sl@0: /* Portions Copyright (c) 2007-2009 Nokia Corporation and/or its subsidiary(-ies).
sl@0: * All rights reserved.
sl@0: */
sl@0:
sl@0: /* inftrees.cpp -- generate Huffman trees for efficient decoding
sl@0: * Copyright (C) 1995-2005 Mark Adler
sl@0: * For conditions of distribution and use, see copyright notice in zlib.h
sl@0: */
sl@0:
sl@0: #include "zutil.h"
sl@0: #include "inftrees.h"
sl@0:
sl@0: #define MAXBITS 15
sl@0:
sl@0:
sl@0: const char inflate_copyright[] =
sl@0: " inflate 1.2.3 Copyright 1995-2005 Mark Adler ";
sl@0: /*
sl@0: If you use the zlib library in a product, an acknowledgment is welcome
sl@0: in the documentation of your product. If for some reason you cannot
sl@0: include such an acknowledgment, I would appreciate that you keep this
sl@0: copyright string in the executable of your product.
sl@0: */
sl@0:
sl@0: /*
sl@0: Build a set of tables to decode the provided canonical Huffman code.
sl@0: The code lengths are lens[0..codes-1]. The result starts at *table,
sl@0: whose indices are 0..2^bits-1. work is a writable array of at least
sl@0: lens shorts, which is used as a work area. type is the type of code
sl@0: to be generated, CODES, LENS, or DISTS. On return, zero is success,
sl@0: -1 is an invalid code, and +1 means that ENOUGH isn't enough. table
sl@0: on return points to the next available entry's address. bits is the
sl@0: requested root table index bits, and on return it is the actual root
sl@0: table index bits. It will differ if the request is greater than the
sl@0: longest code or if it is less than the shortest code.
sl@0: */
sl@0: #ifdef __SYMBIAN32__
sl@0: int inflate_table(codetype type,unsigned short FAR * lens,unsigned codes, code FAR * FAR * table,unsigned FAR * bits,unsigned short FAR * work)
sl@0: #else
sl@0: int inflate_table(type, lens, codes, table, bits, work)
sl@0: codetype type;
sl@0: unsigned short FAR *lens;
sl@0: unsigned codes;
sl@0: code FAR * FAR *table;
sl@0: unsigned FAR *bits;
sl@0: unsigned short FAR *work;
sl@0: #endif //__SYMBIAN32__
sl@0: {
sl@0: // Line to stop compiler warning about unused mandatory global variable 'inflate_copyright'
sl@0: char dontCare = inflate_copyright[0]; dontCare = dontCare;
sl@0:
sl@0: unsigned len; /* a code's length in bits */
sl@0: unsigned sym; /* index of code symbols */
sl@0: unsigned min, max; /* minimum and maximum code lengths */
sl@0: unsigned root; /* number of index bits for root table */
sl@0: unsigned curr; /* number of index bits for current table */
sl@0: unsigned drop; /* code bits to drop for sub-table */
sl@0: int left; /* number of prefix codes available */
sl@0: unsigned used; /* code entries in table used */
sl@0: unsigned huff; /* Huffman code */
sl@0: unsigned incr; /* for incrementing code, index */
sl@0: unsigned fill; /* index for replicating entries */
sl@0: unsigned low; /* low bits for current root entry */
sl@0: unsigned mask; /* mask for low root bits */
sl@0:
sl@0: /* Need to replace "this" variable with "current" as "this" is a reserved
sl@0: * keyword in C++ which is prefectly fine for a c code. As this file
sl@0: * has been changed to C++ "this" needs to be changed.
