1#ifndef DEFLATE_H_
2#define DEFLATE_H_
3/* deflate.h -- internal compression state
4 * Copyright (C) 1995-2016 Jean-loup Gailly
5 * For conditions of distribution and use, see copyright notice in zlib.h
6 */
7
8/* WARNING: this file should *not* be used by applications. It is
9 part of the implementation of the compression library and is
10 subject to change. Applications should only use zlib.h.
11 */
12
13#include "zutil.h"
14#include "zendian.h"
15
16/* define NO_GZIP when compiling if you want to disable gzip header and
17 trailer creation by deflate(). NO_GZIP would be used to avoid linking in
18 the crc code when it is not needed. For shared libraries, gzip encoding
19 should be left enabled. */
20#ifndef NO_GZIP
21# define GZIP
22#endif
23
24/* ===========================================================================
25 * Internal compression state.
26 */
27
28#define LENGTH_CODES 29
29/* number of length codes, not counting the special END_BLOCK code */
30
31#define LITERALS 256
32/* number of literal bytes 0..255 */
33
34#define L_CODES (LITERALS+1+LENGTH_CODES)
35/* number of Literal or Length codes, including the END_BLOCK code */
36
37#define D_CODES 30
38/* number of distance codes */
39
40#define BL_CODES 19
41/* number of codes used to transfer the bit lengths */
42
43#define HEAP_SIZE (2*L_CODES+1)
44/* maximum heap size */
45
46#define MAX_BITS 15
47/* All codes must not exceed MAX_BITS bits */
48
49#define BIT_BUF_SIZE 64
50/* size of bit buffer in bi_buf */
51
52#define END_BLOCK 256
53/* end of block literal code */
54
55#define INIT_STATE 42 /* zlib header -> BUSY_STATE */
56#ifdef GZIP
57# define GZIP_STATE 57 /* gzip header -> BUSY_STATE | EXTRA_STATE */
58#endif
59#define EXTRA_STATE 69 /* gzip extra block -> NAME_STATE */
60#define NAME_STATE 73 /* gzip file name -> COMMENT_STATE */
61#define COMMENT_STATE 91 /* gzip comment -> HCRC_STATE */
62#define HCRC_STATE 103 /* gzip header CRC -> BUSY_STATE */
63#define BUSY_STATE 113 /* deflate -> FINISH_STATE */
64#define FINISH_STATE 666 /* stream complete */
65/* Stream status */
66
67#define HASH_BITS 16u /* log2(HASH_SIZE) */
68#ifndef HASH_SIZE
69# define HASH_SIZE 65536u /* number of elements in hash table */
70#endif
71#define HASH_MASK (HASH_SIZE - 1u) /* HASH_SIZE-1 */
72
73
74/* Data structure describing a single value and its code string. */
75typedef struct ct_data_s {
76 union {
77 uint16_t freq; /* frequency count */
78 uint16_t code; /* bit string */
79 } fc;
80 union {
81 uint16_t dad; /* father node in Huffman tree */
82 uint16_t len; /* length of bit string */
83 } dl;
84} ct_data;
85
86#define Freq fc.freq
87#define Code fc.code
88#define Dad dl.dad
89#define Len dl.len
90
91typedef struct static_tree_desc_s static_tree_desc;
92
93typedef struct tree_desc_s {
94 ct_data *dyn_tree; /* the dynamic tree */
95 int max_code; /* largest code with non zero frequency */
96 const static_tree_desc *stat_desc; /* the corresponding static tree */
97} tree_desc;
98
99typedef uint16_t Pos;
100
101/* A Pos is an index in the character window. We use short instead of int to
102 * save space in the various tables.
103 */
104
105typedef struct internal_state {
106 PREFIX3(stream) *strm; /* pointer back to this zlib stream */
107 unsigned char *pending_buf; /* output still pending */
108 unsigned char *pending_out; /* next pending byte to output to the stream */
109 uint32_t pending_buf_size; /* size of pending_buf */
110 uint32_t pending; /* nb of bytes in the pending buffer */
111 int wrap; /* bit 0 true for zlib, bit 1 true for gzip */
112 uint32_t gzindex; /* where in extra, name, or comment */
113 PREFIX(gz_headerp) gzhead; /* gzip header information to write */
114 int status; /* as the name implies */
115 int last_flush; /* value of flush param for previous deflate call */
116 int reproducible; /* Whether reproducible compression results are required. */
117
118 int block_open;
119 /* Whether or not a block is currently open for the QUICK deflation scheme.
