232 s->window = (Byte *) mem->window_memory;
233 s->prev = (Pos *) mem->prev_memory;
234 s->head = (Pos *) mem->head_memory;
236 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
238 overlay = (ush *) mem->overlay_memory;
239 s->pending_buf = (uch *) overlay;
240 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
242 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
243 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
245 s->level = level;
246 s->strategy = strategy;
247 s->method = (Byte)method;
249 return zlib_deflateReset(strm);
252 /* ========================================================================= */
253 #if 0
254 int zlib_deflateSetDictionary(
255 z_streamp strm,
256 const Byte *dictionary,
257 uInt dictLength
260 deflate_state *s;
261 uInt length = dictLength;
262 uInt n;
263 IPos hash_head = 0;
265 if (strm == NULL || strm->state == NULL || dictionary == NULL)
266 return Z_STREAM_ERROR;
268 s = (deflate_state *) strm->state;
269 if (s->status != INIT_STATE) return Z_STREAM_ERROR;
271 strm->adler = zlib_adler32(strm->adler, dictionary, dictLength);
273 if (length < MIN_MATCH) return Z_OK;
274 if (length > MAX_DIST(s)) {
275 length = MAX_DIST(s);
276 #ifndef USE_DICT_HEAD
277 dictionary += dictLength - length; /* use the tail of the dictionary */
278 #endif
280 memcpy((char *)s->window, dictionary, length);
281 s->strstart = length;
282 s->block_start = (long)length;
284 /* Insert all strings in the hash table (except for the last two bytes).
285 * s->lookahead stays null, so s->ins_h will be recomputed at the next
286 * call of fill_window.
288 s->ins_h = s->window[0];
289 UPDATE_HASH(s, s->ins_h, s->window[1]);
290 for (n = 0; n <= length - MIN_MATCH; n++) {
291 INSERT_STRING(s, n, hash_head);
293 if (hash_head) hash_head = 0; /* to make compiler happy */
294 return Z_OK;
296 #endif /* 0 */
298 /* ========================================================================= */
299 int zlib_deflateReset(
300 z_streamp strm
303 deflate_state *s;
305 if (strm == NULL || strm->state == NULL)
306 return Z_STREAM_ERROR;
308 strm->total_in = strm->total_out = 0;
309 strm->msg = NULL;
310 strm->data_type = Z_UNKNOWN;
312 s = (deflate_state *)strm->state;
313 s->pending = 0;
314 s->pending_out = s->pending_buf;
316 if (s->noheader < 0) {
317 s->noheader = 0; /* was set to -1 by deflate(..., Z_FINISH); */
319 s->status = s->noheader ? BUSY_STATE : INIT_STATE;
320 strm->adler = 1;
321 s->last_flush = Z_NO_FLUSH;
323 zlib_tr_init(s);
324 lm_init(s);
326 return Z_OK;
329 /* ========================================================================= */
330 #if 0
331 int zlib_deflateParams(
332 z_streamp strm,
333 int level,
334 int strategy
337 deflate_state *s;
338 compress_func func;
339 int err = Z_OK;
341 if (strm == NULL || strm->state == NULL) return Z_STREAM_ERROR;
342 s = (deflate_state *) strm->state;
344 if (level == Z_DEFAULT_COMPRESSION) {
345 level = 6;
347 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_HUFFMAN_ONLY) {
348 return Z_STREAM_ERROR;
350 func = configuration_table[s->level].func;
352 if (func != configuration_table[level].func && strm->total_in != 0) {
353 /* Flush the last buffer: */
354 err = zlib_deflate(strm, Z_PARTIAL_FLUSH);
356 if (s->level != level) {
357 s->level = level;
358 s->max_lazy_match = configuration_table[level].max_lazy;
359 s->good_match = configuration_table[level].good_length;
360 s->nice_match = configuration_table[level].nice_length;
361 s->max_chain_length = configuration_table[level].max_chain;
363 s->strategy = strategy;
364 return err;
366 #endif /* 0 */
368 /* =========================================================================
369 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
370 * IN assertion: the stream state is correct and there is enough room in
371 * pending_buf.
