deflate.h revision 146081
117651Speter/* deflate.h -- internal compression state 292114Sgreen * Copyright (C) 1995-2002 Jean-loup Gailly 3131380Stjr * For conditions of distribution and use, see copyright notice in zlib.h 417651Speter */ 517651Speter 617651Speter/* WARNING: this file should *not* be used by applications. It is 717651Speter part of the implementation of the compression library and is 817651Speter subject to change. Applications should only use zlib.h. 917651Speter */ 1017651Speter 11146081Skientzle/* @(#) $Id$ */ 1217651Speter 13131380Stjr#ifndef DEFLATE_H 14131380Stjr#define DEFLATE_H 1517651Speter 1617651Speter#include "zutil.h" 1717651Speter 18131380Stjr/* define NO_GZIP when compiling if you want to disable gzip header and 19131380Stjr trailer creation by deflate(). NO_GZIP would be used to avoid linking in 20131380Stjr the crc code when it is not needed. For shared libraries, gzip encoding 21131380Stjr should be left enabled. */ 22131380Stjr#ifndef NO_GZIP 23131380Stjr# define GZIP 24131380Stjr#endif 25131380Stjr 2617651Speter/* =========================================================================== 2717651Speter * Internal compression state. 2817651Speter */ 2917651Speter 3017651Speter#define LENGTH_CODES 29 3117651Speter/* number of length codes, not counting the special END_BLOCK code */ 3217651Speter 3317651Speter#define LITERALS 256 3417651Speter/* number of literal bytes 0..255 */ 3517651Speter 3617651Speter#define L_CODES (LITERALS+1+LENGTH_CODES) 3717651Speter/* number of Literal or Length codes, including the END_BLOCK code */ 3817651Speter 3917651Speter#define D_CODES 30 4017651Speter/* number of distance codes */ 4117651Speter 4217651Speter#define BL_CODES 19 4317651Speter/* number of codes used to transfer the bit lengths */ 4417651Speter 4517651Speter#define HEAP_SIZE (2*L_CODES+1) 4617651Speter/* maximum heap size */ 4717651Speter 4817651Speter#define MAX_BITS 15 4917651Speter/* All codes must not exceed MAX_BITS bits */ 5017651Speter 5117651Speter#define INIT_STATE 42 5217651Speter#define BUSY_STATE 113 5317651Speter#define FINISH_STATE 666 5417651Speter/* Stream status */ 5517651Speter 5617651Speter 5717651Speter/* Data structure describing a single value and its code string. */ 5817651Spetertypedef struct ct_data_s { 5917651Speter union { 6017651Speter ush freq; /* frequency count */ 6117651Speter ush code; /* bit string */ 6217651Speter } fc; 6317651Speter union { 6417651Speter ush dad; /* father node in Huffman tree */ 6517651Speter ush len; /* length of bit string */ 6617651Speter } dl; 6717651Speter} FAR ct_data; 6817651Speter 6917651Speter#define Freq fc.freq 7017651Speter#define Code fc.code 7117651Speter#define Dad dl.dad 7217651Speter#define Len dl.len 7317651Speter 7417651Spetertypedef struct static_tree_desc_s static_tree_desc; 7517651Speter 7617651Spetertypedef struct tree_desc_s { 7717651Speter ct_data *dyn_tree; /* the dynamic tree */ 7817651Speter int max_code; /* largest code with non zero frequency */ 7917651Speter static_tree_desc *stat_desc; /* the corresponding static tree */ 8017651Speter} FAR tree_desc; 8117651Speter 8217651Spetertypedef ush Pos; 8317651Spetertypedef Pos FAR Posf; 8417651Spetertypedef unsigned IPos; 8517651Speter 8617651Speter/* A Pos is an index in the character window. We use short instead of int to 8717651Speter * save space in the various tables. IPos is used only for parameter passing. 