1/* deflate.c -- compress data using the deflation algorithm
2 * Copyright (C) 1995-2017 Jean-loup Gailly and Mark Adler
3 * For conditions of distribution and use, see copyright notice in zlib.h
4 */
5
6/*
7 * ALGORITHM
8 *
9 * The "deflation" process depends on being able to identify portions
10 * of the input text which are identical to earlier input (within a
11 * sliding window trailing behind the input currently being processed).
12 *
13 * The most straightforward technique turns out to be the fastest for
14 * most input files: try all possible matches and select the longest.
15 * The key feature of this algorithm is that insertions into the string
16 * dictionary are very simple and thus fast, and deletions are avoided
17 * completely. Insertions are performed at each input character, whereas
18 * string matches are performed only when the previous match ends. So it
19 * is preferable to spend more time in matches to allow very fast string
20 * insertions and avoid deletions. The matching algorithm for small
21 * strings is inspired from that of Rabin & Karp. A brute force approach
22 * is used to find longer strings when a small match has been found.
23 * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze
24 * (by Leonid Broukhis).
25 * A previous version of this file used a more sophisticated algorithm
26 * (by Fiala and Greene) which is guaranteed to run in linear amortized
27 * time, but has a larger average cost, uses more memory and is patented.
28 * However the F&G algorithm may be faster for some highly redundant
29 * files if the parameter max_chain_length (described below) is too large.
30 *
31 * ACKNOWLEDGEMENTS
32 *
33 * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and
34 * I found it in 'freeze' written by Leonid Broukhis.
35 * Thanks to many people for bug reports and testing.
36 *
37 * REFERENCES
38 *
39 * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification".
40 * Available in http://tools.ietf.org/html/rfc1951
41 *
42 * A description of the Rabin and Karp algorithm is given in the book
43 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
44 *
45 * Fiala,E.R., and Greene,D.H.
46 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
47 *
48 */
49
50/* @(#) $Id$ */
51#include <assert.h>
52#include "deflate.h"
53#include "x86.h"
54
55#if (defined(__ARM_NEON__) || defined(__ARM_NEON))
56#include "contrib/optimizations/slide_hash_neon.h"
57#endif
58/* We need crypto extension crc32 to implement optimized hash in
59 * insert_string.
60 */
61#if defined(CRC32_ARMV8_CRC32)
62#include "arm_features.h"
63#include "crc32_simd.h"
64#endif
65
66const char deflate_copyright[] =
67 " deflate 1.2.11 Copyright 1995-2017 Jean-loup Gailly and Mark Adler ";
68/*
69 If you use the zlib library in a product, an acknowledgment is welcome
70 in the documentation of your product. If for some reason you cannot
71 include such an acknowledgment, I would appreciate that you keep this
72 copyright string in the executable of your product.
73 */
74
75/* ===========================================================================
76 * Function prototypes.
77 */
78typedef enum {
79 need_more, /* block not completed, need more input or more output */
80 block_done, /* block flush performed */
81 finish_started, /* finish started, need only more output at next deflate */
82 finish_done /* finish done, accept no more input or output */
83} block_state;
84
85typedef block_state (*compress_func) OF((deflate_state *s, int flush));
86/* Compression function. Returns the block state after the call. */
87
88local int deflateStateCheck OF((z_streamp strm));
89local void slide_hash OF((deflate_state *s));
90local void fill_window OF((deflate_state *s));
91local block_state deflate_stored OF((deflate_state *s, int flush));
92local block_state deflate_fast OF((deflate_state *s, int flush));
93#ifndef FASTEST
94local block_state deflate_slow OF((deflate_state *s, int flush));
95#endif
96local block_state deflate_rle OF((deflate_state *s, int flush));
97local block_state deflate_huff OF((deflate_state *s, int flush));
98local void lm_init OF((deflate_state *s));
99local void putShortMSB OF((deflate_state *s, uInt b));
100local void flush_pending OF((z_streamp strm));
101unsigned ZLIB_INTERNAL deflate_read_buf OF((z_streamp strm, Bytef *buf, unsigned size));
102#ifdef ASMV
103# pragma message("Assembler code may have bugs -- use at your own risk")
104 void match_init OF((void)); /* asm code initialization */
105 uInt longest_match OF((deflate_state *s, IPos cur_match));
106#else
107local uInt longest_match OF((deflate_state *s, IPos cur_match));
108#endif
109
110#ifdef ZLIB_DEBUG
111local void check_match OF((deflate_state *s, IPos start, IPos match,
112 int length));
113#endif
114
115/* From crc32.c */
116extern void ZLIB_INTERNAL crc_reset(deflate_state *const s);
117extern void ZLIB_INTERNAL crc_finalize(deflate_state *const s);
118extern void ZLIB_INTERNAL copy_with_crc(z_streamp strm, Bytef *dst, long size);
119
120#ifdef _MSC_VER
121#define INLINE __inline
122#else
123#define INLINE inline
124#endif
125
126/* Inline optimisation */
127local INLINE Pos insert_string_sse(deflate_state *const s, const Pos str);
128
129/* ===========================================================================
130 * Local data
131 */
132
133#define NIL 0
134/* Tail of hash chains */
135
136#ifndef TOO_FAR
137# define TOO_FAR 4096
138#endif
139/* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
140
141/* Values for max_lazy_match, good_match and max_chain_length, depending on
142 * the desired pack level (0..9). The values given below have been tuned to
143 * exclude worst case performance for pathological files. Better values may be
144 * found for specific files.
145 */
146typedef struct config_s {
147 ush good_length; /* reduce lazy search above this match length */
148 ush max_lazy; /* do not perform lazy search above this match length */
149 ush nice_length; /* quit search above this match length */
150 ush max_chain;
151 compress_func func;
152} config;
153
154#ifdef FASTEST
155local const config configuration_table[2] = {
156/* good lazy nice chain */
157/* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
158/* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */
159#else
160local const config configuration_table[10] = {
161/* good lazy nice chain */
162/* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
163/* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */
164/* 2 */ {4, 5, 16, 8, deflate_fast},
165/* 3 */ {4, 6, 32, 32, deflate_fast},
166
167/* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
168/* 5 */ {8, 16, 32, 32, deflate_slow},
169/* 6 */ {8, 16, 128, 128, deflate_slow},
170/* 7 */ {8, 32, 128, 256, deflate_slow},
171/* 8 */ {32, 128, 258, 1024, deflate_slow},
172/* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */
173#endif
174
175/* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
176 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
177 * meaning.
178 */
179
180/* rank Z_BLOCK between Z_NO_FLUSH and Z_PARTIAL_FLUSH */
181#define RANK(f) (((f) * 2) - ((f) > 4 ? 9 : 0))
182
183/* ===========================================================================
184 * Update a hash value with the given input byte
185 * IN assertion: all calls to UPDATE_HASH are made with consecutive input
186 * characters, so that a running hash key can be computed from the previous
187 * key instead of complete recalculation each time.
188 */
189#define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
190
191/* ===========================================================================
192 * Insert string str in the dictionary and set match_head to the previous head
193 * of the hash chain (the most recent string with same hash key). Return
194 * the previous length of the hash chain.
195 * If this file is compiled with -DFASTEST, the compression level is forced
196 * to 1, and no hash chains are maintained.
197 * IN assertion: all calls to INSERT_STRING are made with consecutive input
198 * characters and the first MIN_MATCH bytes of str are valid (except for
199 * the last MIN_MATCH-1 bytes of the input file).
200 */
201local INLINE Pos insert_string_c(deflate_state *const s, const Pos str)
202{
203 Pos ret;
204
205 UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]);
206#ifdef FASTEST
207 ret = s->head[s->ins_h];
208#else
209 ret = s->prev[str & s->w_mask] = s->head[s->ins_h];
210#endif
211 s->head[s->ins_h] = str;
212
213 return ret;
214}
215
216local INLINE Pos insert_string(deflate_state *const s, const Pos str)
217{
218#if defined(CRC32_ARMV8_CRC32)
219 if (arm_cpu_enable_crc32)
220 return insert_string_arm(s, str);
221#endif
222 if (x86_cpu_enable_simd)
223 return insert_string_sse(s, str);
224
225 return insert_string_c(s, str);
226}
227
228/* ===========================================================================
229 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
230 * prev[] will be initialized on the fly.
231 */
232#define CLEAR_HASH(s) \
233 s->head[s->hash_size-1] = NIL; \
234 zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
235
236/* ===========================================================================
237 * Slide the hash table when sliding the window down (could be avoided with 32
238 * bit values at the expense of memory usage). We slide even when level == 0 to
239 * keep the hash table consistent if we switch back to level > 0 later.
240 */
241local void slide_hash(s)
242 deflate_state *s;
243{
244#if (defined(__ARM_NEON__) || defined(__ARM_NEON))
245 /* NEON based hash table rebase. */
246 return neon_slide_hash(s->head, s->prev, s->w_size, s->hash_size);
247#endif
248 unsigned n, m;
249 Posf *p;
250 uInt wsize = s->w_size;
251
252 n = s->hash_size;
253 p = &s->head[n];
254 do {
255 m = *--p;
256 *p = (Pos)(m >= wsize ? m - wsize : NIL);
257 } while (--n);
258 n = wsize;
259#ifndef FASTEST
260 p = &s->prev[n];
261 do {
262 m = *--p;
263 *p = (Pos)(m >= wsize ? m - wsize : NIL);
264 /* If n is not on any hash chain, prev[n] is garbage but
265 * its value will never be used.