373 static void putShortMSB(
374 deflate_state *s,
375 uInt b
378 put_byte(s, (Byte)(b >> 8));
379 put_byte(s, (Byte)(b & 0xff));
382 /* =========================================================================
383 * Flush as much pending output as possible. All deflate() output goes
384 * through this function so some applications may wish to modify it
385 * to avoid allocating a large strm->next_out buffer and copying into it.
386 * (See also read_buf()).
388 static void flush_pending(
389 z_streamp strm
392 deflate_state *s = (deflate_state *) strm->state;
393 unsigned len = s->pending;
395 if (len > strm->avail_out) len = strm->avail_out;
396 if (len == 0) return;
398 if (strm->next_out != NULL) {
399 memcpy(strm->next_out, s->pending_out, len);
400 strm->next_out += len;
402 s->pending_out += len;
403 strm->total_out += len;
404 strm->avail_out -= len;
405 s->pending -= len;
406 if (s->pending == 0) {
407 s->pending_out = s->pending_buf;
411 /* ========================================================================= */
412 int zlib_deflate(
413 z_streamp strm,
414 int flush
417 int old_flush; /* value of flush param for previous deflate call */
418 deflate_state *s;
420 if (strm == NULL || strm->state == NULL ||
421 flush > Z_FINISH || flush < 0) {
422 return Z_STREAM_ERROR;
424 s = (deflate_state *) strm->state;
426 if ((strm->next_in == NULL && strm->avail_in != 0) ||
427 (s->status == FINISH_STATE && flush != Z_FINISH)) {
428 return Z_STREAM_ERROR;
430 if (strm->avail_out == 0) return Z_BUF_ERROR;
432 s->strm = strm; /* just in case */
433 old_flush = s->last_flush;
434 s->last_flush = flush;
436 /* Write the zlib header */
437 if (s->status == INIT_STATE) {
439 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
440 uInt level_flags = (s->level-1) >> 1;
442 if (level_flags > 3) level_flags = 3;
443 header |= (level_flags << 6);
444 if (s->strstart != 0) header |= PRESET_DICT;
445 header += 31 - (header % 31);
447 s->status = BUSY_STATE;
448 putShortMSB(s, header);
450 /* Save the adler32 of the preset dictionary: */
451 if (s->strstart != 0) {
452 putShortMSB(s, (uInt)(strm->adler >> 16));
453 putShortMSB(s, (uInt)(strm->adler & 0xffff));
455 strm->adler = 1L;
458 /* Flush as much pending output as possible */
459 if (s->pending != 0) {
460 flush_pending(strm);
461 if (strm->avail_out == 0) {
462 /* Since avail_out is 0, deflate will be called again with
463 * more output space, but possibly with both pending and
464 * avail_in equal to zero. There won't be anything to do,
465 * but this is not an error situation so make sure we
466 * return OK instead of BUF_ERROR at next call of deflate:
468 s->last_flush = -1;
469 return Z_OK;
472 /* Make sure there is something to do and avoid duplicate consecutive
473 * flushes. For repeated and useless calls with Z_FINISH, we keep
474 * returning Z_STREAM_END instead of Z_BUFF_ERROR.
476 } else if (strm->avail_in == 0 && flush <= old_flush &&
477 flush != Z_FINISH) {
478 return Z_BUF_ERROR;
481 /* User must not provide more input after the first FINISH: */
482 if (s->status == FINISH_STATE && strm->avail_in != 0) {
483 return Z_BUF_ERROR;
486 /* Start a new block or continue the current one.
488 if (strm->avail_in != 0 || s->lookahead != 0 ||
489 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
490 block_state bstate;
492 bstate = (*(configuration_table[s->level].func))(s, flush);
494 if (bstate == finish_started || bstate == finish_done) {
495 s->status = FINISH_STATE;
497 if (bstate == need_more || bstate == finish_started) {
498 if (strm->avail_out == 0) {
499 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
501 return Z_OK;
502 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
503 * of deflate should use the same flush parameter to make sure
504 * that the flush is complete. So we don't have to output an
505 * empty block here, this will be done at next call. This also
506 * ensures that for a very small output buffer, we emit at most
507 * one empty block.