8817651Speter */ 8917651Speter 9017651Spetertypedef struct internal_state { 9117651Speter z_streamp strm; /* pointer back to this zlib stream */ 9217651Speter int status; /* as the name implies */ 9317651Speter Bytef *pending_buf; /* output still pending */ 9433908Ssteve ulg pending_buf_size; /* size of pending_buf */ 9517651Speter Bytef *pending_out; /* next pending byte to output to the stream */ 9617651Speter int pending; /* nb of bytes in the pending buffer */ 97131380Stjr int wrap; /* bit 0 true for zlib, bit 1 true for gzip */ 9817651Speter Byte method; /* STORED (for zip only) or DEFLATED */ 9917651Speter int last_flush; /* value of flush param for previous deflate call */ 10017651Speter 10117651Speter /* used by deflate.c: */ 10217651Speter 10317651Speter uInt w_size; /* LZ77 window size (32K by default) */ 10417651Speter uInt w_bits; /* log2(w_size) (8..16) */ 10517651Speter uInt w_mask; /* w_size - 1 */ 10617651Speter 10717651Speter Bytef *window; 10817651Speter /* Sliding window. Input bytes are read into the second half of the window, 10917651Speter * and move to the first half later to keep a dictionary of at least wSize 11017651Speter * bytes. With this organization, matches are limited to a distance of 11117651Speter * wSize-MAX_MATCH bytes, but this ensures that IO is always 11217651Speter * performed with a length multiple of the block size. Also, it limits 11317651Speter * the window size to 64K, which is quite useful on MSDOS. 11417651Speter * To do: use the user input buffer as sliding window. 11517651Speter */ 11617651Speter 11717651Speter ulg window_size; 11817651Speter /* Actual size of window: 2*wSize, except when the user input buffer 11917651Speter * is directly used as sliding window. 12017651Speter */ 12117651Speter 12217651Speter Posf *prev; 12317651Speter /* Link to older string with same hash index. To limit the size of this 12417651Speter * array to 64K, this link is maintained only for the last 32K strings. 12517651Speter * An index in this array is thus a window index modulo 32K. 12617651Speter */ 12717651Speter 12817651Speter Posf *head; /* Heads of the hash chains or NIL. */ 12917651Speter 13017651Speter uInt ins_h; /* hash index of string to be inserted */ 13117651Speter uInt hash_size; /* number of elements in hash table */ 13217651Speter uInt hash_bits; /* log2(hash_size) */ 13317651Speter uInt hash_mask; /* hash_size-1 */ 13417651Speter 13517651Speter uInt hash_shift; 13617651Speter /* Number of bits by which ins_h must be shifted at each input 13717651Speter * step. It must be such that after MIN_MATCH steps, the oldest 13817651Speter * byte no longer takes part in the hash key, that is: 13917651Speter * hash_shift * MIN_MATCH >= hash_bits 14017651Speter */ 14117651Speter 14217651Speter long block_start; 14317651Speter /* Window position at the beginning of the current output block. Gets 14417651Speter * negative when the window is moved backwards. 14517651Speter */ 14617651Speter 14717651Speter uInt match_length; /* length of best match */ 14817651Speter IPos prev_match; /* previous match */ 14917651Speter int match_available; /* set if previous match exists */ 15017651Speter uInt strstart; /* start of string to insert */ 15117651Speter uInt match_start; /* start of matching string */ 15217651Speter uInt lookahead; /* number of valid bytes ahead in window */ 15317651Speter 15417651Speter uInt prev_length; 15517651Speter /* Length of the best match at previous step. Matches not greater than this 15617651Speter * are discarded. This is used in the lazy match evaluation. 15717651Speter */ 15817651Speter 15917651Speter uInt max_chain_length; 16017651Speter /* To speed up deflation, hash chains are never searched beyond this 16117651Speter * length. A higher limit improves compression ratio but degrades the 16217651Speter * speed. 16317651Speter */ 16417651Speter 16517651Speter uInt max_lazy_match; 16617651Speter /* Attempt to find a better match only when the current match is strictly 16717651Speter * smaller than this value. This mechanism is used only for compression 16817651Speter * levels >= 4. 