Back to index

tetex-bin  3.0
inftrees.c
Go to the documentation of this file.
00001 /* inftrees.c -- generate Huffman trees for efficient decoding
00002  * Copyright (C) 1995-2003 Mark Adler
00003  * For conditions of distribution and use, see copyright notice in zlib.h
00004  */
00005 
00006 #include "zutil.h"
00007 #include "inftrees.h"
00008 
00009 #define MAXBITS 15
00010 
00011 const char inflate_copyright[] =
00012    " inflate 1.2.1 Copyright 1995-2003 Mark Adler ";
00013 /*
00014   If you use the zlib library in a product, an acknowledgment is welcome
00015   in the documentation of your product. If for some reason you cannot
00016   include such an acknowledgment, I would appreciate that you keep this
00017   copyright string in the executable of your product.
00018  */
00019 
00020 /*
00021    Build a set of tables to decode the provided canonical Huffman code.
00022    The code lengths are lens[0..codes-1].  The result starts at *table,
00023    whose indices are 0..2^bits-1.  work is a writable array of at least
00024    lens shorts, which is used as a work area.  type is the type of code
00025    to be generated, CODES, LENS, or DISTS.  On return, zero is success,
00026    -1 is an invalid code, and +1 means that ENOUGH isn't enough.  table
00027    on return points to the next available entry's address.  bits is the
00028    requested root table index bits, and on return it is the actual root
00029    table index bits.  It will differ if the request is greater than the
00030    longest code or if it is less than the shortest code.
00031  */
00032 int inflate_table(type, lens, codes, table, bits, work)
00033 codetype type;
00034 unsigned short FAR *lens;
00035 unsigned codes;
00036 code FAR * FAR *table;
00037 unsigned FAR *bits;
00038 unsigned short FAR *work;
00039 {
00040     unsigned len;               /* a code's length in bits */
00041     unsigned sym;               /* index of code symbols */
00042     unsigned min, max;          /* minimum and maximum code lengths */
00043     unsigned root;              /* number of index bits for root table */
00044     unsigned curr;              /* number of index bits for current table */
00045     unsigned drop;              /* code bits to drop for sub-table */
00046     int left;                   /* number of prefix codes available */
00047     unsigned used;              /* code entries in table used */
00048     unsigned huff;              /* Huffman code */
00049     unsigned incr;              /* for incrementing code, index */
00050     unsigned fill;              /* index for replicating entries */
00051     unsigned low;               /* low bits for current root entry */
00052     unsigned mask;              /* mask for low root bits */
00053     code this;                  /* table entry for duplication */
00054     code FAR *next;             /* next available space in table */
00055     const unsigned short FAR *base;     /* base value table to use */
00056     const unsigned short FAR *extra;    /* extra bits table to use */
00057     int end;                    /* use base and extra for symbol > end */
00058     unsigned short count[MAXBITS+1];    /* number of codes of each length */
00059     unsigned short offs[MAXBITS+1];     /* offsets in table for each length */
00060     static const unsigned short lbase[31] = { /* Length codes 257..285 base */
00061         3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31,
00062         35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0};
00063     static const unsigned short lext[31] = { /* Length codes 257..285 extra */
00064         16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18,
00065         19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 76, 66};
00066     static const unsigned short dbase[32] = { /* Distance codes 0..29 base */
00067         1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,
00068         257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145,
00069         8193, 12289, 16385, 24577, 0, 0};
00070     static const unsigned short dext[32] = { /* Distance codes 0..29 extra */
00071         16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22,
00072         23, 23, 24, 24, 25, 25, 26, 26, 27, 27,
00073         28, 28, 29, 29, 64, 64};
00074 
00075     /*
00076        Process a set of code lengths to create a canonical Huffman code.  The
00077        code lengths are lens[0..codes-1].  Each length corresponds to the
00078        symbols 0..codes-1.  The Huffman code is generated by first sorting the
00079        symbols by length from short to long, and retaining the symbol order
00080        for codes with equal lengths.  Then the code starts with all zero bits
00081        for the first code of the shortest length, and the codes are integer
00082        increments for the same length, and zeros are appended as the length
00083        increases.  For the deflate format, these bits are stored backwards
00084        from their more natural integer increment ordering, and so when the
00085        decoding tables are built in the large loop below, the integer codes
00086        are incremented backwards.
00087 
00088        This routine assumes, but does not check, that all of the entries in
00089        lens[] are in the range 0..MAXBITS.  The caller must assure this.
00090        1..MAXBITS is interpreted as that code length.  zero means that that
00091        symbol does not occur in this code.
00092 
00093        The codes are sorted by computing a count of codes for each length,
00094        creating from that a table of starting indices for each length in the
00095        sorted table, and then entering the symbols in order in the sorted
00096        table.  The sorted table is work[], with that space being provided by
00097        the caller.
00098 
00099        The length counts are used for other purposes as well, i.e. finding
00100        the minimum and maximum length codes, determining if there are any
00101        codes at all, checking for a valid set of lengths, and looking ahead
00102        at length counts to determine sub-table sizes when building the
00103        decoding tables.
00104      */
00105 
00106     /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */
00107     for (len = 0; len <= MAXBITS; len++)
00108         count[len] = 0;
00109     for (sym = 0; sym < codes; sym++)
00110         count[lens[sym]]++;
00111 
00112     /* bound code lengths, force root to be within code lengths */
00113     root = *bits;
00114     for (max = MAXBITS; max >= 1; max--)
00115         if (count[max] != 0) break;
00116     if (root > max) root = max;
00117     if (max == 0) return -1;            /* no codes! */
00118     for (min = 1; min <= MAXBITS; min++)
00119         if (count[min] != 0) break;
00120     if (root < min) root = min;
00121 
00122     /* check for an over-subscribed or incomplete set of lengths */
00123     left = 1;
00124     for (len = 1; len <= MAXBITS; len++) {
00125         left <<= 1;
00126         left -= count[len];
00127         if (left < 0) return -1;        /* over-subscribed */
00128     }
00129     if (left > 0 && (type == CODES || (codes - count[0] != 1)))
00130         return -1;                      /* incomplete set */
00131 
00132     /* generate offsets into symbol table for each length for sorting */
00133     offs[1] = 0;
00134     for (len = 1; len < MAXBITS; len++)
00135         offs[len + 1] = offs[len] + count[len];
00136 
00137     /* sort symbols by length, by symbol order within each length */
00138     for (sym = 0; sym < codes; sym++)
00139         if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym;
00140 
00141     /*
00142        Create and fill in decoding tables.  In this loop, the table being
00143        filled is at next and has curr index bits.  The code being used is huff
00144        with length len.  That code is converted to an index by dropping drop
00145        bits off of the bottom.  For codes where len is less than drop + curr,
00146        those top drop + curr - len bits are incremented through all values to
00147        fill the table with replicated entries.
00148 
00149        root is the number of index bits for the root table.  When len exceeds
00150        root, sub-tables are created pointed to by the root entry with an index
00151        of the low root bits of huff.  This is saved in low to check for when a
00152        new sub-table should be started.  drop is zero when the root table is
00153        being filled, and drop is root when sub-tables are being filled.
00154 
00155        When a new sub-table is needed, it is necessary to look ahead in the
00156        code lengths to determine what size sub-table is needed.  The length
00157        counts are used for this, and so count[] is decremented as codes are
00158        entered in the tables.
00159 
00160        used keeps track of how many table entries have been allocated from the
00161        provided *table space.  It is checked when a LENS table is being made
00162        against the space in *table, ENOUGH, minus the maximum space needed by
00163        the worst case distance code, MAXD.  This should never happen, but the
00164        sufficiency of ENOUGH has not been proven exhaustively, hence the check.
00165        This assumes that when type == LENS, bits == 9.
00166 
00167        sym increments through all symbols, and the loop terminates when
00168        all codes of length max, i.e. all codes, have been processed.  This
00169        routine permits incomplete codes, so another loop after this one fills
00170        in the rest of the decoding tables with invalid code markers.
