1 /* deflate.h -- internal compression state 2 * Copyright (C) 1995-2012 Jean-loup Gailly 3 * For conditions of distribution and use, see copyright notice in zlib.h 4 */ 5 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. 9 */ 10 11 /* @(#) $Id$ */ 12 13 #ifndef DEFLATE_H 14 #define DEFLATE_H 15 16 #include "hammer2_zlib_zutil.h" 17 18 /* =========================================================================== 19 * Internal compression state. 20 */ 21 22 #define LENGTH_CODES 29 23 /* number of length codes, not counting the special END_BLOCK code */ 24 25 #define LITERALS 256 26 /* number of literal bytes 0..255 */ 27 28 #define L_CODES (LITERALS+1+LENGTH_CODES) 29 /* number of Literal or Length codes, including the END_BLOCK code */ 30 31 #define D_CODES 30 32 /* number of distance codes */ 33 34 #define BL_CODES 19 35 /* number of codes used to transfer the bit lengths */ 36 37 #define HEAP_SIZE (2*L_CODES+1) 38 /* maximum heap size */ 39 40 #define MAX_BITS 15 41 /* All codes must not exceed MAX_BITS bits */ 42 43 #define Buf_size 16 44 /* size of bit buffer in bi_buf */ 45 46 #define INIT_STATE 42 47 #define EXTRA_STATE 69 48 #define NAME_STATE 73 49 #define COMMENT_STATE 91 50 #define HCRC_STATE 103 51 #define BUSY_STATE 113 52 #define FINISH_STATE 666 53 /* Stream status */ 54 55 56 /* Data structure describing a single value and its code string. */ 57 typedef struct ct_data_s { 58 union { 59 ush freq; /* frequency count */ 60 ush code; /* bit string */ 61 } fc; 62 union { 63 ush dad; /* father node in Huffman tree */ 64 ush len; /* length of bit string */ 65 } dl; 66 } FAR ct_data; 67 68 #define Freq fc.freq 69 #define Code fc.code 70 #define Dad dl.dad 71 #define Len dl.len 72 73 typedef struct static_tree_desc_s static_tree_desc; 74 75 typedef struct tree_desc_s { 76 ct_data *dyn_tree; /* the dynamic tree */ 77 int max_code; /* largest code with non zero frequency */ 78 static_tree_desc *stat_desc; /* the corresponding static tree */ 79 } FAR tree_desc; 80 81 typedef ush Pos; 82 typedef Pos FAR Posf; 83 typedef unsigned IPos; 84 85 /* A Pos is an index in the character window. We use short instead of int to 86 * save space in the various tables. IPos is used only for parameter passing. 87 */ 88 89 typedef struct internal_state { 90 z_streamp strm; /* pointer back to this zlib stream */ 91 int status; /* as the name implies */ 92 Bytef *pending_buf; /* output still pending */ 93 ulg pending_buf_size; /* size of pending_buf */ 94 Bytef *pending_out; /* next pending byte to output to the stream */ 95 uInt pending; /* nb of bytes in the pending buffer */ 96 int wrap; /* bit 0 true for zlib, bit 1 true for gzip */ 97 uInt gzindex; /* where in extra, name, or comment */ 98 Byte method; /* can only be DEFLATED */ 99 int last_flush; /* value of flush param for previous deflate call */ 100 101 /* used by deflate.c: */ 102 103 uInt w_size; /* LZ77 window size (32K by default) */ 104 uInt w_bits; /* log2(w_size) (8..16) */ 105 uInt w_mask; /* w_size - 1 */ 106 107 Bytef *window; 108 /* Sliding window. Input bytes are read into the second half of the window, 109 * and move to the first half later to keep a dictionary of at least wSize 110 * bytes. With this organization, matches are limited to a distance of 111 * wSize-MAX_MATCH bytes, but this ensures that IO is always 112 * performed with a length multiple of the block size. Also, it limits 113 * the window size to 64K, which is quite useful on MSDOS. 114 * To do: use the user input buffer as sliding window. 115 */ 116 117 ulg window_size; 118 /* Actual size of window: 2*wSize, except when the user input buffer 119 * is directly used as sliding window. 120 */ 121 122 Posf *prev; 123 /* Link to older string with same hash index. To limit the size of this 124 * array to 64K, this link is maintained only for the last 32K strings. 