266 */
267 } while (--n);
268#endif
269}
270
271/* ========================================================================= */
272int ZEXPORT deflateInit_(strm, level, version, stream_size)
273 z_streamp strm;
274 int level;
275 const char *version;
276 int stream_size;
277{
278 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
279 Z_DEFAULT_STRATEGY, version, stream_size);
280 /* To do: ignore strm->next_in if we use it as window */
281}
282
283/* ========================================================================= */
284int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
285 version, stream_size)
286 z_streamp strm;
287 int level;
288 int method;
289 int windowBits;
290 int memLevel;
291 int strategy;
292 const char *version;
293 int stream_size;
294{
295 unsigned window_padding = 8;
296 deflate_state *s;
297 int wrap = 1;
298 static const char my_version[] = ZLIB_VERSION;
299
300 x86_check_features();
301
302 if (version == Z_NULL || version[0] != my_version[0] ||
303 stream_size != sizeof(z_stream)) {
304 return Z_VERSION_ERROR;
305 }
306 if (strm == Z_NULL) return Z_STREAM_ERROR;
307
308 strm->msg = Z_NULL;
309 if (strm->zalloc == (alloc_func)0) {
310#ifdef Z_SOLO
311 return Z_STREAM_ERROR;
312#else
313 strm->zalloc = zcalloc;
314 strm->opaque = (voidpf)0;
315#endif
316 }
317 if (strm->zfree == (free_func)0)
318#ifdef Z_SOLO
319 return Z_STREAM_ERROR;
320#else
321 strm->zfree = zcfree;
322#endif
323
324#ifdef FASTEST
325 if (level != 0) level = 1;
326#else
327 if (level == Z_DEFAULT_COMPRESSION) level = 6;
328#endif
329
330 if (windowBits < 0) { /* suppress zlib wrapper */
331 wrap = 0;
332 windowBits = -windowBits;
333 }
334#ifdef GZIP
335 else if (windowBits > 15) {
336 wrap = 2; /* write gzip wrapper instead */
337 windowBits -= 16;
338 }
339#endif
340 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
341 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
342 strategy < 0 || strategy > Z_FIXED || (windowBits == 8 && wrap != 1)) {
343 return Z_STREAM_ERROR;
344 }
345 if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */
346 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
347 if (s == Z_NULL) return Z_MEM_ERROR;
348 strm->state = (struct internal_state FAR *)s;
349 s->strm = strm;
350 s->status = INIT_STATE; /* to pass state test in deflateReset() */
351
352 s->wrap = wrap;
353 s->gzhead = Z_NULL;
354 s->w_bits = (uInt)windowBits;
355 s->w_size = 1 << s->w_bits;
356 s->w_mask = s->w_size - 1;
357
358 if (x86_cpu_enable_simd) {
359 s->hash_bits = 15;
360 } else {
361 s->hash_bits = memLevel + 7;
362 }
363
364 s->hash_size = 1 << s->hash_bits;
365 s->hash_mask = s->hash_size - 1;
366 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
367
368 s->window = (Bytef *) ZALLOC(strm,
369 s->w_size + window_padding,
370 2*sizeof(Byte));
371 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));
372 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
373
374 s->high_water = 0; /* nothing written to s->window yet */
375
376 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
377
378 /* We overlay pending_buf and sym_buf. This works since the average size
379 * for length/distance pairs over any compressed block is assured to be 31
380 * bits or less.
381 *
382 * Analysis: The longest fixed codes are a length code of 8 bits plus 5
383 * extra bits, for lengths 131 to 257. The longest fixed distance codes are
384 * 5 bits plus 13 extra bits, for distances 16385 to 32768. The longest
385 * possible fixed-codes length/distance pair is then 31 bits total.
386 *
387 * sym_buf starts one-fourth of the way into pending_buf. So there are
388 * three bytes in sym_buf for every four bytes in pending_buf. Each symbol
389 * in sym_buf is three bytes -- two for the distance and one for the
390 * literal/length. As each symbol is consumed, the pointer to the next
391 * sym_buf value to read moves forward three bytes. From that symbol, up to
392 * 31 bits are written to pending_buf. The closest the written pending_buf
393 * bits gets to the next sym_buf symbol to read is just before the last
394 * code is written. At that time, 31*(n-2) bits have been written, just
395 * after 24*(n-2) bits have been consumed from sym_buf. sym_buf starts at
396 * 8*n bits into pending_buf. (Note that the symbol buffer fills when n-1
397 * symbols are written.) The closest the writing gets to what is unread is
398 * then n+14 bits. Here n is lit_bufsize, which is 16384 by default, and
399 * can range from 128 to 32768.
400 *
401 * Therefore, at a minimum, there are 142 bits of space between what is
402 * written and what is read in the overlain buffers, so the symbols cannot
403 * be overwritten by the compressed data. That space is actually 139 bits,
404 * due to the three-bit fixed-code block header.
405 *
406 * That covers the case where either Z_FIXED is specified, forcing fixed
407 * codes, or when the use of fixed codes is chosen, because that choice
408 * results in a smaller compressed block than dynamic codes. That latter
409 * condition then assures that the above analysis also covers all dynamic
410 * blocks. A dynamic-code block will only be chosen to be emitted if it has
411 * fewer bits than a fixed-code block would for the same set of symbols.
412 * Therefore its average symbol length is assured to be less than 31. So
413 * the compressed data for a dynamic block also cannot overwrite the
414 * symbols from which it is being constructed.
415 */
416 s->pending_buf = (uchf *) ZALLOC(strm, s->lit_bufsize, 4);
417 s->pending_buf_size = (ulg)s->lit_bufsize * 4;
418
419 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
420 s->pending_buf == Z_NULL) {
421 s->status = FINISH_STATE;
422 strm->msg = ERR_MSG(Z_MEM_ERROR);
423 deflateEnd (strm);
424 return Z_MEM_ERROR;
425 }
426 s->sym_buf = s->pending_buf + s->lit_bufsize;
427 s->sym_end = (s->lit_bufsize - 1) * 3;
428 /* We avoid equality with lit_bufsize*3 because of wraparound at 64K
429 * on 16 bit machines and because stored blocks are restricted to
430 * 64K-1 bytes.
431 */
432
433 s->level = level;
434 s->strategy = strategy;
435 s->method = (Byte)method;
436
437 return deflateReset(strm);
438}
439
440/* =========================================================================
441 * Check for a valid deflate stream state. Return 0 if ok, 1 if not.
442 */
443local int deflateStateCheck (strm)
444 z_streamp strm;
445{
446 deflate_state *s;
447 if (strm == Z_NULL ||
448 strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0)
449 return 1;
450 s = strm->state;
451 if (s == Z_NULL || s->strm != strm || (s->status != INIT_STATE &&
452#ifdef GZIP
453 s->status != GZIP_STATE &&
454#endif
455 s->status != EXTRA_STATE &&
456 s->status != NAME_STATE &&
457 s->status != COMMENT_STATE &&
458 s->status != HCRC_STATE &&
459 s->status != BUSY_STATE &&
460 s->status != FINISH_STATE))
461 return 1;
462 return 0;
463}
464
465/* ========================================================================= */
466int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
467 z_streamp strm;
468 const Bytef *dictionary;
469 uInt dictLength;
470{
471 deflate_state *s;
472 uInt str, n;
473 int wrap;
474 unsigned avail;
475 z_const unsigned char *next;
476
477 if (deflateStateCheck(strm) || dictionary == Z_NULL)
478 return Z_STREAM_ERROR;
479 s = strm->state;
480 wrap = s->wrap;
481 if (wrap == 2 || (wrap == 1 && s->status != INIT_STATE) || s->lookahead)
482 return Z_STREAM_ERROR;
483
484 /* when using zlib wrappers, compute Adler-32 for provided dictionary */
485 if (wrap == 1)
486 strm->adler = adler32(strm->adler, dictionary, dictLength);
487 s->wrap = 0; /* avoid computing Adler-32 in deflate_read_buf */
488
489 /* if dictionary would fill window, just replace the history */
490 if (dictLength >= s->w_size) {
491 if (wrap == 0) { /* already empty otherwise */
492 CLEAR_HASH(s);
493 s->strstart = 0;
494 s->block_start = 0L;
495 s->insert = 0;
496 }
497 dictionary += dictLength - s->w_size; /* use the tail */
498 dictLength = s->w_size;
499 }
500
501 /* insert dictionary into window and hash */
502 avail = strm->avail_in;
503 next = strm->next_in;
504 strm->avail_in = dictLength;
505 strm->next_in = (z_const Bytef *)dictionary;
506 fill_window(s);
507 while (s->lookahead >= MIN_MATCH) {
508 str = s->strstart;
509 n = s->lookahead - (MIN_MATCH-1);
510 do {
511 insert_string(s, str);
512 str++;
513 } while (--n);
514 s->strstart = str;
515 s->lookahead = MIN_MATCH-1;
516 fill_window(s);
517 }
518 s->strstart += s->lookahead;
519 s->block_start = (long)s->strstart;
520 s->insert = s->lookahead;
521 s->lookahead = 0;
522 s->match_length = s->prev_length = MIN_MATCH-1;
523 s->match_available = 0;
524 strm->next_in = next;
525 strm->avail_in = avail;
526 s->wrap = wrap;
527 return Z_OK;
528}
529
530/* ========================================================================= */
531int ZEXPORT deflateGetDictionary (strm, dictionary, dictLength)
532 z_streamp strm;
533 Bytef *dictionary;
534 uInt *dictLength;
535{
536 deflate_state *s;
537 uInt len;
538
539 if (deflateStateCheck(strm))
540 return Z_STREAM_ERROR;
541 s = strm->state;
542 len = s->strstart + s->lookahead;
543 if (len > s->w_size)
544 len = s->w_size;
545 if (dictionary != Z_NULL && len)
546 zmemcpy(dictionary, s->window + s->strstart + s->lookahead - len, len);
547 if (dictLength != Z_NULL)
548 *dictLength = len;
549 return Z_OK;
550}
551
552/* ========================================================================= */
553int ZEXPORT deflateResetKeep (strm)
554 z_streamp strm;
555{
556 deflate_state *s;
557
558 if (deflateStateCheck(strm)) {
559 return Z_STREAM_ERROR;
560 }
561
562 strm->total_in = strm->total_out = 0;
563 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
