1 /* inftrees.c -- generate Huffman trees for efficient decoding 2 * Copyright (C) 1995-2013 Mark Adler 3 * For conditions of distribution and use, see copyright notice in zlib.h 4 */ 5 6 #include "hammer2_zlib_zutil.h" 7 #include "hammer2_zlib_inftrees.h" 8 9 #define MAXBITS 15 10 11 const char inflate_copyright[] = 12 " inflate 1.2.8 Copyright 1995-2013 Mark Adler "; 13 /* 14 If you use the zlib library in a product, an acknowledgment is welcome 15 in the documentation of your product. If for some reason you cannot 16 include such an acknowledgment, I would appreciate that you keep this 17 copyright string in the executable of your product. 18 */ 19 20 /* 21 Build a set of tables to decode the provided canonical Huffman code. 22 The code lengths are lens[0..codes-1]. The result starts at *table, 23 whose indices are 0..2^bits-1. work is a writable array of at least 24 lens shorts, which is used as a work area. type is the type of code 25 to be generated, CODES, LENS, or DISTS. On return, zero is success, 26 -1 is an invalid code, and +1 means that ENOUGH isn't enough. table 27 on return points to the next available entry's address. bits is the 28 requested root table index bits, and on return it is the actual root 29 table index bits. It will differ if the request is greater than the 30 longest code or if it is less than the shortest code. 31 */ 32 int 33 ZLIB_INTERNAL 34 inflate_table(codetype type, unsigned short FAR *lens, unsigned codes, 35 code FAR * FAR *table, unsigned FAR *bits, 36 unsigned short FAR *work) 37 { 38 unsigned len; /* a code's length in bits */ 39 unsigned sym; /* index of code symbols */ 40 unsigned min, max; /* minimum and maximum code lengths */ 41 unsigned root; /* number of index bits for root table */ 42 unsigned curr; /* number of index bits for current table */ 43 unsigned drop; /* code bits to drop for sub-table */ 44 int left; /* number of prefix codes available */ 45 unsigned used; /* code entries in table used */ 46 unsigned huff; /* Huffman code */ 47 unsigned incr; /* for incrementing code, index */ 48 unsigned fill; /* index for replicating entries */ 49 unsigned low; /* low bits for current root entry */ 50 unsigned mask; /* mask for low root bits */ 51 code here; /* table entry for duplication */ 52 code FAR *next; /* next available space in table */ 53 const unsigned short FAR *base; /* base value table to use */ 54 const unsigned short FAR *extra; /* extra bits table to use */ 55 int end; /* use base and extra for symbol > end */ 56 unsigned short count[MAXBITS+1]; /* number of codes of each length */ 57 unsigned short offs[MAXBITS+1]; /* offsets in table for each length */ 58 static const unsigned short lbase[31] = { /* Length codes 257..285 base */ 59 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31, 60 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0}; 61 static const unsigned short lext[31] = { /* Length codes 257..285 extra */ 62 16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18, 63 19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 72, 78}; 64 static const unsigned short dbase[32] = { /* Distance codes 0..29 base */ 65 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193, 66 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145, 67 8193, 12289, 16385, 24577, 0, 0}; 68 static const unsigned short dext[32] = { /* Distance codes 0..29 extra */ 69 16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22, 70 23, 23, 24, 24, 25, 25, 26, 26, 27, 27, 71 28, 28, 29, 29, 64, 64}; 72 73 /* 74 Process a set of code lengths to create a canonical Huffman code. The 75 code lengths are lens[0..codes-1]. Each length corresponds to the 76 symbols 0..codes-1. The Huffman code is generated by first sorting the 77 symbols by length from short to long, and retaining the symbol order 78 for codes with equal lengths. Then the code starts with all zero bits 79 for the first code of the shortest length, and the codes are integer 80 increments for the same length, and zeros are appended as the length 81 increases. For the deflate format, these bits are stored backwards 82 from their more natural integer increment ordering, and so when the 83 decoding tables are built in the large loop below, the integer codes 84 are incremented backwards. 