blob: d9feaf638608a6de4d148f6fdce19e5a1fc59284 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001
2
3
4#define Assert(err, str)
5#define Trace(dummy)
6#define Tracev(dummy)
7#define Tracecv(err, dummy)
8#define Tracevv(dummy)
9
10
11
12#define LENGTH_CODES 29
13/* number of length codes, not counting the special END_BLOCK code */
14
15#define LITERALS 256
16/* number of literal bytes 0..255 */
17
18#define L_CODES (LITERALS+1+LENGTH_CODES)
19/* number of Literal or Length codes, including the END_BLOCK code */
20
21#define D_CODES 30
22/* number of distance codes */
23
24#define BL_CODES 19
25/* number of codes used to transfer the bit lengths */
26
27#define HEAP_SIZE (2*L_CODES+1)
28/* maximum heap size */
29
30#define MAX_BITS 15
31/* All codes must not exceed MAX_BITS bits */
32
33#define INIT_STATE 42
34#define BUSY_STATE 113
35#define FINISH_STATE 666
36/* Stream status */
37
38
39/* Data structure describing a single value and its code string. */
40typedef struct ct_data_s {
41 union {
42 ush freq; /* frequency count */
43 ush code; /* bit string */
44 } fc;
45 union {
46 ush dad; /* father node in Huffman tree */
47 ush len; /* length of bit string */
48 } dl;
49} ct_data;
50
51#define Freq fc.freq
52#define Code fc.code
53#define Dad dl.dad
54#define Len dl.len
55
56typedef struct static_tree_desc_s static_tree_desc;
57
58typedef struct tree_desc_s {
59 ct_data *dyn_tree; /* the dynamic tree */
60 int max_code; /* largest code with non zero frequency */
61 static_tree_desc *stat_desc; /* the corresponding static tree */
62} tree_desc;
63
64typedef ush Pos;
65typedef unsigned IPos;
66
67/* A Pos is an index in the character window. We use short instead of int to
68 * save space in the various tables. IPos is used only for parameter passing.
69 */
70
71typedef struct deflate_state {
72 z_streamp strm; /* pointer back to this zlib stream */
73 int status; /* as the name implies */
74 Byte *pending_buf; /* output still pending */
75 ulg pending_buf_size; /* size of pending_buf */
76 Byte *pending_out; /* next pending byte to output to the stream */
77 int pending; /* nb of bytes in the pending buffer */
78 int noheader; /* suppress zlib header and adler32 */
79 Byte data_type; /* UNKNOWN, BINARY or ASCII */
80 Byte method; /* STORED (for zip only) or DEFLATED */
81 int last_flush; /* value of flush param for previous deflate call */
82
83 /* used by deflate.c: */
84
85 uInt w_size; /* LZ77 window size (32K by default) */
86 uInt w_bits; /* log2(w_size) (8..16) */
87 uInt w_mask; /* w_size - 1 */
88
89 Byte *window;
90 /* Sliding window. Input bytes are read into the second half of the window,
91 * and move to the first half later to keep a dictionary of at least wSize
92 * bytes. With this organization, matches are limited to a distance of
93 * wSize-MAX_MATCH bytes, but this ensures that IO is always
94 * performed with a length multiple of the block size. Also, it limits
95 * the window size to 64K, which is quite useful on MSDOS.
96 * To do: use the user input buffer as sliding window.
97 */
98
99 ulg window_size;
100 /* Actual size of window: 2*wSize, except when the user input buffer
101 * is directly used as sliding window.
102 */
103
104 Pos *prev;
105 /* Link to older string with same hash index. To limit the size of this
106 * array to 64K, this link is maintained only for the last 32K strings.
107 * An index in this array is thus a window index modulo 32K.
108 */
109
110 Pos *head; /* Heads of the hash chains or NIL. */
111
112 uInt ins_h; /* hash index of string to be inserted */
113 uInt hash_size; /* number of elements in hash table */
114 uInt hash_bits; /* log2(hash_size) */
115 uInt hash_mask; /* hash_size-1 */
116
117 uInt hash_shift;
118 /* Number of bits by which ins_h must be shifted at each input
119 * step. It must be such that after MIN_MATCH steps, the oldest
120 * byte no longer takes part in the hash key, that is:
121 * hash_shift * MIN_MATCH >= hash_bits
122 */
123
124 long block_start;
125 /* Window position at the beginning of the current output block. Gets
126 * negative when the window is moved backwards.