564 strm->data_type = Z_UNKNOWN;
565
566 s = (deflate_state *)strm->state;
567 s->pending = 0;
568 s->pending_out = s->pending_buf;
569
570 if (s->wrap < 0) {
571 s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */
572 }
573 s->status =
574#ifdef GZIP
575 s->wrap == 2 ? GZIP_STATE :
576#endif
577 s->wrap ? INIT_STATE : BUSY_STATE;
578 strm->adler =
579#ifdef GZIP
580 s->wrap == 2 ? crc32(0L, Z_NULL, 0) :
581#endif
582 adler32(0L, Z_NULL, 0);
583 s->last_flush = Z_NO_FLUSH;
584
585 _tr_init(s);
586
587 return Z_OK;
588}
589
590/* ========================================================================= */
591int ZEXPORT deflateReset (strm)
592 z_streamp strm;
593{
594 int ret;
595
596 ret = deflateResetKeep(strm);
597 if (ret == Z_OK)
598 lm_init(strm->state);
599 return ret;
600}
601
602/* ========================================================================= */
603int ZEXPORT deflateSetHeader (strm, head)
604 z_streamp strm;
605 gz_headerp head;
606{
607 if (deflateStateCheck(strm) || strm->state->wrap != 2)
608 return Z_STREAM_ERROR;
609 strm->state->gzhead = head;
610 return Z_OK;
611}
612
613/* ========================================================================= */
614int ZEXPORT deflatePending (strm, pending, bits)
615 unsigned *pending;
616 int *bits;
617 z_streamp strm;
618{
619 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
620 if (pending != Z_NULL)
621 *pending = strm->state->pending;
622 if (bits != Z_NULL)
623 *bits = strm->state->bi_valid;
624 return Z_OK;
625}
626
627/* ========================================================================= */
628int ZEXPORT deflatePrime (strm, bits, value)
629 z_streamp strm;
630 int bits;
631 int value;
632{
633 deflate_state *s;
634 int put;
635
636 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
637 s = strm->state;
638 if (s->sym_buf < s->pending_out + ((Buf_size + 7) >> 3))
639 return Z_BUF_ERROR;
640 do {
641 put = Buf_size - s->bi_valid;
642 if (put > bits)
643 put = bits;
644 s->bi_buf |= (ush)((value & ((1 << put) - 1)) << s->bi_valid);
645 s->bi_valid += put;
646 _tr_flush_bits(s);
647 value >>= put;
648 bits -= put;
649 } while (bits);
650 return Z_OK;
651}
652
653/* ========================================================================= */
654int ZEXPORT deflateParams(strm, level, strategy)
655 z_streamp strm;
656 int level;
657 int strategy;
658{
659 deflate_state *s;
660 compress_func func;
661
662 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
663 s = strm->state;
664
665#ifdef FASTEST
666 if (level != 0) level = 1;
667#else
668 if (level == Z_DEFAULT_COMPRESSION) level = 6;
669#endif
670 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
671 return Z_STREAM_ERROR;
672 }
673 func = configuration_table[s->level].func;
674
675 if ((strategy != s->strategy || func != configuration_table[level].func) &&
676 s->high_water) {
677 /* Flush the last buffer: */
678 int err = deflate(strm, Z_BLOCK);
679 if (err == Z_STREAM_ERROR)
680 return err;
681 if (strm->avail_out == 0)
682 return Z_BUF_ERROR;
683 }
684 if (s->level != level) {
685 if (s->level == 0 && s->matches != 0) {
686 if (s->matches == 1)
687 slide_hash(s);
688 else
689 CLEAR_HASH(s);
690 s->matches = 0;
691 }
692 s->level = level;
693 s->max_lazy_match = configuration_table[level].max_lazy;
694 s->good_match = configuration_table[level].good_length;
695 s->nice_match = configuration_table[level].nice_length;
696 s->max_chain_length = configuration_table[level].max_chain;
697 }
698 s->strategy = strategy;
699 return Z_OK;
700}
701
702/* ========================================================================= */
703int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain)
704 z_streamp strm;
705 int good_length;
706 int max_lazy;
707 int nice_length;
708 int max_chain;
709{
710 deflate_state *s;
711
712 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
713 s = strm->state;
714 s->good_match = (uInt)good_length;
715 s->max_lazy_match = (uInt)max_lazy;
716 s->nice_match = nice_length;
717 s->max_chain_length = (uInt)max_chain;
718 return Z_OK;
719}
720
721/* =========================================================================
722 * For the default windowBits of 15 and memLevel of 8, this function returns
723 * a close to exact, as well as small, upper bound on the compressed size.
724 * They are coded as constants here for a reason--if the #define's are
725 * changed, then this function needs to be changed as well. The return
726 * value for 15 and 8 only works for those exact settings.
727 *
728 * For any setting other than those defaults for windowBits and memLevel,
729 * the value returned is a conservative worst case for the maximum expansion
730 * resulting from using fixed blocks instead of stored blocks, which deflate
731 * can emit on compressed data for some combinations of the parameters.
732 *
733 * This function could be more sophisticated to provide closer upper bounds for
734 * every combination of windowBits and memLevel. But even the conservative
735 * upper bound of about 14% expansion does not seem onerous for output buffer
736 * allocation.
737 */
738uLong ZEXPORT deflateBound(strm, sourceLen)
739 z_streamp strm;
740 uLong sourceLen;
741{
742 deflate_state *s;
743 uLong complen, wraplen;
744
745 /* conservative upper bound for compressed data */
746 complen = sourceLen +
747 ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5;
748
749 /* if can't get parameters, return conservative bound plus zlib wrapper */
750 if (deflateStateCheck(strm))
751 return complen + 6;
752
753 /* compute wrapper length */
754 s = strm->state;
755 switch (s->wrap) {
756 case 0: /* raw deflate */
757 wraplen = 0;
758 break;
759 case 1: /* zlib wrapper */
760 wraplen = 6 + (s->strstart ? 4 : 0);
761 break;
762#ifdef GZIP
763 case 2: /* gzip wrapper */
764 wraplen = 18;
765 if (s->gzhead != Z_NULL) { /* user-supplied gzip header */
766 Bytef *str;
767 if (s->gzhead->extra != Z_NULL)
768 wraplen += 2 + s->gzhead->extra_len;
769 str = s->gzhead->name;
770 if (str != Z_NULL)
771 do {
772 wraplen++;
773 } while (*str++);
774 str = s->gzhead->comment;
775 if (str != Z_NULL)
776 do {
777 wraplen++;
778 } while (*str++);
779 if (s->gzhead->hcrc)
780 wraplen += 2;
781 }
782 break;
783#endif
784 default: /* for compiler happiness */
785 wraplen = 6;
786 }
787
788 /* if not default parameters, return conservative bound */
789 if (s->w_bits != 15 || s->hash_bits != 8 + 7)
790 return complen + wraplen;
791
792 /* default settings: return tight bound for that case */
793 return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) +
794 (sourceLen >> 25) + 13 - 6 + wraplen;
795}
796
797/* =========================================================================
798 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
799 * IN assertion: the stream state is correct and there is enough room in
800 * pending_buf.
801 */
802local void putShortMSB (s, b)
803 deflate_state *s;
804 uInt b;
805{
806 put_byte(s, (Byte)(b >> 8));
807 put_byte(s, (Byte)(b & 0xff));
808}
809
810/* =========================================================================
811 * Flush as much pending output as possible. All deflate() output, except for
812 * some deflate_stored() output, goes through this function so some
813 * applications may wish to modify it to avoid allocating a large
814 * strm->next_out buffer and copying into it. (See also deflate_read_buf()).
815 */
816local void flush_pending(strm)
817 z_streamp strm;
818{
819 unsigned len;
820 deflate_state *s = strm->state;
821
822 _tr_flush_bits(s);
823 len = s->pending;
824 if (len > strm->avail_out) len = strm->avail_out;
825 if (len == 0) return;
826
827 zmemcpy(strm->next_out, s->pending_out, len);
828 strm->next_out += len;
829 s->pending_out += len;
830 strm->total_out += len;
831 strm->avail_out -= len;
832 s->pending -= len;
833 if (s->pending == 0) {
834 s->pending_out = s->pending_buf;
835 }
836}
837
838/* ===========================================================================
839 * Update the header CRC with the bytes s->pending_buf[beg..s->pending - 1].
840 */
841#define HCRC_UPDATE(beg) \
842 do { \
843 if (s->gzhead->hcrc && s->pending > (beg)) \
844 strm->adler = crc32(strm->adler, s->pending_buf + (beg), \
845 s->pending - (beg)); \
846 } while (0)
847
848/* ========================================================================= */
849int ZEXPORT deflate (strm, flush)
850 z_streamp strm;
851 int flush;
852{
853 int old_flush; /* value of flush param for previous deflate call */
854 deflate_state *s;
855
856 if (deflateStateCheck(strm) || flush > Z_BLOCK || flush < 0) {
857 return Z_STREAM_ERROR;
858 }
859 s = strm->state;
860
861 if (strm->next_out == Z_NULL ||
862 (strm->avail_in != 0 && strm->next_in == Z_NULL) ||
863 (s->status == FINISH_STATE && flush != Z_FINISH)) {
864 ERR_RETURN(strm, Z_STREAM_ERROR);
865 }
866 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
867
868 old_flush = s->last_flush;
869 s->last_flush = flush;
870
871 /* Flush as much pending output as possible */
872 if (s->pending != 0) {
873 flush_pending(strm);
874 if (strm->avail_out == 0) {
875 /* Since avail_out is 0, deflate will be called again with
876 * more output space, but possibly with both pending and
877 * avail_in equal to zero. There won't be anything to do,
878 * but this is not an error situation so make sure we
879 * return OK instead of BUF_ERROR at next call of deflate:
880 */
881 s->last_flush = -1;
882 return Z_OK;
883 }
884
885 /* Make sure there is something to do and avoid duplicate consecutive
886 * flushes. For repeated and useless calls with Z_FINISH, we keep
887 * returning Z_STREAM_END instead of Z_BUF_ERROR.