510 if (bstate == block_done) {
511 if (flush == Z_PARTIAL_FLUSH) {
512 zlib_tr_align(s);
513 } else if (flush == Z_PACKET_FLUSH) {
514 /* Output just the 3-bit `stored' block type value,
515 but not a zero length. */
516 zlib_tr_stored_type_only(s);
517 } else { /* FULL_FLUSH or SYNC_FLUSH */
518 zlib_tr_stored_block(s, (char*)0, 0L, 0);
519 /* For a full flush, this empty block will be recognized
520 * as a special marker by inflate_sync().
522 if (flush == Z_FULL_FLUSH) {
523 CLEAR_HASH(s); /* forget history */
526 flush_pending(strm);
527 if (strm->avail_out == 0) {
528 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
529 return Z_OK;
533 Assert(strm->avail_out > 0, "bug2");
535 if (flush != Z_FINISH) return Z_OK;
536 if (s->noheader) return Z_STREAM_END;
538 /* Write the zlib trailer (adler32) */
539 putShortMSB(s, (uInt)(strm->adler >> 16));
540 putShortMSB(s, (uInt)(strm->adler & 0xffff));
541 flush_pending(strm);
542 /* If avail_out is zero, the application will call deflate again
543 * to flush the rest.
545 s->noheader = -1; /* write the trailer only once! */
546 return s->pending != 0 ? Z_OK : Z_STREAM_END;
549 /* ========================================================================= */
550 int zlib_deflateEnd(
551 z_streamp strm
554 int status;
555 deflate_state *s;
557 if (strm == NULL || strm->state == NULL) return Z_STREAM_ERROR;
558 s = (deflate_state *) strm->state;
560 status = s->status;
561 if (status != INIT_STATE && status != BUSY_STATE &&
562 status != FINISH_STATE) {
563 return Z_STREAM_ERROR;
566 strm->state = NULL;
568 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
571 /* =========================================================================
572 * Copy the source state to the destination state.
574 #if 0
575 int zlib_deflateCopy (
576 z_streamp dest,
577 z_streamp source
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;
589 if (source == NULL || dest == NULL || source->state == NULL) {
590 return Z_STREAM_ERROR;
593 ss = (deflate_state *) source->state;
595 *dest = *source;
597 mem = (deflate_workspace *) dest->workspace;
599 ds = &(mem->deflate_memory);
601 dest->state = (struct internal_state *) ds;
602 *ds = *ss;
603 ds->strm = dest;
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;
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);
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;
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;
624 return Z_OK;
625 #endif
627 #endif /* 0 */
629 /* ===========================================================================
630 * Read a new buffer from the current input stream, update the adler32
631 * and total number of bytes read. All deflate() input goes through
632 * this function so some applications may wish to modify it to avoid
633 * allocating a large strm->next_in buffer and copying from it.
634 * (See also flush_pending()).
636 static int read_buf(
637 z_streamp strm,
638 Byte *buf,
639 unsigned size
642 unsigned len = strm->avail_in;
644 if (len > size) len = size;
645 if (len == 0) return 0;
647 strm->avail_in -= len;
649 if (!((deflate_state *)(strm->state))->noheader) {
650 strm->adler = zlib_adler32(strm->adler, strm->next_in, len);
652 memcpy(buf, strm->next_in, len);
653 strm->next_in += len;
654 strm->total_in += len;
656 return (int)len;
659 /* ===========================================================================
660 * Initialize the "longest match" routines for a new zlib stream
662 static void lm_init(
663 deflate_state *s
666 s->window_size = (ulg)2L*s->w_size;
668 CLEAR_HASH(s);
670 /* Set the default configuration parameters:
672 s->max_lazy_match = configuration_table[s->level].max_lazy;
673 s->good_match = configuration_table[s->level].good_length;
674 s->nice_match = configuration_table[s->level].nice_length;
675 s->max_chain_length = configuration_table[s->level].max_chain;
677 s->strstart = 0;
678 s->block_start = 0L;
679 s->lookahead = 0;
680 s->match_length = s->prev_length = MIN_MATCH-1;
681 s->match_available = 0;
682 s->ins_h = 0;
685 /* ===========================================================================
686 * Set match_start to the longest match starting at the given string and
687 * return its length. Matches shorter or equal to prev_length are discarded,
688 * in which case the result is equal to prev_length and match_start is
689 * garbage.