127 */
128
129 uInt match_length; /* length of best match */
130 IPos prev_match; /* previous match */
131 int match_available; /* set if previous match exists */
132 uInt strstart; /* start of string to insert */
133 uInt match_start; /* start of matching string */
134 uInt lookahead; /* number of valid bytes ahead in window */
135
136 uInt prev_length;
137 /* Length of the best match at previous step. Matches not greater than this
138 * are discarded. This is used in the lazy match evaluation.
139 */
140
141 uInt max_chain_length;
142 /* To speed up deflation, hash chains are never searched beyond this
143 * length. A higher limit improves compression ratio but degrades the
144 * speed.
145 */
146
147 uInt max_lazy_match;
148 /* Attempt to find a better match only when the current match is strictly
149 * smaller than this value. This mechanism is used only for compression
150 * levels >= 4.
151 */
152# define max_insert_length max_lazy_match
153 /* Insert new strings in the hash table only if the match length is not
154 * greater than this length. This saves time but degrades compression.
155 * max_insert_length is used only for compression levels <= 3.
156 */
157
158 int level; /* compression level (1..9) */
159 int strategy; /* favor or force Huffman coding*/
160
161 uInt good_match;
162 /* Use a faster search when the previous match is longer than this */
163
164 int nice_match; /* Stop searching when current match exceeds this */
165
166 /* used by trees.c: */
167 /* Didn't use ct_data typedef below to supress compiler warning */
168 struct ct_data_s dyn_ltree[HEAP_SIZE]; /* literal and length tree */
169 struct ct_data_s dyn_dtree[2*D_CODES+1]; /* distance tree */
170 struct ct_data_s bl_tree[2*BL_CODES+1]; /* Huffman tree for bit lengths */
171
172 struct tree_desc_s l_desc; /* desc. for literal tree */
173 struct tree_desc_s d_desc; /* desc. for distance tree */
174 struct tree_desc_s bl_desc; /* desc. for bit length tree */
175
176 ush bl_count[MAX_BITS+1];
177 /* number of codes at each bit length for an optimal tree */
178
179 int heap[2*L_CODES+1]; /* heap used to build the Huffman trees */
180 int heap_len; /* number of elements in the heap */
181 int heap_max; /* element of largest frequency */
182 /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used.
183 * The same heap array is used to build all trees.
184 */
185
186 uch depth[2*L_CODES+1];
187 /* Depth of each subtree used as tie breaker for trees of equal frequency
188 */
189
190 uch *l_buf; /* buffer for literals or lengths */
191
192 uInt lit_bufsize;
193 /* Size of match buffer for literals/lengths. There are 4 reasons for
194 * limiting lit_bufsize to 64K:
195 * - frequencies can be kept in 16 bit counters
196 * - if compression is not successful for the first block, all input
197 * data is still in the window so we can still emit a stored block even
198 * when input comes from standard input. (This can also be done for
199 * all blocks if lit_bufsize is not greater than 32K.)
200 * - if compression is not successful for a file smaller than 64K, we can
201 * even emit a stored file instead of a stored block (saving 5 bytes).
202 * This is applicable only for zip (not gzip or zlib).
203 * - creating new Huffman trees less frequently may not provide fast
204 * adaptation to changes in the input data statistics. (Take for
205 * example a binary file with poorly compressible code followed by
206 * a highly compressible string table.) Smaller buffer sizes give
207 * fast adaptation but have of course the overhead of transmitting
208 * trees more frequently.
209 * - I can't count above 4
210 */
211
212 uInt last_lit; /* running index in l_buf */
213
214 ush *d_buf;
215 /* Buffer for distances. To simplify the code, d_buf and l_buf have
216 * the same number of elements. To use different lengths, an extra flag
217 * array would be necessary.