888 */
889 } else if (strm->avail_in == 0 && RANK(flush) <= RANK(old_flush) &&
890 flush != Z_FINISH) {
891 ERR_RETURN(strm, Z_BUF_ERROR);
892 }
893
894 /* User must not provide more input after the first FINISH: */
895 if (s->status == FINISH_STATE && strm->avail_in != 0) {
896 ERR_RETURN(strm, Z_BUF_ERROR);
897 }
898
899 /* Write the header */
900 if (s->status == INIT_STATE) {
901 /* zlib header */
902 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
903 uInt level_flags;
904
905 if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)
906 level_flags = 0;
907 else if (s->level < 6)
908 level_flags = 1;
909 else if (s->level == 6)
910 level_flags = 2;
911 else
912 level_flags = 3;
913 header |= (level_flags << 6);
914 if (s->strstart != 0) header |= PRESET_DICT;
915 header += 31 - (header % 31);
916
917 putShortMSB(s, header);
918
919 /* Save the adler32 of the preset dictionary: */
920 if (s->strstart != 0) {
921 putShortMSB(s, (uInt)(strm->adler >> 16));
922 putShortMSB(s, (uInt)(strm->adler & 0xffff));
923 }
924 strm->adler = adler32(0L, Z_NULL, 0);
925 s->status = BUSY_STATE;
926
927 /* Compression must start with an empty pending buffer */
928 flush_pending(strm);
929 if (s->pending != 0) {
930 s->last_flush = -1;
931 return Z_OK;
932 }
933 }
934#ifdef GZIP
935 if (s->status == GZIP_STATE) {
936 /* gzip header */
937 crc_reset(s);
938 put_byte(s, 31);
939 put_byte(s, 139);
940 put_byte(s, 8);
941 if (s->gzhead == Z_NULL) {
942 put_byte(s, 0);
943 put_byte(s, 0);
944 put_byte(s, 0);
945 put_byte(s, 0);
946 put_byte(s, 0);
947 put_byte(s, s->level == 9 ? 2 :
948 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
949 4 : 0));
950 put_byte(s, OS_CODE);
951 s->status = BUSY_STATE;
952
953 /* Compression must start with an empty pending buffer */
954 flush_pending(strm);
955 if (s->pending != 0) {
956 s->last_flush = -1;
957 return Z_OK;
958 }
959 }
960 else {
961 put_byte(s, (s->gzhead->text ? 1 : 0) +
962 (s->gzhead->hcrc ? 2 : 0) +
963 (s->gzhead->extra == Z_NULL ? 0 : 4) +
964 (s->gzhead->name == Z_NULL ? 0 : 8) +
965 (s->gzhead->comment == Z_NULL ? 0 : 16)
966 );
967 put_byte(s, (Byte)(s->gzhead->time & 0xff));
968 put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));
969 put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));
970 put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));
971 put_byte(s, s->level == 9 ? 2 :
972 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
973 4 : 0));
974 put_byte(s, s->gzhead->os & 0xff);
975 if (s->gzhead->extra != Z_NULL) {
976 put_byte(s, s->gzhead->extra_len & 0xff);
977 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);
978 }
979 if (s->gzhead->hcrc)
980 strm->adler = crc32(strm->adler, s->pending_buf,
981 s->pending);
982 s->gzindex = 0;
983 s->status = EXTRA_STATE;
984 }
985 }
986 if (s->status == EXTRA_STATE) {
987 if (s->gzhead->extra != Z_NULL) {
988 ulg beg = s->pending; /* start of bytes to update crc */
989 uInt left = (s->gzhead->extra_len & 0xffff) - s->gzindex;
990 while (s->pending + left > s->pending_buf_size) {
991 uInt copy = s->pending_buf_size - s->pending;
992 zmemcpy(s->pending_buf + s->pending,
993 s->gzhead->extra + s->gzindex, copy);
994 s->pending = s->pending_buf_size;
995 HCRC_UPDATE(beg);
996 s->gzindex += copy;
997 flush_pending(strm);
998 if (s->pending != 0) {
999 s->last_flush = -1;
1000 return Z_OK;
1001 }
1002 beg = 0;
1003 left -= copy;
1004 }
1005 zmemcpy(s->pending_buf + s->pending,
1006 s->gzhead->extra + s->gzindex, left);
1007 s->pending += left;
1008 HCRC_UPDATE(beg);
1009 s->gzindex = 0;
1010 }
1011 s->status = NAME_STATE;
1012 }
1013 if (s->status == NAME_STATE) {
1014 if (s->gzhead->name != Z_NULL) {
1015 ulg beg = s->pending; /* start of bytes to update crc */
1016 int val;
1017 do {
1018 if (s->pending == s->pending_buf_size) {
1019 HCRC_UPDATE(beg);
1020 flush_pending(strm);
1021 if (s->pending != 0) {
1022 s->last_flush = -1;
1023 return Z_OK;
1024 }
1025 beg = 0;
1026 }
1027 val = s->gzhead->name[s->gzindex++];
1028 put_byte(s, val);
1029 } while (val != 0);
1030 HCRC_UPDATE(beg);
1031 s->gzindex = 0;
1032 }
1033 s->status = COMMENT_STATE;
1034 }
1035 if (s->status == COMMENT_STATE) {
1036 if (s->gzhead->comment != Z_NULL) {
1037 ulg beg = s->pending; /* start of bytes to update crc */
1038 int val;
1039 do {
1040 if (s->pending == s->pending_buf_size) {
1041 HCRC_UPDATE(beg);
1042 flush_pending(strm);
1043 if (s->pending != 0) {
1044 s->last_flush = -1;
1045 return Z_OK;
1046 }
1047 beg = 0;
1048 }
1049 val = s->gzhead->comment[s->gzindex++];
1050 put_byte(s, val);
1051 } while (val != 0);
1052 HCRC_UPDATE(beg);
1053 }
1054 s->status = HCRC_STATE;
1055 }
1056 if (s->status == HCRC_STATE) {
1057 if (s->gzhead->hcrc) {
1058 if (s->pending + 2 > s->pending_buf_size) {
1059 flush_pending(strm);
1060 if (s->pending != 0) {
1061 s->last_flush = -1;
1062 return Z_OK;
1063 }
1064 }
1065 put_byte(s, (Byte)(strm->adler & 0xff));
1066 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
1067 strm->adler = crc32(0L, Z_NULL, 0);
1068 }
1069 s->status = BUSY_STATE;
1070
1071 /* Compression must start with an empty pending buffer */
1072 flush_pending(strm);
1073 if (s->pending != 0) {
1074 s->last_flush = -1;
1075 return Z_OK;
1076 }
1077 }
1078#endif
1079
1080 /* Start a new block or continue the current one.
1081 */
1082 if (strm->avail_in != 0 || s->lookahead != 0 ||
1083 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
1084 block_state bstate;
1085
1086 bstate = s->level == 0 ? deflate_stored(s, flush) :
1087 s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) :
1088 s->strategy == Z_RLE ? deflate_rle(s, flush) :
1089 (*(configuration_table[s->level].func))(s, flush);
1090
1091 if (bstate == finish_started || bstate == finish_done) {
1092 s->status = FINISH_STATE;
1093 }
1094 if (bstate == need_more || bstate == finish_started) {
1095 if (strm->avail_out == 0) {
1096 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
1097 }
1098 return Z_OK;
1099 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
1100 * of deflate should use the same flush parameter to make sure
1101 * that the flush is complete. So we don't have to output an
1102 * empty block here, this will be done at next call. This also
1103 * ensures that for a very small output buffer, we emit at most
1104 * one empty block.
1105 */
1106 }
1107 if (bstate == block_done) {
1108 if (flush == Z_PARTIAL_FLUSH) {
1109 _tr_align(s);
1110 } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */
1111 _tr_stored_block(s, (char*)0, 0L, 0);
1112 /* For a full flush, this empty block will be recognized
1113 * as a special marker by inflate_sync().
1114 */
1115 if (flush == Z_FULL_FLUSH) {
1116 CLEAR_HASH(s); /* forget history */
1117 if (s->lookahead == 0) {
1118 s->strstart = 0;
1119 s->block_start = 0L;
1120 s->insert = 0;
1121 }
1122 }
1123 }
1124 flush_pending(strm);
1125 if (strm->avail_out == 0) {
1126 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
1127 return Z_OK;
1128 }
1129 }
1130 }
1131
1132 if (flush != Z_FINISH) return Z_OK;
1133 if (s->wrap <= 0) return Z_STREAM_END;
1134
1135 /* Write the trailer */
1136#ifdef GZIP
1137 if (s->wrap == 2) {
1138 crc_finalize(s);
1139 put_byte(s, (Byte)(strm->adler & 0xff));
1140 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
1141 put_byte(s, (Byte)((strm->adler >> 16) & 0xff));
1142 put_byte(s, (Byte)((strm->adler >> 24) & 0xff));
1143 put_byte(s, (Byte)(strm->total_in & 0xff));
1144 put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));
1145 put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));
1146 put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));
1147 }
1148 else
1149#endif
1150 {
1151 putShortMSB(s, (uInt)(strm->adler >> 16));
1152 putShortMSB(s, (uInt)(strm->adler & 0xffff));
1153 }
1154 flush_pending(strm);
1155 /* If avail_out is zero, the application will call deflate again
1156 * to flush the rest.
1157 */
1158 if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
1159 return s->pending != 0 ? Z_OK : Z_STREAM_END;
1160}
1161
1162/* ========================================================================= */
1163int ZEXPORT deflateEnd (strm)
1164 z_streamp strm;
1165{
1166 int status;
1167
1168 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
1169
1170 status = strm->state->status;
1171
1172 /* Deallocate in reverse order of allocations: */
1173 TRY_FREE(strm, strm->state->pending_buf);
1174 TRY_FREE(strm, strm->state->head);
1175 TRY_FREE(strm, strm->state->prev);
1176 TRY_FREE(strm, strm->state->window);
1177
1178 ZFREE(strm, strm->state);
1179 strm->state = Z_NULL;
1180
1181 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
1182}
1183
1184/* =========================================================================
1185 * Copy the source state to the destination state.
1186 * To simplify the source, this is not supported for 16-bit MSDOS (which
1187 * doesn't have enough memory anyway to duplicate compression states).
