Chris@43: /* inftrees.c -- generate Huffman trees for efficient decoding Chris@43: * Copyright (C) 1995-2013 Mark Adler Chris@43: * For conditions of distribution and use, see copyright notice in zlib.h Chris@43: */ Chris@43: Chris@43: #include "zutil.h" Chris@43: #include "inftrees.h" Chris@43: Chris@43: #define MAXBITS 15 Chris@43: Chris@43: const char inflate_copyright[] = Chris@43: " inflate 1.2.8 Copyright 1995-2013 Mark Adler "; Chris@43: /* Chris@43: If you use the zlib library in a product, an acknowledgment is welcome Chris@43: in the documentation of your product. If for some reason you cannot Chris@43: include such an acknowledgment, I would appreciate that you keep this Chris@43: copyright string in the executable of your product. Chris@43: */ Chris@43: Chris@43: /* Chris@43: Build a set of tables to decode the provided canonical Huffman code. Chris@43: The code lengths are lens[0..codes-1]. The result starts at *table, Chris@43: whose indices are 0..2^bits-1. work is a writable array of at least Chris@43: lens shorts, which is used as a work area. type is the type of code Chris@43: to be generated, CODES, LENS, or DISTS. On return, zero is success, Chris@43: -1 is an invalid code, and +1 means that ENOUGH isn't enough. table Chris@43: on return points to the next available entry's address. bits is the Chris@43: requested root table index bits, and on return it is the actual root Chris@43: table index bits. It will differ if the request is greater than the Chris@43: longest code or if it is less than the shortest code. Chris@43: */ Chris@43: int ZLIB_INTERNAL inflate_table(type, lens, codes, table, bits, work) Chris@43: codetype type; Chris@43: unsigned short FAR *lens; Chris@43: unsigned codes; Chris@43: code FAR * FAR *table; Chris@43: unsigned FAR *bits; Chris@43: unsigned short FAR *work; Chris@43: { Chris@43: unsigned len; /* a code's length in bits */ Chris@43: unsigned sym; /* index of code symbols */ Chris@43: unsigned min, max; /* minimum and maximum code lengths */ Chris@43: unsigned root; /* number of index bits for root table */ Chris@43: unsigned curr; /* number of index bits for current table */ Chris@43: unsigned drop; /* code bits to drop for sub-table */ Chris@43: int left; /* number of prefix codes available */ Chris@43: unsigned used; /* code entries in table used */ Chris@43: unsigned huff; /* Huffman code */ Chris@43: unsigned incr; /* for incrementing code, index */ Chris@43: unsigned fill; /* index for replicating entries */ Chris@43: unsigned low; /* low bits for current root entry */ Chris@43: unsigned mask; /* mask for low root bits */ Chris@43: code here; /* table entry for duplication */ Chris@43: code FAR *next; /* next available space in table */ Chris@43: const unsigned short FAR *base; /* base value table to use */ Chris@43: const unsigned short FAR *extra; /* extra bits table to use */ Chris@43: int end; /* use base and extra for symbol > end */ Chris@43: unsigned short count[MAXBITS+1]; /* number of codes of each length */ Chris@43: unsigned short offs[MAXBITS+1]; /* offsets in table for each length */ Chris@43: static const unsigned short lbase[31] = { /* Length codes 257..285 base */ Chris@43: 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31, Chris@43: 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0}; Chris@43: static const unsigned short lext[31] = { /* Length codes 257..285 extra */ Chris@43: 16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18, Chris@43: 19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 72, 78}; Chris@43: static const unsigned short dbase[32] = { /* Distance codes 0..29 base */ Chris@43: 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193, Chris@43: 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145, Chris@43: 8193, 12289, 16385, 24577, 0, 0}; Chris@43: static const unsigned short dext[32] = { /* Distance codes 0..29 extra */ Chris@43: 16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22, Chris@43: 23, 23, 24, 24, 25, 25, 26, 26, 27, 27, Chris@43: 28, 28, 29, 29, 64, 64}; Chris@43: Chris@43: /* Chris@43: Process a set of code lengths to create a canonical Huffman code. The Chris@43: code lengths are lens[0..codes-1]. Each length corresponds to the Chris@43: symbols 0..codes-1. The Huffman code is generated by first sorting the Chris@43: symbols by length from short to long, and retaining the symbol order Chris@43: for codes with equal lengths. Then the code starts with all zero bits Chris@43: for the first code of the shortest length, and the codes are integer Chris@43: increments