218 */
219
220 ulg opt_len; /* bit length of current block with optimal trees */
221 ulg static_len; /* bit length of current block with static trees */
222 ulg compressed_len; /* total bit length of compressed file */
223 uInt matches; /* number of string matches in current block */
224 int last_eob_len; /* bit length of EOB code for last block */
225
226#ifdef DEBUG_ZLIB
227 ulg bits_sent; /* bit length of the compressed data */
228#endif
229
230 ush bi_buf;
231 /* Output buffer. bits are inserted starting at the bottom (least
232 * significant bits).
233 */
234 int bi_valid;
235 /* Number of valid bits in bi_buf. All bits above the last valid bit
236 * are always zero.
237 */
238
239} deflate_state;
240
241typedef struct deflate_workspace {
242 /* State memory for the deflator */
243 deflate_state deflate_memory;
244 Byte window_memory[2 * (1 << MAX_WBITS)];
245 Pos prev_memory[1 << MAX_WBITS];
246 Pos head_memory[1 << (MAX_MEM_LEVEL + 7)];
247 char overlay_memory[(1 << (MAX_MEM_LEVEL + 6)) * (sizeof(ush)+2)];
248} deflate_workspace;
249
250/* Output a byte on the stream.
251 * IN assertion: there is enough room in pending_buf.
252 */
253#define put_byte(s, c) {s->pending_buf[s->pending++] = (c);}
254
255
256#define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
257/* Minimum amount of lookahead, except at the end of the input file.
258 * See deflate.c for comments about the MIN_MATCH+1.
259 */
260
261#define MAX_DIST(s) ((s)->w_size-MIN_LOOKAHEAD)
262/* In order to simplify the code, particularly on 16 bit machines, match
263 * distances are limited to MAX_DIST instead of WSIZE.
264 */
265
266 /* in trees.c */
267void zlib_tr_init (deflate_state *s);
268int zlib_tr_tally (deflate_state *s, unsigned dist, unsigned lc);
269ulg zlib_tr_flush_block (deflate_state *s, char *buf, ulg stored_len,
270 int eof);
271void zlib_tr_align (deflate_state *s);
272void zlib_tr_stored_block (deflate_state *s, char *buf, ulg stored_len,
273 int eof);
274void zlib_tr_stored_type_only (deflate_state *);
275
276
277/* ===========================================================================
278 * Output a short LSB first on the stream.
279 * IN assertion: there is enough room in pendingBuf.
280 */
281#define put_short(s, w) { \
282 put_byte(s, (uch)((w) & 0xff)); \
283 put_byte(s, (uch)((ush)(w) >> 8)); \
284}
285
286/* ===========================================================================
287 * Reverse the first len bits of a code, using straightforward code (a faster
288 * method would use a table)
289 * IN assertion: 1 <= len <= 15
290 */
291static inline unsigned bi_reverse(unsigned code, /* the value to invert */
292 int len) /* its bit length */
293{
294 register unsigned res = 0;
295 do {
296 res |= code & 1;
297 code >>= 1, res <<= 1;
298 } while (--len > 0);
299 return res >> 1;
300}
301
302/* ===========================================================================
303 * Flush the bit buffer, keeping at most 7 bits in it.
304 */
305static inline void bi_flush(deflate_state *s)
306{
307 if (s->bi_valid == 16) {
308 put_short(s, s->bi_buf);
309 s->bi_buf = 0;
310 s->bi_valid = 0;
311 } else if (s->bi_valid >= 8) {
312 put_byte(s, (Byte)s->bi_buf);
313 s->bi_buf >>= 8;
314 s->bi_valid -= 8;
315 }
316}
317
318/* ===========================================================================
319 * Flush the bit buffer and align the output on a byte boundary
320 */
321static inline void bi_windup(deflate_state *s)
322{
323 if (s->bi_valid > 8) {
324 put_short(s, s->bi_buf);
325 } else if (s->bi_valid > 0) {
326 put_byte(s, (Byte)s->bi_buf);
327 }
328 s->bi_buf = 0;
329 s->bi_valid = 0;
330#ifdef DEBUG_ZLIB
331 s->bits_sent = (s->bits_sent+7) & ~7;
332#endif
333}
334