1188 */
1189int ZEXPORT deflateCopy (dest, source)
1190 z_streamp dest;
1191 z_streamp source;
1192{
1193#ifdef MAXSEG_64K
1194 return Z_STREAM_ERROR;
1195#else
1196 deflate_state *ds;
1197 deflate_state *ss;
1198
1199
1200 if (deflateStateCheck(source) || dest == Z_NULL) {
1201 return Z_STREAM_ERROR;
1202 }
1203
1204 ss = source->state;
1205
1206 zmemcpy((voidpf)dest, (voidpf)source, sizeof(z_stream));
1207
1208 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
1209 if (ds == Z_NULL) return Z_MEM_ERROR;
1210 dest->state = (struct internal_state FAR *) ds;
1211 zmemcpy((voidpf)ds, (voidpf)ss, sizeof(deflate_state));
1212 ds->strm = dest;
1213
1214 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
1215 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));
1216 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));
1217 ds->pending_buf = (uchf *) ZALLOC(dest, ds->lit_bufsize, 4);
1218
1219 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
1220 ds->pending_buf == Z_NULL) {
1221 deflateEnd (dest);
1222 return Z_MEM_ERROR;
1223 }
1224 /* following zmemcpy do not work for 16-bit MSDOS */
1225 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
1226 zmemcpy((voidpf)ds->prev, (voidpf)ss->prev, ds->w_size * sizeof(Pos));
1227 zmemcpy((voidpf)ds->head, (voidpf)ss->head, ds->hash_size * sizeof(Pos));
1228 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
1229
1230 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
1231 ds->sym_buf = ds->pending_buf + ds->lit_bufsize;
1232
1233 ds->l_desc.dyn_tree = ds->dyn_ltree;
1234 ds->d_desc.dyn_tree = ds->dyn_dtree;
1235 ds->bl_desc.dyn_tree = ds->bl_tree;
1236
1237 return Z_OK;
1238#endif /* MAXSEG_64K */
1239}
1240
1241/* ===========================================================================
1242 * Read a new buffer from the current input stream, update the adler32
1243 * and total number of bytes read. All deflate() input goes through
1244 * this function so some applications may wish to modify it to avoid
1245 * allocating a large strm->next_in buffer and copying from it.
1246 * (See also flush_pending()).
1247 */
1248ZLIB_INTERNAL unsigned deflate_read_buf(strm, buf, size)
1249 z_streamp strm;
1250 Bytef *buf;
1251 unsigned size;
1252{
1253 unsigned len = strm->avail_in;
1254
1255 if (len > size) len = size;
1256 if (len == 0) return 0;
1257
1258 strm->avail_in -= len;
1259
1260#ifdef GZIP
1261 if (strm->state->wrap == 2)
1262 copy_with_crc(strm, buf, len);
1263 else
1264#endif
1265 {
1266 zmemcpy(buf, strm->next_in, len);
1267 if (strm->state->wrap == 1)
1268 strm->adler = adler32(strm->adler, buf, len);
1269 }
1270 strm->next_in += len;
1271 strm->total_in += len;
1272
1273 return len;
1274}
1275
1276/* ===========================================================================
1277 * Initialize the "longest match" routines for a new zlib stream
1278 */
1279local void lm_init (s)
1280 deflate_state *s;
1281{
1282 s->window_size = (ulg)2L*s->w_size;
1283
1284 CLEAR_HASH(s);
1285
1286 /* Set the default configuration parameters:
1287 */
1288 s->max_lazy_match = configuration_table[s->level].max_lazy;
1289 s->good_match = configuration_table[s->level].good_length;
1290 s->nice_match = configuration_table[s->level].nice_length;
1291 s->max_chain_length = configuration_table[s->level].max_chain;
1292
1293 s->strstart = 0;
1294 s->block_start = 0L;
1295 s->lookahead = 0;
1296 s->insert = 0;
1297 s->match_length = s->prev_length = MIN_MATCH-1;
1298 s->match_available = 0;
1299 s->ins_h = 0;
1300#ifndef FASTEST
1301#ifdef ASMV
1302 match_init(); /* initialize the asm code */
1303#endif
1304#endif
1305}
1306
1307#ifndef FASTEST
1308/* ===========================================================================
1309 * Set match_start to the longest match starting at the given string and
1310 * return its length. Matches shorter or equal to prev_length are discarded,
1311 * in which case the result is equal to prev_length and match_start is
1312 * garbage.
1313 * IN assertions: cur_match is the head of the hash chain for the current
1314 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
1315 * OUT assertion: the match length is not greater than s->lookahead.
1316 */
1317#ifndef ASMV
1318/* For 80x86 and 680x0, an optimized version will be provided in match.asm or
1319 * match.S. The code will be functionally equivalent.
1320 */
1321local uInt longest_match(s, cur_match)
1322 deflate_state *s;
1323 IPos cur_match; /* current match */
1324{
1325 unsigned chain_length = s->max_chain_length;/* max hash chain length */
1326 register Bytef *scan = s->window + s->strstart; /* current string */
1327 register Bytef *match; /* matched string */
1328 register int len; /* length of current match */
1329 int best_len = (int)s->prev_length; /* best match length so far */
1330 int nice_match = s->nice_match; /* stop if match long enough */
1331 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
1332 s->strstart - (IPos)MAX_DIST(s) : NIL;
1333 /* Stop when cur_match becomes <= limit. To simplify the code,
1334 * we prevent matches with the string of window index 0.
1335 */
1336 Posf *prev = s->prev;
1337 uInt wmask = s->w_mask;
1338
1339#ifdef UNALIGNED_OK
1340 /* Compare two bytes at a time. Note: this is not always beneficial.
1341 * Try with and without -DUNALIGNED_OK to check.
1342 */
1343 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
1344 register ush scan_start = *(ushf*)scan;
1345 register ush scan_end = *(ushf*)(scan+best_len-1);
1346#else
1347 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1348 register Byte scan_end1 = scan[best_len-1];
1349 register Byte scan_end = scan[best_len];
1350#endif
1351
1352 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1353 * It is easy to get rid of this optimization if necessary.
1354 */
1355 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1356
1357 /* Do not waste too much time if we already have a good match: */
1358 if (s->prev_length >= s->good_match) {
1359 chain_length >>= 2;
1360 }
1361 /* Do not look for matches beyond the end of the input. This is necessary
1362 * to make deflate deterministic.
1363 */
1364 if ((uInt)nice_match > s->lookahead) nice_match = (int)s->lookahead;
1365
1366 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1367
1368 do {
1369 Assert(cur_match < s->strstart, "no future");
1370 match = s->window + cur_match;
1371
1372 /* Skip to next match if the match length cannot increase
1373 * or if the match length is less than 2. Note that the checks below
1374 * for insufficient lookahead only occur occasionally for performance
1375 * reasons. Therefore uninitialized memory will be accessed, and
1376 * conditional jumps will be made that depend on those values.
1377 * However the length of the match is limited to the lookahead, so
1378 * the output of deflate is not affected by the uninitialized values.
1379 */
1380#if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
1381 /* This code assumes sizeof(unsigned short) == 2. Do not use
1382 * UNALIGNED_OK if your compiler uses a different size.
1383 */
1384 if (*(ushf*)(match+best_len-1) != scan_end ||
1385 *(ushf*)match != scan_start) continue;
1386
1387 /* It is not necessary to compare scan[2] and match[2] since they are
1388 * always equal when the other bytes match, given that the hash keys
1389 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
1390 * strstart+3, +5, ... up to strstart+257. We check for insufficient
1391 * lookahead only every 4th comparison; the 128th check will be made
1392 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
1393 * necessary to put more guard bytes at the end of the window, or
1394 * to check more often for insufficient lookahead.
1395 */
1396 Assert(scan[2] == match[2], "scan[2]?");
1397 scan++, match++;
1398 do {
1399 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1400 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1401 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1402 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1403 scan < strend);
1404 /* The funny "do {}" generates better code on most compilers */
1405
1406 /* Here, scan <= window+strstart+257 */
1407 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1408 if (*scan == *match) scan++;
1409
1410 len = (MAX_MATCH - 1) - (int)(strend-scan);
1411 scan = strend - (MAX_MATCH-1);
1412
1413#else /* UNALIGNED_OK */
1414
1415 if (match[best_len] != scan_end ||
1416 match[best_len-1] != scan_end1 ||
1417 *match != *scan ||
1418 *++match != scan[1]) continue;
1419
1420 /* The check at best_len-1 can be removed because it will be made
1421 * again later. (This heuristic is not always a win.)
1422 * It is not necessary to compare scan[2] and match[2] since they
1423 * are always equal when the other bytes match, given that
1424 * the hash keys are equal and that HASH_BITS >= 8.
1425 */
1426 scan += 2, match++;
1427 Assert(*scan == *match, "match[2]?");
1428
1429 /* We check for insufficient lookahead only every 8th comparison;
1430 * the 256th check will be made at strstart+258.
1431 */
1432 do {
1433 } while (*++scan == *++match && *++scan == *++match &&
1434 *++scan == *++match && *++scan == *++match &&
1435 *++scan == *++match && *++scan == *++match &&
1436 *++scan == *++match && *++scan == *++match &&
1437 scan < strend);
1438
1439 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1440
1441 len = MAX_MATCH - (int)(strend - scan);
1442 scan = strend - MAX_MATCH;
1443
1444#endif /* UNALIGNED_OK */
1445
1446 if (len > best_len) {
1447 s->match_start = cur_match;
1448 best_len = len;
1449 if (len >= nice_match) break;
1450#ifdef UNALIGNED_OK
1451 scan_end = *(ushf*)(scan+best_len-1);
1452#else
1453 scan_end1 = scan[best_len-1];
1454 scan_end = scan[best_len];
1455#endif
1456 }
1457 } while ((cur_match = prev[cur_match & wmask]) > limit
1458 && --chain_length != 0);
1459
1460 if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
1461 return s->lookahead;
1462}
1463#endif /* ASMV */
1464
1465#else /* FASTEST */
1466
1467/* ---------------------------------------------------------------------------
1468 * Optimized version for FASTEST only
1469 */
1470local uInt longest_match(s, cur_match)
1471 deflate_state *s;
1472 IPos cur_match; /* current match */
1473{
1474 register Bytef *scan = s->window + s->strstart; /* current string */
1475 register Bytef *match; /* matched string */
1476 register int len; /* length of current match */
1477 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1478
1479 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1480 * It is easy to get rid of this optimization if necessary.