85 86 This routine assumes, but does not check, that all of the entries in 87 lens[] are in the range 0..MAXBITS. The caller must assure this. 88 1..MAXBITS is interpreted as that code length. zero means that that 89 symbol does not occur in this code. 90 91 The codes are sorted by computing a count of codes for each length, 92 creating from that a table of starting indices for each length in the 93 sorted table, and then entering the symbols in order in the sorted 94 table. The sorted table is work[], with that space being provided by 95 the caller. 96 97 The length counts are used for other purposes as well, i.e. finding 98 the minimum and maximum length codes, determining if there are any 99 codes at all, checking for a valid set of lengths, and looking ahead 100 at length counts to determine sub-table sizes when building the 101 decoding tables. 102 */ 103 104 /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */ 105 for (len = 0; len <= MAXBITS; len++) 106 count[len] = 0; 107 for (sym = 0; sym < codes; sym++) 108 count[lens[sym]]++; 109 110 /* bound code lengths, force root to be within code lengths */ 111 root = *bits; 112 for (max = MAXBITS; max >= 1; max--) 113 if (count[max] != 0) break; 114 if (root > max) root = max; 115 if (max == 0) { /* no symbols to code at all */ 116 here.op = (unsigned char)64; /* invalid code marker */ 117 here.bits = (unsigned char)1; 118 here.val = (unsigned short)0; 119 *(*table)++ = here; /* make a table to force an error */ 120 *(*table)++ = here; 121 *bits = 1; 122 return 0; /* no symbols, but wait for decoding to report error */ 123 } 124 for (min = 1; min < max; min++) 125 if (count[min] != 0) break; 126 if (root < min) root = min; 127 128 /* check for an over-subscribed or incomplete set of lengths */ 129 left = 1; 130 for (len = 1; len <= MAXBITS; len++) { 131 left <<= 1; 132 left -= count[len]; 133 if (left < 0) return -1; /* over-subscribed */ 134 } 135 if (left > 0 && (type == CODES || max != 1)) 136 return -1; /* incomplete set */ 137 138 /* generate offsets into symbol table for each length for sorting */ 139 offs[1] = 0; 140 for (len = 1; len < MAXBITS; len++) 141 offs[len + 1] = offs[len] + count[len]; 142 143 /* sort symbols by length, by symbol order within each length */ 144 for (sym = 0; sym < codes; sym++) 145 if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym; 146 147 /* 148 Create and fill in decoding tables. In this loop, the table being 149 filled is at next and has curr index bits. The code being used is huff 150 with length len. That code is converted to an index by dropping drop 151 bits off of the bottom. For codes where len is less than drop + curr, 152 those top drop + curr - len bits are incremented through all values to 153 fill the table with replicated entries. 154 155 root is the number of index bits for the root table. When len exceeds 156 root, sub-tables are created pointed to by the root entry with an index 157 of the low root bits of huff. This is saved in low to check for when a 158 new sub-table should be started. drop is zero when the root table is 159 being filled, and drop is root when sub-tables are being filled. 160 161 When a new sub-table is needed, it is necessary to look ahead in the 162 code lengths to determine what size sub-table is needed. The length 163 counts are used for this, and so count[] is decremented as codes are 164 entered in the tables. 