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