for the same length, and zeros are appended as the length Chris@43: increases. For the deflate format, these bits are stored backwards Chris@43: from their more natural integer increment ordering, and so when the Chris@43: decoding tables are built in the large loop below, the integer codes Chris@43: are incremented backwards. Chris@43: Chris@43: This routine assumes, but does not check, that all of the entries in Chris@43: lens[] are in the range 0..MAXBITS. The caller must assure this. Chris@43: 1..MAXBITS is interpreted as that code length. zero means that that Chris@43: symbol does not occur in this code. Chris@43: Chris@43: The codes are sorted by computing a count of codes for each length, Chris@43: creating from that a table of starting indices for each length in the Chris@43: sorted table, and then entering the symbols in order in the sorted Chris@43: table. The sorted table is work[], with that space being provided by Chris@43: the caller. Chris@43: Chris@43: The length counts are used for other purposes as well, i.e. finding Chris@43: the minimum and maximum length codes, determining if there are any Chris@43: codes at all, checking for a valid set of lengths, and looking ahead Chris@43: at length counts to determine sub-table sizes when building the Chris@43: decoding tables. Chris@43: */ Chris@43: Chris@43: /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */ Chris@43: for (len = 0; len <= MAXBITS; len++) Chris@43: count[len] = 0; Chris@43: for (sym = 0; sym < codes; sym++) Chris@43: count[lens[sym]]++; Chris@43: Chris@43: /* bound code lengths, force root to be within code lengths */ Chris@43: root = *bits; Chris@43: for (max = MAXBITS; max >= 1; max--) Chris@43: if (count[max] != 0) break; Chris@43: if (root > max) root = max; Chris@43: if (max == 0) { /* no symbols to code at all */ Chris@43: here.op = (unsigned char)64; /* invalid code marker */ Chris@43: here.bits = (unsigned char)1; Chris@43: here.val = (unsigned short)0; Chris@43: *(*table)++ = here; /* make a table to force an error */ Chris@43: *(*table)++ = here; Chris@43: *bits = 1; Chris@43: return 0; /* no symbols, but wait for decoding to report error */ Chris@43: } Chris@43: for (min = 1; min < max; min++) Chris@43: if (count[min] != 0) break; Chris@43: if (root < min) root = min; Chris@43: Chris@43: /* check for an over-subscribed or incomplete set of lengths */ Chris@43: left = 1; Chris@43: for (len = 1; len <= MAXBITS; len++) { Chris@43: left <<= 1; Chris@43: left -= count[len]; Chris@43: if (left < 0) return -1; /* over-subscribed */ Chris@43: } Chris@43: if (left > 0 && (type == CODES || max != 1)) Chris@43: return -1; /* incomplete set */ Chris@43: Chris@43: /* generate offsets into symbol table for each length for sorting */ Chris@43: offs[1] = 0; Chris@43: for (len = 1; len < MAXBITS; len++) Chris@43: offs[len + 1] = offs[len] + count[len]; Chris@43: Chris@43: /* sort symbols by length, by symbol order within each length */ Chris@43: for (sym = 0; sym < codes; sym++) Chris@43: if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym; Chris@43: Chris@43: /* Chris@43: Create and fill in decoding tables. In this loop, the table being Chris@43: filled is at next and has curr index bits. The code being used is huff Chris@43: with length len. That code is converted to an index by dropping drop Chris@43: bits off of the bottom. For codes where len is less than drop + curr, Chris@43: those top drop + curr - len bits are incremented through all values to Chris@43: fill the table with replicated entries. Chris@43: Chris@43: root is the number of index bits for the root table. When len exceeds Chris@43: root, sub-tables are created pointed to by the root entry with an index Chris@43: of the low root bits of huff. This is saved in low to check for when a Chris@43: new sub-table should be started. drop is zero when the root table is Chris@43: being filled, and drop is root when sub-tables are being filled. Chris@43: Chris@43: When a new sub-table is needed, it is necessary to look ahead in the Chris@43: code lengths to determine what size sub-table is needed. The length Chris@43: counts are used for this, and so count[] is decremented as codes are Chris@43: entered in the tables. Chris@43: Chris@43: used keeps track of how many table entries have been allocated from the Chris@43: provided *table space. It is checked for LENS and DIST tables against Chris@43: the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in Chris@43: the initial root table size constants. See the comments in inftrees.h Chris@43: for more