165 166 used keeps track of how many table entries have been allocated from the 167 provided *table space. It is checked for LENS and DIST tables against 168 the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in 169 the initial root table size constants. See the comments in inftrees.h 170 for more information. 171 172 sym increments through all symbols, and the loop terminates when 173 all codes of length max, i.e. all codes, have been processed. This 174 routine permits incomplete codes, so another loop after this one fills 175 in the rest of the decoding tables with invalid code markers. 176 */ 177 178 /* set up for code type */ 179 switch (type) { 180 case CODES: 181 base = extra = work; /* dummy value--not used */ 182 end = 19; 183 break; 184 case LENS: 185 base = lbase; 186 base -= 257; 187 extra = lext; 188 extra -= 257; 189 end = 256; 190 break; 191 default: /* DISTS */ 192 base = dbase; 193 extra = dext; 194 end = -1; 195 } 196 197 /* initialize state for loop */ 198 huff = 0; /* starting code */ 199 sym = 0; /* starting code symbol */ 200 len = min; /* starting code length */ 201 next = *table; /* current table to fill in */ 202 curr = root; /* current table index bits */ 203 drop = 0; /* current bits to drop from code for index */ 204 low = (unsigned)(-1); /* trigger new sub-table when len > root */ 205 used = 1U << root; /* use root table entries */ 206 mask = used - 1; /* mask for comparing low */ 207 208 /* check available table space */ 209 if ((type == LENS && used > ENOUGH_LENS) || 210 (type == DISTS && used > ENOUGH_DISTS)) 211 return 1; 212 213 /* process all codes and make table entries */ 214 for (;;) { 215 /* create table entry */ 216 here.bits = (unsigned char)(len - drop); 217 if ((int)(work[sym]) < end) { 218 here.op = (unsigned char)0; 219 here.val = work[sym]; 220 } 221 else if ((int)(work[sym]) > end) { 222 here.op = (unsigned char)(extra[work[sym]]); 223 here.val = base[work[sym]]; 224 } 225 else { 226 here.op = (unsigned char)(32 + 64); /* end of block */ 227 here.val = 0; 228 } 229 230 /* replicate for those indices with low len bits equal to huff */ 231 incr = 1U << (len - drop); 232 fill = 1U << curr; 233 min = fill; /* save offset to next table */ 234 do { 235 fill -= incr; 236 next[(huff >> drop) + fill] = here; 237 } while (fill != 0); 238 239 /* backwards increment the len-bit code huff */ 240 incr = 1U << (len - 1); 241 while (huff & incr) 242 incr >>= 1; 243 if (incr != 0) { 244 huff &= incr - 1; 245 huff += incr; 246 } 247 else 248 huff = 0; 249 250 /* go to next symbol, update count, len */ 251 sym++; 252 if (--(count[len]) == 0) { 253 if (len == max) break; 254 len = lens[work[sym]]; 255 } 256 257 /* create new sub-table if needed */ 258 if (len > root && (huff & mask) != low) { 259 /* if first time, transition to sub-tables */ 260 if (drop == 0) 261 drop = root; 262 263 /* increment past last table */ 264 next += min; /* here min is 1 << curr */ 265 266 /* determine length of next table */ 267 curr = len - drop; 268 left = (int)(1 << curr); 269 while (curr + drop < max) { 270 left -= count[curr + drop]; 271 if (left <= 0) break; 272 curr++; 273 left <<= 1; 274 } 275 276 /* check for enough space */ 277 used += 1U << curr; 278 if ((type == LENS && used > ENOUGH_LENS) || 279 (type == DISTS && used > ENOUGH_DISTS)) 280 return 1; 281 282 /* point entry in root table to sub-table */ 283 low = huff & mask; 284 (*table)[low].op = (unsigned char)curr; 285 (*table)[low].bits = (unsigned char)root; 286 (*table)[low].val = (unsigned short)(next - *table); 287 } 288 } 289 290 /* fill in remaining table entry if code is incomplete (guaranteed to have 291 at most one remaining entry, since if the code is incomplete, the 292 maximum code length that was allowed to get this far is one bit) */ 293 if (huff != 0) { 294 here.op = (unsigned char)64; /* invalid code marker */ 295 here.bits = (unsigned char)(len - drop); 296 here.val = (unsigned short)0; 297 next[huff] = here; 298 } 299 300 /* set return parameters */ 301 *table += used; 302 *bits = root; 303 return 0; 304 } 305