1481 */
1482 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1483
1484 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1485
1486 Assert(cur_match < s->strstart, "no future");
1487
1488 match = s->window + cur_match;
1489
1490 /* Return failure if the match length is less than 2:
1491 */
1492 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
1493
1494 /* The check at best_len-1 can be removed because it will be made
1495 * again later. (This heuristic is not always a win.)
1496 * It is not necessary to compare scan[2] and match[2] since they
1497 * are always equal when the other bytes match, given that
1498 * the hash keys are equal and that HASH_BITS >= 8.
1499 */
1500 scan += 2, match += 2;
1501 Assert(*scan == *match, "match[2]?");
1502
1503 /* We check for insufficient lookahead only every 8th comparison;
1504 * the 256th check will be made at strstart+258.
1505 */
1506 do {
1507 } while (*++scan == *++match && *++scan == *++match &&
1508 *++scan == *++match && *++scan == *++match &&
1509 *++scan == *++match && *++scan == *++match &&
1510 *++scan == *++match && *++scan == *++match &&
1511 scan < strend);
1512
1513 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1514
1515 len = MAX_MATCH - (int)(strend - scan);
1516
1517 if (len < MIN_MATCH) return MIN_MATCH - 1;
1518
1519 s->match_start = cur_match;
1520 return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;
1521}
1522
1523#endif /* FASTEST */
1524
1525#ifdef ZLIB_DEBUG
1526
1527#define EQUAL 0
1528/* result of memcmp for equal strings */
1529
1530/* ===========================================================================
1531 * Check that the match at match_start is indeed a match.
1532 */
1533local void check_match(s, start, match, length)
1534 deflate_state *s;
1535 IPos start, match;
1536 int length;
1537{
1538 /* check that the match is indeed a match */
1539 if (zmemcmp(s->window + match,
1540 s->window + start, length) != EQUAL) {
1541 fprintf(stderr, " start %u, match %u, length %d\n",
1542 start, match, length);
1543 do {
1544 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
1545 } while (--length != 0);
1546 z_error("invalid match");
1547 }
1548 if (z_verbose > 1) {
1549 fprintf(stderr,"\\[%d,%d]", start-match, length);
1550 do { putc(s->window[start++], stderr); } while (--length != 0);
1551 }
1552}
1553#else
1554# define check_match(s, start, match, length)
1555#endif /* ZLIB_DEBUG */
1556
1557/* ===========================================================================
1558 * Fill the window when the lookahead becomes insufficient.
1559 * Updates strstart and lookahead.
1560 *
1561 * IN assertion: lookahead < MIN_LOOKAHEAD
1562 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
1563 * At least one byte has been read, or avail_in == 0; reads are
1564 * performed for at least two bytes (required for the zip translate_eol
1565 * option -- not supported here).
1566 */
1567local void fill_window_c(deflate_state *s);
1568
1569local void fill_window(deflate_state *s)
1570{
1571 if (x86_cpu_enable_simd) {
1572 fill_window_sse(s);
1573 return;
1574 }
1575
1576 fill_window_c(s);
1577}
1578
1579local void fill_window_c(s)
1580 deflate_state *s;
1581{
1582 unsigned n;
1583 unsigned more; /* Amount of free space at the end of the window. */
1584 uInt wsize = s->w_size;
1585
1586 Assert(s->lookahead < MIN_LOOKAHEAD, "already enough lookahead");
1587
1588 do {
1589 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
1590
1591 /* Deal with !@#$% 64K limit: */
1592 if (sizeof(int) <= 2) {
1593 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
1594 more = wsize;
1595
1596 } else if (more == (unsigned)(-1)) {
1597 /* Very unlikely, but possible on 16 bit machine if
1598 * strstart == 0 && lookahead == 1 (input done a byte at time)
1599 */
1600 more--;
1601 }
1602 }
1603
1604 /* If the window is almost full and there is insufficient lookahead,
1605 * move the upper half to the lower one to make room in the upper half.
1606 */
1607 if (s->strstart >= wsize+MAX_DIST(s)) {
1608
1609 zmemcpy(s->window, s->window+wsize, (unsigned)wsize - more);
1610 s->match_start -= wsize;
1611 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
1612 s->block_start -= (long) wsize;
1613 slide_hash(s);
1614 more += wsize;
1615 }
1616 if (s->strm->avail_in == 0) break;
1617
1618 /* If there was no sliding:
1619 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1620 * more == window_size - lookahead - strstart
1621 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1622 * => more >= window_size - 2*WSIZE + 2
1623 * In the BIG_MEM or MMAP case (not yet supported),
1624 * window_size == input_size + MIN_LOOKAHEAD &&
1625 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1626 * Otherwise, window_size == 2*WSIZE so more >= 2.
1627 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1628 */
1629 Assert(more >= 2, "more < 2");
1630
1631 n = deflate_read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1632 s->lookahead += n;
1633
1634 /* Initialize the hash value now that we have some input: */
1635 if (s->lookahead + s->insert >= MIN_MATCH) {
1636 uInt str = s->strstart - s->insert;
1637 s->ins_h = s->window[str];
1638 UPDATE_HASH(s, s->ins_h, s->window[str + 1]);
1639#if MIN_MATCH != 3
1640 Call UPDATE_HASH() MIN_MATCH-3 more times
1641#endif
1642 while (s->insert) {
1643 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
1644#ifndef FASTEST
1645 s->prev[str & s->w_mask] = s->head[s->ins_h];
1646#endif
1647 s->head[s->ins_h] = (Pos)str;
1648 str++;
1649 s->insert--;
1650 if (s->lookahead + s->insert < MIN_MATCH)
1651 break;
1652 }
1653 }
1654 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1655 * but this is not important since only literal bytes will be emitted.
1656 */
1657
1658 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1659
1660 /* If the WIN_INIT bytes after the end of the current data have never been
1661 * written, then zero those bytes in order to avoid memory check reports of
1662 * the use of uninitialized (or uninitialised as Julian writes) bytes by
1663 * the longest match routines. Update the high water mark for the next
1664 * time through here. WIN_INIT is set to MAX_MATCH since the longest match
1665 * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead.
1666 */
1667 if (s->high_water < s->window_size) {
1668 ulg curr = s->strstart + (ulg)(s->lookahead);
1669 ulg init;
1670
1671 if (s->high_water < curr) {
1672 /* Previous high water mark below current data -- zero WIN_INIT
1673 * bytes or up to end of window, whichever is less.
1674 */
1675 init = s->window_size - curr;
1676 if (init > WIN_INIT)
1677 init = WIN_INIT;
1678 zmemzero(s->window + curr, (unsigned)init);
1679 s->high_water = curr + init;
1680 }
1681 else if (s->high_water < (ulg)curr + WIN_INIT) {
1682 /* High water mark at or above current data, but below current data
1683 * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up
1684 * to end of window, whichever is less.
1685 */
1686 init = (ulg)curr + WIN_INIT - s->high_water;
1687 if (init > s->window_size - s->high_water)
1688 init = s->window_size - s->high_water;
1689 zmemzero(s->window + s->high_water, (unsigned)init);
1690 s->high_water += init;
1691 }
1692 }
1693
1694 Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD,
1695 "not enough room for search");
1696}
1697
1698/* ===========================================================================
1699 * Flush the current block, with given end-of-file flag.
1700 * IN assertion: strstart is set to the end of the current match.
1701 */
1702#define FLUSH_BLOCK_ONLY(s, last) { \
1703 _tr_flush_block(s, (s->block_start >= 0L ? \
1704 (charf *)&s->window[(unsigned)s->block_start] : \
1705 (charf *)Z_NULL), \
1706 (ulg)((long)s->strstart - s->block_start), \
1707 (last)); \
1708 s->block_start = s->strstart; \
1709 flush_pending(s->strm); \
1710 Tracev((stderr,"[FLUSH]")); \
1711}
1712
1713/* Same but force premature exit if necessary. */
1714#define FLUSH_BLOCK(s, last) { \
1715 FLUSH_BLOCK_ONLY(s, last); \
1716 if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \
1717}
1718
1719/* Maximum stored block length in deflate format (not including header). */
1720#define MAX_STORED 65535
1721
1722/* Minimum of a and b. */
1723#define MIN(a, b) ((a) > (b) ? (b) : (a))
1724
1725/* ===========================================================================
1726 * Copy without compression as much as possible from the input stream, return
1727 * the current block state.
1728 *
1729 * In case deflateParams() is used to later switch to a non-zero compression
1730 * level, s->matches (otherwise unused when storing) keeps track of the number
1731 * of hash table slides to perform. If s->matches is 1, then one hash table
1732 * slide will be done when switching. If s->matches is 2, the maximum value
1733 * allowed here, then the hash table will be cleared, since two or more slides
1734 * is the same as a clear.
1735 *
1736 * deflate_stored() is written to minimize the number of times an input byte is
1737 * copied. It is most efficient with large input and output buffers, which
1738 * maximizes the opportunites to have a single copy from next_in to next_out.
1739 */
1740local block_state deflate_stored(s, flush)
1741 deflate_state *s;
1742 int flush;
1743{
1744 /* Smallest worthy block size when not flushing or finishing. By default
1745 * this is 32K. This can be as small as 507 bytes for memLevel == 1. For
1746 * large input and output buffers, the stored block size will be larger.
1747 */
1748 unsigned min_block = MIN(s->pending_buf_size - 5, s->w_size);
1749
1750 /* Copy as many min_block or larger stored blocks directly to next_out as
1751 * possible. If flushing, copy the remaining available input to next_out as
1752 * stored blocks, if there is enough space.
1753 */
1754 unsigned len, left, have, last = 0;
1755 unsigned used = s->strm->avail_in;
1756 do {
1757 /* Set len to the maximum size block that we can copy directly with the
1758 * available input data and output space. Set left to how much of that
1759 * would be copied from what's left in the window.