sl@0: */
sl@0: # define this current
sl@0: code this; /* table entry for duplication */
sl@0: code FAR *next; /* next available space in table */
sl@0: const unsigned short FAR *base; /* base value table to use */
sl@0: const unsigned short FAR *extra; /* extra bits table to use */
sl@0: int end; /* use base and extra for symbol > end */
sl@0: unsigned short count[MAXBITS+1]; /* number of codes of each length */
sl@0: unsigned short offs[MAXBITS+1]; /* offsets in table for each length */
sl@0: static const unsigned short lbase[31] = { /* Length codes 257..285 base */
sl@0: 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31,
sl@0: 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0};
sl@0: static const unsigned short lext[31] = { /* Length codes 257..285 extra */
sl@0: 16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18,
sl@0: 19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 201, 196};
sl@0: static const unsigned short dbase[32] = { /* Distance codes 0..29 base */
sl@0: 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,
sl@0: 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145,
sl@0: 8193, 12289, 16385, 24577, 0, 0};
sl@0: static const unsigned short dext[32] = { /* Distance codes 0..29 extra */
sl@0: 16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22,
sl@0: 23, 23, 24, 24, 25, 25, 26, 26, 27, 27,
sl@0: 28, 28, 29, 29, 64, 64};
sl@0:
sl@0: /*
sl@0: Process a set of code lengths to create a canonical Huffman code. The
sl@0: code lengths are lens[0..codes-1]. Each length corresponds to the
sl@0: symbols 0..codes-1. The Huffman code is generated by first sorting the
sl@0: symbols by length from short to long, and retaining the symbol order
sl@0: for codes with equal lengths. Then the code starts with all zero bits
sl@0: for the first code of the shortest length, and the codes are integer
sl@0: increments for the same length, and zeros are appended as the length
sl@0: increases. For the deflate format, these bits are stored backwards
sl@0: from their more natural integer increment ordering, and so when the
sl@0: decoding tables are built in the large loop below, the integer codes
sl@0: are incremented backwards.
sl@0:
sl@0: This routine assumes, but does not check, that all of the entries in
sl@0: lens[] are in the range 0..MAXBITS. The caller must assure this.
sl@0: 1..MAXBITS is interpreted as that code length. zero means that that
sl@0: symbol does not occur in this code.
sl@0:
sl@0: The codes are sorted by computing a count of codes for each length,
sl@0: creating from that a table of starting indices for each length in the
sl@0: sorted table, and then entering the symbols in order in the sorted
sl@0: table. The sorted table is work[], with that space being provided by
sl@0: the caller.
sl@0:
sl@0: The length counts are used for other purposes as well, i.e. finding
sl@0: the minimum and maximum length codes, determining if there are any
sl@0: codes at all, checking for a valid set of lengths, and looking ahead
sl@0: at length counts to determine sub-table sizes when building the
sl@0: decoding tables.
sl@0: */
sl@0:
sl@0: /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */
sl@0: for (len = 0; len <= MAXBITS; len++)
sl@0: count[len] = 0;
sl@0: for (sym = 0; sym < codes; sym++)
sl@0: count[lens[sym]]++;
sl@0:
sl@0: /* bound code lengths, force root to be within code lengths */
sl@0: root = *bits;
sl@0: for (max = MAXBITS; max >= 1; max--)
sl@0: if (count[max] != 0) break;
sl@0: if (root > max) root = max;
sl@0: if (max == 0) { /* no symbols to code at all */
sl@0: this.op = (unsigned char)64; /* invalid code marker */
sl@0: this.bits = (unsigned char)1;
sl@0: this.val = (unsigned short)0;
sl@0: *(*table)++ = this; /* make a table to force an error */
sl@0: *(*table)++ = this;
sl@0: *bits = 1;
sl@0: return 0; /* no symbols, but wait for decoding to report error */
sl@0: }
sl@0: for (min = 1; min <= MAXBITS; min++)
sl@0: if (count[min] != 0) break;
sl@0: if (root < min) root = min;
sl@0:
sl@0: /* check for an over-subscribed or incomplete set of lengths */
sl@0: left = 1;
sl@0: for (len = 1; len <= MAXBITS; len++) {
sl@0: left <<= 1;
sl@0: left -= count[len];
sl@0: if (left < 0) return -1; /* over-subscribed */
sl@0: }
sl@0: if (left > 0 && (type == CODES || max != 1))
sl@0: return -1; /* incomplete set */
sl@0:
sl@0: /* generate offsets into symbol table for each length for sorting */
sl@0: offs[1] = 0;
sl@0: for (len = 1; len < MAXBITS; len++)
sl@0: offs[len + 1] = offs[len] + count[len];
sl@0:
sl@0: /* sort symbols by length, by symbol order within each length */
sl@0: for (sym = 0; sym < codes; sym++)
sl@0: if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym;
sl@0:
sl@0: /*
sl@0: Create and fill in decoding tables. In this loop, the table being
sl@0: filled is at next and has curr index bits. The code being used is huff
sl@0: with length len. That code is converted to an index by dropping drop
sl@0: bits off of the bottom. For codes where len is less than drop + curr,
sl@0: those top drop + curr - len bits are incremented through all values to
sl@0: fill the table with replicated entries.
