michael@0: /* inftrees.c -- generate Huffman trees for efficient decoding
michael@0:  * Copyright (C) 1995-2010 Mark Adler
michael@0:  * For conditions of distribution and use, see copyright notice in zlib.h
michael@0:  */
michael@0: 
michael@0: #include "zutil.h"
michael@0: #include "inftrees.h"
michael@0: 
michael@0: #define MAXBITS 15
michael@0: 
michael@0: const char inflate_copyright[] =
michael@0:    " inflate 1.2.5 Copyright 1995-2010 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:    Build a set of tables to decode the provided canonical Huffman code.
michael@0:    The code lengths are lens[0..codes-1].  The result starts at *table,
michael@0:    whose indices are 0..2^bits-1.  work is a writable array of at least
michael@0:    lens shorts, which is used as a work area.  type is the type of code
michael@0:    to be generated, CODES, LENS, or DISTS.  On return, zero is success,
michael@0:    -1 is an invalid code, and +1 means that ENOUGH isn't enough.  table
michael@0:    on return points to the next available entry's address.  bits is the
michael@0:    requested root table index bits, and on return it is the actual root
michael@0:    table index bits.  It will differ if the request is greater than the
michael@0:    longest code or if it is less than the shortest code.
michael@0:  */
michael@0: int ZLIB_INTERNAL inflate_table(type, lens, codes, table, bits, work)
michael@0: codetype type;
michael@0: unsigned short FAR *lens;
michael@0: unsigned codes;
michael@0: code FAR * FAR *table;
michael@0: unsigned FAR *bits;
michael@0: unsigned short FAR *work;
michael@0: {
michael@0:     unsigned len;               /* a code's length in bits */
michael@0:     unsigned sym;               /* index of code symbols */
michael@0:     unsigned min, max;          /* minimum and maximum code lengths */
michael@0:     unsigned root;              /* number of index bits for root table */
michael@0:     unsigned curr;              /* number of index bits for current table */
michael@0:     unsigned drop;              /* code bits to drop for sub-table */
michael@0:     int left;                   /* number of prefix codes available */
michael@0:     unsigned used;              /* code entries in table used */
michael@0:     unsigned huff;              /* Huffman code */
michael@0:     unsigned incr;              /* for incrementing code, index */
michael@0:     unsigned fill;              /* index for replicating entries */
michael@0:     unsigned low;               /* low bits for current root entry */
michael@0:     unsigned mask;              /* mask for low root bits */
michael@0:     code here;                  /* table entry for duplication */
michael@0:     code FAR *next;             /* next available space in table */
michael@0:     const unsigned short FAR *base;     /* base value table to use */
michael@0:     const unsigned short FAR *extra;    /* extra bits table to use */
michael@0:     int end;                    /* use base and extra for symbol > end */
michael@0:     unsigned short count[MAXBITS+1];    /* number of codes of each length */
michael@0:     unsigned short offs[MAXBITS+1];     /* offsets in table for each length */
michael@0:     static const unsigned short lbase[31] = { /* Length codes 257..285 base */
michael@0:         3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31,
michael@0:         35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0};
michael@0:     static const unsigned short lext[31] = { /* Length codes 257..285 extra */
michael@0:         16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18,
michael@0:         19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 73, 195};
michael@0:     static const unsigned short dbase[32] = { /* Distance codes 0..29 base */
michael@0:         1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,
michael@0:         257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145,
michael@0:         8193, 12289, 16385, 24577, 0, 0};
michael@0:     static const unsigned short dext[32] = { /* Distance codes 0..29 extra */
michael@0:         16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22,
michael@0:         23, 23, 24, 24, 25, 25, 26, 26, 27, 27,
michael@0:         28, 28, 29, 29, 64, 64};
michael@0: 
michael@0:     /*
michael@0:        Process a set of code lengths to create a canonical Huffman code.  The
michael@0:        code lengths are lens[0..codes-1].  Each length corresponds to the
michael@0:        symbols 0..codes-1.  The Huffman code is generated by first sorting the
michael@0:        symbols by length from short to long, and retaining the symbol order
michael@0:        for codes with equal lengths.  Then the code starts with all zero bits
michael@0:        for the first code of the shortest length, and the codes are integer
michael@0:        increments for the same length, and zeros are appended as the length
michael@0:        increases.  For the deflate format, these bits are stored backwards
michael@0:        from their more natural integer increment ordering, and so when the
michael@0:        decoding tables are built in the large loop below, the integer codes
michael@0:        are incremented backwards.
michael@0: 
michael@0:        This routine assumes, but does not check, that all of the entries in
michael@0:        lens[] are in the range 0..MAXBITS.  The caller must assure this.
michael@0:        1..MAXBITS is interpreted as that code length.  zero means that that
michael@0:        symbol does not occur in this code.
michael@0: 
michael@0:        The codes are sorted by computing a count of codes for each length,
michael@0:        creating from that a table of starting indices for each length in the
michael@0:        sorted table, and then entering the symbols in order in the sorted
michael@0:        table.  The sorted table is work[], with that space being provided by
michael@0:        the caller.
michael@0: 
michael@0:        The length counts are used for other purposes as well, i.e. finding
michael@0:        the minimum and maximum length codes, determining if there are any
michael@0:        codes at all, checking for a valid set of lengths, and looking ahead
michael@0:        at length counts to determine sub-table sizes when building the
michael@0:        decoding tables.
michael@0:      */
michael@0: 
michael@0:     /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */
michael@0:     for (len = 0; len <= MAXBITS; len++)
michael@0:         count[len] = 0;
michael@0:     for (sym = 0; sym < codes; sym++)
michael@0:         count[lens[sym]]++;
michael@0: 
michael@0:     /* bound code lengths, force root to be within code lengths */
michael@0:     root = *bits;
michael@0:     for (max = MAXBITS; max >= 1; max--)
michael@0:         if (count[max] != 0) break;
michael@0:     if (root > max) root = max;
michael@0:     if (max == 0) {                     /* no symbols to code at all */
michael@0:         here.op = (unsigned char)64;    /* invalid code marker */
michael@0:         here.bits = (unsigned char)1;
michael@0:         here.val = (unsigned short)0;
michael@0:         *(*table)++ = here;             /* make a table to force an error */
michael@0:         *(*table)++ = here;
michael@0:         *bits = 1;
michael@0:         return 0;     /* no symbols, but wait for decoding to report error */
michael@0:     }
michael@0:     for (min = 1; min < max; min++)
michael@0:         if (count[min] != 0) break;
michael@0:     if (root < min) root = min;
michael@0: 
michael@0:     /* check for an over-subscribed or incomplete set of lengths */
michael@0:     left = 1;
michael@0:     for (len = 1; len <= MAXBITS; len++) {
michael@0:         left <<= 1;
michael@0:         left -= count[len];
michael@0:         if (left < 0) return -1;        /* over-subscribed */
michael@0:     }
michael@0:     if (left > 0 && (type == CODES || max != 1))
michael@0:         return -1;                      /* incomplete set */
michael@0: 
michael@0:     /* generate offsets into symbol table for each length for sorting */
michael@0:     offs[1] = 0;
michael@0:     for (len = 1; len < MAXBITS; len++)
michael@0:         offs[len + 1] = offs[len] + count[len];
michael@0: 
michael@0:     /* sort symbols by length, by symbol order within each length */
michael@0:     for (sym = 0; sym < codes; sym++)
michael@0:         if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym;
michael@0: 
michael@0:     /*
michael@0:        Create and fill in decoding tables.  In this loop, the table being
michael@0:        filled is at next and has curr index bits.  The code being used is huff
michael@0:        with length len.  That code is converted to an index by dropping drop
michael@0:        bits off of the bottom.  For codes where len is less than drop + curr,
michael@0:        those top drop + curr - len bits are incremented through all values to
michael@0:        fill the table with replicated entries.
michael@0: 
michael@0:        root is the number of index bits for the root table.  When len exceeds
michael@0:        root, sub-tables are created pointed to by the root entry with an index
michael@0:        of the low root bits of huff.  This is saved in low to check for when a
michael@0:        new sub-table should be started.  drop is zero when the root table is
michael@0:        being filled, and drop is root when sub-tables are being filled.
michael@0: 
michael@0:        When a new sub-table is needed, it is necessary to look ahead in the
michael@0:        code lengths to determine what size sub-table is needed.  The length
michael@0:        counts are used for this, and so count[] is decremented as codes are
michael@0:        entered in the tables.
michael@0: 
michael@0:        used keeps track of how many table entries have been allocated from the
michael@0:        provided *table space.  It is checked for LENS and DIST tables against
michael@0:        the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in
michael@0:        the initial root table size constants.  See the comments in inftrees.h
michael@0:        for more information.
michael@0: 
michael@0:        sym increments through all symbols, and the loop terminates when
michael@0:        all codes of length max, i.e. all codes, have been processed.  This
michael@0:        routine permits incomplete codes, so another loop after this one fills
michael@0:        in the rest of the decoding tables with invalid code markers.
