cannam@89: /* inftree9.c -- generate Huffman trees for efficient decoding
cannam@89:  * Copyright (C) 1995-2012 Mark Adler
cannam@89:  * For conditions of distribution and use, see copyright notice in zlib.h
cannam@89:  */
cannam@89: 
cannam@89: #include "zutil.h"
cannam@89: #include "inftree9.h"
cannam@89: 
cannam@89: #define MAXBITS 15
cannam@89: 
cannam@89: const char inflate9_copyright[] =
cannam@89:    " inflate9 1.2.7 Copyright 1995-2012 Mark Adler ";
cannam@89: /*
cannam@89:   If you use the zlib library in a product, an acknowledgment is welcome
cannam@89:   in the documentation of your product. If for some reason you cannot
cannam@89:   include such an acknowledgment, I would appreciate that you keep this
cannam@89:   copyright string in the executable of your product.
cannam@89:  */
cannam@89: 
cannam@89: /*
cannam@89:    Build a set of tables to decode the provided canonical Huffman code.
cannam@89:    The code lengths are lens[0..codes-1].  The result starts at *table,
cannam@89:    whose indices are 0..2^bits-1.  work is a writable array of at least
cannam@89:    lens shorts, which is used as a work area.  type is the type of code
cannam@89:    to be generated, CODES, LENS, or DISTS.  On return, zero is success,
cannam@89:    -1 is an invalid code, and +1 means that ENOUGH isn't enough.  table
cannam@89:    on return points to the next available entry's address.  bits is the
cannam@89:    requested root table index bits, and on return it is the actual root
cannam@89:    table index bits.  It will differ if the request is greater than the
cannam@89:    longest code or if it is less than the shortest code.
cannam@89:  */
cannam@89: int inflate_table9(type, lens, codes, table, bits, work)
cannam@89: codetype type;
cannam@89: unsigned short FAR *lens;
cannam@89: unsigned codes;
cannam@89: code FAR * FAR *table;
cannam@89: unsigned FAR *bits;
cannam@89: unsigned short FAR *work;
cannam@89: {
cannam@89:     unsigned len;               /* a code's length in bits */
cannam@89:     unsigned sym;               /* index of code symbols */
cannam@89:     unsigned min, max;          /* minimum and maximum code lengths */
cannam@89:     unsigned root;              /* number of index bits for root table */
cannam@89:     unsigned curr;              /* number of index bits for current table */
cannam@89:     unsigned drop;              /* code bits to drop for sub-table */
cannam@89:     int left;                   /* number of prefix codes available */
cannam@89:     unsigned used;              /* code entries in table used */
cannam@89:     unsigned huff;              /* Huffman code */
cannam@89:     unsigned incr;              /* for incrementing code, index */
cannam@89:     unsigned fill;              /* index for replicating entries */
cannam@89:     unsigned low;               /* low bits for current root entry */
cannam@89:     unsigned mask;              /* mask for low root bits */
cannam@89:     code this;                  /* table entry for duplication */
cannam@89:     code FAR *next;             /* next available space in table */
cannam@89:     const unsigned short FAR *base;     /* base value table to use */
cannam@89:     const unsigned short FAR *extra;    /* extra bits table to use */
cannam@89:     int end;                    /* use base and extra for symbol > end */
cannam@89:     unsigned short count[MAXBITS+1];    /* number of codes of each length */
cannam@89:     unsigned short offs[MAXBITS+1];     /* offsets in table for each length */
cannam@89:     static const unsigned short lbase[31] = { /* Length codes 257..285 base */
cannam@89:         3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17,
cannam@89:         19, 23, 27, 31, 35, 43, 51, 59, 67, 83, 99, 115,
cannam@89:         131, 163, 195, 227, 3, 0, 0};
cannam@89:     static const unsigned short lext[31] = { /* Length codes 257..285 extra */
cannam@89:         128, 128, 128, 128, 128, 128, 128, 128, 129, 129, 129, 129,
cannam@89:         130, 130, 130, 130, 131, 131, 131, 131, 132, 132, 132, 132,
cannam@89:         133, 133, 133, 133, 144, 78, 68};
cannam@89:     static const unsigned short dbase[32] = { /* Distance codes 0..31 base */
cannam@89:         1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49,
cannam@89:         65, 97, 129, 193, 257, 385, 513, 769, 1025, 1537, 2049, 3073,
cannam@89:         4097, 6145, 8193, 12289, 16385, 24577, 32769, 49153};
cannam@89:     static const unsigned short dext[32] = { /* Distance codes 0..31 extra */