1760 */
1761 len = MAX_STORED; /* maximum deflate stored block length */
1762 have = (s->bi_valid + 42) >> 3; /* number of header bytes */
1763 if (s->strm->avail_out < have) /* need room for header */
1764 break;
1765 /* maximum stored block length that will fit in avail_out: */
1766 have = s->strm->avail_out - have;
1767 left = s->strstart - s->block_start; /* bytes left in window */
1768 if (len > (ulg)left + s->strm->avail_in)
1769 len = left + s->strm->avail_in; /* limit len to the input */
1770 if (len > have)
1771 len = have; /* limit len to the output */
1772
1773 /* If the stored block would be less than min_block in length, or if
1774 * unable to copy all of the available input when flushing, then try
1775 * copying to the window and the pending buffer instead. Also don't
1776 * write an empty block when flushing -- deflate() does that.
1777 */
1778 if (len < min_block && ((len == 0 && flush != Z_FINISH) ||
1779 flush == Z_NO_FLUSH ||
1780 len != left + s->strm->avail_in))
1781 break;
1782
1783 /* Make a dummy stored block in pending to get the header bytes,
1784 * including any pending bits. This also updates the debugging counts.
1785 */
1786 last = flush == Z_FINISH && len == left + s->strm->avail_in ? 1 : 0;
1787 _tr_stored_block(s, (char *)0, 0L, last);
1788
1789 /* Replace the lengths in the dummy stored block with len. */
1790 s->pending_buf[s->pending - 4] = len;
1791 s->pending_buf[s->pending - 3] = len >> 8;
1792 s->pending_buf[s->pending - 2] = ~len;
1793 s->pending_buf[s->pending - 1] = ~len >> 8;
1794
1795 /* Write the stored block header bytes. */
1796 flush_pending(s->strm);
1797
1798#ifdef ZLIB_DEBUG
1799 /* Update debugging counts for the data about to be copied. */
1800 s->compressed_len += len << 3;
1801 s->bits_sent += len << 3;
1802#endif
1803
1804 /* Copy uncompressed bytes from the window to next_out. */
1805 if (left) {
1806 if (left > len)
1807 left = len;
1808 zmemcpy(s->strm->next_out, s->window + s->block_start, left);
1809 s->strm->next_out += left;
1810 s->strm->avail_out -= left;
1811 s->strm->total_out += left;
1812 s->block_start += left;
1813 len -= left;
1814 }
1815
1816 /* Copy uncompressed bytes directly from next_in to next_out, updating
1817 * the check value.
1818 */
1819 if (len) {
1820 deflate_read_buf(s->strm, s->strm->next_out, len);
1821 s->strm->next_out += len;
1822 s->strm->avail_out -= len;
1823 s->strm->total_out += len;
1824 }
1825 } while (last == 0);
1826
1827 /* Update the sliding window with the last s->w_size bytes of the copied
1828 * data, or append all of the copied data to the existing window if less
1829 * than s->w_size bytes were copied. Also update the number of bytes to
1830 * insert in the hash tables, in the event that deflateParams() switches to
1831 * a non-zero compression level.
1832 */
1833 used -= s->strm->avail_in; /* number of input bytes directly copied */
1834 if (used) {
1835 /* If any input was used, then no unused input remains in the window,
1836 * therefore s->block_start == s->strstart.
1837 */
1838 if (used >= s->w_size) { /* supplant the previous history */
1839 s->matches = 2; /* clear hash */
1840 zmemcpy(s->window, s->strm->next_in - s->w_size, s->w_size);
1841 s->strstart = s->w_size;
1842 }
1843 else {
1844 if (s->window_size - s->strstart <= used) {
1845 /* Slide the window down. */
1846 s->strstart -= s->w_size;
1847 zmemcpy(s->window, s->window + s->w_size, s->strstart);
1848 if (s->matches < 2)
1849 s->matches++; /* add a pending slide_hash() */
1850 }
1851 zmemcpy(s->window + s->strstart, s->strm->next_in - used, used);
1852 s->strstart += used;
1853 }
1854 s->block_start = s->strstart;
1855 s->insert += MIN(used, s->w_size - s->insert);
1856 }
1857 if (s->high_water < s->strstart)
1858 s->high_water = s->strstart;
1859
1860 /* If the last block was written to next_out, then done. */
1861 if (last)
1862 return finish_done;
1863
1864 /* If flushing and all input has been consumed, then done. */
1865 if (flush != Z_NO_FLUSH && flush != Z_FINISH &&
1866 s->strm->avail_in == 0 && (long)s->strstart == s->block_start)
1867 return block_done;
1868
1869 /* Fill the window with any remaining input. */
1870 have = s->window_size - s->strstart - 1;
1871 if (s->strm->avail_in > have && s->block_start >= (long)s->w_size) {
1872 /* Slide the window down. */
1873 s->block_start -= s->w_size;
1874 s->strstart -= s->w_size;
1875 zmemcpy(s->window, s->window + s->w_size, s->strstart);
1876 if (s->matches < 2)
1877 s->matches++; /* add a pending slide_hash() */
1878 have += s->w_size; /* more space now */
1879 }
1880 if (have > s->strm->avail_in)
1881 have = s->strm->avail_in;
1882 if (have) {
1883 deflate_read_buf(s->strm, s->window + s->strstart, have);
1884 s->strstart += have;
1885 }
1886 if (s->high_water < s->strstart)
1887 s->high_water = s->strstart;
1888
1889 /* There was not enough avail_out to write a complete worthy or flushed
1890 * stored block to next_out. Write a stored block to pending instead, if we
1891 * have enough input for a worthy block, or if flushing and there is enough
1892 * room for the remaining input as a stored block in the pending buffer.
1893 */
1894 have = (s->bi_valid + 42) >> 3; /* number of header bytes */
1895 /* maximum stored block length that will fit in pending: */
1896 have = MIN(s->pending_buf_size - have, MAX_STORED);
1897 min_block = MIN(have, s->w_size);
1898 left = s->strstart - s->block_start;
1899 if (left >= min_block ||
1900 ((left || flush == Z_FINISH) && flush != Z_NO_FLUSH &&
1901 s->strm->avail_in == 0 && left <= have)) {
1902 len = MIN(left, have);
1903 last = flush == Z_FINISH && s->strm->avail_in == 0 &&
1904 len == left ? 1 : 0;
1905 _tr_stored_block(s, (charf *)s->window + s->block_start, len, last);
1906 s->block_start += len;
1907 flush_pending(s->strm);
1908 }
1909
1910 /* We've done all we can with the available input and output. */
1911 return last ? finish_started : need_more;
1912}
1913
1914/* ===========================================================================
1915 * Compress as much as possible from the input stream, return the current
1916 * block state.
1917 * This function does not perform lazy evaluation of matches and inserts
1918 * new strings in the dictionary only for unmatched strings or for short
1919 * matches. It is used only for the fast compression options.
1920 */
1921local block_state deflate_fast(s, flush)
1922 deflate_state *s;
1923 int flush;
1924{
1925 IPos hash_head; /* head of the hash chain */
1926 int bflush; /* set if current block must be flushed */
1927
1928 for (;;) {
1929 /* Make sure that we always have enough lookahead, except
1930 * at the end of the input file. We need MAX_MATCH bytes
1931 * for the next match, plus MIN_MATCH bytes to insert the
1932 * string following the next match.
1933 */
1934 if (s->lookahead < MIN_LOOKAHEAD) {
1935 fill_window(s);
1936 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1937 return need_more;
1938 }
1939 if (s->lookahead == 0) break; /* flush the current block */
1940 }
1941
1942 /* Insert the string window[strstart .. strstart+2] in the
1943 * dictionary, and set hash_head to the head of the hash chain:
1944 */
1945 hash_head = NIL;
1946 if (s->lookahead >= MIN_MATCH) {
1947 hash_head = insert_string(s, s->strstart);
1948 }
1949
1950 /* Find the longest match, discarding those <= prev_length.
1951 * At this point we have always match_length < MIN_MATCH
1952 */
1953 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1954 /* To simplify the code, we prevent matches with the string
1955 * of window index 0 (in particular we have to avoid a match
1956 * of the string with itself at the start of the input file).
1957 */
1958 s->match_length = longest_match (s, hash_head);
1959 /* longest_match() sets match_start */
1960 }
1961 if (s->match_length >= MIN_MATCH) {
1962 check_match(s, s->strstart, s->match_start, s->match_length);
1963
1964 _tr_tally_dist(s, s->strstart - s->match_start,
1965 s->match_length - MIN_MATCH, bflush);
1966
1967 s->lookahead -= s->match_length;
1968
1969 /* Insert new strings in the hash table only if the match length
1970 * is not too large. This saves time but degrades compression.
1971 */
1972#ifndef FASTEST
1973 if (s->match_length <= s->max_insert_length &&
1974 s->lookahead >= MIN_MATCH) {
1975 s->match_length--; /* string at strstart already in table */
1976 do {
1977 s->strstart++;
1978 hash_head = insert_string(s, s->strstart);
1979 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1980 * always MIN_MATCH bytes ahead.
1981 */
1982 } while (--s->match_length != 0);
1983 s->strstart++;
1984 } else
1985#endif
1986 {
1987 s->strstart += s->match_length;
1988 s->match_length = 0;
1989 s->ins_h = s->window[s->strstart];
1990 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1991#if MIN_MATCH != 3
1992 Call UPDATE_HASH() MIN_MATCH-3 more times
1993#endif
1994 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1995 * matter since it will be recomputed at next deflate call.
1996 */
1997 }
1998 } else {
1999 /* No match, output a literal byte */
2000 Tracevv((stderr,"%c", s->window[s->strstart]));
2001 _tr_tally_lit (s, s->window[s->strstart], bflush);
2002 s->lookahead--;
2003 s->strstart++;
2004 }
2005 if (bflush) FLUSH_BLOCK(s, 0);
2006 }
2007 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
2008 if (flush == Z_FINISH) {
2009 FLUSH_BLOCK(s, 1);
2010 return finish_done;
2011 }
2012 if (s->sym_next)
2013 FLUSH_BLOCK(s, 0);
2014 return block_done;
2015}
2016
2017#ifndef FASTEST
2018/* ===========================================================================
2019 * Same as above, but achieves better compression. We use a lazy
2020 * evaluation for matches: a match is finally adopted only if there is
2021 * no better match at the next window position.
