1 /* deflate.h -- internal compression state
2 * Copyright (C) 1995-2002 Jean-loup Gailly
3 * For conditions of distribution and use, see copyright notice in zlib.h
6 /* WARNING: this file should *not* be used by applications. It is
7 part of the implementation of the compression library and is
8 subject to change. Applications should only use zlib.h.
18 /* define NO_GZIP when compiling if you want to disable gzip header and
19 trailer creation by deflate(). NO_GZIP would be used to avoid linking in
20 the crc code when it is not needed. For shared libraries, gzip encoding
21 should be left enabled. */
26 /* ===========================================================================
27 * Internal compression state.
30 #define LENGTH_CODES 29
31 /* number of length codes, not counting the special END_BLOCK code */
34 /* number of literal bytes 0..255 */
36 #define L_CODES (LITERALS+1+LENGTH_CODES)
37 /* number of Literal or Length codes, including the END_BLOCK code */
40 /* number of distance codes */
43 /* number of codes used to transfer the bit lengths */
45 #define HEAP_SIZE (2*L_CODES+1)
46 /* maximum heap size */
49 /* All codes must not exceed MAX_BITS bits */
52 #define BUSY_STATE 113
53 #define FINISH_STATE 666
57 /* Data structure describing a single value and its code string. */
58 typedef struct ct_data_s {
60 ush freq; /* frequency count */
61 ush code; /* bit string */
64 ush dad; /* father node in Huffman tree */
65 ush len; /* length of bit string */
74 typedef struct static_tree_desc_s static_tree_desc;
76 typedef struct tree_desc_s {
78 ct_data *dyn_tree; /* the dynamic tree */
79 int max_code; /* largest code with non zero frequency */
81 static_tree_desc *stat_desc; /* the corresponding static tree */
86 typedef unsigned IPos;
88 /* A Pos is an index in the character window. We use short instead of int to
89 * save space in the various tables. IPos is used only for parameter passing.
92 typedef struct internal_state {
93 z_streamp strm; /* pointer back to this zlib stream */
94 int status; /* as the name implies */
96 Bytef *pending_buf; /* output still pending */
97 ulg pending_buf_size; /* size of pending_buf */
99 Bytef *pending_out; /* next pending byte to output to the stream */
100 int pending; /* nb of bytes in the pending buffer */
101 int wrap; /* bit 0 true for zlib, bit 1 true for gzip */
102 Byte method; /* STORED (for zip only) or DEFLATED */
103 int last_flush; /* value of flush param for previous deflate call */
105 /* used by deflate.c: */
107 uInt w_size; /* LZ77 window size (32K by default) */
108 uInt w_bits; /* log2(w_size) (8..16) */
109 uInt w_mask; /* w_size - 1 */
112 /* Sliding window. Input bytes are read into the second half of the window,
113 * and move to the first half later to keep a dictionary of at least wSize
114 * bytes. With this organization, matches are limited to a distance of
115 * wSize-MAX_MATCH bytes, but this ensures that IO is always
116 * performed with a length multiple of the block size. Also, it limits
117 * the window size to 64K, which is quite useful on MSDOS.
118 * To do: use the user input buffer as sliding window.
122 /* Actual size of window: 2*wSize, except when the user input buffer
123 * is directly used as sliding window.
127 /* Link to older string with same hash index. To limit the size of this
128 * array to 64K, this link is maintained only for the last 32K strings.
129 * An index in this array is thus a window index modulo 32K.
132 Posf *head; /* Heads of the hash chains or NIL. */
134 uInt ins_h; /* hash index of string to be inserted */
135 uInt hash_size; /* number of elements in hash table */
136 uInt hash_bits; /* log2(hash_size) */
137 uInt hash_mask; /* hash_size-1 */
140 /* Number of bits by which ins_h must be shifted at each input
141 * step. It must be such that after MIN_MATCH steps, the oldest
142 * byte no longer takes part in the hash key, that is:
143 * hash_shift * MIN_MATCH >= hash_bits
147 /* Window position at the beginning of the current output block. Gets
148 * negative when the window is moved backwards.
151 uInt match_length; /* length of best match */
152 IPos prev_match; /* previous match */
153 int match_available; /* set if previous match exists */
154 uInt strstart; /* start of string to insert */
155 uInt match_start; /* start of matching string */
156 uInt lookahead; /* number of valid bytes ahead in window */
159 /* Length of the best match at previous step. Matches not greater than this
160 * are discarded. This is used in the lazy match evaluation.
163 uInt max_chain_length;
164 /* To speed up deflation, hash chains are never searched beyond this
165 * length. A higher limit improves compression ratio but degrades the
170 /* Attempt to find a better match only when the current match is strictly
171 * smaller than this value. This mechanism is used only for compression
174 # define max_insert_length max_lazy_match
175 /* Insert new strings in the hash table only if the match length is not
176 * greater than this length. This saves time but degrades compression.
177 * max_insert_length is used only for compression levels <= 3.