cannam@89:         128, 128, 128, 128, 129, 129, 130, 130, 131, 131, 132, 132,
cannam@89:         133, 133, 134, 134, 135, 135, 136, 136, 137, 137, 138, 138,
cannam@89:         139, 139, 140, 140, 141, 141, 142, 142};
cannam@89: 
cannam@89:     /*
cannam@89:        Process a set of code lengths to create a canonical Huffman code.  The
cannam@89:        code lengths are lens[0..codes-1].  Each length corresponds to the
cannam@89:        symbols 0..codes-1.  The Huffman code is generated by first sorting the
cannam@89:        symbols by length from short to long, and retaining the symbol order
cannam@89:        for codes with equal lengths.  Then the code starts with all zero bits
cannam@89:        for the first code of the shortest length, and the codes are integer
cannam@89:        increments for the same length, and zeros are appended as the length
cannam@89:        increases.  For the deflate format, these bits are stored backwards
cannam@89:        from their more natural integer increment ordering, and so when the
cannam@89:        decoding tables are built in the large loop below, the integer codes
cannam@89:        are incremented backwards.
cannam@89: 
cannam@89:        This routine assumes, but does not check, that all of the entries in
cannam@89:        lens[] are in the range 0..MAXBITS.  The caller must assure this.
cannam@89:        1..MAXBITS is interpreted as that code length.  zero means that that
cannam@89:        symbol does not occur in this code.
cannam@89: 
cannam@89:        The codes are sorted by computing a count of codes for each length,
cannam@89:        creating from that a table of starting indices for each length in the
cannam@89:        sorted table, and then entering the symbols in order in the sorted
cannam@89:        table.  The sorted table is work[], with that space being provided by
cannam@89:        the caller.
cannam@89: 
cannam@89:        The length counts are used for other purposes as well, i.e. finding
cannam@89:        the minimum and maximum length codes, determining if there are any
cannam@89:        codes at all, checking for a valid set of lengths, and looking ahead
cannam@89:        at length counts to determine sub-table sizes when building the
cannam@89:        decoding tables.
cannam@89:      */
cannam@89: 
cannam@89:     /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */
cannam@89:     for (len = 0; len <= MAXBITS; len++)
cannam@89:         count[len] = 0;
cannam@89:     for (sym = 0; sym < codes; sym++)
cannam@89:         count[lens[sym]]++;
cannam@89: 
cannam@89:     /* bound code lengths, force root to be within code lengths */
cannam@89:     root = *bits;
cannam@89:     for (max = MAXBITS; max >= 1; max--)
cannam@89:         if (count[max] != 0) break;
cannam@89:     if (root > max) root = max;
cannam@89:     if (max == 0) return -1;            /* no codes! */
cannam@89:     for (min = 1; min <= MAXBITS; min++)
cannam@89:         if (count[min] != 0) break;
cannam@89:     if (root < min) root = min;
cannam@89: 
cannam@89:     /* check for an over-subscribed or incomplete set of lengths */
cannam@89:     left = 1;
cannam@89:     for (len = 1; len <= MAXBITS; len++) {
cannam@89:         left <<= 1;
cannam@89:         left -= count[len];
cannam@89:         if (left < 0) return -1;        /* over-subscribed */
cannam@89:     }
cannam@89:     if (left > 0 && (type == CODES || max != 1))
cannam@89:         return -1;                      /* incomplete set */
cannam@89: 
cannam@89:     /* generate offsets into symbol table for each length for sorting */
cannam@89:     offs[1] = 0;
cannam@89:     for (len = 1; len < MAXBITS; len++)
cannam@89:         offs[len + 1] = offs[len] + count[len];
cannam@89: 
cannam@89:     /* sort symbols by length, by symbol order within each length */
cannam@89:     for (sym = 0; sym < codes; sym++)
cannam@89:         if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym;
cannam@89: 
cannam@89:     /*
cannam@89:        Create and fill in decoding tables.  In this loop, the table being
cannam@89:        filled is at next and has curr index bits.  The code being used is huff
cannam@89:        with length len.  That code is converted to an index by dropping drop
cannam@89:        bits off of the bottom.  For codes where len is less than drop + curr,
cannam@89:        those top drop + curr - len bits are incremented through all values to
cannam@89:        fill the table with replicated entries.
