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