690 * IN assertions: cur_match is the head of the hash chain for the current
691 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
692 * OUT assertion: the match length is not greater than s->lookahead.
694 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
695 * match.S. The code will be functionally equivalent.
697 static uInt longest_match(
698 deflate_state *s,
699 IPos cur_match /* current match */
702 unsigned chain_length = s->max_chain_length;/* max hash chain length */
703 register Byte *scan = s->window + s->strstart; /* current string */
704 register Byte *match; /* matched string */
705 register int len; /* length of current match */
706 int best_len = s->prev_length; /* best match length so far */
707 int nice_match = s->nice_match; /* stop if match long enough */
708 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
709 s->strstart - (IPos)MAX_DIST(s) : NIL;
710 /* Stop when cur_match becomes <= limit. To simplify the code,
711 * we prevent matches with the string of window index 0.
713 Pos *prev = s->prev;
714 uInt wmask = s->w_mask;
716 #ifdef UNALIGNED_OK
717 /* Compare two bytes at a time. Note: this is not always beneficial.
718 * Try with and without -DUNALIGNED_OK to check.
720 register Byte *strend = s->window + s->strstart + MAX_MATCH - 1;
721 register ush scan_start = *(ush*)scan;
722 register ush scan_end = *(ush*)(scan+best_len-1);
723 #else
724 register Byte *strend = s->window + s->strstart + MAX_MATCH;
725 register Byte scan_end1 = scan[best_len-1];
726 register Byte scan_end = scan[best_len];
727 #endif
729 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
730 * It is easy to get rid of this optimization if necessary.
732 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
734 /* Do not waste too much time if we already have a good match: */
735 if (s->prev_length >= s->good_match) {
736 chain_length >>= 2;
738 /* Do not look for matches beyond the end of the input. This is necessary
739 * to make deflate deterministic.
741 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
743 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
745 do {
746 Assert(cur_match < s->strstart, "no future");
747 match = s->window + cur_match;
749 /* Skip to next match if the match length cannot increase
750 * or if the match length is less than 2:
752 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
753 /* This code assumes sizeof(unsigned short) == 2. Do not use
754 * UNALIGNED_OK if your compiler uses a different size.
756 if (*(ush*)(match+best_len-1) != scan_end ||
757 *(ush*)match != scan_start) continue;
759 /* It is not necessary to compare scan[2] and match[2] since they are
760 * always equal when the other bytes match, given that the hash keys
761 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
762 * strstart+3, +5, ... up to strstart+257. We check for insufficient
763 * lookahead only every 4th comparison; the 128th check will be made
764 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
765 * necessary to put more guard bytes at the end of the window, or
766 * to check more often for insufficient lookahead.
768 Assert(scan[2] == match[2], "scan[2]?");
769 scan++, match++;
770 do {
771 } while (*(ush*)(scan+=2) == *(ush*)(match+=2) &&
772 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
773 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
774 *(ush*)(scan+=2) == *(ush*)(match+=2) &&
775 scan < strend);
776 /* The funny "do {}" generates better code on most compilers */
778 /* Here, scan <= window+strstart+257 */
779 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
780 if (*scan == *match) scan++;
782 len = (MAX_MATCH - 1) - (int)(strend-scan);
783 scan = strend - (MAX_MATCH-1);
785 #else /* UNALIGNED_OK */
787 if (match[best_len] != scan_end ||
788 match[best_len-1] != scan_end1 ||
789 *match != *scan ||
790 *++match != scan[1]) continue;
792 /* The check at best_len-1 can be removed because it will be made
793 * again later. (This heuristic is not always a win.)
