From d15dbca39dc924ce6ddb809c0a0315c292fb4d19 Mon Sep 17 00:00:00 2001 From: Andrey Kiselev Date: Sat, 24 Jul 2004 19:01:15 +0000 Subject: [PATCH] LZW compression code is merged back from the separate package. --- libtiff/tif_lzw.c | 397 ++++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 365 insertions(+), 32 deletions(-) diff --git a/libtiff/tif_lzw.c b/libtiff/tif_lzw.c index b45bf36e..e7652002 100644 --- a/libtiff/tif_lzw.c +++ b/libtiff/tif_lzw.c @@ -1,4 +1,4 @@ -/* $Header: /cvs/maptools/cvsroot/libtiff/libtiff/tif_lzw.c,v 1.20 2003-11-03 14:45:38 dron Exp $ */ +/* $Id: tif_lzw.c,v 1.21 2004-07-24 19:01:15 dron Exp $ */ /* * Copyright (c) 1988-1997 Sam Leffler @@ -97,6 +97,8 @@ typedef struct { u_short free_ent; /* next free entry in hash table */ long nextdata; /* next bits of i/o */ long nextbits; /* # of valid bits in lzw_nextdata */ + + int rw_mode; /* preserve rw_mode from init */ } LZWBaseState; #define lzw_nbits base.nbits @@ -109,6 +111,10 @@ typedef struct { * Encoding-specific state. */ typedef uint16 hcode_t; /* codes fit in 16 bits */ +typedef struct { + long hash; + hcode_t code; +} hash_t; /* * Decoding-specific state. @@ -125,7 +131,7 @@ typedef int (*decodeFunc)(TIFF*, tidata_t, tsize_t, tsample_t); typedef struct { LZWBaseState base; - /* Decoding specific data */ + /* Decoding specific data */ long dec_nbitsmask; /* lzw_nbits 1 bits, right adjusted */ long dec_restart; /* restart count */ #ifdef LZW_CHECKEOS @@ -137,15 +143,27 @@ typedef struct { code_t* dec_free_entp; /* next free entry */ code_t* dec_maxcodep; /* max available entry */ code_t* dec_codetab; /* kept separate for small machines */ + + /* Encoding specific data */ + int enc_oldcode; /* last code encountered */ + long enc_checkpoint; /* point at which to clear table */ +#define CHECK_GAP 10000 /* enc_ratio check interval */ + long enc_ratio; /* current compression ratio */ + long enc_incount; /* (input) data bytes encoded */ + long enc_outcount; /* encoded (output) bytes */ + tidata_t enc_rawlimit; /* bound on tif_rawdata buffer */ + hash_t* enc_hashtab; /* kept separate for small machines */ } LZWCodecState; #define LZWState(tif) ((LZWBaseState*) (tif)->tif_data) #define DecoderState(tif) ((LZWCodecState*) LZWState(tif)) +#define EncoderState(tif) ((LZWCodecState*) LZWState(tif)) static int LZWDecode(TIFF*, tidata_t, tsize_t, tsample_t); #ifdef LZW_COMPAT static int LZWDecodeCompat(TIFF*, tidata_t, tsize_t, tsample_t); #endif +static void cl_hash(LZWCodecState*); /* * LZW Decoder. @@ -175,9 +193,33 @@ static int LZWSetupDecode(TIFF* tif) { LZWCodecState* sp = DecoderState(tif); - static const char module[] = "LZWSetupDecode"; + static const char module[] = " LZWSetupDecode"; int code; + if( sp == NULL ) + { + /* + * Allocate state block so tag methods have storage to record + * values. + */ + tif->tif_data = (tidata_t) _TIFFmalloc(sizeof(LZWCodecState)); + if (tif->tif_data == NULL) + { + TIFFError("LZWPreDecode", "No space for LZW state block"); + return (0); + } + + DecoderState(tif)->dec_codetab = NULL; + DecoderState(tif)->dec_decode = NULL; + + /* + * Setup predictor setup. + */ + (void) TIFFPredictorInit(tif); + + sp = DecoderState(tif); + } + assert(sp != NULL); if (sp->dec_codetab == NULL) { @@ -191,8 +233,8 @@ LZWSetupDecode(TIFF* tif) */ code = 255; do { - sp->dec_codetab[code].value = (u_char) code; - sp->dec_codetab[code].firstchar = (u_char) code; + sp->dec_codetab[code].value = code; + sp->dec_codetab[code].firstchar = code; sp->dec_codetab[code].length = 1; sp->dec_codetab[code].next = NULL; } while (code--); @@ -347,7 +389,7 @@ LZWDecode(TIFF* tif, tidata_t op0, tsize_t occ0, tsample_t s) --tp; t = codep->value; codep = codep->next; - *tp = (char) t; + *tp = t; } while (--residue && codep); sp->dec_restart = 0; } @@ -373,7 +415,7 @@ LZWDecode(TIFF* tif, tidata_t op0, tsize_t occ0, tsample_t s) NextCode(tif, sp, bp, code, GetNextCode); if (code == CODE_EOI) break; - *op++ = (char)code, occ--; + *op++ = code, occ--; oldcodep = sp->dec_codetab + code; continue; } @@ -451,7 +493,7 @@ LZWDecode(TIFF* tif, tidata_t op0, tsize_t occ0, tsample_t s) --tp; t = codep->value; codep = codep->next; - *tp = (char) t; + *tp = t; } while (codep && tp > op); if (codep) { codeLoop(tif); @@ -459,7 +501,7 @@ LZWDecode(TIFF* tif, tidata_t op0, tsize_t occ0, tsample_t s) } op += len, occ -= len; } else - *op++ = (char)code, occ--; + *op++ = code, occ--; } tif->tif_rawcp = (tidata_t) bp; @@ -569,7 +611,7 @@ LZWDecodeCompat(TIFF* tif, tidata_t op0, tsize_t occ0, tsample_t s) NextCode(tif, sp, bp, code, GetNextCodeCompat); if (code == CODE_EOI) break; - *op++ = (char) code, occ--; + *op++ = code, occ--; oldcodep = sp->dec_codetab + code; continue; } @@ -642,11 +684,11 @@ LZWDecodeCompat(TIFF* tif, tidata_t op0, tsize_t occ0, tsample_t s) *--tp = codep->value; } while( (codep = codep->next) != NULL); } else - *op++ = (char) code, occ--; + *op++ = code, occ--; } tif->tif_rawcp = (tidata_t) bp; - sp->lzw_nbits = (u_short) nbits; + sp->lzw_nbits = nbits; sp->lzw_nextdata = nextdata; sp->lzw_nextbits = nextbits; sp->dec_nbitsmask = nbitsmask; @@ -664,32 +706,319 @@ LZWDecodeCompat(TIFF* tif, tidata_t op0, tsize_t occ0, tsample_t s) } #endif /* LZW_COMPAT */ - - -static void -LZWCleanup(TIFF* tif) -{ - if (tif->tif_data) { - if (DecoderState(tif)->dec_codetab) - _TIFFfree(DecoderState(tif)->dec_codetab); - _TIFFfree(tif->tif_data); - tif->tif_data = NULL; - } -} +/* + * LZW Encoding. + */ static int LZWSetupEncode(TIFF* tif) { - TIFFError(tif->tif_name, - "LZW compression is not available to due to Unisys patent enforcement"); - return (0); + LZWCodecState* sp = EncoderState(tif); + static const char module[] = "LZWSetupEncode"; + + assert(sp != NULL); + sp->enc_hashtab = (hash_t*) _TIFFmalloc(HSIZE*sizeof (hash_t)); + if (sp->enc_hashtab == NULL) { + TIFFError(module, "No space for LZW hash table"); + return (0); + } + return (1); +} + +/* + * Reset encoding state at the start of a strip. + */ +static int +LZWPreEncode(TIFF* tif, tsample_t s) +{ + LZWCodecState *sp = EncoderState(tif); + + (void) s; + assert(sp != NULL); + sp->lzw_nbits = BITS_MIN; + sp->lzw_maxcode = MAXCODE(BITS_MIN); + sp->lzw_free_ent = CODE_FIRST; + sp->lzw_nextbits = 0; + sp->lzw_nextdata = 0; + sp->enc_checkpoint = CHECK_GAP; + sp->enc_ratio = 0; + sp->enc_incount = 0; + sp->enc_outcount = 0; + /* + * The 4 here insures there is space for 2 max-sized + * codes in LZWEncode and LZWPostDecode. + */ + sp->enc_rawlimit = tif->tif_rawdata + tif->tif_rawdatasize-1 - 4; + cl_hash(sp); /* clear hash table */ + sp->enc_oldcode = (hcode_t) -1; /* generates CODE_CLEAR in LZWEncode */ + return (1); +} + +#define CALCRATIO(sp, rat) { \ + if (incount > 0x007fffff) { /* NB: shift will overflow */\ + rat = outcount >> 8; \ + rat = (rat == 0 ? 