information. Chris@43: Chris@43: sym increments through all symbols, and the loop terminates when Chris@43: all codes of length max, i.e. all codes, have been processed. This Chris@43: routine permits incomplete codes, so another loop after this one fills Chris@43: in the rest of the decoding tables with invalid code markers. Chris@43: */ Chris@43: Chris@43: /* set up for code type */ Chris@43: switch (type) { Chris@43: case CODES: Chris@43: base = extra = work; /* dummy value--not used */ Chris@43: end = 19; Chris@43: break; Chris@43: case LENS: Chris@43: base = lbase; Chris@43: base -= 257; Chris@43: extra = lext; Chris@43: extra -= 257; Chris@43: end = 256; Chris@43: break; Chris@43: default: /* DISTS */ Chris@43: base = dbase; Chris@43: extra = dext; Chris@43: end = -1; Chris@43: } Chris@43: Chris@43: /* initialize state for loop */ Chris@43: huff = 0; /* starting code */ Chris@43: sym = 0; /* starting code symbol */ Chris@43: len = min; /* starting code length */ Chris@43: next = *table; /* current table to fill in */ Chris@43: curr = root; /* current table index bits */ Chris@43: drop = 0; /* current bits to drop from code for index */ Chris@43: low = (unsigned)(-1); /* trigger new sub-table when len > root */ Chris@43: used = 1U << root; /* use root table entries */ Chris@43: mask = used - 1; /* mask for comparing low */ Chris@43: Chris@43: /* check available table space */ Chris@43: if ((type == LENS && used > ENOUGH_LENS) || Chris@43: (type == DISTS && used > ENOUGH_DISTS)) Chris@43: return 1; Chris@43: Chris@43: /* process all codes and make table entries */ Chris@43: for (;;) { Chris@43: /* create table entry */ Chris@43: here.bits = (unsigned char)(len - drop); Chris@43: if ((int)(work[sym]) < end) { Chris@43: here.op = (unsigned char)0; Chris@43: here.val = work[sym]; Chris@43: } Chris@43: else if ((int)(work[sym]) > end) { Chris@43: here.op = (unsigned char)(extra[work[sym]]); Chris@43: here.val = base[work[sym]]; Chris@43: } Chris@43: else { Chris@43: here.op = (unsigned char)(32 + 64); /* end of block */ Chris@43: here.val = 0; Chris@43: } Chris@43: Chris@43: /* replicate for those indices with low len bits equal to huff */ Chris@43: incr = 1U << (len - drop); Chris@43: fill = 1U << curr; Chris@43: min = fill; /* save offset to next table */ Chris@43: do { Chris@43: fill -= incr; Chris@43: next[(huff >> drop) + fill] = here; Chris@43: } while (fill != 0); Chris@43: Chris@43: /* backwards increment the len-bit code huff */ Chris@43: incr = 1U << (len - 1); Chris@43: while (huff & incr) Chris@43: incr >>= 1; Chris@43: if (incr != 0) { Chris@43: huff &= incr - 1; Chris@43: huff += incr; Chris@43: } Chris@43: else Chris@43: huff = 0; Chris@43: Chris@43: /* go to next symbol, update count, len */ Chris@43: sym++; Chris@43: if (--(count[len]) == 0) { Chris@43: if (len == max) break; Chris@43: len = lens[work[sym]]; Chris@43: } Chris@43: Chris@43: /* create new sub-table if needed */ Chris@43: if (len > root && (huff & mask) != low) { Chris@43: /* if first time, transition to sub-tables */ Chris@43: if (drop == 0) Chris@43: drop = root; Chris@43: Chris@43: /* increment past last table */ Chris@43: next += min; /* here min is 1 << curr */ Chris@43: Chris@43: /* determine length of next table */ Chris@43: curr = len - drop; Chris@43: left = (int)(1 << curr); Chris@43: while (curr + drop < max) { Chris@43: left -= count[curr + drop]; Chris@43: if (left <= 0) break; Chris@43: curr++; Chris@43: left <<= 1; Chris@43: } Chris@43: Chris@43: /* check for enough space */ Chris@43: used += 1U << curr; Chris@43: if ((type == LENS && used > ENOUGH_LENS) || Chris@43: (type == DISTS && used > ENOUGH_DISTS)) Chris@43: return 1; Chris@43: Chris@43: /* point entry in root table to sub-table */ Chris@43: low = huff & mask; Chris@43: (*table)[low].op = (unsigned char)curr; Chris@43: (*table)[low].bits = (unsigned char)root; Chris@43: (*table)[low].val = (unsigned short)(next - *table); Chris@43: } Chris@43: } Chris@43: Chris@43: /* fill in remaining table entry if code is incomplete (guaranteed to have Chris@43: at most one remaining entry, since if the code is incomplete, the Chris@43: maximum code length that was allowed to get this far is one bit) */ Chris@43: if (huff != 0) { Chris@43: here.op = (unsigned char)64; /* invalid code marker */ Chris@43: here.bits = (unsigned char)(len - drop); Chris@43: here.val = (unsigned short)0; Chris@43: next[huff] = here; Chris@43: } Chris@43: Chris@43: /* set return parameters */ Chris@43: *table += used; Chris@43: *bits = root; Chris@43: return 0; Chris@43: }