794 * It is not necessary to compare scan[2] and match[2] since they
795 * are always equal when the other bytes match, given that
796 * the hash keys are equal and that HASH_BITS >= 8.
798 scan += 2, match++;
799 Assert(*scan == *match, "match[2]?");
801 /* We check for insufficient lookahead only every 8th comparison;
802 * the 256th check will be made at strstart+258.
804 do {
805 } while (*++scan == *++match && *++scan == *++match &&
806 *++scan == *++match && *++scan == *++match &&
807 *++scan == *++match && *++scan == *++match &&
808 *++scan == *++match && *++scan == *++match &&
809 scan < strend);
811 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
813 len = MAX_MATCH - (int)(strend - scan);
814 scan = strend - MAX_MATCH;
816 #endif /* UNALIGNED_OK */
818 if (len > best_len) {
819 s->match_start = cur_match;
820 best_len = len;
821 if (len >= nice_match) break;
822 #ifdef UNALIGNED_OK
823 scan_end = *(ush*)(scan+best_len-1);
824 #else
825 scan_end1 = scan[best_len-1];
826 scan_end = scan[best_len];
827 #endif
829 } while ((cur_match = prev[cur_match & wmask]) > limit
830 && --chain_length != 0);
832 if ((uInt)best_len <= s->lookahead) return best_len;
833 return s->lookahead;
836 #ifdef DEBUG_ZLIB
837 /* ===========================================================================
838 * Check that the match at match_start is indeed a match.
840 static void check_match(
841 deflate_state *s,
842 IPos start,
843 IPos match,
844 int length
847 /* check that the match is indeed a match */
848 if (memcmp((char *)s->window + match,
849 (char *)s->window + start, length) != EQUAL) {
850 fprintf(stderr, " start %u, match %u, length %d\n",
851 start, match, length);
852 do {
853 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
854 } while (--length != 0);
855 z_error("invalid match");
857 if (z_verbose > 1) {
858 fprintf(stderr,"\\[%d,%d]", start-match, length);
859 do { putc(s->window[start++], stderr); } while (--length != 0);
862 #else
863 # define check_match(s, start, match, length)
864 #endif
866 /* ===========================================================================
867 * Fill the window when the lookahead becomes insufficient.
868 * Updates strstart and lookahead.
870 * IN assertion: lookahead < MIN_LOOKAHEAD
871 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
872 * At least one byte has been read, or avail_in == 0; reads are
873 * performed for at least two bytes (required for the zip translate_eol
874 * option -- not supported here).
876 static void fill_window(
877 deflate_state *s
880 register unsigned n, m;
881 register Pos *p;
882 unsigned more; /* Amount of free space at the end of the window. */
883 uInt wsize = s->w_size;
885 do {
886 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
888 /* Deal with !@#$% 64K limit: */
889 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
890 more = wsize;
892 } else if (more == (unsigned)(-1)) {
893 /* Very unlikely, but possible on 16 bit machine if strstart == 0
894 * and lookahead == 1 (input done one byte at time)
896 more--;
898 /* If the window is almost full and there is insufficient lookahead,
899 * move the upper half to the lower one to make room in the upper half.
901 } else if (s->strstart >= wsize+MAX_DIST(s)) {
903 memcpy((char *)s->window, (char *)s->window+wsize,
904 (unsigned)wsize);
905 s->match_start -= wsize;
906 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
907 s->block_start -= (long) wsize;
909 /* Slide the hash table (could be avoided with 32 bit values
910 at the expense of memory usage). We slide even when level == 0
911 to keep the hash table consistent if we switch back to level > 0
912 later. (Using level 0 permanently is not an optimal usage of
913 zlib, so we don't care about this pathological case.)
915 n = s->hash_size;
916 p = &s->head[n];
917 do {
918 m = *--p;
919 *p = (Pos)(m >= wsize ? m-wsize : NIL);
920 } while (--n);
922 n = wsize;
923 p = &s->prev[n];
924 do {
925 m = *--p;
926 *p = (Pos)(m >= wsize ? m-wsize : NIL);
927 /* If n is not on any hash chain, prev[n] is garbage but
928 * its value will never be used.