120 * This is set to 1 if there is an active block, or 0 if the block was just closed.
121 */
122
123 /* used by deflate.c: */
124
125 unsigned int w_size; /* LZ77 window size (32K by default) */
126 unsigned int w_bits; /* log2(w_size) (8..16) */
127 unsigned int w_mask; /* w_size - 1 */
128 unsigned int lookahead; /* number of valid bytes ahead in window */
129
130 unsigned int high_water;
131 /* High water mark offset in window for initialized bytes -- bytes above
132 * this are set to zero in order to avoid memory check warnings when
133 * longest match routines access bytes past the input. This is then
134 * updated to the new high water mark.
135 */
136
137 unsigned int window_size;
138 /* Actual size of window: 2*wSize, except when the user input buffer
139 * is directly used as sliding window.
140 */
141
142 unsigned char *window;
143 /* Sliding window. Input bytes are read into the second half of the window,
144 * and move to the first half later to keep a dictionary of at least wSize
145 * bytes. With this organization, matches are limited to a distance of
146 * wSize-MAX_MATCH bytes, but this ensures that IO is always
147 * performed with a length multiple of the block size. Also, it limits
148 * the window size to 64K, which is quite useful on MSDOS.
149 * To do: use the user input buffer as sliding window.
150 */
151
152 Pos *prev;
153 /* Link to older string with same hash index. To limit the size of this
154 * array to 64K, this link is maintained only for the last 32K strings.
155 * An index in this array is thus a window index modulo 32K.
156 */
157
158 Pos *head; /* Heads of the hash chains or 0. */
159
160 int block_start;
161 /* Window position at the beginning of the current output block. Gets
162 * negative when the window is moved backwards.
163 */
164
165 unsigned int match_length; /* length of best match */
166 Pos prev_match; /* previous match */
167 int match_available; /* set if previous match exists */
168 unsigned int strstart; /* start of string to insert */
169 unsigned int match_start; /* start of matching string */
170
171 unsigned int prev_length;
172 /* Length of the best match at previous step. Matches not greater than this
173 * are discarded. This is used in the lazy match evaluation.
174 */
175
176 unsigned int max_chain_length;
177 /* To speed up deflation, hash chains are never searched beyond this length.
178 * A higher limit improves compression ratio but degrades the speed.
179 */
180
181 unsigned int max_lazy_match;
182 /* Attempt to find a better match only when the current match is strictly smaller
183 * than this value. This mechanism is used only for compression levels >= 4.
184 */
185# define max_insert_length max_lazy_match
186 /* Insert new strings in the hash table only if the match length is not
187 * greater than this length. This saves time but degrades compression.
188 * max_insert_length is used only for compression levels <= 3.
189 */
190
191 int level; /* compression level (1..9) */
192 int strategy; /* favor or force Huffman coding*/
193
194 unsigned int good_match;
195 /* Use a faster search when the previous match is longer than this */
196
197 int nice_match; /* Stop searching when current match exceeds this */
198
199#if defined(__x86_64__) || defined(_M_X64) || defined(__i386) || defined(_M_IX86)
200 /* Only used if X86_PCLMULQDQ_CRC is defined */
201 unsigned crc0[4 * 5];
202#endif
203
204 /* used by trees.c: */
205 /* Didn't use ct_data typedef below to suppress compiler warning */
206 struct ct_data_s dyn_ltree[HEAP_SIZE]; /* literal and length tree */
207 struct ct_data_s dyn_dtree[2*D_CODES+1]; /* distance tree */
208 struct ct_data_s bl_tree[2*BL_CODES+1]; /* Huffman tree for bit lengths */
209
210 struct tree_desc_s l_desc; /* desc. for literal tree */
211 struct tree_desc_s d_desc; /* desc. for distance tree */
212 struct tree_desc_s bl_desc; /* desc. for bit length tree */
213
214 uint16_t bl_count[MAX_BITS+1];
215 /* number of codes at each bit length for an optimal tree */
216
217 int heap[2*L_CODES+1]; /* heap used to build the Huffman trees */
218 int heap_len; /* number of elements in the heap */
219 int heap_max; /* element of largest frequency */
220 /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used.