michael@0:      */
michael@0: 
michael@0:     /* set up for code type */
michael@0:     switch (type) {
michael@0:     case CODES:
michael@0:         base = extra = work;    /* dummy value--not used */
michael@0:         end = 19;
michael@0:         break;
michael@0:     case LENS:
michael@0:         base = lbase;
michael@0:         base -= 257;
michael@0:         extra = lext;
michael@0:         extra -= 257;
michael@0:         end = 256;
michael@0:         break;
michael@0:     default:            /* DISTS */
michael@0:         base = dbase;
michael@0:         extra = dext;
michael@0:         end = -1;
michael@0:     }
michael@0: 
michael@0:     /* initialize state for loop */
michael@0:     huff = 0;                   /* starting code */
michael@0:     sym = 0;                    /* starting code symbol */
michael@0:     len = min;                  /* starting code length */
michael@0:     next = *table;              /* current table to fill in */
michael@0:     curr = root;                /* current table index bits */
michael@0:     drop = 0;                   /* current bits to drop from code for index */
michael@0:     low = (unsigned)(-1);       /* trigger new sub-table when len > root */
michael@0:     used = 1U << root;          /* use root table entries */
michael@0:     mask = used - 1;            /* mask for comparing low */
michael@0: 
michael@0:     /* check available table space */
michael@0:     if ((type == LENS && used >= ENOUGH_LENS) ||
michael@0:         (type == DISTS && used >= ENOUGH_DISTS))
michael@0:         return 1;
michael@0: 
michael@0:     /* process all codes and make table entries */
michael@0:     for (;;) {
michael@0:         /* create table entry */
michael@0:         here.bits = (unsigned char)(len - drop);
michael@0:         if ((int)(work[sym]) < end) {
michael@0:             here.op = (unsigned char)0;
michael@0:             here.val = work[sym];
michael@0:         }
michael@0:         else if ((int)(work[sym]) > end) {
michael@0:             here.op = (unsigned char)(extra[work[sym]]);
michael@0:             here.val = base[work[sym]];
michael@0:         }
michael@0:         else {
michael@0:             here.op = (unsigned char)(32 + 64);         /* end of block */
michael@0:             here.val = 0;
michael@0:         }
michael@0: 
michael@0:         /* replicate for those indices with low len bits equal to huff */
michael@0:         incr = 1U << (len - drop);
michael@0:         fill = 1U << curr;
michael@0:         min = fill;                 /* save offset to next table */
michael@0:         do {
michael@0:             fill -= incr;
michael@0:             next[(huff >> drop) + fill] = here;
michael@0:         } while (fill != 0);
michael@0: 
michael@0:         /* backwards increment the len-bit code huff */
michael@0:         incr = 1U << (len - 1);
michael@0:         while (huff & incr)
michael@0:             incr >>= 1;
michael@0:         if (incr != 0) {
michael@0:             huff &= incr - 1;
michael@0:             huff += incr;
michael@0:         }
michael@0:         else
michael@0:             huff = 0;
michael@0: 
michael@0:         /* go to next symbol, update count, len */
michael@0:         sym++;
michael@0:         if (--(count[len]) == 0) {
michael@0:             if (len == max) break;
michael@0:             len = lens[work[sym]];
michael@0:         }
michael@0: 
michael@0:         /* create new sub-table if needed */
michael@0:         if (len > root && (huff & mask) != low) {
michael@0:             /* if first time, transition to sub-tables */
michael@0:             if (drop == 0)
michael@0:                 drop = root;
michael@0: 
michael@0:             /* increment past last table */
michael@0:             next += min;            /* here min is 1 << curr */
michael@0: 
michael@0:             /* determine length of next table */
michael@0:             curr = len - drop;
michael@0:             left = (int)(1 << curr);
michael@0:             while (curr + drop < max) {
michael@0:                 left -= count[curr + drop];
michael@0:                 if (left <= 0) break;
michael@0:                 curr++;
michael@0:                 left <<= 1;
michael@0:             }
michael@0: 
michael@0:             /* check for enough space */
michael@0:             used += 1U << curr;
michael@0:             if ((type == LENS && used >= ENOUGH_LENS) ||
michael@0:                 (type == DISTS && used >= ENOUGH_DISTS))
michael@0:                 return 1;
michael@0: 
michael@0:             /* point entry in root table to sub-table */
michael@0:             low = huff & mask;
michael@0:             (*table)[low].op = (unsigned char)curr;
michael@0:             (*table)[low].bits = (unsigned char)root;
michael@0:             (*table)[low].val = (unsigned short)(next - *table);
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     /*
michael@0:        Fill in rest of table for incomplete codes.  This loop is similar to the
michael@0:        loop above in incrementing huff for table indices.  It is assumed that
michael@0:        len is equal to curr + drop, so there is no loop needed to increment
michael@0:        through high index bits.  When the current sub-table is filled, the loop
michael@0:        drops back to the root table to fill in any remaining entries there.
michael@0:      */
michael@0:     here.op = (unsigned char)64;                /* invalid code marker */
michael@0:     here.bits = (unsigned char)(len - drop);
michael@0:     here.val = (unsigned short)0;
michael@0:     while (huff != 0) {
michael@0:         /* when done with sub-table, drop back to root table */
michael@0:         if (drop != 0 && (huff & mask) != low) {
michael@0:             drop = 0;
michael@0:             len = root;
michael@0:             next = *table;
michael@0:             here.bits = (unsigned char)len;
michael@0:         }
michael@0: 
michael@0:         /* put invalid code marker in table */
michael@0:         next[huff >> drop] = here;
michael@0: 
michael@0:         /* backwards increment the len-bit code huff */
michael@0:         incr = 1U << (len - 1);
michael@0:         while (huff & incr)
michael@0:             incr >>= 1;
michael@0:         if (incr != 0) {
michael@0:             huff &= incr - 1;
michael@0:             huff += incr;
michael@0:         }
michael@0:         else
michael@0:             huff = 0;
michael@0:     }
michael@0: 
michael@0:     /* set return parameters */
michael@0:     *table += used;
michael@0:     *bits = root;
michael@0:     return 0;
michael@0: }