930 } while (--n);
931 more += wsize;
933 if (s->strm->avail_in == 0) return;
935 /* If there was no sliding:
936 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
937 * more == window_size - lookahead - strstart
938 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
939 * => more >= window_size - 2*WSIZE + 2
940 * In the BIG_MEM or MMAP case (not yet supported),
941 * window_size == input_size + MIN_LOOKAHEAD &&
942 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
943 * Otherwise, window_size == 2*WSIZE so more >= 2.
944 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
946 Assert(more >= 2, "more < 2");
948 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
949 s->lookahead += n;
951 /* Initialize the hash value now that we have some input: */
952 if (s->lookahead >= MIN_MATCH) {
953 s->ins_h = s->window[s->strstart];
954 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
955 #if MIN_MATCH != 3
956 Call UPDATE_HASH() MIN_MATCH-3 more times
957 #endif
959 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
960 * but this is not important since only literal bytes will be emitted.
963 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
966 /* ===========================================================================
967 * Flush the current block, with given end-of-file flag.
968 * IN assertion: strstart is set to the end of the current match.
970 #define FLUSH_BLOCK_ONLY(s, eof) { \
971 zlib_tr_flush_block(s, (s->block_start >= 0L ? \
972 (char *)&s->window[(unsigned)s->block_start] : \
973 NULL), \
974 (ulg)((long)s->strstart - s->block_start), \
975 (eof)); \
976 s->block_start = s->strstart; \
977 flush_pending(s->strm); \
978 Tracev((stderr,"[FLUSH]")); \
981 /* Same but force premature exit if necessary. */
982 #define FLUSH_BLOCK(s, eof) { \
983 FLUSH_BLOCK_ONLY(s, eof); \
984 if (s->strm->avail_out == 0) return (eof) ? finish_started : need_more; \
987 /* ===========================================================================
988 * Copy without compression as much as possible from the input stream, return
989 * the current block state.
990 * This function does not insert new strings in the dictionary since
991 * uncompressible data is probably not useful. This function is used
992 * only for the level=0 compression option.
993 * NOTE: this function should be optimized to avoid extra copying from
994 * window to pending_buf.
996 static block_state deflate_stored(
997 deflate_state *s,
998 int flush
1001 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited
1002 * to pending_buf_size, and each stored block has a 5 byte header:
1004 ulg max_block_size = 0xffff;
1005 ulg max_start;
1007 if (max_block_size > s->pending_buf_size - 5) {
1008 max_block_size = s->pending_buf_size - 5;
1011 /* Copy as much as possible from input to output: */
1012 for (;;) {
1013 /* Fill the window as much as possible: */
1014 if (s->lookahead <= 1) {
1016 Assert(s->strstart < s->w_size+MAX_DIST(s) ||
1017 s->block_start >= (long)s->w_size, "slide too late");
1019 fill_window(s);
1020 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
1022 if (s->lookahead == 0) break; /* flush the current block */
1024 Assert(s->block_start >= 0L, "block gone");
1026 s->strstart += s->lookahead;
1027 s->lookahead = 0;
1029 /* Emit a stored block if pending_buf will be full: */
1030 max_start = s->block_start + max_block_size;
1031 if (s->strstart == 0 || (ulg)s->strstart >= max_start) {
1032 /* strstart == 0 is possible when wraparound on 16-bit machine */
1033 s->lookahead = (uInt)(s->strstart - max_start);
1034 s->strstart = (uInt)max_start;
1035 FLUSH_BLOCK(s, 0);
1037 /* Flush if we may have to slide, otherwise block_start may become
1038 * negative and the data will be gone:
1040 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
1041 FLUSH_BLOCK(s, 0);
1044 FLUSH_BLOCK(s, flush == Z_FINISH);
1045 return flush == Z_FINISH ? finish_done : block_done;
1048 /* ===========================================================================
1049 * Compress as much as possible from the input stream, return the current
1050 * block state.