cannam@89: 
cannam@89:        root is the number of index bits for the root table.  When len exceeds
cannam@89:        root, sub-tables are created pointed to by the root entry with an index
cannam@89:        of the low root bits of huff.  This is saved in low to check for when a
cannam@89:        new sub-table should be started.  drop is zero when the root table is
cannam@89:        being filled, and drop is root when sub-tables are being filled.
cannam@89: 
cannam@89:        When a new sub-table is needed, it is necessary to look ahead in the
cannam@89:        code lengths to determine what size sub-table is needed.  The length
cannam@89:        counts are used for this, and so count[] is decremented as codes are
cannam@89:        entered in the tables.
cannam@89: 
cannam@89:        used keeps track of how many table entries have been allocated from the
cannam@89:        provided *table space.  It is checked for LENS and DIST tables against
cannam@89:        the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in
cannam@89:        the initial root table size constants.  See the comments in inftree9.h
cannam@89:        for more information.
cannam@89: 
cannam@89:        sym increments through all symbols, and the loop terminates when
cannam@89:        all codes of length max, i.e. all codes, have been processed.  This
cannam@89:        routine permits incomplete codes, so another loop after this one fills
cannam@89:        in the rest of the decoding tables with invalid code markers.
cannam@89:      */
cannam@89: 
cannam@89:     /* set up for code type */
cannam@89:     switch (type) {
cannam@89:     case CODES:
cannam@89:         base = extra = work;    /* dummy value--not used */
cannam@89:         end = 19;
cannam@89:         break;
cannam@89:     case LENS:
cannam@89:         base = lbase;
cannam@89:         base -= 257;
cannam@89:         extra = lext;
cannam@89:         extra -= 257;
cannam@89:         end = 256;
cannam@89:         break;
cannam@89:     default:            /* DISTS */
cannam@89:         base = dbase;
cannam@89:         extra = dext;
cannam@89:         end = -1;
cannam@89:     }
cannam@89: 
cannam@89:     /* initialize state for loop */
cannam@89:     huff = 0;                   /* starting code */
cannam@89:     sym = 0;                    /* starting code symbol */
cannam@89:     len = min;                  /* starting code length */
cannam@89:     next = *table;              /* current table to fill in */
cannam@89:     curr = root;                /* current table index bits */
cannam@89:     drop = 0;                   /* current bits to drop from code for index */
cannam@89:     low = (unsigned)(-1);       /* trigger new sub-table when len > root */
cannam@89:     used = 1U << root;          /* use root table entries */
cannam@89:     mask = used - 1;            /* mask for comparing low */
cannam@89: 
cannam@89:     /* check available table space */
cannam@89:     if ((type == LENS && used >= ENOUGH_LENS) ||
cannam@89:         (type == DISTS && used >= ENOUGH_DISTS))
cannam@89:         return 1;
cannam@89: 
cannam@89:     /* process all codes and make table entries */
cannam@89:     for (;;) {
cannam@89:         /* create table entry */
cannam@89:         this.bits = (unsigned char)(len - drop);
cannam@89:         if ((int)(work[sym]) < end) {
cannam@89:             this.op = (unsigned char)0;
cannam@89:             this.val = work[sym];
cannam@89:         }
cannam@89:         else if ((int)(work[sym]) > end) {
cannam@89:             this.op = (unsigned char)(extra[work[sym]]);
cannam@89:             this.val = base[work[sym]];
cannam@89:         }
cannam@89:         else {
cannam@89:             this.op = (unsigned char)(32 + 64);         /* end of block */
cannam@89:             this.val = 0;
cannam@89:         }
cannam@89: 
cannam@89:         /* replicate for those indices with low len bits equal to huff */
cannam@89:         incr = 1U << (len - drop);
cannam@89:         fill = 1U << curr;