125 * An index in this array is thus a window index modulo 32K. 126 */ 127 128 Posf *head; /* Heads of the hash chains or NIL. */ 129 130 uInt ins_h; /* hash index of string to be inserted */ 131 uInt hash_size; /* number of elements in hash table */ 132 uInt hash_bits; /* log2(hash_size) */ 133 uInt hash_mask; /* hash_size-1 */ 134 135 uInt hash_shift; 136 /* Number of bits by which ins_h must be shifted at each input 137 * step. It must be such that after MIN_MATCH steps, the oldest 138 * byte no longer takes part in the hash key, that is: 139 * hash_shift * MIN_MATCH >= hash_bits 140 */ 141 142 long block_start; 143 /* Window position at the beginning of the current output block. Gets 144 * negative when the window is moved backwards. 145 */ 146 147 uInt match_length; /* length of best match */ 148 IPos prev_match; /* previous match */ 149 int match_available; /* set if previous match exists */ 150 uInt strstart; /* start of string to insert */ 151 uInt match_start; /* start of matching string */ 152 uInt lookahead; /* number of valid bytes ahead in window */ 153 154 uInt prev_length; 155 /* Length of the best match at previous step. Matches not greater than this 156 * are discarded. This is used in the lazy match evaluation. 157 */ 158 159 uInt max_chain_length; 160 /* To speed up deflation, hash chains are never searched beyond this 161 * length. A higher limit improves compression ratio but degrades the 162 * speed. 163 */ 164 165 uInt max_lazy_match; 166 /* Attempt to find a better match only when the current match is strictly 167 * smaller than this value. This mechanism is used only for compression 168 * levels >= 4. 169 */ 170 # define max_insert_length max_lazy_match 171 /* Insert new strings in the hash table only if the match length is not 172 * greater than this length. This saves time but degrades compression. 173 * max_insert_length is used only for compression levels <= 3. 174 */ 175 176 int level; /* compression level (1..9) */ 177 int strategy; /* favor or force Huffman coding*/ 178 179 uInt good_match; 180 /* Use a faster search when the previous match is longer than this */ 181 182 int nice_match; /* Stop searching when current match exceeds this */ 183 184 /* used by trees.c: */ 185 /* Didn't use ct_data typedef below to suppress compiler warning */ 186 struct ct_data_s dyn_ltree[HEAP_SIZE]; /* literal and length tree */ 187 struct ct_data_s dyn_dtree[2*D_CODES+1]; /* distance tree */ 188 struct ct_data_s bl_tree[2*BL_CODES+1]; /* Huffman tree for bit lengths */ 189 190 struct tree_desc_s l_desc; /* desc. for literal tree */ 191 struct tree_desc_s d_desc; /* desc. for distance tree */ 192 struct tree_desc_s bl_desc; /* desc. for bit length tree */ 193 194 ush bl_count[MAX_BITS+1]; 195 /* number of codes at each bit length for an optimal tree */ 196 197 int heap[2*L_CODES+1]; /* heap used to build the Huffman trees */ 198 int heap_len; /* number of elements in the heap */ 199 int heap_max; /* element of largest frequency */ 200 /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used. 201 * The same heap array is used to build all trees. 202 */ 203 204 uch depth[2*L_CODES+1]; 205 /* Depth of each subtree used as tie breaker for trees of equal frequency 206 */ 207 208 uchf *l_buf; /* buffer for literals or lengths */ 209 210 uInt lit_bufsize; 211 /* Size of match buffer for literals/lengths. There are 4 reasons for 212 * limiting lit_bufsize to 64K: 213 * - frequencies can be kept in 16 bit counters 214 * - if compression is not successful for the first block, all input 215 * data is still in the window so we can still emit a stored block even 216 * when input comes from standard input. (This can also be done for 217 * all blocks if lit_bufsize is not greater than 32K.) 218 * - if compression is not successful for a file smaller than 64K, we can 219 * even emit a stored file instead of a stored block (saving 5 bytes). 220 * This is applicable only for zip (not gzip or zlib). 