00171      */
00172 
00173     /* set up for code type */
00174     switch (type) {
00175     case CODES:
00176         base = extra = work;    /* dummy value--not used */
00177         end = 19;
00178         break;
00179     case LENS:
00180         base = lbase;
00181         base -= 257;
00182         extra = lext;
00183         extra -= 257;
00184         end = 256;
00185         break;
00186     default:            /* DISTS */
00187         base = dbase;
00188         extra = dext;
00189         end = -1;
00190     }
00191 
00192     /* initialize state for loop */
00193     huff = 0;                   /* starting code */
00194     sym = 0;                    /* starting code symbol */
00195     len = min;                  /* starting code length */
00196     next = *table;              /* current table to fill in */
00197     curr = root;                /* current table index bits */
00198     drop = 0;                   /* current bits to drop from code for index */
00199     low = (unsigned)(-1);       /* trigger new sub-table when len > root */
00200     used = 1U << root;          /* use root table entries */
00201     mask = used - 1;            /* mask for comparing low */
00202 
00203     /* check available table space */
00204     if (type == LENS && used >= ENOUGH - MAXD)
00205         return 1;
00206 
00207     /* process all codes and make table entries */
00208     for (;;) {
00209         /* create table entry */
00210         this.bits = (unsigned char)(len - drop);
00211         if ((int)(work[sym]) < end) {
00212             this.op = (unsigned char)0;
00213             this.val = work[sym];
00214         }
00215         else if ((int)(work[sym]) > end) {
00216             this.op = (unsigned char)(extra[work[sym]]);
00217             this.val = base[work[sym]];
00218         }
00219         else {
00220             this.op = (unsigned char)(32 + 64);         /* end of block */
00221             this.val = 0;
00222         }
00223 
00224         /* replicate for those indices with low len bits equal to huff */
00225         incr = 1U << (len - drop);
00226         fill = 1U << curr;
00227         do {
00228             fill -= incr;
00229             next[(huff >> drop) + fill] = this;
00230         } while (fill != 0);
00231 
00232         /* backwards increment the len-bit code huff */
00233         incr = 1U << (len - 1);
00234         while (huff & incr)
00235             incr >>= 1;
00236         if (incr != 0) {
00237             huff &= incr - 1;
00238             huff += incr;
00239         }
00240         else
00241             huff = 0;
00242 
00243         /* go to next symbol, update count, len */
00244         sym++;
00245         if (--(count[len]) == 0) {
00246             if (len == max) break;
00247             len = lens[work[sym]];
00248         }
00249 
00250         /* create new sub-table if needed */
00251         if (len > root && (huff & mask) != low) {
00252             /* if first time, transition to sub-tables */
00253             if (drop == 0)
00254                 drop = root;
00255 
00256             /* increment past last table */
00257             next += 1U << curr;
00258 
00259             /* determine length of next table */
00260             curr = len - drop;
00261             left = (int)(1 << curr);
00262             while (curr + drop < max) {
00263                 left -= count[curr + drop];
00264                 if (left <= 0) break;
00265                 curr++;
00266                 left <<= 1;
00267             }
00268 
00269             /* check for enough space */
00270             used += 1U << curr;
00271             if (type == LENS && used >= ENOUGH - MAXD)
00272                 return 1;
00273 
00274             /* point entry in root table to sub-table */
00275             low = huff & mask;
00276             (*table)[low].op = (unsigned char)curr;
00277             (*table)[low].bits = (unsigned char)root;
00278             (*table)[low].val = (unsigned short)(next - *table);
00279         }
00280     }
00281 
00282     /*
00283        Fill in rest of table for incomplete codes.  This loop is similar to the
00284        loop above in incrementing huff for table indices.  It is assumed that
00285        len is equal to curr + drop, so there is no loop needed to increment
00286        through high index bits.  When the current sub-table is filled, the loop
00287        drops back to the root table to fill in any remaining entries there.
00288      */
00289     this.op = (unsigned char)64;                /* invalid code marker */
00290     this.bits = (unsigned char)(len - drop);
00291     this.val = (unsigned short)0;
00292     while (huff != 0) {
00293         /* when done with sub-table, drop back to root table */
00294         if (drop != 0 && (huff & mask) != low) {
00295             drop = 0;
00296             len = root;
00297             next = *table;
00298             curr = root;
00299             this.bits = (unsigned char)len;
00300         }
00301 
00302         /* put invalid code marker in table */
00303         next[huff >> drop] = this;
00304 
00305         /* backwards increment the len-bit code huff */
00306         incr = 1U << (len - 1);
00307         while (huff & incr)
00308             incr >>= 1;
00309         if (incr != 0) {
00310             huff &= incr - 1;
00311             huff += incr;
00312         }
00313         else
00314             huff = 0;
00315     }
00316 
00317     /* set return parameters */
00318     *table += used;
00319     *bits = root;
00320     return 0;
00321 }