cannam@89:         do {
cannam@89:             fill -= incr;
cannam@89:             next[(huff >> drop) + fill] = this;
cannam@89:         } while (fill != 0);
cannam@89: 
cannam@89:         /* backwards increment the len-bit code huff */
cannam@89:         incr = 1U << (len - 1);
cannam@89:         while (huff & incr)
cannam@89:             incr >>= 1;
cannam@89:         if (incr != 0) {
cannam@89:             huff &= incr - 1;
cannam@89:             huff += incr;
cannam@89:         }
cannam@89:         else
cannam@89:             huff = 0;
cannam@89: 
cannam@89:         /* go to next symbol, update count, len */
cannam@89:         sym++;
cannam@89:         if (--(count[len]) == 0) {
cannam@89:             if (len == max) break;
cannam@89:             len = lens[work[sym]];
cannam@89:         }
cannam@89: 
cannam@89:         /* create new sub-table if needed */
cannam@89:         if (len > root && (huff & mask) != low) {
cannam@89:             /* if first time, transition to sub-tables */
cannam@89:             if (drop == 0)
cannam@89:                 drop = root;
cannam@89: 
cannam@89:             /* increment past last table */
cannam@89:             next += 1U << curr;
cannam@89: 
cannam@89:             /* determine length of next table */
cannam@89:             curr = len - drop;
cannam@89:             left = (int)(1 << curr);
cannam@89:             while (curr + drop < max) {
cannam@89:                 left -= count[curr + drop];
cannam@89:                 if (left <= 0) break;
cannam@89:                 curr++;
cannam@89:                 left <<= 1;
cannam@89:             }
cannam@89: 
cannam@89:             /* check for enough space */
cannam@89:             used += 1U << curr;
cannam@89:             if ((type == LENS && used >= ENOUGH_LENS) ||
cannam@89:                 (type == DISTS && used >= ENOUGH_DISTS))
cannam@89:                 return 1;
cannam@89: 
cannam@89:             /* point entry in root table to sub-table */
cannam@89:             low = huff & mask;
cannam@89:             (*table)[low].op = (unsigned char)curr;
cannam@89:             (*table)[low].bits = (unsigned char)root;
cannam@89:             (*table)[low].val = (unsigned short)(next - *table);
cannam@89:         }
cannam@89:     }
cannam@89: 
cannam@89:     /*
cannam@89:        Fill in rest of table for incomplete codes.  This loop is similar to the
cannam@89:        loop above in incrementing huff for table indices.  It is assumed that
cannam@89:        len is equal to curr + drop, so there is no loop needed to increment
cannam@89:        through high index bits.  When the current sub-table is filled, the loop
cannam@89:        drops back to the root table to fill in any remaining entries there.
cannam@89:      */
cannam@89:     this.op = (unsigned char)64;                /* invalid code marker */
cannam@89:     this.bits = (unsigned char)(len - drop);
cannam@89:     this.val = (unsigned short)0;
cannam@89:     while (huff != 0) {
cannam@89:         /* when done with sub-table, drop back to root table */
cannam@89:         if (drop != 0 && (huff & mask) != low) {
cannam@89:             drop = 0;
cannam@89:             len = root;
cannam@89:             next = *table;
cannam@89:             curr = root;
cannam@89:             this.bits = (unsigned char)len;
cannam@89:         }
cannam@89: 
cannam@89:         /* put invalid code marker in table */
cannam@89:         next[huff >> drop] = this;
cannam@89: 
cannam@89:         /* backwards increment the len-bit code huff */
cannam@89:         incr = 1U << (len - 1);
cannam@89:         while (huff & incr)
cannam@89:             incr >>= 1;
cannam@89:         if (incr != 0) {
cannam@89:             huff &= incr - 1;
cannam@89:             huff += incr;
cannam@89:         }
cannam@89:         else
cannam@89:             huff = 0;
cannam@89:     }
cannam@89: 
cannam@89:     /* set return parameters */
cannam@89:     *table += used;
cannam@89:     *bits = root;
cannam@89:     return 0;
cannam@89: }