1051 * This function does not perform lazy evaluation of matches and inserts
1052 * new strings in the dictionary only for unmatched strings or for short
1053 * matches. It is used only for the fast compression options.
1055 static block_state deflate_fast(
1056 deflate_state *s,
1057 int flush
1060 IPos hash_head = NIL; /* head of the hash chain */
1061 int bflush; /* set if current block must be flushed */
1063 for (;;) {
1064 /* Make sure that we always have enough lookahead, except
1065 * at the end of the input file. We need MAX_MATCH bytes
1066 * for the next match, plus MIN_MATCH bytes to insert the
1067 * string following the next match.
1069 if (s->lookahead < MIN_LOOKAHEAD) {
1070 fill_window(s);
1071 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1072 return need_more;
1074 if (s->lookahead == 0) break; /* flush the current block */
1077 /* Insert the string window[strstart .. strstart+2] in the
1078 * dictionary, and set hash_head to the head of the hash chain:
1080 if (s->lookahead >= MIN_MATCH) {
1081 INSERT_STRING(s, s->strstart, hash_head);
1084 /* Find the longest match, discarding those <= prev_length.
1085 * At this point we have always match_length < MIN_MATCH
1087 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1088 /* To simplify the code, we prevent matches with the string
1089 * of window index 0 (in particular we have to avoid a match
1090 * of the string with itself at the start of the input file).
1092 if (s->strategy != Z_HUFFMAN_ONLY) {
1093 s->match_length = longest_match (s, hash_head);
1095 /* longest_match() sets match_start */
1097 if (s->match_length >= MIN_MATCH) {
1098 check_match(s, s->strstart, s->match_start, s->match_length);
1100 bflush = zlib_tr_tally(s, s->strstart - s->match_start,
1101 s->match_length - MIN_MATCH);
1103 s->lookahead -= s->match_length;
1105 /* Insert new strings in the hash table only if the match length
1106 * is not too large. This saves time but degrades compression.
1108 if (s->match_length <= s->max_insert_length &&
1109 s->lookahead >= MIN_MATCH) {
1110 s->match_length--; /* string at strstart already in hash table */
1111 do {
1112 s->strstart++;
1113 INSERT_STRING(s, s->strstart, hash_head);
1114 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1115 * always MIN_MATCH bytes ahead.
1117 } while (--s->match_length != 0);
1118 s->strstart++;
1119 } else {
1120 s->strstart += s->match_length;
1121 s->match_length = 0;
1122 s->ins_h = s->window[s->strstart];
1123 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1124 #if MIN_MATCH != 3
1125 Call UPDATE_HASH() MIN_MATCH-3 more times
1126 #endif
1127 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1128 * matter since it will be recomputed at next deflate call.
1131 } else {
1132 /* No match, output a literal byte */
1133 Tracevv((stderr,"%c", s->window[s->strstart]));
1134 bflush = zlib_tr_tally (s, 0, s->window[s->strstart]);
1135 s->lookahead--;
1136 s->strstart++;
1138 if (bflush) FLUSH_BLOCK(s, 0);
1140 FLUSH_BLOCK(s, flush == Z_FINISH);
1141 return flush == Z_FINISH ? finish_done : block_done;
1144 /* ===========================================================================
1145 * Same as above, but achieves better compression. We use a lazy
1146 * evaluation for matches: a match is finally adopted only if there is
1147 * no better match at the next window position.
1149 static block_state deflate_slow(
1150 deflate_state *s,
1151 int flush
1154 IPos hash_head = NIL; /* head of hash chain */
1155 int bflush; /* set if current block must be flushed */
1157 /* Process the input block. */
1158 for (;;) {
1159 /* Make sure that we always have enough lookahead, except
1160 * at the end of the input file. We need MAX_MATCH bytes
1161 * for the next match, plus MIN_MATCH bytes to insert the
1162 * string following the next match.