0x7fffffff : incount/rat); \ + } else \ + rat = (incount<<8) / outcount; \ +} +#define PutNextCode(op, c) { \ + nextdata = (nextdata << nbits) | c; \ + nextbits += nbits; \ + *op++ = (u_char)(nextdata >> (nextbits-8)); \ + nextbits -= 8; \ + if (nextbits >= 8) { \ + *op++ = (u_char)(nextdata >> (nextbits-8)); \ + nextbits -= 8; \ + } \ + outcount += nbits; \ +} + +/* + * Encode a chunk of pixels. + * + * Uses an open addressing double hashing (no chaining) on the + * prefix code/next character combination. We do a variant of + * Knuth's algorithm D (vol. 3, sec. 6.4) along with G. Knott's + * relatively-prime secondary probe. Here, the modular division + * first probe is gives way to a faster exclusive-or manipulation. + * Also do block compression with an adaptive reset, whereby the + * code table is cleared when the compression ratio decreases, + * but after the table fills. The variable-length output codes + * are re-sized at this point, and a CODE_CLEAR is generated + * for the decoder. + */ +static int +LZWEncode(TIFF* tif, tidata_t bp, tsize_t cc, tsample_t s) +{ + register LZWCodecState *sp = EncoderState(tif); + register long fcode; + register hash_t *hp; + register int h, c; + hcode_t ent; + long disp; + long incount, outcount, checkpoint; + long nextdata, nextbits; + int free_ent, maxcode, nbits; + tidata_t op, limit; + + (void) s; + if (sp == NULL) + return (0); + /* + * Load local state. + */ + incount = sp->enc_incount; + outcount = sp->enc_outcount; + checkpoint = sp->enc_checkpoint; + nextdata = sp->lzw_nextdata; + nextbits = sp->lzw_nextbits; + free_ent = sp->lzw_free_ent; + maxcode = sp->lzw_maxcode; + nbits = sp->lzw_nbits; + op = tif->tif_rawcp; + limit = sp->enc_rawlimit; + ent = sp->enc_oldcode; + + if (ent == (hcode_t) -1 && cc > 0) { + /* + * NB: This is safe because it can only happen + * at the start of a strip where we know there + * is space in the data buffer. + */ + PutNextCode(op, CODE_CLEAR); + ent = *bp++; cc--; incount++; + } + while (cc > 0) { + c = *bp++; cc--; incount++; + fcode = ((long)c << BITS_MAX) + ent; + h = (c << HSHIFT) ^ ent; /* xor hashing */ +#ifdef _WINDOWS + /* + * Check hash index for an overflow. + */ + if (h >= HSIZE) + h -= HSIZE; +#endif + hp = &sp->enc_hashtab[h]; + if (hp->hash == fcode) { + ent = hp->code; + continue; + } + if (hp->hash >= 0) { + /* + * Primary hash failed, check secondary hash. + */ + disp = HSIZE - h; + if (h == 0) + disp = 1; + do { + /* + * Avoid pointer arithmetic 'cuz of + * wraparound problems with segments. + */ + if ((h -= disp) < 0) + h += HSIZE; + hp = &sp->enc_hashtab[h]; + if (hp->hash == fcode) { + ent = hp->code; + goto hit; + } + } while (hp->hash >= 0); + } + /* + * New entry, emit code and add to table. + */ + /* + * Verify there is space in the buffer for the code + * and any potential Clear code that might be emitted + * below. The value of limit is setup so that there + * are at least 4 bytes free--room for 2 codes. + */ + if (op > limit) { + tif->tif_rawcc = (tsize_t)(op - tif->tif_rawdata); + TIFFFlushData1(tif); + op = tif->tif_rawdata; + } + PutNextCode(op, ent); + ent = c; + hp->code = free_ent++; + hp->hash = fcode; + if (free_ent == CODE_MAX-1) { + /* table is full, emit clear code and reset */ + cl_hash(sp); + sp->enc_ratio = 0; + incount = 0; + outcount = 0; + free_ent = CODE_FIRST; + PutNextCode(op, CODE_CLEAR); + nbits = BITS_MIN; + maxcode = MAXCODE(BITS_MIN); + } else { + /* + * If the next entry is going to be too big for + * the code size, then increase it, if possible. + */ + if (free_ent > maxcode) { + nbits++; + assert(nbits <= BITS_MAX); + maxcode = (int) MAXCODE(nbits); + } else if (incount >= checkpoint) { + long rat; + /* + * Check compression ratio and, if things seem + * to be slipping, clear the hash table and + * reset state. The compression ratio is a + * 24+8-bit fractional number. + */ + checkpoint = incount+CHECK_GAP; + CALCRATIO(sp, rat); + if (rat <= sp->enc_ratio) { + cl_hash(sp); + sp->enc_ratio = 0; + incount = 0; + outcount = 0; + free_ent = CODE_FIRST; + PutNextCode(op, CODE_CLEAR); + nbits = BITS_MIN; + maxcode = MAXCODE(BITS_MIN); + } else + sp->enc_ratio = rat; + } + } + hit: + ; + } + + /* + * Restore global state. + */ + sp->enc_incount = incount; + sp->enc_outcount = outcount; + sp->enc_checkpoint = checkpoint; + sp->enc_oldcode = ent; + sp->lzw_nextdata = nextdata; + sp->lzw_nextbits = nextbits; + sp->lzw_free_ent = free_ent; + sp->lzw_maxcode = maxcode; + sp->lzw_nbits = nbits; + tif->tif_rawcp = op; + return (1); +} + +/* + * Finish off an encoded strip by flushing the last + * string and tacking on an End Of Information code. + */ +static int +LZWPostEncode(TIFF* tif) +{ + register LZWCodecState *sp = EncoderState(tif); + tidata_t op = tif->tif_rawcp; + long nextbits = sp->lzw_nextbits; + long nextdata = sp->lzw_nextdata; + long outcount = sp->enc_outcount; + int nbits = sp->lzw_nbits; + + if (op > sp->enc_rawlimit) { + tif->tif_rawcc = (tsize_t)(op - tif->tif_rawdata); + TIFFFlushData1(tif); + op = tif->tif_rawdata; + } + if (sp->enc_oldcode != (hcode_t) -1) { + PutNextCode(op, sp->enc_oldcode); + sp->enc_oldcode = (hcode_t) -1; + } + PutNextCode(op, CODE_EOI); + if (nextbits > 0) + *op++ = (u_char)(nextdata << (8-nextbits)); + tif->tif_rawcc = (tsize_t)(op - tif->tif_rawdata); + return (1); +} + +/* + * Reset encoding hash table. + */ +static void +cl_hash(LZWCodecState* sp) +{ + register hash_t *hp = &sp->enc_hashtab[HSIZE-1]; + register long i = HSIZE-8; + + do { + i -= 8; + hp[-7].hash = -1; + hp[-6].hash = -1; + hp[-5].hash = -1; + hp[-4].hash = -1; + hp[-3].hash = -1; + hp[-2].hash = -1; + hp[-1].hash = -1; + hp[ 0].hash = -1; + hp -= 8; + } while (i >= 0); + for (i += 8; i > 0; i--, hp--) + hp->hash = -1; +} + +static void +LZWCleanup(TIFF* tif) +{ + if (tif->tif_data) { + if (DecoderState(tif)->dec_codetab) + _TIFFfree(DecoderState(tif)->dec_codetab); + + if (EncoderState(tif)->enc_hashtab) + _TIFFfree(EncoderState(tif)->enc_hashtab); + + _TIFFfree(tif->tif_data); + tif->tif_data = NULL; + } } int TIFFInitLZW(TIFF* tif, int scheme) { assert(scheme == COMPRESSION_LZW); - /* * Allocate state block so tag methods have storage to record values. */ @@ -698,25 +1027,29 @@ TIFFInitLZW(TIFF* tif, int scheme) goto bad; DecoderState(tif)->dec_codetab = NULL; DecoderState(tif)->dec_decode = NULL; + EncoderState(tif)->enc_hashtab = NULL; + LZWState(tif)->rw_mode = tif->tif_mode; /* * Install codec methods. */ - tif->tif_setupencode = LZWSetupEncode; tif->tif_setupdecode = LZWSetupDecode; tif->tif_predecode = LZWPreDecode; tif->tif_decoderow = LZWDecode; tif->tif_decodestrip = LZWDecode; tif->tif_decodetile = LZWDecode; + tif->tif_setupencode = LZWSetupEncode; + tif->tif_preencode = LZWPreEncode; + tif->tif_postencode = LZWPostEncode; + tif->tif_encoderow = LZWEncode; + tif->tif_encodestrip = LZWEncode; + tif->tif_encodetile = LZWEncode; tif->tif_cleanup = LZWCleanup; - /* * Setup predictor setup. */ (void) TIFFPredictorInit(tif); - return (1); - bad: TIFFError("TIFFInitLZW", "No space for LZW state block"); return (0);