16917651Speter */ 17017651Speter# define max_insert_length max_lazy_match 17117651Speter /* Insert new strings in the hash table only if the match length is not 17217651Speter * greater than this length. This saves time but degrades compression. 17317651Speter * max_insert_length is used only for compression levels <= 3. 17417651Speter */ 17517651Speter 17617651Speter int level; /* compression level (1..9) */ 17717651Speter int strategy; /* favor or force Huffman coding*/ 17817651Speter 17917651Speter uInt good_match; 18017651Speter /* Use a faster search when the previous match is longer than this */ 18117651Speter 18217651Speter int nice_match; /* Stop searching when current match exceeds this */ 18317651Speter 18417651Speter /* used by trees.c: */ 18517651Speter /* Didn't use ct_data typedef below to supress compiler warning */ 18617651Speter struct ct_data_s dyn_ltree[HEAP_SIZE]; /* literal and length tree */ 18717651Speter struct ct_data_s dyn_dtree[2*D_CODES+1]; /* distance tree */ 18817651Speter struct ct_data_s bl_tree[2*BL_CODES+1]; /* Huffman tree for bit lengths */ 18917651Speter 19017651Speter struct tree_desc_s l_desc; /* desc. for literal tree */ 19117651Speter struct tree_desc_s d_desc; /* desc. for distance tree */ 19217651Speter struct tree_desc_s bl_desc; /* desc. for bit length tree */ 19317651Speter 19417651Speter ush bl_count[MAX_BITS+1]; 19517651Speter /* number of codes at each bit length for an optimal tree */ 19617651Speter 19717651Speter int heap[2*L_CODES+1]; /* heap used to build the Huffman trees */ 19817651Speter int heap_len; /* number of elements in the heap */ 19917651Speter int heap_max; /* element of largest frequency */ 20017651Speter /* The sons of heap[n] are heap[2*n] and heap[2*n+1]. heap[0] is not used. 20117651Speter * The same heap array is used to build all trees. 20217651Speter */ 20317651Speter 20417651Speter uch depth[2*L_CODES+1]; 20517651Speter /* Depth of each subtree used as tie breaker for trees of equal frequency 20617651Speter */ 20717651Speter 20817651Speter uchf *l_buf; /* buffer for literals or lengths */ 20917651Speter 21017651Speter uInt lit_bufsize; 21117651Speter /* Size of match buffer for literals/lengths. There are 4 reasons for 21217651Speter * limiting lit_bufsize to 64K: 21317651Speter * - frequencies can be kept in 16 bit counters 21417651Speter * - if compression is not successful for the first block, all input 21517651Speter * data is still in the window so we can still emit a stored block even 21617651Speter * when input comes from standard input. (This can also be done for 21717651Speter * all blocks if lit_bufsize is not greater than 32K.) 21817651Speter * - if compression is not successful for a file smaller than 64K, we can 21917651Speter * even emit a stored file instead of a stored block (saving 5 bytes). 22017651Speter * This is applicable only for zip (not gzip or zlib). 22117651Speter * - creating new Huffman trees less frequently may not provide fast 22217651Speter * adaptation to changes in the input data statistics. (Take for 22317651Speter * example a binary file with poorly compressible code followed by 22417651Speter * a highly compressible string table.) Smaller buffer sizes give 22517651Speter * fast adaptation but have of course the overhead of transmitting 22617651Speter * trees more frequently. 22717651Speter * - I can't count above 4 22817651Speter */ 22917651Speter 23017651Speter uInt last_lit; /* running index in l_buf */ 23117651Speter 23217651Speter ushf *d_buf; 23317651Speter /* Buffer for distances. To simplify the code, d_buf and l_buf have 23417651Speter * the same number of elements. To use different lengths, an extra flag 23517651Speter * array would be necessary. 