221 * The same heap array is used to build all trees.
222 */
223
224 unsigned char depth[2*L_CODES+1];
225 /* Depth of each subtree used as tie breaker for trees of equal frequency
226 */
227
228 unsigned int lit_bufsize;
229 /* Size of match buffer for literals/lengths. There are 4 reasons for
230 * limiting lit_bufsize to 64K:
231 * - frequencies can be kept in 16 bit counters
232 * - if compression is not successful for the first block, all input
233 * data is still in the window so we can still emit a stored block even
234 * when input comes from standard input. (This can also be done for
235 * all blocks if lit_bufsize is not greater than 32K.)
236 * - if compression is not successful for a file smaller than 64K, we can
237 * even emit a stored file instead of a stored block (saving 5 bytes).
238 * This is applicable only for zip (not gzip or zlib).
239 * - creating new Huffman trees less frequently may not provide fast
240 * adaptation to changes in the input data statistics. (Take for
241 * example a binary file with poorly compressible code followed by
242 * a highly compressible string table.) Smaller buffer sizes give
243 * fast adaptation but have of course the overhead of transmitting
244 * trees more frequently.
245 * - I can't count above 4
246 */
247
248 unsigned char *sym_buf; /* buffer for distances and literals/lengths */
249 unsigned int sym_next; /* running index in sym_buf */
250 unsigned int sym_end; /* symbol table full when sym_next reaches this */
251
252 unsigned long opt_len; /* bit length of current block with optimal trees */
253 unsigned long static_len; /* bit length of current block with static trees */
254 unsigned int matches; /* number of string matches in current block */
255 unsigned int insert; /* bytes at end of window left to insert */
256
257 /* compressed_len and bits_sent are only used if ZLIB_DEBUG is defined */
258 unsigned long compressed_len; /* total bit length of compressed file mod 2^32 */
259 unsigned long bits_sent; /* bit length of compressed data sent mod 2^32 */
260
261 /* Reserved for future use and alignment purposes */
262 char *reserved_p;
263
264 uint64_t bi_buf;
265 /* Output buffer. bits are inserted starting at the bottom (least significant bits). */
266
267 int32_t bi_valid;
268 /* Number of valid bits in bi_buf. All bits above the last valid bit are always zero. */
269
270 /* Reserved for future use and alignment purposes */
271 int32_t reserved[11];
272} ALIGNED_(8) deflate_state;
273
274typedef enum {
275 need_more, /* block not completed, need more input or more output */
276 block_done, /* block flush performed */
277 finish_started, /* finish started, need only more output at next deflate */
278 finish_done /* finish done, accept no more input or output */
279} block_state;
280
281/* Output a byte on the stream.
282 * IN assertion: there is enough room in pending_buf.
283 */
284#define put_byte(s, c) { \
285 s->pending_buf[s->pending++] = (unsigned char)(c); \
286}
287
288/* ===========================================================================
289 * Output a short LSB first on the stream.
290 * IN assertion: there is enough room in pending_buf.
291 */
292static inline void put_short(deflate_state *s, uint16_t w) {
293#if defined(UNALIGNED_OK)
294 *(uint16_t *)(&s->pending_buf[s->pending]) = w;
295 s->pending += 2;
296#else
297 put_byte(s, (w & 0xff));
298 put_byte(s, ((w >> 8) & 0xff));
299#endif
300}
301
302/* ===========================================================================
303 * Output a short MSB first on the stream.
304 * IN assertion: there is enough room in pending_buf.
305 */
306static inline void put_short_msb(deflate_state *s, uint16_t w) {
307 put_byte(s, ((w >> 8) & 0xff));
308 put_byte(s, (w & 0xff));
309}
310
311/* ===========================================================================
312 * Output a 32-bit unsigned int LSB first on the stream.
313 * IN assertion: there is enough room in pending_buf.
314 */
315static inline void put_uint32(deflate_state *s, uint32_t dw) {
316#if defined(UNALIGNED_OK)
317 *(uint32_t *)(&s->pending_buf[s->pending]) = dw;
318 s->pending += 4;
319#else
320 put_byte(s, (dw & 0xff));
321 put_byte(s, ((dw >> 8) & 0xff));
322 put_byte(s, ((dw >> 16) & 0xff));
323 put_byte(s, ((dw >> 24) & 0xff));
324#endif
325}
326
327/* ===========================================================================
328 * Output a 32-bit unsigned int MSB first on the stream.
