libkern/zlib/inftrees.c

   1 /*
   2  * Copyright (c) 2008 Apple Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28 /* inftrees.c -- generate Huffman trees for efficient decoding
  29  * Copyright (C) 1995-2005 Mark Adler
  30  * For conditions of distribution and use, see copyright notice in zlib.h
  31  */
  32
  33 #include "zutil.h"
  34 #include "inftrees.h"
  35
  36 #define MAXBITS 15
  37
  38 const char inflate_copyright[] =
  39    " inflate 1.2.3 Copyright 1995-2005 Mark Adler ";
  40 /*
  41   If you use the zlib library in a product, an acknowledgment is welcome
  42   in the documentation of your product. If for some reason you cannot
  43   include such an acknowledgment, I would appreciate that you keep this
  44   copyright string in the executable of your product.
  45  */
  46
  47 /*
  48    Build a set of tables to decode the provided canonical Huffman code.
  49    The code lengths are lens[0..codes-1].  The result starts at *table,
  50    whose indices are 0..2^bits-1.  work is a writable array of at least
  51    lens shorts, which is used as a work area.  type is the type of code
  52    to be generated, CODES, LENS, or DISTS.  On return, zero is success,
  53    -1 is an invalid code, and +1 means that ENOUGH isn't enough.  table
  54    on return points to the next available entry's address.  bits is the
  55    requested root table index bits, and on return it is the actual root
  56    table index bits.  It will differ if the request is greater than the
  57    longest code or if it is less than the shortest code.
  58  */
  59 int inflate_table(type, lens, codes, table, bits, work)
  60 codetype type;
  61 unsigned short FAR *lens;
  62 unsigned codes;
  63 code FAR * FAR *table;
  64 unsigned FAR *bits;
  65 unsigned short FAR *work;
  66 {
  67     unsigned len;               /* a code's length in bits */
  68     unsigned sym;               /* index of code symbols */
  69     unsigned min, max;          /* minimum and maximum code lengths */
  70     unsigned root;              /* number of index bits for root table */
  71     unsigned curr;              /* number of index bits for current table */
  72     unsigned drop;              /* code bits to drop for sub-table */
  73     int left;                   /* number of prefix codes available */
  74     unsigned used;              /* code entries in table used */
  75     unsigned huff;              /* Huffman code */
  76     unsigned incr;              /* for incrementing code, index */
  77     unsigned fill;              /* index for replicating entries */
  78     unsigned low;               /* low bits for current root entry */
  79     unsigned mask;              /* mask for low root bits */
  80     code this;                  /* table entry for duplication */
  81     code FAR *next;             /* next available space in table */
  82     const unsigned short FAR *base;     /* base value table to use */
  83     const unsigned short FAR *extra;    /* extra bits table to use */
  84     int end;                    /* use base and extra for symbol > end */
  85     unsigned short count[MAXBITS+1];    /* number of codes of each length */
  86     unsigned short offs[MAXBITS+1];     /* offsets in table for each length */
  87     static const unsigned short lbase[31] = { /* Length codes 257..285 base */
  88         3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31,
  89         35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0};
  90     static const unsigned short lext[31] = { /* Length codes 257..285 extra */
  91         16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18,
  92         19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 201, 196};
  93     static const unsigned short dbase[32] = { /* Distance codes 0..29 base */
  94         1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,
  95         257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145,
  96         8193, 12289, 16385, 24577, 0, 0};
  97     static const unsigned short dext[32] = { /* Distance codes 0..29 extra */
  98         16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22,
  99         23, 23, 24, 24, 25, 25, 26, 26, 27, 27,
 100         28, 28, 29, 29, 64, 64};
 101
 102     /*
 103        Process a set of code lengths to create a canonical Huffman code.  The
 104        code lengths are lens[0..codes-1].  Each length corresponds to the
 105        symbols 0..codes-1.  The Huffman code is generated by first sorting the
 106        symbols by length from short to long, and retaining the symbol order
 107        for codes with equal lengths.  Then the code starts with all zero bits
 108        for the first code of the shortest length, and the codes are integer
 109        increments for the same length, and zeros are appended as the length
 110        increases.  For the deflate format, these bits are stored backwards
 111        from their more natural integer increment ordering, and so when the
 112        decoding tables are built in the large loop below, the integer codes
 113        are incremented backwards.
 114
 115        This routine assumes, but does not check, that all of the entries in
 116        lens[] are in the range 0..MAXBITS.  The caller must assure this.