23617651Speter */ 23717651Speter 23817651Speter ulg opt_len; /* bit length of current block with optimal trees */ 23917651Speter ulg static_len; /* bit length of current block with static trees */ 24017651Speter uInt matches; /* number of string matches in current block */ 24117651Speter int last_eob_len; /* bit length of EOB code for last block */ 24217651Speter 24317651Speter#ifdef DEBUG 24442471Speter ulg compressed_len; /* total bit length of compressed file mod 2^32 */ 24542471Speter ulg bits_sent; /* bit length of compressed data sent mod 2^32 */ 24617651Speter#endif 24717651Speter 24817651Speter ush bi_buf; 24917651Speter /* Output buffer. bits are inserted starting at the bottom (least 25017651Speter * significant bits). 25117651Speter */ 25217651Speter int bi_valid; 25317651Speter /* Number of valid bits in bi_buf. All bits above the last valid bit 25417651Speter * are always zero. 25517651Speter */ 25617651Speter 25717651Speter} FAR deflate_state; 25817651Speter 25917651Speter/* Output a byte on the stream. 26017651Speter * IN assertion: there is enough room in pending_buf. 26117651Speter */ 26217651Speter#define put_byte(s, c) {s->pending_buf[s->pending++] = (c);} 26317651Speter 26417651Speter 26517651Speter#define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1) 26617651Speter/* Minimum amount of lookahead, except at the end of the input file. 26717651Speter * See deflate.c for comments about the MIN_MATCH+1. 26817651Speter */ 26917651Speter 27017651Speter#define MAX_DIST(s) ((s)->w_size-MIN_LOOKAHEAD) 27117651Speter/* In order to simplify the code, particularly on 16 bit machines, match 27217651Speter * distances are limited to MAX_DIST instead of WSIZE. 27317651Speter */ 27417651Speter 27517651Speter /* in trees.c */ 27617651Spetervoid _tr_init OF((deflate_state *s)); 27717651Speterint _tr_tally OF((deflate_state *s, unsigned dist, unsigned lc)); 27842471Spetervoid _tr_flush_block OF((deflate_state *s, charf *buf, ulg stored_len, 279131380Stjr int eof)); 28017651Spetervoid _tr_align OF((deflate_state *s)); 28117651Spetervoid _tr_stored_block OF((deflate_state *s, charf *buf, ulg stored_len, 28217651Speter int eof)); 28333908Ssteve 28433908Ssteve#define d_code(dist) \ 28533908Ssteve ((dist) < 256 ? _dist_code[dist] : _dist_code[256+((dist)>>7)]) 28633908Ssteve/* Mapping from a distance to a distance code. dist is the distance - 1 and 28733908Ssteve * must not have side effects. _dist_code[256] and _dist_code[257] are never 28833908Ssteve * used. 28933908Ssteve */ 29033908Ssteve 29133908Ssteve#ifndef DEBUG 29233908Ssteve/* Inline versions of _tr_tally for speed: */ 29333908Ssteve 29433908Ssteve#if defined(GEN_TREES_H) || !defined(STDC) 29533908Ssteve extern uch _length_code[]; 29633908Ssteve extern uch _dist_code[]; 29733908Ssteve#else 29833908Ssteve extern const uch _length_code[]; 29933908Ssteve extern const uch _dist_code[]; 30017651Speter#endif 30133908Ssteve 30233908Ssteve# define _tr_tally_lit(s, c, flush) \ 30333908Ssteve { uch cc = (c); \ 30433908Ssteve s->d_buf[s->last_lit] = 0; \ 30533908Ssteve s->l_buf[s->last_lit++] = cc; \ 30633908Ssteve s->dyn_ltree[cc].Freq++; \ 30733908Ssteve flush = (s->last_lit == s->lit_bufsize-1); \ 30833908Ssteve } 30933908Ssteve# define _tr_tally_dist(s, distance, length, flush) \ 31033908Ssteve { uch len = (length); \ 31133908Ssteve ush dist = (distance); \ 31233908Ssteve s->d_buf[s->last_lit] = dist; \ 31333908Ssteve s->l_buf[s->last_lit++] = len; \ 31433908Ssteve dist--; \ 31533908Ssteve s->dyn_ltree[_length_code[len]+LITERALS+1].Freq++; \ 31633908Ssteve s->dyn_dtree[d_code(dist)].Freq++; \ 31733908Ssteve flush = (s->last_lit == s->lit_bufsize-1); \ 31833908Ssteve } 31933908Ssteve#else 32033908Ssteve# define _tr_tally_lit(s, c, flush) flush = _tr_tally(s, 0, c) 32133908Ssteve# define _tr_tally_dist(s, distance, length, flush) \ 322131380Stjr flush = _tr_tally(s, distance, length) 32333908Ssteve#endif 32433908Ssteve 325131380Stjr#endif /* DEFLATE_H */ 326