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