 117        1..MAXBITS is interpreted as that code length.  zero means that that
 118        symbol does not occur in this code.
 119
 120        The codes are sorted by computing a count of codes for each length,
 121        creating from that a table of starting indices for each length in the
 122        sorted table, and then entering the symbols in order in the sorted
 123        table.  The sorted table is work[], with that space being provided by
 124        the caller.
 125
 126        The length counts are used for other purposes as well, i.e. finding
 127        the minimum and maximum length codes, determining if there are any
 128        codes at all, checking for a valid set of lengths, and looking ahead
 129        at length counts to determine sub-table sizes when building the
 130        decoding tables.
 131      */
 132
 133     /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */
 134     for (len = 0; len <= MAXBITS; len++)
 135         count[len] = 0;
 136     for (sym = 0; sym < codes; sym++)
 137         count[lens[sym]]++;
 138
 139     /* bound code lengths, force root to be within code lengths */
 140     root = *bits;
 141     for (max = MAXBITS; max >= 1; max--)
 142         if (count[max] != 0) break;
 143     if (root > max) root = max;
 144     if (max == 0) {                     /* no symbols to code at all */
 145         this.op = (unsigned char)64;    /* invalid code marker */
 146         this.bits = (unsigned char)1;
 147         this.val = (unsigned short)0;
 148         *(*table)++ = this;             /* make a table to force an error */
 149         *(*table)++ = this;
 150         *bits = 1;
 151         return 0;     /* no symbols, but wait for decoding to report error */
 152     }
 153     for (min = 1; min <= MAXBITS; min++)
 154         if (count[min] != 0) break;
 155     if (root < min) root = min;
 156
 157     /* check for an over-subscribed or incomplete set of lengths */
 158     left = 1;
 159     for (len = 1; len <= MAXBITS; len++) {
 160         left <<= 1;
 161         left -= count[len];
 162         if (left < 0) return -1;        /* over-subscribed */
 163     }
 164     if (left > 0 && (type == CODES || max != 1))
 165         return -1;                      /* incomplete set */
 166
 167     /* generate offsets into symbol table for each length for sorting */
 168     offs[1] = 0;
 169     for (len = 1; len < MAXBITS; len++)
 170         offs[len + 1] = offs[len] + count[len];
 171
 172     /* sort symbols by length, by symbol order within each length */
 173     for (sym = 0; sym < codes; sym++)
 174         if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym;
 175
 176     /*
 177        Create and fill in decoding tables.  In this loop, the table being
 178        filled is at next and has curr index bits.  The code being used is huff
 179        with length len.  That code is converted to an index by dropping drop
 180        bits off of the bottom.  For codes where len is less than drop + curr,
 181        those top drop + curr - len bits are incremented through all values to
 182        fill the table with replicated entries.
 183
 184        root is the number of index bits for the root table.  When len exceeds
 185        root, sub-tables are created pointed to by the root entry with an index
 186        of the low root bits of huff.  This is saved in low to check for when a
 187        new sub-table should be started.  drop is zero when the root table is
 188        being filled, and drop is root when sub-tables are being filled.
 189
 190        When a new sub-table is needed, it is necessary to look ahead in the
 191        code lengths to determine what size sub-table is needed.  The length
 192        counts are used for this, and so count[] is decremented as codes are
 193        entered in the tables.
 194
 195        used keeps track of how many table entries have been allocated from the
 196        provided *table space.  It is checked when a LENS table is being made
 197        against the space in *table, ENOUGH, minus the maximum space needed by
 198        the worst case distance code, MAXD.  This should never happen, but the
 199        sufficiency of ENOUGH has not been proven exhaustively, hence the check.
 200        This assumes that when type == LENS, bits == 9.
 201
 202        sym increments through all symbols, and the loop terminates when
 203        all codes of length max, i.e. all codes, have been processed.  This
 204        routine permits incomplete codes, so another loop after this one fills
 205        in the rest of the decoding tables with invalid code markers.