221 * - creating new Huffman trees less frequently may not provide fast 222 * adaptation to changes in the input data statistics. (Take for 223 * example a binary file with poorly compressible code followed by 224 * a highly compressible string table.) Smaller buffer sizes give 225 * fast adaptation but have of course the overhead of transmitting 226 * trees more frequently. 227 * - I can't count above 4 228 */ 229 230 uInt last_lit; /* running index in l_buf */ 231 232 ushf *d_buf; 233 /* Buffer for distances. To simplify the code, d_buf and l_buf have 234 * the same number of elements. To use different lengths, an extra flag 235 * array would be necessary. 236 */ 237 238 ulg opt_len; /* bit length of current block with optimal trees */ 239 ulg static_len; /* bit length of current block with static trees */ 240 uInt matches; /* number of string matches in current block */ 241 uInt insert; /* bytes at end of window left to insert */ 242 243 #ifdef DEBUG 244 ulg compressed_len; /* total bit length of compressed file mod 2^32 */ 245 ulg bits_sent; /* bit length of compressed data sent mod 2^32 */ 246 #endif 247 248 ush bi_buf; 249 /* Output buffer. bits are inserted starting at the bottom (least 250 * significant bits). 251 */ 252 int bi_valid; 253 /* Number of valid bits in bi_buf. All bits above the last valid bit 254 * are always zero. 255 */ 256 257 ulg high_water; 258 /* High water mark offset in window for initialized bytes -- bytes above 259 * this are set to zero in order to avoid memory check warnings when 260 * longest match routines access bytes past the input. This is then 261 * updated to the new high water mark. 262 */ 263 264 } FAR deflate_state; 265 266 /* Output a byte on the stream. 267 * IN assertion: there is enough room in pending_buf. 268 */ 269 #define put_byte(s, c) {s->pending_buf[s->pending++] = (c);} 270 271 272 #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1) 273 /* Minimum amount of lookahead, except at the end of the input file. 274 * See deflate.c for comments about the MIN_MATCH+1. 275 */ 276 277 #define MAX_DIST(s) ((s)->w_size-MIN_LOOKAHEAD) 278 /* In order to simplify the code, particularly on 16 bit machines, match 279 * distances are limited to MAX_DIST instead of WSIZE. 280 */ 281 282 #define WIN_INIT MAX_MATCH 283 /* Number of bytes after end of data in window to initialize in order to avoid 284 memory checker errors from longest match routines */ 285 286 /* in trees.c */ 287 void ZLIB_INTERNAL _tr_init(deflate_state *s); 288 int ZLIB_INTERNAL _tr_tally(deflate_state *s, unsigned dist, unsigned lc); 289 void ZLIB_INTERNAL _tr_flush_block(deflate_state *s, charf *buf, 290 ulg stored_len, int last); 291 void ZLIB_INTERNAL _tr_flush_bits(deflate_state *s); 292 void ZLIB_INTERNAL _tr_align(deflate_state *s); 293 void ZLIB_INTERNAL _tr_stored_block(deflate_state *s, charf *buf, 294 ulg stored_len, int last); 295 296 #define d_code(dist) \ 297 ((dist) < 256 ? _dist_code[dist] : _dist_code[256+((dist)>>7)]) 298 /* Mapping from a distance to a distance code. dist is the distance - 1 and 299 * must not have side effects. _dist_code[256] and _dist_code[257] are never 300 * used. 301 */ 302 303 #ifndef DEBUG 304 /* Inline versions of _tr_tally for speed: */ 305 306 #if defined(GEN_TREES_H) || !defined(STDC) 307 extern uch ZLIB_INTERNAL _length_code[]; 308 extern uch ZLIB_INTERNAL _dist_code[]; 309 #else 310 extern const uch ZLIB_INTERNAL _length_code[]; 311 extern const uch ZLIB_INTERNAL _dist_code[]; 312 #endif 313 314 # define _tr_tally_lit(s, c, flush) \ 315 { uch cc = (c); \ 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); \ 320 } 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; \ 326 dist--; \ 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); \ 330 } 331 #else 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) 335 #endif 336 337 #endif /* DEFLATE_H */ 338