sl@0:
sl@0: root is the number of index bits for the root table. When len exceeds
sl@0: root, sub-tables are created pointed to by the root entry with an index
sl@0: of the low root bits of huff. This is saved in low to check for when a
sl@0: new sub-table should be started. drop is zero when the root table is
sl@0: being filled, and drop is root when sub-tables are being filled.
sl@0:
sl@0: When a new sub-table is needed, it is necessary to look ahead in the
sl@0: code lengths to determine what size sub-table is needed. The length
sl@0: counts are used for this, and so count[] is decremented as codes are
sl@0: entered in the tables.
sl@0:
sl@0: used keeps track of how many table entries have been allocated from the
sl@0: provided *table space. It is checked when a LENS table is being made
sl@0: against the space in *table, ENOUGH, minus the maximum space needed by
sl@0: the worst case distance code, MAXD. This should never happen, but the
sl@0: sufficiency of ENOUGH has not been proven exhaustively, hence the check.
sl@0: This assumes that when type == LENS, bits == 9.
sl@0:
sl@0: sym increments through all symbols, and the loop terminates when
sl@0: all codes of length max, i.e. all codes, have been processed. This
sl@0: routine permits incomplete codes, so another loop after this one fills
sl@0: in the rest of the decoding tables with invalid code markers.
sl@0: */
sl@0:
sl@0: /* set up for code type */
sl@0: switch (type) {
sl@0: case CODES:
sl@0: base = extra = work; /* dummy value--not used */
sl@0: end = 19;
sl@0: break;
sl@0: case LENS:
sl@0: base = lbase;
sl@0: base -= 257;
sl@0: extra = lext;
sl@0: extra -= 257;
sl@0: end = 256;
sl@0: break;
sl@0: default: /* DISTS */
sl@0: base = dbase;
sl@0: extra = dext;
sl@0: end = -1;
sl@0: }
sl@0:
sl@0: /* initialize state for loop */
sl@0: huff = 0; /* starting code */
sl@0: sym = 0; /* starting code symbol */
sl@0: len = min; /* starting code length */
sl@0: next = *table; /* current table to fill in */
sl@0: curr = root; /* current table index bits */
sl@0: drop = 0; /* current bits to drop from code for index */
sl@0: low = (unsigned)(-1); /* trigger new sub-table when len > root */
sl@0: used = 1U << root; /* use root table entries */
sl@0: mask = used - 1; /* mask for comparing low */
sl@0:
sl@0: /* check available table space */
sl@0: if (type == LENS && used >= ENOUGH - MAXD)
sl@0: return 1;
sl@0:
sl@0: /* process all codes and make table entries */
sl@0: for (;;) {
sl@0: /* create table entry */
sl@0: this.bits = (unsigned char)(len - drop);
sl@0: if ((int)(work[sym]) < end) {
sl@0: this.op = (unsigned char)0;
sl@0: this.val = work[sym];
sl@0: }
sl@0: else if ((int)(work[sym]) > end) {
sl@0: this.op = (unsigned char)(extra[work[sym]]);
sl@0: this.val = base[work[sym]];
sl@0: }
sl@0: else {
sl@0: this.op = (unsigned char)(32 + 64); /* end of block */