 206      */
 207
 208     /* set up for code type */
 209     switch (type) {
 210     case CODES:
 211         base = extra = work;    /* dummy value--not used */
 212         end = 19;
 213         break;
 214     case LENS:
 215         base = lbase;
 216         base -= 257;
 217         extra = lext;
 218         extra -= 257;
 219         end = 256;
 220         break;
 221     default:            /* DISTS */
 222         base = dbase;
 223         extra = dext;
 224         end = -1;
 225     }
 226
 227     /* initialize state for loop */
 228     huff = 0;                   /* starting code */
 229     sym = 0;                    /* starting code symbol */
 230     len = min;                  /* starting code length */
 231     next = *table;              /* current table to fill in */
 232     curr = root;                /* current table index bits */
 233     drop = 0;                   /* current bits to drop from code for index */
 234     low = (unsigned)(-1);       /* trigger new sub-table when len > root */
 235     used = 1U << root;          /* use root table entries */
 236     mask = used - 1;            /* mask for comparing low */
 237
 238     /* check available table space */
 239     if (type == LENS && used >= ENOUGH - MAXD)
 240         return 1;
 241
 242     /* process all codes and make table entries */
 243     for (;;) {
 244         /* create table entry */
 245         this.bits = (unsigned char)(len - drop);
 246         if ((int)(work[sym]) < end) {
 247             this.op = (unsigned char)0;
 248             this.val = work[sym];
 249         }
 250         else if ((int)(work[sym]) > end) {
 251             this.op = (unsigned char)(extra[work[sym]]);
 252             this.val = base[work[sym]];
 253         }
 254         else {
 255             this.op = (unsigned char)(32 + 64);         /* end of block */
 256             this.val = 0;
 257         }
 258
 259         /* replicate for those indices with low len bits equal to huff */
 260         incr = 1U << (len - drop);
 261         fill = 1U << curr;
 262         min = fill;                 /* save offset to next table */
 263         do {
 264             fill -= incr;
 265             next[(huff >> drop) + fill] = this;
 266         } while (fill != 0);
 267
 268         /* backwards increment the len-bit code huff */
 269         incr = 1U << (len - 1);
 270         while (huff & incr)
 271             incr >>= 1;
 272         if (incr != 0) {
 273             huff &= incr - 1;
 274             huff += incr;
 275         }
 276         else
 277             huff = 0;
 278
 279         /* go to next symbol, update count, len */
 280         sym++;
 281         if (--(count[len]) == 0) {
 282             if (len == max) break;
 283             len = lens[work[sym]];
 284         }
 285
 286         /* create new sub-table if needed */
 287         if (len > root && (huff & mask) != low) {
 288             /* if first time, transition to sub-tables */
 289             if (drop == 0)
 290                 drop = root;
 291
 292             /* increment past last table */
 293             next += min;            /* here min is 1 << curr */
 294
 295             /* determine length of next table */
 296             curr = len - drop;
 297             left = (int)(1 << curr);
 298             while (curr + drop < max) {
 299                 left -= count[curr + drop];
 300                 if (left <= 0) break;
 301                 curr++;
 302                 left <<= 1;
 303             }
 304
 305             /* check for enough space */
 306             used += 1U << curr;
 307             if (type == LENS && used >= ENOUGH - MAXD)
 308                 return 1;
 309
 310             /* point entry in root table to sub-table */
 311             low = huff & mask;
 312             (*table)[low].op = (unsigned char)curr;
 313             (*table)[low].bits = (unsigned char)root;
 314             (*table)[low].val = (unsigned short)(next - *table);
 315         }
 316     }
 317
 318     /*
 319        Fill in rest of table for incomplete codes.  This loop is similar to the
 320        loop above in incrementing huff for table indices.  It is assumed that
 321        len is equal to curr + drop, so there is no loop needed to increment
 322        through high index bits.  When the current sub-table is filled, the loop
 323        drops back to the root table to fill in any remaining entries there.
 324      */
 325     this.op = (unsigned char)64;                /* invalid code marker */
 326     this.bits = (unsigned char)(len - drop);
 327     this.val = (unsigned short)0;
 328     while (huff != 0) {
 329         /* when done with sub-table, drop back to root table */
 330         if (drop != 0 && (huff & mask) != low) {
 331             drop = 0;
 332             len = root;
 333             next = *table;
 334             this.bits = (unsigned char)len;
 335         }
 336
 337         /* put invalid code marker in table */
 338         next[huff >> drop] = this;
 339
 340         /* backwards increment the len-bit code huff */
 341         incr = 1U << (len - 1);
 342         while (huff & incr)
 343             incr >>= 1;
 344         if (incr != 0) {
 345             huff &= incr - 1;
 346             huff += incr;
 347         }
 348         else
 349             huff = 0;
 350     }
 351
 352     /* set return parameters */
 353     *table += used;
 354     *bits = root;
 355     return 0;
 356 }