1164 if (s->lookahead < MIN_LOOKAHEAD) {
1165 fill_window(s);
1166 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1167 return need_more;
1169 if (s->lookahead == 0) break; /* flush the current block */
1172 /* Insert the string window[strstart .. strstart+2] in the
1173 * dictionary, and set hash_head to the head of the hash chain:
1175 if (s->lookahead >= MIN_MATCH) {
1176 INSERT_STRING(s, s->strstart, hash_head);
1179 /* Find the longest match, discarding those <= prev_length.
1181 s->prev_length = s->match_length, s->prev_match = s->match_start;
1182 s->match_length = MIN_MATCH-1;
1184 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1185 s->strstart - hash_head <= MAX_DIST(s)) {
1186 /* To simplify the code, we prevent matches with the string
1187 * of window index 0 (in particular we have to avoid a match
1188 * of the string with itself at the start of the input file).
1190 if (s->strategy != Z_HUFFMAN_ONLY) {
1191 s->match_length = longest_match (s, hash_head);
1193 /* longest_match() sets match_start */
1195 if (s->match_length <= 5 && (s->strategy == Z_FILTERED ||
1196 (s->match_length == MIN_MATCH &&
1197 s->strstart - s->match_start > TOO_FAR))) {
1199 /* If prev_match is also MIN_MATCH, match_start is garbage
1200 * but we will ignore the current match anyway.
1202 s->match_length = MIN_MATCH-1;
1205 /* If there was a match at the previous step and the current
1206 * match is not better, output the previous match:
1208 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1209 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1210 /* Do not insert strings in hash table beyond this. */
1212 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1214 bflush = zlib_tr_tally(s, s->strstart -1 - s->prev_match,
1215 s->prev_length - MIN_MATCH);
1217 /* Insert in hash table all strings up to the end of the match.
1218 * strstart-1 and strstart are already inserted. If there is not
1219 * enough lookahead, the last two strings are not inserted in
1220 * the hash table.
1222 s->lookahead -= s->prev_length-1;
1223 s->prev_length -= 2;
1224 do {
1225 if (++s->strstart <= max_insert) {
1226 INSERT_STRING(s, s->strstart, hash_head);
1228 } while (--s->prev_length != 0);
1229 s->match_available = 0;
1230 s->match_length = MIN_MATCH-1;
1231 s->strstart++;
1233 if (bflush) FLUSH_BLOCK(s, 0);
1235 } else if (s->match_available) {
1236 /* If there was no match at the previous position, output a
1237 * single literal. If there was a match but the current match
1238 * is longer, truncate the previous match to a single literal.
1240 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1241 if (zlib_tr_tally (s, 0, s->window[s->strstart-1])) {
1242 FLUSH_BLOCK_ONLY(s, 0);
1244 s->strstart++;
1245 s->lookahead--;
1246 if (s->strm->avail_out == 0) return need_more;
1247 } else {
1248 /* There is no previous match to compare with, wait for
1249 * the next step to decide.
1251 s->match_available = 1;
1252 s->strstart++;
1253 s->lookahead--;
1256 Assert (flush != Z_NO_FLUSH, "no flush?");
1257 if (s->match_available) {
1258 Tracevv((stderr,"%c", s->window[s->strstart-1]));
1259 zlib_tr_tally (s, 0, s->window[s->strstart-1]);
1260 s->match_available = 0;
1262 FLUSH_BLOCK(s, flush == Z_FINISH);
1263 return flush == Z_FINISH ? finish_done : block_done;
1266 int zlib_deflate_workspacesize(int windowBits, int memLevel)
1268 if (windowBits < 0) /* undocumented feature: suppress zlib header */
1269 windowBits = -windowBits;
1271 /* Since the return value is typically passed to vmalloc() unchecked... */
1272 BUG_ON(memLevel < 1 || memLevel > MAX_MEM_LEVEL || windowBits < 9 ||
1273 windowBits > 15);
1275 return sizeof(deflate_workspace)
1276 + zlib_deflate_window_memsize(windowBits)
1277 + zlib_deflate_prev_memsize(windowBits)
1278 + zlib_deflate_head_memsize(memLevel)
1279 + zlib_deflate_overlay_memsize(memLevel);