michael@0: /* deflate.c -- compress data using the deflation algorithm michael@0: * Copyright (C) 1995-2010 Jean-loup Gailly and Mark Adler michael@0: * For conditions of distribution and use, see copyright notice in zlib.h michael@0: */ michael@0: michael@0: /* michael@0: * ALGORITHM michael@0: * michael@0: * The "deflation" process depends on being able to identify portions michael@0: * of the input text which are identical to earlier input (within a michael@0: * sliding window trailing behind the input currently being processed). michael@0: * michael@0: * The most straightforward technique turns out to be the fastest for michael@0: * most input files: try all possible matches and select the longest. michael@0: * The key feature of this algorithm is that insertions into the string michael@0: * dictionary are very simple and thus fast, and deletions are avoided michael@0: * completely. Insertions are performed at each input character, whereas michael@0: * string matches are performed only when the previous match ends. So it michael@0: * is preferable to spend more time in matches to allow very fast string michael@0: * insertions and avoid deletions. The matching algorithm for small michael@0: * strings is inspired from that of Rabin & Karp. A brute force approach michael@0: * is used to find longer strings when a small match has been found. michael@0: * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze michael@0: * (by Leonid Broukhis). michael@0: * A previous version of this file used a more sophisticated algorithm michael@0: * (by Fiala and Greene) which is guaranteed to run in linear amortized michael@0: * time, but has a larger average cost, uses more memory and is patented. michael@0: * However the F&G algorithm may be faster for some highly redundant michael@0: * files if the parameter max_chain_length (described below) is too large. michael@0: * michael@0: * ACKNOWLEDGEMENTS michael@0: * michael@0: * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and michael@0: * I found it in 'freeze' written by Leonid Broukhis. michael@0: * Thanks to many people for bug reports and testing. michael@0: * michael@0: * REFERENCES michael@0: * michael@0: * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification". michael@0: * Available in http://www.ietf.org/rfc/rfc1951.txt michael@0: * michael@0: * A description of the Rabin and Karp algorithm is given in the book michael@0: * "Algorithms" by R. Sedgewick, Addison-Wesley, p252. michael@0: * michael@0: * Fiala,E.R., and Greene,D.H. michael@0: * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595 michael@0: * michael@0: */ michael@0: michael@0: /* @(#) $Id$ */ michael@0: michael@0: #include "deflate.h" michael@0: michael@0: const char deflate_copyright[] = michael@0: " deflate 1.2.5 Copyright 1995-2010 Jean-loup Gailly and Mark Adler "; michael@0: /* michael@0: If you use the zlib library in a product, an acknowledgment is welcome michael@0: in the documentation of your product. If for some reason you cannot michael@0: include such an acknowledgment, I would appreciate that you keep this michael@0: copyright string in the executable of your product. michael@0: */ michael@0: michael@0: /* =========================================================================== michael@0: * Function prototypes. michael@0: */ michael@0: typedef enum { michael@0: need_more, /* block not completed, need more input or more output */ michael@0: block_done, /* block flush performed */ michael@0: finish_started, /* finish started, need only more output at next deflate */ michael@0: finish_done /* finish done, accept no more input or output */ michael@0: } block_state; michael@0: michael@0: typedef block_state (*compress_func) OF((deflate_state *s, int flush)); michael@0: /* Compression function. Returns the block state after the call. */ michael@0: michael@0: local void fill_window OF((deflate_state *s)); michael@0: local block_state deflate_stored OF((deflate_state *s, int flush)); michael@0: local block_state deflate_fast OF((deflate_state *s, int flush)); michael@0: #ifndef FASTEST michael@0: local block_state deflate_slow OF((deflate_state *s, int flush)); michael@0: #endif michael@0: local block_state deflate_rle OF((deflate_state *s, int flush)); michael@0: local block_state deflate_huff OF((deflate_state *s, int flush)); michael@0: local void lm_init OF((deflate_state *s)); michael@0: local void putShortMSB OF((deflate_state *s, uInt b)); michael@0: local void flush_pending OF((z_streamp strm)); michael@0: local int read_buf OF((z_streamp strm, Bytef *buf, unsigned size)); michael@0: #ifdef ASMV michael@0: void match_init OF((void)); /* asm code initialization */ michael@0: uInt longest_match OF((deflate_state *s, IPos cur_match)); michael@0: #else michael@0: local uInt longest_match OF((deflate_state *s, IPos cur_match)); michael@0: #endif michael@0: michael@0: #ifdef DEBUG michael@0: local void check_match OF((deflate_state *s, IPos start, IPos match, michael@0: int length)); michael@0: #endif michael@0: michael@0: /* =========================================================================== michael@0: * Local data michael@0: */ michael@0: michael@0: #define NIL 0 michael@0: /* Tail of hash chains */ michael@0: michael@0: #ifndef TOO_FAR michael@0: # define TOO_FAR 4096 michael@0: #endif michael@0: /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */ michael@0: michael@0: /* Values for max_lazy_match, good_match and max_chain_length, depending on michael@0: * the desired pack level (0..9). The values given below have been tuned to michael@0: * exclude worst case performance for pathological files. Better values may be michael@0: * found for specific files. michael@0: */ michael@0: typedef struct config_s { michael@0: ush good_length; /* reduce lazy search above this match length */ michael@0: ush max_lazy; /* do not perform lazy search above this match length */ michael@0: ush nice_length; /* quit search above this match length */ michael@0: ush max_chain; michael@0: compress_func func; michael@0: } config; michael@0: michael@0: #ifdef FASTEST michael@0: local const config configuration_table[2] = { michael@0: /* good lazy nice chain */ michael@0: /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */ michael@0: /* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */ michael@0: #else michael@0: local const config configuration_table[10] = { michael@0: /* good lazy nice chain */ michael@0: /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */ michael@0: /* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */ michael@0: /* 2 */ {4, 5, 16, 8, deflate_fast}, michael@0: /* 3 */ {4, 6, 32, 32, deflate_fast}, michael@0: michael@0: /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */ michael@0: /* 5 */ {8, 16, 32, 32, deflate_slow}, michael@0: /* 6 */ {8, 16, 128, 128, deflate_slow}, michael@0: /* 7 */ {8, 32, 128, 256, deflate_slow}, michael@0: /* 8 */ {32, 128, 258, 1024, deflate_slow}, michael@0: /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */ michael@0: #endif michael@0: michael@0: /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4 michael@0: * For deflate_fast() (levels <= 3) good is ignored and lazy has a different michael@0: * meaning. michael@0: */ michael@0: michael@0: #define EQUAL 0 michael@0: /* result of memcmp for equal strings */ michael@0: michael@0: #ifndef NO_DUMMY_DECL michael@0: struct static_tree_desc_s {int dummy;}; /* for buggy compilers */ michael@0: #endif michael@0: michael@0: /* =========================================================================== michael@0: * Update a hash value with the given input byte michael@0: * IN assertion: all calls to to UPDATE_HASH are made with consecutive michael@0: * input characters, so that a running hash key can be computed from the michael@0: * previous key instead of complete recalculation each time. michael@0: */ michael@0: #define UPDATE_HASH(s,h,c) (h = (((h)<hash_shift) ^ (c)) & s->hash_mask) michael@0: michael@0: michael@0: /* =========================================================================== michael@0: * Insert string str in the dictionary and set match_head to the previous head michael@0: * of the hash chain (the most recent string with same hash key). Return michael@0: * the previous length of the hash chain. michael@0: * If this file is compiled with -DFASTEST, the compression level is forced michael@0: * to 1, and no hash chains are maintained. michael@0: * IN assertion: all calls to to INSERT_STRING are made with consecutive michael@0: * input characters and the first MIN_MATCH bytes of str are valid michael@0: * (except for the last MIN_MATCH-1 bytes of the input file). michael@0: */ michael@0: #ifdef FASTEST michael@0: #define INSERT_STRING(s, str, match_head) \ michael@0: (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \ michael@0: match_head = s->head[s->ins_h], \ michael@0: s->head[s->ins_h] = (Pos)(str)) michael@0: #else michael@0: #define INSERT_STRING(s, str, match_head) \ michael@0: (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \ michael@0: match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \ michael@0: s->head[s->ins_h] = (Pos)(str)) michael@0: #endif michael@0: michael@0: /* =========================================================================== michael@0: * Initialize the hash table (avoiding 64K overflow for 16 bit systems). michael@0: * prev[] will be initialized on the fly. michael@0: */ michael@0: #define CLEAR_HASH(s) \ michael@0: s->head[s->hash_size-1] = NIL; \ michael@0: zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head)); michael@0: michael@0: /* ========================================================================= */ michael@0: int ZEXPORT deflateInit_(strm, level, version, stream_size) michael@0: z_streamp strm; michael@0: int level; michael@0: const char *version; michael@0: int stream_size; michael@0: { michael@0: return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL, michael@0: Z_DEFAULT_STRATEGY, version, stream_size); michael@0: /* To do: ignore strm->next_in if we use it as window */ michael@0: } michael@0: michael@0: /* ========================================================================= */ michael@0: int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy, michael@0: version, stream_size) michael@0: z_streamp strm; michael@0: int level; michael@0: int method; michael@0: int windowBits; michael@0: int memLevel; michael@0: int strategy; michael@0: const char *version; michael@0: int stream_size; michael@0: { michael@0: deflate_state *s; michael@0: int wrap = 1; michael@0: static const char my_version[] = ZLIB_VERSION; michael@0: michael@0: ushf *overlay; michael@0: /* We overlay pending_buf and d_buf+l_buf. This works since the average michael@0: * output size for (length,distance) codes is <= 24 bits. michael@0: */ michael@0: michael@0: if (version == Z_NULL || version[0] != my_version[0] || michael@0: stream_size != sizeof(z_stream)) { michael@0: return Z_VERSION_ERROR; michael@0: } michael@0: if (strm == Z_NULL) return Z_STREAM_ERROR; michael@0: michael@0: strm->msg = Z_NULL; michael@0: if (strm->zalloc == (alloc_func)0) { michael@0: strm->zalloc = zcalloc; michael@0: strm->opaque = (voidpf)0; michael@0: } michael@0: if (strm->zfree == (free_func)0) strm->zfree = zcfree; michael@0: michael@0: #ifdef FASTEST michael@0: if (level != 0) level = 1; michael@0: #else michael@0: if (level == Z_DEFAULT_COMPRESSION) level = 6; michael@0: #endif michael@0: michael@0: if (windowBits < 0) { /* suppress zlib wrapper */ michael@0: wrap = 0; michael@0: windowBits = -windowBits; michael@0: } michael@0: #ifdef GZIP michael@0: else if (windowBits > 15) { michael@0: wrap = 2; /* write gzip wrapper instead */ michael@0: windowBits -= 16; michael@0: } michael@0: #endif michael@0: if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED || michael@0: windowBits < 8 || windowBits > 15 || level < 0 || level > 9 || michael@0: strategy < 0 || strategy > Z_FIXED) { michael@0: return Z_STREAM_ERROR; michael@0: } michael@0: if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */ michael@0: s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state)); michael@0: if (s == Z_NULL) return Z_MEM_ERROR; michael@0: strm->state = (struct internal_state FAR *)s; michael@0: s->strm = strm; michael@0: michael@0: s->wrap = wrap; michael@0: s->gzhead = Z_NULL; michael@0: s->w_bits = windowBits; michael@0: s->w_size = 1 << s->w_bits; michael@0: s->w_mask = s->w_size - 1; michael@0: michael@0: s->hash_bits = memLevel + 7; michael@0: s->hash_size = 1 << s->hash_bits; michael@0: s->hash_mask = s->hash_size - 1; michael@0: s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH); michael@0: michael@0: s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte)); michael@0: s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos)); michael@0: s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos)); michael@0: michael@0: s->high_water = 0; /* nothing written to s->window yet */ michael@0: michael@0: s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */ michael@0: michael@0: overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2); michael@0: s->pending_buf = (uchf *) overlay; michael@0: s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L); michael@0: michael@0: if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL || michael@0: s->pending_buf == Z_NULL) { michael@0: s->status = FINISH_STATE; michael@0: strm->msg = (char*)ERR_MSG(Z_MEM_ERROR); michael@0: deflateEnd (strm); michael@0: return Z_MEM_ERROR; michael@0: } michael@0: s->d_buf = overlay + s->lit_bufsize/sizeof(ush); michael@0: s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize; michael@0: michael@0: s->level = level; michael@0: s->strategy = strategy; michael@0: s->method = (Byte)method; michael@0: michael@0: return deflateReset(strm); michael@0: } michael@0: michael@0: /* ========================================================================= */ michael@0: int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength) michael@0: z_streamp strm; michael@0: const Bytef *dictionary; michael@0: uInt dictLength; michael@0: { michael@0: deflate_state *s; michael@0: uInt length = dictLength; michael@0: uInt n; michael@0: IPos hash_head = 0; michael@0: michael@0: if (strm == Z_NULL || strm->state == Z_NULL || dictionary == Z_NULL || michael@0: strm->state->wrap == 2 || michael@0: (strm->state->wrap == 1 && strm->state->status != INIT_STATE)) michael@0: return Z_STREAM_ERROR; michael@0: michael@0: s = strm->state; michael@0: if (s->wrap) michael@0: strm->adler = adler32(strm->adler, dictionary, dictLength); michael@0: michael@0: if (length < MIN_MATCH) return Z_OK; michael@0: if (length > s->w_size) { michael@0: length = s->w_size; michael@0: dictionary += dictLength - length; /* use the tail of the dictionary */ michael@0: } michael@0: zmemcpy(s->window, dictionary, length); michael@0: s->strstart = length; michael@0: s->block_start = (long)length; michael@0: michael@0: /* Insert all strings in the hash table (except for the last two bytes). michael@0: * s->lookahead stays null, so s->ins_h will be recomputed at the next michael@0: * call of fill_window. michael@0: */ michael@0: s->ins_h = s->window[0]; michael@0: UPDATE_HASH(s, s->ins_h, s->window[1]); michael@0: for (n = 0; n <= length - MIN_MATCH; n++) { michael@0: INSERT_STRING(s, n, hash_head); michael@0: } michael@0: if (hash_head) hash_head = 0; /* to make compiler happy */ michael@0: return Z_OK; michael@0: } michael@0: michael@0: /* ========================================================================= */ michael@0: int ZEXPORT deflateReset (strm) michael@0: z_streamp strm; michael@0: { michael@0: deflate_state *s; michael@0: michael@0: if (strm == Z_NULL || strm->state == Z_NULL || michael@0: strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0) { michael@0: return Z_STREAM_ERROR; michael@0: } michael@0: michael@0: strm->total_in = strm->total_out = 0; michael@0: strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */ michael@0: strm->data_type = Z_UNKNOWN; michael@0: michael@0: s = (deflate_state *)strm->state; michael@0: s->pending = 0; michael@0: s->pending_out = s->pending_buf; michael@0: michael@0: if (s->wrap < 0) { michael@0: s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */ michael@0: } michael@0: s->status = s->wrap ? INIT_STATE : BUSY_STATE; michael@0: strm->adler = michael@0: #ifdef GZIP michael@0: s->wrap == 2 ? crc32(0L, Z_NULL, 0) : michael@0: #endif michael@0: adler32(0L, Z_NULL, 0); michael@0: s->last_flush = Z_NO_FLUSH; michael@0: michael@0: _tr_init(s); michael@0: lm_init(s); michael@0: michael@0: return Z_OK; michael@0: } michael@0: michael@0: /* ========================================================================= */ michael@0: int ZEXPORT deflateSetHeader (strm, head) michael@0: z_streamp strm; michael@0: gz_headerp head; michael@0: { michael@0: if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; michael@0: if (strm->state->wrap != 2) return Z_STREAM_ERROR; michael@0: strm->state->gzhead = head; michael@0: return Z_OK; michael@0: } michael@0: michael@0: /* ========================================================================= */ michael@0: int ZEXPORT deflatePrime (strm, bits, value) michael@0: z_streamp strm; michael@0: int bits; michael@0: int value; michael@0: { michael@0: if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; michael@0: strm->state->bi_valid = bits; michael@0: strm->state->bi_buf = (ush)(value & ((1 << bits) - 1)); michael@0: return Z_OK; michael@0: } michael@0: michael@0: /* ========================================================================= */ michael@0: int ZEXPORT deflateParams(strm, level, strategy) michael@0: z_streamp strm; michael@0: int level; michael@0: int strategy; michael@0: { michael@0: deflate_state *s; michael@0: compress_func func; michael@0: int err = Z_OK; michael@0: michael@0: if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; michael@0: s = strm->state; michael@0: michael@0: #ifdef FASTEST michael@0: if (level != 0) level = 1; michael@0: #else michael@0: if (level == Z_DEFAULT_COMPRESSION) level = 6; michael@0: #endif michael@0: if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) { michael@0: return Z_STREAM_ERROR; michael@0: } michael@0: func = configuration_table[s->level].func; michael@0: michael@0: if ((strategy != s->strategy || func != configuration_table[level].func) && michael@0: strm->total_in != 0) { michael@0: /* Flush the last buffer: */ michael@0: err = deflate(strm, Z_BLOCK); michael@0: } michael@0: if (s->level != level) { michael@0: s->level = level; michael@0: s->max_lazy_match = configuration_table[level].max_lazy; michael@0: s->good_match = configuration_table[level].good_length; michael@0: s->nice_match = configuration_table[level].nice_length; michael@0: s->max_chain_length = configuration_table[level].max_chain; michael@0: } michael@0: s->strategy = strategy; michael@0: return err; michael@0: } michael@0: michael@0: /* ========================================================================= */ michael@0: int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain) michael@0: z_streamp strm; michael@0: int good_length; michael@0: int max_lazy; michael@0: int nice_length; michael@0: int max_chain; michael@0: { michael@0: deflate_state *s; michael@0: michael@0: if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; michael@0: s = strm->state; michael@0: s->good_match = good_length; michael@0: s->max_lazy_match = max_lazy; michael@0: s->nice_match = nice_length; michael@0: s->max_chain_length = max_chain; michael@0: return Z_OK; michael@0: } michael@0: michael@0: /* ========================================================================= michael@0: * For the default windowBits of 15 and memLevel of 8, this function returns michael@0: * a close to exact, as well as small, upper bound on the compressed size. michael@0: * They are coded as constants here for a reason--if the #define's are michael@0: * changed, then this function needs to be changed as well. The return michael@0: * value for 15 and 8 only works for those exact settings. michael@0: * michael@0: * For any setting other than those defaults for windowBits and memLevel, michael@0: * the value returned is a conservative worst case for the maximum expansion michael@0: * resulting from using fixed blocks instead of stored blocks, which deflate michael@0: * can emit on compressed data for some combinations of the parameters. michael@0: * michael@0: * This function could be more sophisticated to provide closer upper bounds for michael@0: * every combination of windowBits and memLevel. But even the conservative michael@0: * upper bound of about 14% expansion does not seem onerous for output buffer michael@0: * allocation. michael@0: */ michael@0: uLong ZEXPORT deflateBound(strm, sourceLen) michael@0: z_streamp strm; michael@0: uLong sourceLen; michael@0: { michael@0: deflate_state *s; michael@0: uLong complen, wraplen; michael@0: Bytef *str; michael@0: michael@0: /* conservative upper bound for compressed data */ michael@0: complen = sourceLen + michael@0: ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5; michael@0: michael@0: /* if can't get parameters, return conservative bound plus zlib wrapper */ michael@0: if (strm == Z_NULL || strm->state == Z_NULL) michael@0: return complen + 6; michael@0: michael@0: /* compute wrapper length */ michael@0: s = strm->state; michael@0: switch (s->wrap) { michael@0: case 0: /* raw deflate */ michael@0: wraplen = 0; michael@0: break; michael@0: case 1: /* zlib wrapper */ michael@0: wraplen = 6 + (s->strstart ? 4 : 0); michael@0: break; michael@0: case 2: /* gzip wrapper */ michael@0: wraplen = 18; michael@0: if (s->gzhead != Z_NULL) { /* user-supplied gzip header */ michael@0: if (s->gzhead->extra != Z_NULL) michael@0: wraplen += 2 + s->gzhead->extra_len; michael@0: str = s->gzhead->name; michael@0: if (str != Z_NULL) michael@0: do { michael@0: wraplen++; michael@0: } while (*str++); michael@0: str = s->gzhead->comment; michael@0: if (str != Z_NULL) michael@0: do { michael@0: wraplen++; michael@0: } while (*str++); michael@0: if (s->gzhead->hcrc) michael@0: wraplen += 2; michael@0: } michael@0: break; michael@0: default: /* for compiler happiness */ michael@0: wraplen = 6; michael@0: } michael@0: michael@0: /* if not default parameters, return conservative bound */ michael@0: if (s->w_bits != 15 || s->hash_bits != 8 + 7) michael@0: return complen + wraplen; michael@0: michael@0: /* default settings: return tight bound for that case */ michael@0: return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) + michael@0: (sourceLen >> 25) + 13 - 6 + wraplen; michael@0: } michael@0: michael@0: /* ========================================================================= michael@0: * Put a short in the pending buffer. The 16-bit value is put in MSB order. michael@0: * IN assertion: the stream state is correct and there is enough room in michael@0: * pending_buf. michael@0: */ michael@0: local void putShortMSB (s, b) michael@0: deflate_state *s; michael@0: uInt b; michael@0: { michael@0: put_byte(s, (Byte)(b >> 8)); michael@0: put_byte(s, (Byte)(b & 0xff)); michael@0: } michael@0: michael@0: /* ========================================================================= michael@0: * Flush as much pending output as possible. All deflate() output goes michael@0: * through this function so some applications may wish to modify it michael@0: * to avoid allocating a large strm->next_out buffer and copying into it. michael@0: * (See also read_buf()). michael@0: */ michael@0: local void flush_pending(strm) michael@0: z_streamp strm; michael@0: { michael@0: unsigned len = strm->state->pending; michael@0: michael@0: if (len > strm->avail_out) len = strm->avail_out; michael@0: if (len == 0) return; michael@0: michael@0: zmemcpy(strm->next_out, strm->state->pending_out, len); michael@0: strm->next_out += len; michael@0: strm->state->pending_out += len; michael@0: strm->total_out += len; michael@0: strm->avail_out -= len; michael@0: strm->state->pending -= len; michael@0: if (strm->state->pending == 0) { michael@0: strm->state->pending_out = strm->state->pending_buf; michael@0: } michael@0: } michael@0: michael@0: /* ========================================================================= */ michael@0: int ZEXPORT deflate (strm, flush) michael@0: z_streamp strm; michael@0: int flush; michael@0: { michael@0: int old_flush; /* value of flush param for previous deflate call */ michael@0: deflate_state *s; michael@0: michael@0: if (strm == Z_NULL || strm->state == Z_NULL || michael@0: flush > Z_BLOCK || flush < 0) { michael@0: return Z_STREAM_ERROR; michael@0: } michael@0: s = strm->state; michael@0: michael@0: if (strm->next_out == Z_NULL || michael@0: (strm->next_in == Z_NULL && strm->avail_in != 0) || michael@0: (s->status == FINISH_STATE && flush != Z_FINISH)) { michael@0: ERR_RETURN(strm, Z_STREAM_ERROR); michael@0: } michael@0: if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR); michael@0: michael@0: s->strm = strm; /* just in case */ michael@0: old_flush = s->last_flush; michael@0: s->last_flush = flush; michael@0: michael@0: /* Write the header */ michael@0: if (s->status == INIT_STATE) { michael@0: #ifdef GZIP michael@0: if (s->wrap == 2) { michael@0: strm->adler = crc32(0L, Z_NULL, 0); michael@0: put_byte(s, 31); michael@0: put_byte(s, 139); michael@0: put_byte(s, 8); michael@0: if (s->gzhead == Z_NULL) { michael@0: put_byte(s, 0); michael@0: put_byte(s, 0); michael@0: put_byte(s, 0); michael@0: put_byte(s, 0); michael@0: put_byte(s, 0); michael@0: put_byte(s, s->level == 9 ? 2 : michael@0: (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ? michael@0: 4 : 0)); michael@0: put_byte(s, OS_CODE); michael@0: s->status = BUSY_STATE; michael@0: } michael@0: else { michael@0: put_byte(s, (s->gzhead->text ? 1 : 0) + michael@0: (s->gzhead->hcrc ? 2 : 0) + michael@0: (s->gzhead->extra == Z_NULL ? 0 : 4) + michael@0: (s->gzhead->name == Z_NULL ? 0 : 8) + michael@0: (s->gzhead->comment == Z_NULL ? 0 : 16) michael@0: ); michael@0: put_byte(s, (Byte)(s->gzhead->time & 0xff)); michael@0: put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff)); michael@0: put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff)); michael@0: put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff)); michael@0: put_byte(s, s->level == 9 ? 2 : michael@0: (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ? michael@0: 4 : 0)); michael@0: put_byte(s, s->gzhead->os & 0xff); michael@0: if (s->gzhead->extra != Z_NULL) { michael@0: put_byte(s, s->gzhead->extra_len & 0xff); michael@0: put_byte(s, (s->gzhead->extra_len >> 8) & 0xff); michael@0: } michael@0: if (s->gzhead->hcrc) michael@0: strm->adler = crc32(strm->adler, s->pending_buf, michael@0: s->pending); michael@0: s->gzindex = 0; michael@0: s->status = EXTRA_STATE; michael@0: } michael@0: } michael@0: else michael@0: #endif michael@0: { michael@0: uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8; michael@0: uInt level_flags; michael@0: michael@0: if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2) michael@0: level_flags = 0; michael@0: else if (s->level < 6) michael@0: level_flags = 1; michael@0: else if (s->level == 6) michael@0: level_flags = 2; michael@0: else michael@0: level_flags = 3; michael@0: header |= (level_flags << 6); michael@0: if (s->strstart != 0) header |= PRESET_DICT; michael@0: header += 31 - (header % 31); michael@0: michael@0: s->status = BUSY_STATE; michael@0: putShortMSB(s, header); michael@0: michael@0: /* Save the adler32 of the preset dictionary: */ michael@0: if (s->strstart != 0) { michael@0: putShortMSB(s, (uInt)(strm->adler >> 16)); michael@0: putShortMSB(s, (uInt)(strm->adler & 0xffff)); michael@0: } michael@0: strm->adler = adler32(0L, Z_NULL, 0); michael@0: } michael@0: } michael@0: #ifdef GZIP michael@0: if (s->status == EXTRA_STATE) { michael@0: if (s->gzhead->extra != Z_NULL) { michael@0: uInt beg = s->pending; /* start of bytes to update crc */ michael@0: michael@0: while (s->gzindex < (s->gzhead->extra_len & 0xffff)) { michael@0: if (s->pending == s->pending_buf_size) { michael@0: if (s->gzhead->hcrc && s->pending > beg) michael@0: strm->adler = crc32(strm->adler, s->pending_buf + beg, michael@0: s->pending - beg); michael@0: flush_pending(strm); michael@0: beg = s->pending; michael@0: if (s->pending == s->pending_buf_size) michael@0: break; michael@0: } michael@0: put_byte(s, s->gzhead->extra[s->gzindex]); michael@0: s->gzindex++; michael@0: } michael@0: if (s->gzhead->hcrc && s->pending > beg) michael@0: strm->adler = crc32(strm->adler, s->pending_buf + beg, michael@0: s->pending - beg); michael@0: if (s->gzindex == s->gzhead->extra_len) { michael@0: s->gzindex = 0; michael@0: s->status = NAME_STATE; michael@0: } michael@0: } michael@0: else michael@0: s->status = NAME_STATE; michael@0: } michael@0: if (s->status == NAME_STATE) { michael@0: if (s->gzhead->name != Z_NULL) { michael@0: uInt beg = s->pending; /* start of bytes to update crc */ michael@0: int val; michael@0: michael@0: do { michael@0: if (s->pending == s->pending_buf_size) { michael@0: if (s->gzhead->hcrc && s->pending > beg) michael@0: strm->adler = crc32(strm->adler, s->pending_buf + beg, michael@0: s->pending - beg); michael@0: flush_pending(strm); michael@0: beg = s->pending; michael@0: if (s->pending == s->pending_buf_size) { michael@0: val = 1; michael@0: break; michael@0: } michael@0: } michael@0: val = s->gzhead->name[s->gzindex++]; michael@0: put_byte(s, val); michael@0: } while (val != 0); michael@0: if (s->gzhead->hcrc && s->pending > beg) michael@0: strm->adler = crc32(strm->adler, s->pending_buf + beg, michael@0: s->pending - beg); michael@0: if (val == 0) { michael@0: s->gzindex = 0; michael@0: s->status = COMMENT_STATE; michael@0: } michael@0: } michael@0: else michael@0: s->status = COMMENT_STATE; michael@0: } michael@0: if (s->status == COMMENT_STATE) { michael@0: if (s->gzhead->comment != Z_NULL) { michael@0: uInt beg = s->pending; /* start of bytes to update crc */ michael@0: int val; michael@0: michael@0: do { michael@0: if (s->pending == s->pending_buf_size) { michael@0: if (s->gzhead->hcrc && s->pending > beg) michael@0: strm->adler = crc32(strm->adler, s->pending_buf + beg, michael@0: s->pending - beg); michael@0: flush_pending(strm); michael@0: beg = s->pending; michael@0: if (s->pending == s->pending_buf_size) { michael@0: val = 1; michael@0: break; michael@0: } michael@0: } michael@0: val = s->gzhead->comment[s->gzindex++]; michael@0: put_byte(s, val); michael@0: } while (val != 0); michael@0: if (s->gzhead->hcrc && s->pending > beg) michael@0: strm->adler = crc32(strm->adler, s->pending_buf + beg, michael@0: s->pending - beg); michael@0: if (val == 0) michael@0: s->status = HCRC_STATE; michael@0: } michael@0: else michael@0: s->status = HCRC_STATE; michael@0: } michael@0: if (s->status == HCRC_STATE) { michael@0: if (s->gzhead->hcrc) { michael@0: if (s->pending + 2 > s->pending_buf_size) michael@0: flush_pending(strm); michael@0: if (s->pending + 2 <= s->pending_buf_size) { michael@0: put_byte(s, (Byte)(strm->adler & 0xff)); michael@0: put_byte(s, (Byte)((strm->adler >> 8) & 0xff)); michael@0: strm->adler = crc32(0L, Z_NULL, 0); michael@0: s->status = BUSY_STATE; michael@0: } michael@0: } michael@0: else michael@0: s->status = BUSY_STATE; michael@0: } michael@0: #endif michael@0: michael@0: /* Flush as much pending output as possible */ michael@0: if (s->pending != 0) { michael@0: flush_pending(strm); michael@0: if (strm->avail_out == 0) { michael@0: /* Since avail_out is 0, deflate will be called again with michael@0: * more output space, but possibly with both pending and michael@0: * avail_in equal to zero. There won't be anything to do, michael@0: * but this is not an error situation so make sure we michael@0: * return OK instead of BUF_ERROR at next call of deflate: michael@0: */ michael@0: s->last_flush = -1; michael@0: return Z_OK; michael@0: } michael@0: michael@0: /* Make sure there is something to do and avoid duplicate consecutive michael@0: * flushes. For repeated and useless calls with Z_FINISH, we keep michael@0: * returning Z_STREAM_END instead of Z_BUF_ERROR. michael@0: */ michael@0: } else if (strm->avail_in == 0 && flush <= old_flush && michael@0: flush != Z_FINISH) { michael@0: ERR_RETURN(strm, Z_BUF_ERROR); michael@0: } michael@0: michael@0: /* User must not provide more input after the first FINISH: */ michael@0: if (s->status == FINISH_STATE && strm->avail_in != 0) { michael@0: ERR_RETURN(strm, Z_BUF_ERROR); michael@0: } michael@0: michael@0: /* Start a new block or continue the current one. michael@0: */ michael@0: if (strm->avail_in != 0 || s->lookahead != 0 || michael@0: (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) { michael@0: block_state bstate; michael@0: michael@0: bstate = s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) : michael@0: (s->strategy == Z_RLE ? deflate_rle(s, flush) : michael@0: (*(configuration_table[s->level].func))(s, flush)); michael@0: michael@0: if (bstate == finish_started || bstate == finish_done) { michael@0: s->status = FINISH_STATE; michael@0: } michael@0: if (bstate == need_more || bstate == finish_started) { michael@0: if (strm->avail_out == 0) { michael@0: s->last_flush = -1; /* avoid BUF_ERROR next call, see above */ michael@0: } michael@0: return Z_OK; michael@0: /* If flush != Z_NO_FLUSH && avail_out == 0, the next call michael@0: * of deflate should use the same flush parameter to make sure michael@0: * that the flush is complete. So we don't have to output an michael@0: * empty block here, this will be done at next call. This also michael@0: * ensures that for a very small output buffer, we emit at most michael@0: * one empty block. michael@0: */ michael@0: } michael@0: if (bstate == block_done) { michael@0: if (flush == Z_PARTIAL_FLUSH) { michael@0: _tr_align(s); michael@0: } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */ michael@0: _tr_stored_block(s, (char*)0, 0L, 0); michael@0: /* For a full flush, this empty block will be recognized michael@0: * as a special marker by inflate_sync(). michael@0: */ michael@0: if (flush == Z_FULL_FLUSH) { michael@0: CLEAR_HASH(s); /* forget history */ michael@0: if (s->lookahead == 0) { michael@0: s->strstart = 0; michael@0: s->block_start = 0L; michael@0: } michael@0: } michael@0: } michael@0: flush_pending(strm); michael@0: if (strm->avail_out == 0) { michael@0: s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */ michael@0: return Z_OK; michael@0: } michael@0: } michael@0: } michael@0: Assert(strm->avail_out > 0, "bug2"); michael@0: michael@0: if (flush != Z_FINISH) return Z_OK; michael@0: if (s->wrap <= 0) return Z_STREAM_END; michael@0: michael@0: /* Write the trailer */ michael@0: #ifdef GZIP michael@0: if (s->wrap == 2) { michael@0: put_byte(s, (Byte)(strm->adler & 0xff)); michael@0: put_byte(s, (Byte)((strm->adler >> 8) & 0xff)); michael@0: put_byte(s, (Byte)((strm->adler >> 16) & 0xff)); michael@0: put_byte(s, (Byte)((strm->adler >> 24) & 0xff)); michael@0: put_byte(s, (Byte)(strm->total_in & 0xff)); michael@0: put_byte(s, (Byte)((strm->total_in >> 8) & 0xff)); michael@0: put_byte(s, (Byte)((strm->total_in >> 16) & 0xff)); michael@0: put_byte(s, (Byte)((strm->total_in >> 24) & 0xff)); michael@0: } michael@0: else michael@0: #endif michael@0: { michael@0: putShortMSB(s, (uInt)(strm->adler >> 16)); michael@0: putShortMSB(s, (uInt)(strm->adler & 0xffff)); michael@0: } michael@0: flush_pending(strm); michael@0: /* If avail_out is zero, the application will call deflate again michael@0: * to flush the rest. michael@0: */ michael@0: if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */ michael@0: return s->pending != 0 ? Z_OK : Z_STREAM_END; michael@0: } michael@0: michael@0: /* ========================================================================= */ michael@0: int ZEXPORT deflateEnd (strm) michael@0: z_streamp strm; michael@0: { michael@0: int status; michael@0: michael@0: if (strm == Z_NULL || strm->state == Z_NULL) return Z_STREAM_ERROR; michael@0: michael@0: status = strm->state->status; michael@0: if (status != INIT_STATE && michael@0: status != EXTRA_STATE && michael@0: status != NAME_STATE && michael@0: status != COMMENT_STATE && michael@0: status != HCRC_STATE && michael@0: status != BUSY_STATE && michael@0: status != FINISH_STATE) { michael@0: return Z_STREAM_ERROR; michael@0: } michael@0: michael@0: /* Deallocate in reverse order of allocations: */ michael@0: TRY_FREE(strm, strm->state->pending_buf); michael@0: TRY_FREE(strm, strm->state->head); michael@0: TRY_FREE(strm, strm->state->prev); michael@0: TRY_FREE(strm, strm->state->window); michael@0: michael@0: ZFREE(strm, strm->state); michael@0: strm->state = Z_NULL; michael@0: michael@0: return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK; michael@0: } michael@0: michael@0: /* ========================================================================= michael@0: * Copy the source state to the destination state. michael@0: * To simplify the source, this is not supported for 16-bit MSDOS (which michael@0: * doesn't have enough memory anyway to duplicate compression states). michael@0: */ michael@0: int ZEXPORT deflateCopy (dest, source) michael@0: z_streamp dest; michael@0: z_streamp source; michael@0: { michael@0: #ifdef MAXSEG_64K michael@0: return Z_STREAM_ERROR; michael@0: #else michael@0: deflate_state *ds; michael@0: deflate_state *ss; michael@0: ushf *overlay; michael@0: michael@0: michael@0: if (source == Z_NULL || dest == Z_NULL || source->state == Z_NULL) { michael@0: return Z_STREAM_ERROR; michael@0: } michael@0: michael@0: ss = source->state; michael@0: michael@0: zmemcpy(dest, source, sizeof(z_stream)); michael@0: michael@0: ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state)); michael@0: if (ds == Z_NULL) return Z_MEM_ERROR; michael@0: dest->state = (struct internal_state FAR *) ds; michael@0: zmemcpy(ds, ss, sizeof(deflate_state)); michael@0: ds->strm = dest; michael@0: michael@0: ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte)); michael@0: ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos)); michael@0: ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos)); michael@0: overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2); michael@0: ds->pending_buf = (uchf *) overlay; michael@0: michael@0: if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL || michael@0: ds->pending_buf == Z_NULL) { michael@0: deflateEnd (dest); michael@0: return Z_MEM_ERROR; michael@0: } michael@0: /* following zmemcpy do not work for 16-bit MSDOS */ michael@0: zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte)); michael@0: zmemcpy(ds->prev, ss->prev, ds->w_size * sizeof(Pos)); michael@0: zmemcpy(ds->head, ss->head, ds->hash_size * sizeof(Pos)); michael@0: zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size); michael@0: michael@0: ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf); michael@0: ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush); michael@0: ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize; michael@0: michael@0: ds->l_desc.dyn_tree = ds->dyn_ltree; michael@0: ds->d_desc.dyn_tree = ds->dyn_dtree; michael@0: ds->bl_desc.dyn_tree = ds->bl_tree; michael@0: michael@0: return Z_OK; michael@0: #endif /* MAXSEG_64K */ michael@0: } michael@0: michael@0: /* =========================================================================== michael@0: * Read a new buffer from the current input stream, update the adler32 michael@0: * and total number of bytes read. All deflate() input goes through michael@0: * this function so some applications may wish to modify it to avoid michael@0: * allocating a large strm->next_in buffer and copying from it. michael@0: * (See also flush_pending()). michael@0: */ michael@0: local int read_buf(strm, buf, size) michael@0: z_streamp strm; michael@0: Bytef *buf; michael@0: unsigned size; michael@0: { michael@0: unsigned len = strm->avail_in; michael@0: michael@0: if (len > size) len = size; michael@0: if (len == 0) return 0; michael@0: michael@0: strm->avail_in -= len; michael@0: michael@0: if (strm->state->wrap == 1) { michael@0: strm->adler = adler32(strm->adler, strm->next_in, len); michael@0: } michael@0: #ifdef GZIP michael@0: else if (strm->state->wrap == 2) { michael@0: strm->adler = crc32(strm->adler, strm->next_in, len); michael@0: } michael@0: #endif michael@0: zmemcpy(buf, strm->next_in, len); michael@0: strm->next_in += len; michael@0: strm->total_in += len; michael@0: michael@0: return (int)len; michael@0: } michael@0: michael@0: /* =========================================================================== michael@0: * Initialize the "longest match" routines for a new zlib stream michael@0: */ michael@0: local void lm_init (s) michael@0: deflate_state *s; michael@0: { michael@0: s->window_size = (ulg)2L*s->w_size; michael@0: michael@0: CLEAR_HASH(s); michael@0: michael@0: /* Set the default configuration parameters: michael@0: */ michael@0: s->max_lazy_match = configuration_table[s->level].max_lazy; michael@0: s->good_match = configuration_table[s->level].good_length; michael@0: s->nice_match = configuration_table[s->level].nice_length; michael@0: s->max_chain_length = configuration_table[s->level].max_chain; michael@0: michael@0: s->strstart = 0; michael@0: s->block_start = 0L; michael@0: s->lookahead = 0; michael@0: s->match_length = s->prev_length = MIN_MATCH-1; michael@0: s->match_available = 0; michael@0: s->ins_h = 0; michael@0: #ifndef FASTEST michael@0: #ifdef ASMV michael@0: match_init(); /* initialize the asm code */ michael@0: #endif michael@0: #endif michael@0: } michael@0: michael@0: #ifndef FASTEST michael@0: /* =========================================================================== michael@0: * Set match_start to the longest match starting at the given string and michael@0: * return its length. Matches shorter or equal to prev_length are discarded, michael@0: * in which case the result is equal to prev_length and match_start is michael@0: * garbage. michael@0: * IN assertions: cur_match is the head of the hash chain for the current michael@0: * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1 michael@0: * OUT assertion: the match length is not greater than s->lookahead. michael@0: */ michael@0: #ifndef ASMV michael@0: /* For 80x86 and 680x0, an optimized version will be provided in match.asm or michael@0: * match.S. The code will be functionally equivalent. michael@0: */ michael@0: local uInt longest_match(s, cur_match) michael@0: deflate_state *s; michael@0: IPos cur_match; /* current match */ michael@0: { michael@0: unsigned chain_length = s->max_chain_length;/* max hash chain length */ michael@0: register Bytef *scan = s->window + s->strstart; /* current string */ michael@0: register Bytef *match; /* matched string */ michael@0: register int len; /* length of current match */ michael@0: int best_len = s->prev_length; /* best match length so far */ michael@0: int nice_match = s->nice_match; /* stop if match long enough */ michael@0: IPos limit = s->strstart > (IPos)MAX_DIST(s) ? michael@0: s->strstart - (IPos)MAX_DIST(s) : NIL; michael@0: /* Stop when cur_match becomes <= limit. To simplify the code, michael@0: * we prevent matches with the string of window index 0. michael@0: */ michael@0: Posf *prev = s->prev; michael@0: uInt wmask = s->w_mask; michael@0: michael@0: #ifdef UNALIGNED_OK michael@0: /* Compare two bytes at a time. Note: this is not always beneficial. michael@0: * Try with and without -DUNALIGNED_OK to check. michael@0: */ michael@0: register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1; michael@0: register ush scan_start = *(ushf*)scan; michael@0: register ush scan_end = *(ushf*)(scan+best_len-1); michael@0: #else michael@0: register Bytef *strend = s->window + s->strstart + MAX_MATCH; michael@0: register Byte scan_end1 = scan[best_len-1]; michael@0: register Byte scan_end = scan[best_len]; michael@0: #endif michael@0: michael@0: /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. michael@0: * It is easy to get rid of this optimization if necessary. michael@0: */ michael@0: Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever"); michael@0: michael@0: /* Do not waste too much time if we already have a good match: */ michael@0: if (s->prev_length >= s->good_match) { michael@0: chain_length >>= 2; michael@0: } michael@0: /* Do not look for matches beyond the end of the input. This is necessary michael@0: * to make deflate deterministic. michael@0: */ michael@0: if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead; michael@0: michael@0: Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead"); michael@0: michael@0: do { michael@0: Assert(cur_match < s->strstart, "no future"); michael@0: match = s->window + cur_match; michael@0: michael@0: /* Skip to next match if the match length cannot increase michael@0: * or if the match length is less than 2. Note that the checks below michael@0: * for insufficient lookahead only occur occasionally for performance michael@0: * reasons. Therefore uninitialized memory will be accessed, and michael@0: * conditional jumps will be made that depend on those values. michael@0: * However the length of the match is limited to the lookahead, so michael@0: * the output of deflate is not affected by the uninitialized values. michael@0: */ michael@0: #if (defined(UNALIGNED_OK) && MAX_MATCH == 258) michael@0: /* This code assumes sizeof(unsigned short) == 2. Do not use michael@0: * UNALIGNED_OK if your compiler uses a different size. michael@0: */ michael@0: if (*(ushf*)(match+best_len-1) != scan_end || michael@0: *(ushf*)match != scan_start) continue; michael@0: michael@0: /* It is not necessary to compare scan[2] and match[2] since they are michael@0: * always equal when the other bytes match, given that the hash keys michael@0: * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at michael@0: * strstart+3, +5, ... up to strstart+257. We check for insufficient michael@0: * lookahead only every 4th comparison; the 128th check will be made michael@0: * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is michael@0: * necessary to put more guard bytes at the end of the window, or michael@0: * to check more often for insufficient lookahead. michael@0: */ michael@0: Assert(scan[2] == match[2], "scan[2]?"); michael@0: scan++, match++; michael@0: do { michael@0: } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) && michael@0: *(ushf*)(scan+=2) == *(ushf*)(match+=2) && michael@0: *(ushf*)(scan+=2) == *(ushf*)(match+=2) && michael@0: *(ushf*)(scan+=2) == *(ushf*)(match+=2) && michael@0: scan < strend); michael@0: /* The funny "do {}" generates better code on most compilers */ michael@0: michael@0: /* Here, scan <= window+strstart+257 */ michael@0: Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); michael@0: if (*scan == *match) scan++; michael@0: michael@0: len = (MAX_MATCH - 1) - (int)(strend-scan); michael@0: scan = strend - (MAX_MATCH-1); michael@0: michael@0: #else /* UNALIGNED_OK */ michael@0: michael@0: if (match[best_len] != scan_end || michael@0: match[best_len-1] != scan_end1 || michael@0: *match != *scan || michael@0: *++match != scan[1]) continue; michael@0: michael@0: /* The check at best_len-1 can be removed because it will be made michael@0: * again later. (This heuristic is not always a win.) michael@0: * It is not necessary to compare scan[2] and match[2] since they michael@0: * are always equal when the other bytes match, given that michael@0: * the hash keys are equal and that HASH_BITS >= 8. michael@0: */ michael@0: scan += 2, match++; michael@0: Assert(*scan == *match, "match[2]?"); michael@0: michael@0: /* We check for insufficient lookahead only every 8th comparison; michael@0: * the 256th check will be made at strstart+258. michael@0: */ michael@0: do { michael@0: } while (*++scan == *++match && *++scan == *++match && michael@0: *++scan == *++match && *++scan == *++match && michael@0: *++scan == *++match && *++scan == *++match && michael@0: *++scan == *++match && *++scan == *++match && michael@0: scan < strend); michael@0: michael@0: Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); michael@0: michael@0: len = MAX_MATCH - (int)(strend - scan); michael@0: scan = strend - MAX_MATCH; michael@0: michael@0: #endif /* UNALIGNED_OK */ michael@0: michael@0: if (len > best_len) { michael@0: s->match_start = cur_match; michael@0: best_len = len; michael@0: if (len >= nice_match) break; michael@0: #ifdef UNALIGNED_OK michael@0: scan_end = *(ushf*)(scan+best_len-1); michael@0: #else michael@0: scan_end1 = scan[best_len-1]; michael@0: scan_end = scan[best_len]; michael@0: #endif michael@0: } michael@0: } while ((cur_match = prev[cur_match & wmask]) > limit michael@0: && --chain_length != 0); michael@0: michael@0: if ((uInt)best_len <= s->lookahead) return (uInt)best_len; michael@0: return s->lookahead; michael@0: } michael@0: #endif /* ASMV */ michael@0: michael@0: #else /* FASTEST */ michael@0: michael@0: /* --------------------------------------------------------------------------- michael@0: * Optimized version for FASTEST only michael@0: */ michael@0: local uInt longest_match(s, cur_match) michael@0: deflate_state *s; michael@0: IPos cur_match; /* current match */ michael@0: { michael@0: register Bytef *scan = s->window + s->strstart; /* current string */ michael@0: register Bytef *match; /* matched string */ michael@0: register int len; /* length of current match */ michael@0: register Bytef *strend = s->window + s->strstart + MAX_MATCH; michael@0: michael@0: /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. michael@0: * It is easy to get rid of this optimization if necessary. michael@0: */ michael@0: Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever"); michael@0: michael@0: Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead"); michael@0: michael@0: Assert(cur_match < s->strstart, "no future"); michael@0: michael@0: match = s->window + cur_match; michael@0: michael@0: /* Return failure if the match length is less than 2: michael@0: */ michael@0: if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1; michael@0: michael@0: /* The check at best_len-1 can be removed because it will be made michael@0: * again later. (This heuristic is not always a win.) michael@0: * It is not necessary to compare scan[2] and match[2] since they michael@0: * are always equal when the other bytes match, given that michael@0: * the hash keys are equal and that HASH_BITS >= 8. michael@0: */ michael@0: scan += 2, match += 2; michael@0: Assert(*scan == *match, "match[2]?"); michael@0: michael@0: /* We check for insufficient lookahead only every 8th comparison; michael@0: * the 256th check will be made at strstart+258. michael@0: */ michael@0: do { michael@0: } while (*++scan == *++match && *++scan == *++match && michael@0: *++scan == *++match && *++scan == *++match && michael@0: *++scan == *++match && *++scan == *++match && michael@0: *++scan == *++match && *++scan == *++match && michael@0: scan < strend); michael@0: michael@0: Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); michael@0: michael@0: len = MAX_MATCH - (int)(strend - scan); michael@0: michael@0: if (len < MIN_MATCH) return MIN_MATCH - 1; michael@0: michael@0: s->match_start = cur_match; michael@0: return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead; michael@0: } michael@0: michael@0: #endif /* FASTEST */ michael@0: michael@0: #ifdef DEBUG michael@0: /* =========================================================================== michael@0: * Check that the match at match_start is indeed a match. michael@0: */ michael@0: local void check_match(s, start, match, length) michael@0: deflate_state *s; michael@0: IPos start, match; michael@0: int length; michael@0: { michael@0: /* check that the match is indeed a match */ michael@0: if (zmemcmp(s->window + match, michael@0: s->window + start, length) != EQUAL) { michael@0: fprintf(stderr, " start %u, match %u, length %d\n", michael@0: start, match, length); michael@0: do { michael@0: fprintf(stderr, "%c%c", s->window[match++], s->window[start++]); michael@0: } while (--length != 0); michael@0: z_error("invalid match"); michael@0: } michael@0: if (z_verbose > 1) { michael@0: fprintf(stderr,"\\[%d,%d]", start-match, length); michael@0: do { putc(s->window[start++], stderr); } while (--length != 0); michael@0: } michael@0: } michael@0: #else michael@0: # define check_match(s, start, match, length) michael@0: #endif /* DEBUG */ michael@0: michael@0: /* =========================================================================== michael@0: * Fill the window when the lookahead becomes insufficient. michael@0: * Updates strstart and lookahead. michael@0: * michael@0: * IN assertion: lookahead < MIN_LOOKAHEAD michael@0: * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD michael@0: * At least one byte has been read, or avail_in == 0; reads are michael@0: * performed for at least two bytes (required for the zip translate_eol michael@0: * option -- not supported here). michael@0: */ michael@0: local void fill_window(s) michael@0: deflate_state *s; michael@0: { michael@0: register unsigned n, m; michael@0: register Posf *p; michael@0: unsigned more; /* Amount of free space at the end of the window. */ michael@0: uInt wsize = s->w_size; michael@0: michael@0: do { michael@0: more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart); michael@0: michael@0: /* Deal with !@#$% 64K limit: */ michael@0: if (sizeof(int) <= 2) { michael@0: if (more == 0 && s->strstart == 0 && s->lookahead == 0) { michael@0: more = wsize; michael@0: michael@0: } else if (more == (unsigned)(-1)) { michael@0: /* Very unlikely, but possible on 16 bit machine if michael@0: * strstart == 0 && lookahead == 1 (input done a byte at time) michael@0: */ michael@0: more--; michael@0: } michael@0: } michael@0: michael@0: /* If the window is almost full and there is insufficient lookahead, michael@0: * move the upper half to the lower one to make room in the upper half. michael@0: */ michael@0: if (s->strstart >= wsize+MAX_DIST(s)) { michael@0: michael@0: zmemcpy(s->window, s->window+wsize, (unsigned)wsize); michael@0: s->match_start -= wsize; michael@0: s->strstart -= wsize; /* we now have strstart >= MAX_DIST */ michael@0: s->block_start -= (long) wsize; michael@0: michael@0: /* Slide the hash table (could be avoided with 32 bit values michael@0: at the expense of memory usage). We slide even when level == 0 michael@0: to keep the hash table consistent if we switch back to level > 0 michael@0: later. (Using level 0 permanently is not an optimal usage of michael@0: zlib, so we don't care about this pathological case.) michael@0: */ michael@0: n = s->hash_size; michael@0: p = &s->head[n]; michael@0: do { michael@0: m = *--p; michael@0: *p = (Pos)(m >= wsize ? m-wsize : NIL); michael@0: } while (--n); michael@0: michael@0: n = wsize; michael@0: #ifndef FASTEST michael@0: p = &s->prev[n]; michael@0: do { michael@0: m = *--p; michael@0: *p = (Pos)(m >= wsize ? m-wsize : NIL); michael@0: /* If n is not on any hash chain, prev[n] is garbage but michael@0: * its value will never be used. michael@0: */ michael@0: } while (--n); michael@0: #endif michael@0: more += wsize; michael@0: } michael@0: if (s->strm->avail_in == 0) return; michael@0: michael@0: /* If there was no sliding: michael@0: * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 && michael@0: * more == window_size - lookahead - strstart michael@0: * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1) michael@0: * => more >= window_size - 2*WSIZE + 2 michael@0: * In the BIG_MEM or MMAP case (not yet supported), michael@0: * window_size == input_size + MIN_LOOKAHEAD && michael@0: * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD. michael@0: * Otherwise, window_size == 2*WSIZE so more >= 2. michael@0: * If there was sliding, more >= WSIZE. So in all cases, more >= 2. michael@0: */ michael@0: Assert(more >= 2, "more < 2"); michael@0: michael@0: n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more); michael@0: s->lookahead += n; michael@0: michael@0: /* Initialize the hash value now that we have some input: */ michael@0: if (s->lookahead >= MIN_MATCH) { michael@0: s->ins_h = s->window[s->strstart]; michael@0: UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]); michael@0: #if MIN_MATCH != 3 michael@0: Call UPDATE_HASH() MIN_MATCH-3 more times michael@0: #endif michael@0: } michael@0: /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage, michael@0: * but this is not important since only literal bytes will be emitted. michael@0: */ michael@0: michael@0: } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0); michael@0: michael@0: /* If the WIN_INIT bytes after the end of the current data have never been michael@0: * written, then zero those bytes in order to avoid memory check reports of michael@0: * the use of uninitialized (or uninitialised as Julian writes) bytes by michael@0: * the longest match routines. Update the high water mark for the next michael@0: * time through here. WIN_INIT is set to MAX_MATCH since the longest match michael@0: * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead. michael@0: */ michael@0: if (s->high_water < s->window_size) { michael@0: ulg curr = s->strstart + (ulg)(s->lookahead); michael@0: ulg init; michael@0: michael@0: if (s->high_water < curr) { michael@0: /* Previous high water mark below current data -- zero WIN_INIT michael@0: * bytes or up to end of window, whichever is less. michael@0: */ michael@0: init = s->window_size - curr; michael@0: if (init > WIN_INIT) michael@0: init = WIN_INIT; michael@0: zmemzero(s->window + curr, (unsigned)init); michael@0: s->high_water = curr + init; michael@0: } michael@0: else if (s->high_water < (ulg)curr + WIN_INIT) { michael@0: /* High water mark at or above current data, but below current data michael@0: * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up michael@0: * to end of window, whichever is less. michael@0: */ michael@0: init = (ulg)curr + WIN_INIT - s->high_water; michael@0: if (init > s->window_size - s->high_water) michael@0: init = s->window_size - s->high_water; michael@0: zmemzero(s->window + s->high_water, (unsigned)init); michael@0: s->high_water += init; michael@0: } michael@0: } michael@0: } michael@0: michael@0: /* =========================================================================== michael@0: * Flush the current block, with given end-of-file flag. michael@0: * IN assertion: strstart is set to the end of the current match. michael@0: */ michael@0: #define FLUSH_BLOCK_ONLY(s, last) { \ michael@0: _tr_flush_block(s, (s->block_start >= 0L ? \ michael@0: (charf *)&s->window[(unsigned)s->block_start] : \ michael@0: (charf *)Z_NULL), \ michael@0: (ulg)((long)s->strstart - s->block_start), \ michael@0: (last)); \ michael@0: s->block_start = s->strstart; \ michael@0: flush_pending(s->strm); \ michael@0: Tracev((stderr,"[FLUSH]")); \ michael@0: } michael@0: michael@0: /* Same but force premature exit if necessary. */ michael@0: #define FLUSH_BLOCK(s, last) { \ michael@0: FLUSH_BLOCK_ONLY(s, last); \ michael@0: if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \ michael@0: } michael@0: michael@0: /* =========================================================================== michael@0: * Copy without compression as much as possible from the input stream, return michael@0: * the current block state. michael@0: * This function does not insert new strings in the dictionary since michael@0: * uncompressible data is probably not useful. This function is used michael@0: * only for the level=0 compression option. michael@0: * NOTE: this function should be optimized to avoid extra copying from michael@0: * window to pending_buf. michael@0: */ michael@0: local block_state deflate_stored(s, flush) michael@0: deflate_state *s; michael@0: int flush; michael@0: { michael@0: /* Stored blocks are limited to 0xffff bytes, pending_buf is limited michael@0: * to pending_buf_size, and each stored block has a 5 byte header: michael@0: */ michael@0: ulg max_block_size = 0xffff; michael@0: ulg max_start; michael@0: michael@0: if (max_block_size > s->pending_buf_size - 5) { michael@0: max_block_size = s->pending_buf_size - 5; michael@0: } michael@0: michael@0: /* Copy as much as possible from input to output: */ michael@0: for (;;) { michael@0: /* Fill the window as much as possible: */ michael@0: if (s->lookahead <= 1) { michael@0: michael@0: Assert(s->strstart < s->w_size+MAX_DIST(s) || michael@0: s->block_start >= (long)s->w_size, "slide too late"); michael@0: michael@0: fill_window(s); michael@0: if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more; michael@0: michael@0: if (s->lookahead == 0) break; /* flush the current block */ michael@0: } michael@0: Assert(s->block_start >= 0L, "block gone"); michael@0: michael@0: s->strstart += s->lookahead; michael@0: s->lookahead = 0; michael@0: michael@0: /* Emit a stored block if pending_buf will be full: */ michael@0: max_start = s->block_start + max_block_size; michael@0: if (s->strstart == 0 || (ulg)s->strstart >= max_start) { michael@0: /* strstart == 0 is possible when wraparound on 16-bit machine */ michael@0: s->lookahead = (uInt)(s->strstart - max_start); michael@0: s->strstart = (uInt)max_start; michael@0: FLUSH_BLOCK(s, 0); michael@0: } michael@0: /* Flush if we may have to slide, otherwise block_start may become michael@0: * negative and the data will be gone: michael@0: */ michael@0: if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) { michael@0: FLUSH_BLOCK(s, 0); michael@0: } michael@0: } michael@0: FLUSH_BLOCK(s, flush == Z_FINISH); michael@0: return flush == Z_FINISH ? finish_done : block_done; michael@0: } michael@0: michael@0: /* =========================================================================== michael@0: * Compress as much as possible from the input stream, return the current michael@0: * block state. michael@0: * This function does not perform lazy evaluation of matches and inserts michael@0: * new strings in the dictionary only for unmatched strings or for short michael@0: * matches. It is used only for the fast compression options. michael@0: */ michael@0: local block_state deflate_fast(s, flush) michael@0: deflate_state *s; michael@0: int flush; michael@0: { michael@0: IPos hash_head; /* head of the hash chain */ michael@0: int bflush; /* set if current block must be flushed */ michael@0: michael@0: for (;;) { michael@0: /* Make sure that we always have enough lookahead, except michael@0: * at the end of the input file. We need MAX_MATCH bytes michael@0: * for the next match, plus MIN_MATCH bytes to insert the michael@0: * string following the next match. michael@0: */ michael@0: if (s->lookahead < MIN_LOOKAHEAD) { michael@0: fill_window(s); michael@0: if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { michael@0: return need_more; michael@0: } michael@0: if (s->lookahead == 0) break; /* flush the current block */ michael@0: } michael@0: michael@0: /* Insert the string window[strstart .. strstart+2] in the michael@0: * dictionary, and set hash_head to the head of the hash chain: michael@0: */ michael@0: hash_head = NIL; michael@0: if (s->lookahead >= MIN_MATCH) { michael@0: INSERT_STRING(s, s->strstart, hash_head); michael@0: } michael@0: michael@0: /* Find the longest match, discarding those <= prev_length. michael@0: * At this point we have always match_length < MIN_MATCH michael@0: */ michael@0: if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) { michael@0: /* To simplify the code, we prevent matches with the string michael@0: * of window index 0 (in particular we have to avoid a match michael@0: * of the string with itself at the start of the input file). michael@0: */ michael@0: s->match_length = longest_match (s, hash_head); michael@0: /* longest_match() sets match_start */ michael@0: } michael@0: if (s->match_length >= MIN_MATCH) { michael@0: check_match(s, s->strstart, s->match_start, s->match_length); michael@0: michael@0: _tr_tally_dist(s, s->strstart - s->match_start, michael@0: s->match_length - MIN_MATCH, bflush); michael@0: michael@0: s->lookahead -= s->match_length; michael@0: michael@0: /* Insert new strings in the hash table only if the match length michael@0: * is not too large. This saves time but degrades compression. michael@0: */ michael@0: #ifndef FASTEST michael@0: if (s->match_length <= s->max_insert_length && michael@0: s->lookahead >= MIN_MATCH) { michael@0: s->match_length--; /* string at strstart already in table */ michael@0: do { michael@0: s->strstart++; michael@0: INSERT_STRING(s, s->strstart, hash_head); michael@0: /* strstart never exceeds WSIZE-MAX_MATCH, so there are michael@0: * always MIN_MATCH bytes ahead. michael@0: */ michael@0: } while (--s->match_length != 0); michael@0: s->strstart++; michael@0: } else michael@0: #endif michael@0: { michael@0: s->strstart += s->match_length; michael@0: s->match_length = 0; michael@0: s->ins_h = s->window[s->strstart]; michael@0: UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]); michael@0: #if MIN_MATCH != 3 michael@0: Call UPDATE_HASH() MIN_MATCH-3 more times michael@0: #endif michael@0: /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not michael@0: * matter since it will be recomputed at next deflate call. michael@0: */ michael@0: } michael@0: } else { michael@0: /* No match, output a literal byte */ michael@0: Tracevv((stderr,"%c", s->window[s->strstart])); michael@0: _tr_tally_lit (s, s->window[s->strstart], bflush); michael@0: s->lookahead--; michael@0: s->strstart++; michael@0: } michael@0: if (bflush) FLUSH_BLOCK(s, 0); michael@0: } michael@0: FLUSH_BLOCK(s, flush == Z_FINISH); michael@0: return flush == Z_FINISH ? finish_done : block_done; michael@0: } michael@0: michael@0: #ifndef FASTEST michael@0: /* =========================================================================== michael@0: * Same as above, but achieves better compression. We use a lazy michael@0: * evaluation for matches: a match is finally adopted only if there is michael@0: * no better match at the next window position. michael@0: */ michael@0: local block_state deflate_slow(s, flush) michael@0: deflate_state *s; michael@0: int flush; michael@0: { michael@0: IPos hash_head; /* head of hash chain */ michael@0: int bflush; /* set if current block must be flushed */ michael@0: michael@0: /* Process the input block. */ michael@0: for (;;) { michael@0: /* Make sure that we always have enough lookahead, except michael@0: * at the end of the input file. We need MAX_MATCH bytes michael@0: * for the next match, plus MIN_MATCH bytes to insert the michael@0: * string following the next match. michael@0: */ michael@0: if (s->lookahead < MIN_LOOKAHEAD) { michael@0: fill_window(s); michael@0: if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { michael@0: return need_more; michael@0: } michael@0: if (s->lookahead == 0) break; /* flush the current block */ michael@0: } michael@0: michael@0: /* Insert the string window[strstart .. strstart+2] in the michael@0: * dictionary, and set hash_head to the head of the hash chain: michael@0: */ michael@0: hash_head = NIL; michael@0: if (s->lookahead >= MIN_MATCH) { michael@0: INSERT_STRING(s, s->strstart, hash_head); michael@0: } michael@0: michael@0: /* Find the longest match, discarding those <= prev_length. michael@0: */ michael@0: s->prev_length = s->match_length, s->prev_match = s->match_start; michael@0: s->match_length = MIN_MATCH-1; michael@0: michael@0: if (hash_head != NIL && s->prev_length < s->max_lazy_match && michael@0: s->strstart - hash_head <= MAX_DIST(s)) { michael@0: /* To simplify the code, we prevent matches with the string michael@0: * of window index 0 (in particular we have to avoid a match michael@0: * of the string with itself at the start of the input file). michael@0: */ michael@0: s->match_length = longest_match (s, hash_head); michael@0: /* longest_match() sets match_start */ michael@0: michael@0: if (s->match_length <= 5 && (s->strategy == Z_FILTERED michael@0: #if TOO_FAR <= 32767 michael@0: || (s->match_length == MIN_MATCH && michael@0: s->strstart - s->match_start > TOO_FAR) michael@0: #endif michael@0: )) { michael@0: michael@0: /* If prev_match is also MIN_MATCH, match_start is garbage michael@0: * but we will ignore the current match anyway. michael@0: */ michael@0: s->match_length = MIN_MATCH-1; michael@0: } michael@0: } michael@0: /* If there was a match at the previous step and the current michael@0: * match is not better, output the previous match: michael@0: */ michael@0: if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) { michael@0: uInt max_insert = s->strstart + s->lookahead - MIN_MATCH; michael@0: /* Do not insert strings in hash table beyond this. */ michael@0: michael@0: check_match(s, s->strstart-1, s->prev_match, s->prev_length); michael@0: michael@0: _tr_tally_dist(s, s->strstart -1 - s->prev_match, michael@0: s->prev_length - MIN_MATCH, bflush); michael@0: michael@0: /* Insert in hash table all strings up to the end of the match. michael@0: * strstart-1 and strstart are already inserted. If there is not michael@0: * enough lookahead, the last two strings are not inserted in michael@0: * the hash table. michael@0: */ michael@0: s->lookahead -= s->prev_length-1; michael@0: s->prev_length -= 2; michael@0: do { michael@0: if (++s->strstart <= max_insert) { michael@0: INSERT_STRING(s, s->strstart, hash_head); michael@0: } michael@0: } while (--s->prev_length != 0); michael@0: s->match_available = 0; michael@0: s->match_length = MIN_MATCH-1; michael@0: s->strstart++; michael@0: michael@0: if (bflush) FLUSH_BLOCK(s, 0); michael@0: michael@0: } else if (s->match_available) { michael@0: /* If there was no match at the previous position, output a michael@0: * single literal. If there was a match but the current match michael@0: * is longer, truncate the previous match to a single literal. michael@0: */ michael@0: Tracevv((stderr,"%c", s->window[s->strstart-1])); michael@0: _tr_tally_lit(s, s->window[s->strstart-1], bflush); michael@0: if (bflush) { michael@0: FLUSH_BLOCK_ONLY(s, 0); michael@0: } michael@0: s->strstart++; michael@0: s->lookahead--; michael@0: if (s->strm->avail_out == 0) return need_more; michael@0: } else { michael@0: /* There is no previous match to compare with, wait for michael@0: * the next step to decide. michael@0: */ michael@0: s->match_available = 1; michael@0: s->strstart++; michael@0: s->lookahead--; michael@0: } michael@0: } michael@0: Assert (flush != Z_NO_FLUSH, "no flush?"); michael@0: if (s->match_available) { michael@0: Tracevv((stderr,"%c", s->window[s->strstart-1])); michael@0: _tr_tally_lit(s, s->window[s->strstart-1], bflush); michael@0: s->match_available = 0; michael@0: } michael@0: FLUSH_BLOCK(s, flush == Z_FINISH); michael@0: return flush == Z_FINISH ? finish_done : block_done; michael@0: } michael@0: #endif /* FASTEST */ michael@0: michael@0: /* =========================================================================== michael@0: * For Z_RLE, simply look for runs of bytes, generate matches only of distance michael@0: * one. Do not maintain a hash table. (It will be regenerated if this run of michael@0: * deflate switches away from Z_RLE.) michael@0: */ michael@0: local block_state deflate_rle(s, flush) michael@0: deflate_state *s; michael@0: int flush; michael@0: { michael@0: int bflush; /* set if current block must be flushed */ michael@0: uInt prev; /* byte at distance one to match */ michael@0: Bytef *scan, *strend; /* scan goes up to strend for length of run */ michael@0: michael@0: for (;;) { michael@0: /* Make sure that we always have enough lookahead, except michael@0: * at the end of the input file. We need MAX_MATCH bytes michael@0: * for the longest encodable run. michael@0: */ michael@0: if (s->lookahead < MAX_MATCH) { michael@0: fill_window(s); michael@0: if (s->lookahead < MAX_MATCH && flush == Z_NO_FLUSH) { michael@0: return need_more; michael@0: } michael@0: if (s->lookahead == 0) break; /* flush the current block */ michael@0: } michael@0: michael@0: /* See how many times the previous byte repeats */ michael@0: s->match_length = 0; michael@0: if (s->lookahead >= MIN_MATCH && s->strstart > 0) { michael@0: scan = s->window + s->strstart - 1; michael@0: prev = *scan; michael@0: if (prev == *++scan && prev == *++scan && prev == *++scan) { michael@0: strend = s->window + s->strstart + MAX_MATCH; michael@0: do { michael@0: } while (prev == *++scan && prev == *++scan && michael@0: prev == *++scan && prev == *++scan && michael@0: prev == *++scan && prev == *++scan && michael@0: prev == *++scan && prev == *++scan && michael@0: scan < strend); michael@0: s->match_length = MAX_MATCH - (int)(strend - scan); michael@0: if (s->match_length > s->lookahead) michael@0: s->match_length = s->lookahead; michael@0: } michael@0: } michael@0: michael@0: /* Emit match if have run of MIN_MATCH or longer, else emit literal */ michael@0: if (s->match_length >= MIN_MATCH) { michael@0: check_match(s, s->strstart, s->strstart - 1, s->match_length); michael@0: michael@0: _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush); michael@0: michael@0: s->lookahead -= s->match_length; michael@0: s->strstart += s->match_length; michael@0: s->match_length = 0; michael@0: } else { michael@0: /* No match, output a literal byte */ michael@0: Tracevv((stderr,"%c", s->window[s->strstart])); michael@0: _tr_tally_lit (s, s->window[s->strstart], bflush); michael@0: s->lookahead--; michael@0: s->strstart++; michael@0: } michael@0: if (bflush) FLUSH_BLOCK(s, 0); michael@0: } michael@0: FLUSH_BLOCK(s, flush == Z_FINISH); michael@0: return flush == Z_FINISH ? finish_done : block_done; michael@0: } michael@0: michael@0: /* =========================================================================== michael@0: * For Z_HUFFMAN_ONLY, do not look for matches. Do not maintain a hash table. michael@0: * (It will be regenerated if this run of deflate switches away from Huffman.) michael@0: */ michael@0: local block_state deflate_huff(s, flush) michael@0: deflate_state *s; michael@0: int flush; michael@0: { michael@0: int bflush; /* set if current block must be flushed */ michael@0: michael@0: for (;;) { michael@0: /* Make sure that we have a literal to write. */ michael@0: if (s->lookahead == 0) { michael@0: fill_window(s); michael@0: if (s->lookahead == 0) { michael@0: if (flush == Z_NO_FLUSH) michael@0: return need_more; michael@0: break; /* flush the current block */ michael@0: } michael@0: } michael@0: michael@0: /* Output a literal byte */ michael@0: s->match_length = 0; michael@0: Tracevv((stderr,"%c", s->window[s->strstart])); michael@0: _tr_tally_lit (s, s->window[s->strstart], bflush); michael@0: s->lookahead--; michael@0: s->strstart++; michael@0: if (bflush) FLUSH_BLOCK(s, 0); michael@0: } michael@0: FLUSH_BLOCK(s, flush == Z_FINISH); michael@0: return flush == Z_FINISH ? finish_done : block_done; michael@0: }