2022 */
2023local block_state deflate_slow(s, flush)
2024 deflate_state *s;
2025 int flush;
2026{
2027 IPos hash_head; /* head of hash chain */
2028 int bflush; /* set if current block must be flushed */
2029
2030 /* Process the input block. */
2031 for (;;) {
2032 /* Make sure that we always have enough lookahead, except
2033 * at the end of the input file. We need MAX_MATCH bytes
2034 * for the next match, plus MIN_MATCH bytes to insert the
2035 * string following the next match.
2036 */
2037 if (s->lookahead < MIN_LOOKAHEAD) {
2038 fill_window(s);
2039 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
2040 return need_more;
2041 }
2042 if (s->lookahead == 0) break; /* flush the current block */
2043 }
2044
2045 /* Insert the string window[strstart .. strstart+2] in the
2046 * dictionary, and set hash_head to the head of the hash chain:
2047 */
2048 hash_head = NIL;
2049 if (s->lookahead >= MIN_MATCH) {
2050 hash_head = insert_string(s, s->strstart);
2051 }
2052
2053 /* Find the longest match, discarding those <= prev_length.
2054 */
2055 s->prev_length = s->match_length, s->prev_match = s->match_start;
2056 s->match_length = MIN_MATCH-1;
2057
2058 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
2059 s->strstart - hash_head <= MAX_DIST(s)) {
2060 /* To simplify the code, we prevent matches with the string
2061 * of window index 0 (in particular we have to avoid a match
2062 * of the string with itself at the start of the input file).
2063 */
2064 s->match_length = longest_match (s, hash_head);
2065 /* longest_match() sets match_start */
2066
2067 if (s->match_length <= 5 && (s->strategy == Z_FILTERED
2068#if TOO_FAR <= 32767
2069 || (s->match_length == MIN_MATCH &&
2070 s->strstart - s->match_start > TOO_FAR)
2071#endif
2072 )) {
2073
2074 /* If prev_match is also MIN_MATCH, match_start is garbage
2075 * but we will ignore the current match anyway.
2076 */
2077 s->match_length = MIN_MATCH-1;
2078 }
2079 }
2080 /* If there was a match at the previous step and the current
2081 * match is not better, output the previous match:
2082 */
2083 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
2084 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
2085 /* Do not insert strings in hash table beyond this. */
2086
2087 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
2088
2089 _tr_tally_dist(s, s->strstart -1 - s->prev_match,
2090 s->prev_length - MIN_MATCH, bflush);
2091
2092 /* Insert in hash table all strings up to the end of the match.
2093 * strstart-1 and strstart are already inserted. If there is not
2094 * enough lookahead, the last two strings are not inserted in
2095 * the hash table.
2096 */
2097 s->lookahead -= s->prev_length-1;
2098 s->prev_length -= 2;
2099 do {
2100 if (++s->strstart <= max_insert) {
2101 hash_head = insert_string(s, s->strstart);
2102 }
2103 } while (--s->prev_length != 0);
2104 s->match_available = 0;
2105 s->match_length = MIN_MATCH-1;
2106 s->strstart++;
2107
2108 if (bflush) FLUSH_BLOCK(s, 0);
2109
2110 } else if (s->match_available) {
2111 /* If there was no match at the previous position, output a
2112 * single literal. If there was a match but the current match
2113 * is longer, truncate the previous match to a single literal.
2114 */
2115 Tracevv((stderr,"%c", s->window[s->strstart-1]));
2116 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
2117 if (bflush) {
2118 FLUSH_BLOCK_ONLY(s, 0);
2119 }
2120 s->strstart++;
2121 s->lookahead--;
2122 if (s->strm->avail_out == 0) return need_more;
2123 } else {
2124 /* There is no previous match to compare with, wait for
2125 * the next step to decide.
2126 */
2127 s->match_available = 1;
2128 s->strstart++;
2129 s->lookahead--;
2130 }
2131 }
2132 Assert (flush != Z_NO_FLUSH, "no flush?");
2133 if (s->match_available) {
2134 Tracevv((stderr,"%c", s->window[s->strstart-1]));
2135 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
2136 s->match_available = 0;
2137 }
2138 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
2139 if (flush == Z_FINISH) {
2140 FLUSH_BLOCK(s, 1);
2141 return finish_done;
2142 }
2143 if (s->sym_next)
2144 FLUSH_BLOCK(s, 0);
2145 return block_done;
2146}
2147#endif /* FASTEST */
2148
2149/* ===========================================================================
2150 * For Z_RLE, simply look for runs of bytes, generate matches only of distance
2151 * one. Do not maintain a hash table. (It will be regenerated if this run of
2152 * deflate switches away from Z_RLE.)
2153 */
2154local block_state deflate_rle(s, flush)
2155 deflate_state *s;
2156 int flush;
2157{
2158 int bflush; /* set if current block must be flushed */
2159 uInt prev; /* byte at distance one to match */
2160 Bytef *scan, *strend; /* scan goes up to strend for length of run */
2161
2162 for (;;) {
2163 /* Make sure that we always have enough lookahead, except
2164 * at the end of the input file. We need MAX_MATCH bytes
2165 * for the longest run, plus one for the unrolled loop.
2166 */
2167 if (s->lookahead <= MAX_MATCH) {
2168 fill_window(s);
2169 if (s->lookahead <= MAX_MATCH && flush == Z_NO_FLUSH) {
2170 return need_more;
2171 }
2172 if (s->lookahead == 0) break; /* flush the current block */
2173 }
2174
2175 /* See how many times the previous byte repeats */
2176 s->match_length = 0;
2177 if (s->lookahead >= MIN_MATCH && s->strstart > 0) {
2178 scan = s->window + s->strstart - 1;
2179 prev = *scan;
2180 if (prev == *++scan && prev == *++scan && prev == *++scan) {
2181 strend = s->window + s->strstart + MAX_MATCH;
2182 do {
2183 } while (prev == *++scan && prev == *++scan &&
2184 prev == *++scan && prev == *++scan &&
2185 prev == *++scan && prev == *++scan &&
2186 prev == *++scan && prev == *++scan &&
2187 scan < strend);
2188 s->match_length = MAX_MATCH - (uInt)(strend - scan);
2189 if (s->match_length > s->lookahead)
2190 s->match_length = s->lookahead;
2191 }
2192 Assert(scan <= s->window+(uInt)(s->window_size-1), "wild scan");
2193 }
2194
2195 /* Emit match if have run of MIN_MATCH or longer, else emit literal */
2196 if (s->match_length >= MIN_MATCH) {
2197 check_match(s, s->strstart, s->strstart - 1, s->match_length);
2198
2199 _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush);
2200
2201 s->lookahead -= s->match_length;
2202 s->strstart += s->match_length;
2203 s->match_length = 0;
2204 } else {
2205 /* No match, output a literal byte */
2206 Tracevv((stderr,"%c", s->window[s->strstart]));
2207 _tr_tally_lit (s, s->window[s->strstart], bflush);
2208 s->lookahead--;
2209 s->strstart++;
2210 }
2211 if (bflush) FLUSH_BLOCK(s, 0);
2212 }
2213 s->insert = 0;
2214 if (flush == Z_FINISH) {
2215 FLUSH_BLOCK(s, 1);
2216 return finish_done;
2217 }
2218 if (s->sym_next)
2219 FLUSH_BLOCK(s, 0);
2220 return block_done;
2221}
2222
2223/* ===========================================================================
2224 * For Z_HUFFMAN_ONLY, do not look for matches. Do not maintain a hash table.
2225 * (It will be regenerated if this run of deflate switches away from Huffman.)
2226 */
2227local block_state deflate_huff(s, flush)
2228 deflate_state *s;
2229 int flush;
2230{
2231 int bflush; /* set if current block must be flushed */
2232
2233 for (;;) {
2234 /* Make sure that we have a literal to write. */
2235 if (s->lookahead == 0) {
2236 fill_window(s);
2237 if (s->lookahead == 0) {
2238 if (flush == Z_NO_FLUSH)
2239 return need_more;
2240 break; /* flush the current block */
2241 }
2242 }
2243
2244 /* Output a literal byte */
2245 s->match_length = 0;
2246 Tracevv((stderr,"%c", s->window[s->strstart]));
2247 _tr_tally_lit (s, s->window[s->strstart], bflush);
2248 s->lookahead--;
2249 s->strstart++;
2250 if (bflush) FLUSH_BLOCK(s, 0);
2251 }
2252 s->insert = 0;
2253 if (flush == Z_FINISH) {
2254 FLUSH_BLOCK(s, 1);
2255 return finish_done;
2256 }
2257 if (s->sym_next)
2258 FLUSH_BLOCK(s, 0);
2259 return block_done;
2260}
2261
2262/* Safe to inline this as GCC/clang will use inline asm and Visual Studio will
2263 * use intrinsic without extra params
2264 */
2265local INLINE Pos insert_string_sse(deflate_state *const s, const Pos str)
2266{
2267 Pos ret;
2268 unsigned *ip, val, h = 0;
2269
2270 ip = (unsigned *)&s->window[str];
2271 val = *ip;
2272
2273 if (s->level >= 6)
2274 val &= 0xFFFFFF;
2275
2276/* Windows clang should use inline asm */
2277#if defined(_MSC_VER) && !defined(__clang__) && (defined(_M_IX86) || defined(_M_X64))
2278 h = _mm_crc32_u32(h, val);
2279#elif defined(__i386__) || defined(__amd64__)
2280 __asm__ __volatile__ (
2281 "crc32 %1,%0\n\t"
2282 : "+r" (h)
2283 : "r" (val)
2284 );
2285#else
2286 /* This should never happen */
2287 assert(0);
2288#endif
2289
2290 ret = s->head[h & s->hash_mask];
2291 s->head[h & s->hash_mask] = str;
2292 s->prev[str & s->w_mask] = ret;
2293 return ret;
2294}
2295