180 int level; /* compression level (1..9) */
181 int strategy; /* favor or force Huffman coding*/
184 /* Use a faster search when the previous match is longer than this */
186 int nice_match; /* Stop searching when current match exceeds this */
188 /* used by trees.c: */
189 /* Didn't use ct_data typedef below to supress compiler warning */
190 struct ct_data_s dyn_ltree[HEAP_SIZE]; /* literal and length tree */
191 struct ct_data_s dyn_dtree[2*D_CODES+1]; /* distance tree */
192 struct ct_data_s bl_tree[2*BL_CODES+1]; /* Huffman tree for bit lengths */
194 struct tree_desc_s l_desc; /* desc. for literal tree */
195 struct tree_desc_s d_desc; /* desc. for distance tree */
196 struct tree_desc_s bl_desc; /* desc. for bit length tree */
198 ush bl_count[MAX_BITS+1];
199 /* number of codes at each bit length for an optimal tree */
201 int heap[2*L_CODES+1]; /* heap used to build the Huffman trees */
202 int heap_len; /* number of elements in the heap */
203 int heap_max; /* element of largest frequency */
204 /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used.
205 * The same heap array is used to build all trees.
208 uch depth[2*L_CODES+1];
209 /* Depth of each subtree used as tie breaker for trees of equal frequency
213 uchf *l_buf; /* buffer for literals or lengths */
216 /* Size of match buffer for literals/lengths. There are 4 reasons for
217 * limiting lit_bufsize to 64K:
218 * - frequencies can be kept in 16 bit counters
219 * - if compression is not successful for the first block, all input
220 * data is still in the window so we can still emit a stored block even
221 * when input comes from standard input. (This can also be done for
222 * all blocks if lit_bufsize is not greater than 32K.)
223 * - if compression is not successful for a file smaller than 64K, we can
224 * even emit a stored file instead of a stored block (saving 5 bytes).
225 * This is applicable only for zip (not gzip or zlib).
226 * - creating new Huffman trees less frequently may not provide fast
227 * adaptation to changes in the input data statistics. (Take for
228 * example a binary file with poorly compressible code followed by
229 * a highly compressible string table.) Smaller buffer sizes give
230 * fast adaptation but have of course the overhead of transmitting
231 * trees more frequently.
232 * - I can't count above 4
235 uInt last_lit; /* running index in l_buf */
238 /* Buffer for distances. To simplify the code, d_buf and l_buf have
239 * the same number of elements. To use different lengths, an extra flag
240 * array would be necessary.
243 ulg opt_len; /* bit length of current block with optimal trees */
244 ulg static_len; /* bit length of current block with static trees */
245 uInt matches; /* number of string matches in current block */
246 int last_eob_len; /* bit length of EOB code for last block */
249 ulg compressed_len; /* total bit length of compressed file mod 2^32 */
250 ulg bits_sent; /* bit length of compressed data sent mod 2^32 */
254 /* Output buffer. bits are inserted starting at the bottom (least
258 /* Number of valid bits in bi_buf. All bits above the last valid bit
264 /* Output a byte on the stream.
265 * IN assertion: there is enough room in pending_buf.
267 #define put_byte(s, c) {s->pending_buf[s->pending++] = (c);}
270 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
271 /* Minimum amount of lookahead, except at the end of the input file.
272 * See deflate.c for comments about the MIN_MATCH+1.
275 #define MAX_DIST(s) ((s)->w_size-MIN_LOOKAHEAD)
276 /* In order to simplify the code, particularly on 16 bit machines, match
277 * distances are limited to MAX_DIST instead of WSIZE.
281 void _tr_init OF((deflate_state *s))
283 int _tr_tally OF((deflate_state *s, unsigned dist, unsigned lc))
285 void _tr_flush_block OF((deflate_state *s, charf *buf, ulg stored_len,
288 void _tr_align OF((deflate_state *s))
290 void _tr_stored_block OF((deflate_state *s, charf *buf, ulg stored_len,
294 #define d_code(dist) \
295 ((dist) < 256 ? _dist_code[dist] : _dist_code[256+((dist)>>7)])
296 /* Mapping from a distance to a distance code. dist is the distance - 1 and
297 * must not have side effects. _dist_code[256] and _dist_code[257] are never
302 /* Inline versions of _tr_tally for speed: */
304 #if defined(GEN_TREES_H) || !defined(STDC)
305 extern uch _length_code[];
306 extern uch _dist_code[];
308 /*@unchecked@*/ /*@observer@*/
309 extern const uch _length_code[];
310 /*@unchecked@*/ /*@observer@*/
311 extern const uch _dist_code[];
314 # define _tr_tally_lit(s, c, flush) \
316 s->d_buf[s->last_lit] = 0; \
317 s->l_buf[s->last_lit++] = cc; \
318 s->dyn_ltree[cc].Freq++; \
319 flush = (s->last_lit == s->lit_bufsize-1); \
321 # define _tr_tally_dist(s, distance, length, flush) \
322 { uch len = (length); \
323 ush dist = (distance); \
324 s->d_buf[s->last_lit] = dist; \
325 s->l_buf[s->last_lit++] = len; \
327 s->dyn_ltree[_length_code[len]+LITERALS+1].Freq++; \
328 s->dyn_dtree[d_code(dist)].Freq++; \
329 flush = (s->last_lit == s->lit_bufsize-1); \
332 # define _tr_tally_lit(s, c, flush) flush = _tr_tally(s, 0, c)
333 # define _tr_tally_dist(s, distance, length, flush) \
334 flush = _tr_tally(s, distance, length)
337 #endif /* DEFLATE_H */