sl@0: this.val = 0;
sl@0: }
sl@0:
sl@0: /* replicate for those indices with low len bits equal to huff */
sl@0: incr = 1U << (len - drop);
sl@0: fill = 1U << curr;
sl@0: min = fill; /* save offset to next table */
sl@0: do {
sl@0: fill -= incr;
sl@0: next[(huff >> drop) + fill] = this;
sl@0: } while (fill != 0);
sl@0:
sl@0: /* backwards increment the len-bit code huff */
sl@0: incr = 1U << (len - 1);
sl@0: while (huff & incr)
sl@0: incr >>= 1;
sl@0: if (incr != 0) {
sl@0: huff &= incr - 1;
sl@0: huff += incr;
sl@0: }
sl@0: else
sl@0: huff = 0;
sl@0:
sl@0: /* go to next symbol, update count, len */
sl@0: sym++;
sl@0: if (--(count[len]) == 0) {
sl@0: if (len == max) break;
sl@0: len = lens[work[sym]];
sl@0: }
sl@0:
sl@0: /* create new sub-table if needed */
sl@0: if (len > root && (huff & mask) != low) {
sl@0: /* if first time, transition to sub-tables */
sl@0: if (drop == 0)
sl@0: drop = root;
sl@0:
sl@0: /* increment past last table */
sl@0: next += min; /* here min is 1 << curr */
sl@0:
sl@0: /* determine length of next table */
sl@0: curr = len - drop;
sl@0: left = (int)(1 << curr);
sl@0: while (curr + drop < max) {
sl@0: left -= count[curr + drop];
sl@0: if (left <= 0) break;
sl@0: curr++;
sl@0: left <<= 1;
sl@0: }
sl@0:
sl@0: /* check for enough space */
sl@0: used += 1U << curr;
sl@0: if (type == LENS && used >= ENOUGH - MAXD)
sl@0: return 1;
sl@0:
sl@0: /* point entry in root table to sub-table */
sl@0: low = huff & mask;
sl@0: (*table)[low].op = (unsigned char)curr;
sl@0: (*table)[low].bits = (unsigned char)root;
sl@0: (*table)[low].val = (unsigned short)(next - *table);
sl@0: }
sl@0: }
sl@0:
sl@0: /*
sl@0: Fill in rest of table for incomplete codes. This loop is similar to the
sl@0: loop above in incrementing huff for table indices. It is assumed that
sl@0: len is equal to curr + drop, so there is no loop needed to increment
sl@0: through high index bits. When the current sub-table is filled, the loop
sl@0: drops back to the root table to fill in any remaining entries there.
sl@0: */
sl@0: this.op = (unsigned char)64; /* invalid code marker */
sl@0: this.bits = (unsigned char)(len - drop);
sl@0: this.val = (unsigned short)0;
sl@0: while (huff != 0) {
sl@0: /* when done with sub-table, drop back to root table */
sl@0: if (drop != 0 && (huff & mask) != low) {
sl@0: drop = 0;
sl@0: len = root;
sl@0: next = *table;
sl@0: this.bits = (unsigned char)len;
sl@0: }
sl@0:
sl@0: /* put invalid code marker in table */
sl@0: next[huff >> drop] = this;
sl@0:
sl@0: /* backwards increment the len-bit code huff */
sl@0: incr = 1U << (len - 1);
sl@0: while (huff & incr)
sl@0: incr >>= 1;
sl@0: if (incr != 0) {
sl@0: huff &= incr - 1;
sl@0: huff += incr;
sl@0: }
sl@0: else
sl@0: huff = 0;
sl@0: }
sl@0:
sl@0: /* set return parameters */
sl@0: *table += used;
sl@0: *bits = root;
sl@0: return 0;
sl@0: }
sl@0:
sl@0:
sl@0:
sl@0:
sl@0:
sl@0: