common/dist/zlib/contrib/infback9/inftree9.c

   1 /*      $NetBSD: inftree9.c,v 1.1.1.1 2006/01/14 20:10:52 christos Exp $        */
   2
   3 /* inftree9.c -- generate Huffman trees for efficient decoding
   4  * Copyright (C) 1995-2005 Mark Adler
   5  * For conditions of distribution and use, see copyright notice in zlib.h
   6  */
   7
   8 #include "zutil.h"
   9 #include "inftree9.h"
  10
  11 #define MAXBITS 15
  12
  13 const char inflate9_copyright[] =
  14    " inflate9 1.2.3 Copyright 1995-2005 Mark Adler ";
  15 /*
  16   If you use the zlib library in a product, an acknowledgment is welcome
  17   in the documentation of your product. If for some reason you cannot
  18   include such an acknowledgment, I would appreciate that you keep this
  19   copyright string in the executable of your product.
  20  */
  21
  22 /*
  23    Build a set of tables to decode the provided canonical Huffman code.
  24    The code lengths are lens[0..codes-1].  The result starts at *table,
  25    whose indices are 0..2^bits-1.  work is a writable array of at least
  26    lens shorts, which is used as a work area.  type is the type of code
  27    to be generated, CODES, LENS, or DISTS.  On return, zero is success,
  28    -1 is an invalid code, and +1 means that ENOUGH isn't enough.  table
  29    on return points to the next available entry's address.  bits is the
  30    requested root table index bits, and on return it is the actual root
  31    table index bits.  It will differ if the request is greater than the
  32    longest code or if it is less than the shortest code.
  33  */
  34 int inflate_table9(type, lens, codes, table, bits, work)
  35 codetype type;
  36 unsigned short FAR *lens;
  37 unsigned codes;
  38 code FAR * FAR *table;
  39 unsigned FAR *bits;
  40 unsigned short FAR *work;
  41 {
  42     unsigned len;               /* a code's length in bits */
  43     unsigned sym;               /* index of code symbols */
  44     unsigned min, max;          /* minimum and maximum code lengths */
  45     unsigned root;              /* number of index bits for root table */
  46     unsigned curr;              /* number of index bits for current table */
  47     unsigned drop;              /* code bits to drop for sub-table */
  48     int left;                   /* number of prefix codes available */
  49     unsigned used;              /* code entries in table used */
  50     unsigned huff;              /* Huffman code */
  51     unsigned incr;              /* for incrementing code, index */
  52     unsigned fill;              /* index for replicating entries */
  53     unsigned low;               /* low bits for current root entry */
  54     unsigned mask;              /* mask for low root bits */
  55     code this;                  /* table entry for duplication */
  56     code FAR *next;             /* next available space in table */
  57     const unsigned short FAR *base;     /* base value table to use */
  58     const unsigned short FAR *extra;    /* extra bits table to use */
  59     int end;                    /* use base and extra for symbol > end */
  60     unsigned short count[MAXBITS+1];    /* number of codes of each length */
  61     unsigned short offs[MAXBITS+1];     /* offsets in table for each length */
  62     static const unsigned short lbase[31] = { /* Length codes 257..285 base */
  63         3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17,
  64         19, 23, 27, 31, 35, 43, 51, 59, 67, 83, 99, 115,
  65         131, 163, 195, 227, 3, 0, 0};
  66     static const unsigned short lext[31] = { /* Length codes 257..285 extra */
  67         128, 128, 128, 128, 128, 128, 128, 128, 129, 129, 129, 129,
  68         130, 130, 130, 130, 131, 131, 131, 131, 132, 132, 132, 132,
  69         133, 133, 133, 133, 144, 201, 196};
  70     static const unsigned short dbase[32] = { /* Distance codes 0..31 base */
  71         1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49,
  72         65, 97, 129, 193, 257, 385, 513, 769, 1025, 1537, 2049, 3073,
  73         4097, 6145, 8193, 12289, 16385, 24577, 32769, 49153};
  74     static const unsigned short dext[32] = { /* Distance codes 0..31 extra */
  75         128, 128, 128, 128, 129, 129, 130, 130, 131, 131, 132, 132,
  76         133, 133, 134, 134, 135, 135, 136, 136, 137, 137, 138, 138,
  77         139, 139, 140, 140, 141, 141, 142, 142};
  78
  79     /*
  80        Process a set of code lengths to create a canonical Huffman code.  The
  81        code lengths are lens[0..codes-1].  Each length corresponds to the
  82        symbols 0..codes-1.  The Huffman code is generated by first sorting the
  83        symbols by length from short to long, and retaining the symbol order
  84        for codes with equal lengths.  Then the code starts with all zero bits
  85        for the first code of the shortest length, and the codes are integer
  86        increments for the same length, and zeros are appended as the length
  87        increases.  For the deflate format, these bits are stored backwards
  88        from their more natural integer increment ordering, and so when the
  89        decoding tables are built in the large loop below, the integer codes
  90        are incremented backwards.
  91
  92        This routine assumes, but does not check, that all of the entries in
  93        lens[] are in the range 0..MAXBITS.  The caller must assure this.
  94        1..MAXBITS is interpreted as that code length.  zero means that that
  95        symbol does not occur in this code.
  96
  97        The codes are sorted by computing a count of codes for each length,
  98        creating from that a table of starting indices for each length in the
  99        sorted table, and then entering the symbols in order in the sorted
 100        table.  The sorted table is work[], with that space being provided by
 101        the caller.
 102
 103        The length counts are used for other purposes as well, i.e. finding
 104        the minimum and maximum length codes, determining if there are any
 105        codes at all, checking for a valid set of lengths, and looking ahead
 106        at length counts to determine sub-table sizes when building the
 107        decoding tables.
 108      */
 109
 110     /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */
 111     for (len = 0; len <= MAXBITS; len++)
 112         count[len] = 0;
 113     for (sym = 0; sym < codes; sym++)
 114         count[lens[sym]]++;
 115
 116     /* bound code lengths, force root to be within code lengths */
 117     root = *bits;
 118     for (max = MAXBITS; max >= 1; max--)
 119         if (count[max] != 0) break;
 120     if (root > max) root = max;
 121     if (max == 0) return -1;            /* no codes! */
 122     for (min = 1; min <= MAXBITS; min++)
 123         if (count[min] != 0) break;
 124     if (root < min) root = min;
 125
 126     /* check for an over-subscribed or incomplete set of lengths */
 127     left = 1;
 128     for (len = 1; len <= MAXBITS; len++) {
 129         left <<= 1;
 130         left -= count[len];
 131         if (left < 0) return -1;        /* over-subscribed */
 132     }
 133     if (left > 0 && (type == CODES || max != 1))
 134         return -1;                      /* incomplete set */
 135
 136     /* generate offsets into symbol table for each length for sorting */
 137     offs[1] = 0;
 138     for (len = 1; len < MAXBITS; len++)
 139         offs[len + 1] = offs[len] + count[len];
 140
 141     /* sort symbols by length, by symbol order within each length */
 142     for (sym = 0; sym < codes; sym++)
 143         if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym;
 144
 145     /*
 146        Create and fill in decoding tables.  In this loop, the table being
 147        filled is at next and has curr index bits.  The code being used is huff
 148        with length len.  That code is converted to an index by dropping drop
 149        bits off of the bottom.  For codes where len is less than drop + curr,
 150        those top drop + curr - len bits are incremented through all values to
 151        fill the table with replicated entries.
 152
 153        root is the number of index bits for the root table.  When len exceeds
 154        root, sub-tables are created pointed to by the root entry with an index
 155        of the low root bits of huff.  This is saved in low to check for when a
 156        new sub-table should be started.  drop is zero when the root table is
 157        being filled, and drop is root when sub-tables are being filled.
 158
 159        When a new sub-table is needed, it is necessary to look ahead in the
 160        code lengths to determine what size sub-table is needed.  The length
 161        counts are used for this, and so count[] is decremented as codes are
 162        entered in the tables.
 163
 164        used keeps track of how many table entries have been allocated from the
 165        provided *table space.  It is checked when a LENS table is being made
 166        against the space in *table, ENOUGH, minus the maximum space needed by
 167        the worst case distance code, MAXD.  This should never happen, but the
 168        sufficiency of ENOUGH has not been proven exhaustively, hence the check.
 169        This assumes that when type == LENS, bits == 9.
 170
 171        sym increments through all symbols, and the loop terminates when
 172        all codes of length max, i.e. all codes, have been processed.  This
 173        routine permits incomplete codes, so another loop after this one fills
 174        in the rest of the decoding tables with invalid code markers.
 175      */
 176
 177     /* set up for code type */
 178     switch (type) {
 179     case CODES:
 180         base = extra = work;    /* dummy value--not used */
 181         end = 19;
 182         break;
 183     case LENS:
 184         base = lbase;
 185         base -= 257;
 186         extra = lext;
 187         extra -= 257;
 188         end = 256;
 189         break;
 190     default:            /* DISTS */
 191         base = dbase;
 192         extra = dext;
 193         end = -1;
 194     }
 195
 196     /* initialize state for loop */
 197     huff = 0;                   /* starting code */
 198     sym = 0;                    /* starting code symbol */
 199     len = min;                  /* starting code length */
 200     next = *table;              /* current table to fill in */
 201     curr = root;                /* current table index bits */
 202     drop = 0;                   /* current bits to drop from code for index */
 203     low = (unsigned)(-1);       /* trigger new sub-table when len > root */
 204     used = 1U << root;          /* use root table entries */
 205     mask = used - 1;            /* mask for comparing low */
 206
 207     /* check available table space */
 208     if (type == LENS && used >= ENOUGH - MAXD)
 209         return 1;
 210
 211     /* process all codes and make table entries */
 212     for (;;) {
 213         /* create table entry */
 214         this.bits = (unsigned char)(len - drop);
 215         if ((int)(work[sym]) < end) {
 216             this.op = (unsigned char)0;
 217             this.val = work[sym];
 218         }
 219         else if ((int)(work[sym]) > end) {
 220             this.op = (unsigned char)(extra[work[sym]]);
 221             this.val = base[work[sym]];
 222         }
 223         else {
 224             this.op = (unsigned char)(32 + 64);         /* end of block */
 225             this.val = 0;
 226         }
 227
 228         /* replicate for those indices with low len bits equal to huff */
 229         incr = 1U << (len - drop);
 230         fill = 1U << curr;
 231         do {
 232             fill -= incr;
 233             next[(huff >> drop) + fill] = this;
 234         } while (fill != 0);
 235
 236         /* backwards increment the len-bit code huff */
 237         incr = 1U << (len - 1);
 238         while (huff & incr)
 239             incr >>= 1;
 240         if (incr != 0) {
 241             huff &= incr - 1;
 242             huff += incr;
 243         }
 244         else
 245             huff = 0;
 246
 247         /* go to next symbol, update count, len */
 248         sym++;
 249         if (--(count[len]) == 0) {
 250             if (len == max) break;
 251             len = lens[work[sym]];
 252         }
 253
 254         /* create new sub-table if needed */
 255         if (len > root && (huff & mask) != low) {
 256             /* if first time, transition to sub-tables */
 257             if (drop == 0)
 258                 drop = root;
 259
 260             /* increment past last table */
 261             next += 1U << curr;
 262
 263             /* determine length of next table */
 264             curr = len - drop;
 265             left = (int)(1 << curr);
 266             while (curr + drop < max) {
 267                 left -= count[curr + drop];
 268                 if (left <= 0) break;
 269                 curr++;
 270                 left <<= 1;
 271             }
 272
 273             /* check for enough space */
 274             used += 1U << curr;
 275             if (type == LENS && used >= ENOUGH - MAXD)
 276                 return 1;
 277
 278             /* point entry in root table to sub-table */
 279             low = huff & mask;
 280             (*table)[low].op = (unsigned char)curr;
 281             (*table)[low].bits = (unsigned char)root;
 282             (*table)[low].val = (unsigned short)(next - *table);
 283         }
 284     }
 285
 286     /*
 287        Fill in rest of table for incomplete codes.  This loop is similar to the
 288        loop above in incrementing huff for table indices.  It is assumed that
 289        len is equal to curr + drop, so there is no loop needed to increment
 290        through high index bits.  When the current sub-table is filled, the loop
 291        drops back to the root table to fill in any remaining entries there.
 292      */
 293     this.op = (unsigned char)64;                /* invalid code marker */
 294     this.bits = (unsigned char)(len - drop);
 295     this.val = (unsigned short)0;
 296     while (huff != 0) {
 297         /* when done with sub-table, drop back to root table */
 298         if (drop != 0 && (huff & mask) != low) {
 299             drop = 0;
 300             len = root;
 301             next = *table;
 302             curr = root;
 303             this.bits = (unsigned char)len;
 304         }
 305
 306         /* put invalid code marker in table */
 307         next[huff >> drop] = this;
 308
 309         /* backwards increment the len-bit code huff */
 310         incr = 1U << (len - 1);
 311         while (huff & incr)
 312             incr >>= 1;
 313         if (incr != 0) {
 314             huff &= incr - 1;
 315             huff += incr;
 316         }
 317         else
 318             huff = 0;
 319     }
 320
 321     /* set return parameters */
 322     *table += used;
 323     *bits = root;
 324     return 0;
 325 }