329 * IN assertion: there is enough room in pending_buf.
330 */
331static inline void put_uint32_msb(deflate_state *s, uint32_t dw) {
332#if defined(UNALIGNED_OK)
333 *(uint32_t *)(&s->pending_buf[s->pending]) = ZSWAP32(dw);
334 s->pending += 4;
335#else
336 put_byte(s, ((dw >> 24) & 0xff));
337 put_byte(s, ((dw >> 16) & 0xff));
338 put_byte(s, ((dw >> 8) & 0xff));
339 put_byte(s, (dw & 0xff));
340#endif
341}
342
343/* ===========================================================================
344 * Output a 64-bit unsigned int LSB first on the stream.
345 * IN assertion: there is enough room in pending_buf.
346 */
347static inline void put_uint64(deflate_state *s, uint64_t lld) {
348#if defined(UNALIGNED64_OK)
349 *(uint64_t *)(&s->pending_buf[s->pending]) = lld;
350 s->pending += 8;
351#elif defined(UNALIGNED_OK)
352 *(uint32_t *)(&s->pending_buf[s->pending]) = lld & 0xffffffff;
353 s->pending += 4;
354 *(uint32_t *)(&s->pending_buf[s->pending]) = (lld >> 32) & 0xffffffff;
355 s->pending += 4;
356#else
357 put_byte(s, (lld & 0xff));
358 put_byte(s, ((lld >> 8) & 0xff));
359 put_byte(s, ((lld >> 16) & 0xff));
360 put_byte(s, ((lld >> 24) & 0xff));
361 put_byte(s, ((lld >> 32) & 0xff));
362 put_byte(s, ((lld >> 40) & 0xff));
363 put_byte(s, ((lld >> 48) & 0xff));
364 put_byte(s, ((lld >> 56) & 0xff));
365#endif
366}
367
368#define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
369/* Minimum amount of lookahead, except at the end of the input file.
370 * See deflate.c for comments about the MIN_MATCH+1.
371 */
372
373#define MAX_DIST(s) ((s)->w_size-MIN_LOOKAHEAD)
374/* In order to simplify the code, particularly on 16 bit machines, match
375 * distances are limited to MAX_DIST instead of WSIZE.
376 */
377
378#define WIN_INIT MAX_MATCH
379/* Number of bytes after end of data in window to initialize in order to avoid
380 memory checker errors from longest match routines */
381
382
383void Z_INTERNAL fill_window(deflate_state *s);
384void Z_INTERNAL slide_hash_c(deflate_state *s);
385
386 /* in trees.c */
387void Z_INTERNAL zng_tr_init(deflate_state *s);
388void Z_INTERNAL zng_tr_flush_block(deflate_state *s, char *buf, uint32_t stored_len, int last);
389void Z_INTERNAL zng_tr_flush_bits(deflate_state *s);
390void Z_INTERNAL zng_tr_align(deflate_state *s);
391void Z_INTERNAL zng_tr_stored_block(deflate_state *s, char *buf, uint32_t stored_len, int last);
392unsigned Z_INTERNAL bi_reverse(unsigned code, int len);
393void Z_INTERNAL flush_pending(PREFIX3(streamp) strm);
394#define d_code(dist) ((dist) < 256 ? zng_dist_code[dist] : zng_dist_code[256+((dist)>>7)])
395/* Mapping from a distance to a distance code. dist is the distance - 1 and
396 * must not have side effects. zng_dist_code[256] and zng_dist_code[257] are never
397 * used.
398 */
399
400/* Bit buffer and compress bits calculation debugging */
401#ifdef ZLIB_DEBUG
402# define cmpr_bits_add(s, len) s->compressed_len += (len)
403# define cmpr_bits_align(s) s->compressed_len = (s->compressed_len + 7) & ~7L
404# define sent_bits_add(s, bits) s->bits_sent += (bits)
405# define sent_bits_align(s) s->bits_sent = (s->bits_sent + 7) & ~7L
406#else
407# define cmpr_bits_add(s, len) Z_UNUSED(len)
408# define cmpr_bits_align(s)
409# define sent_bits_add(s, bits) Z_UNUSED(bits)
410# define sent_bits_align(s)
411#endif
412
413#endif /* DEFLATE_H_ */
414