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"zutil.h" 7#include"inftrees.h" 8 9#defineMAXBITS 15 10 11constcharinflate_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 */ 32intZLIB_INTERNAL inflate_table( 33codetype type, 34unsignedshortFAR *lens, 35unsignedcodes, 36code FAR * FAR *table, 37unsignedFAR *bits, 38unsignedshortFAR *work) 39{ 40unsignedlen; /* a code's length in bits */ 41unsignedsym; /* index of code symbols */ 42unsignedmin, max; /* minimum and maximum code lengths */ 43unsignedroot; /* number of index bits for root table */ 44unsignedcurr; /* number of index bits for current table */ 45unsigneddrop; /* code bits to drop for sub-table */ 46intleft; /* number of prefix codes available */ 47unsignedused; /* code entries in table used */ 48unsignedhuff; /* Huffman code */ 49unsignedincr; /* for incrementing code, index */ 50unsignedfill; /* index for replicating entries */ 51unsignedlow; /* low bits for current root entry */ 52unsignedmask; /* mask for low root bits */ 53 code here; /* table entry for duplication */ 54 code FAR *next; /* next available space in table */ 55constunsignedshortFAR *base; /* base value table to use */ 56constunsignedshortFAR *extra; /* extra bits table to use */ 57intend; /* use base and extra for symbol > end */ 58unsignedshortcount[MAXBITS+1]; /* number of codes of each length */ 59unsignedshortoffs[MAXBITS+1]; /* offsets in table for each length */ 60staticconstunsignedshortlbase[31] = { /* Length codes 257..285 base */ 61 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31, 62 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0}; 63staticconstunsignedshortlext[31] = { /* Length codes 257..285 extra */ 64 16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18, 65 19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 72, 78}; 66staticconstunsignedshortdbase[32] = { /* Distance codes 0..29 base */ 67 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193, 68 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145, 69 8193, 12289, 16385, 24577, 0, 0}; 70staticconstunsignedshortdext[32] = { /* Distance codes 0..29 extra */ 71 16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22, 72 23, 23, 24, 24, 25, 25, 26, 26, 27, 27, 73 28, 28, 29, 29, 64, 64}; 74 75 /* 76 Process a set of code lengths to create a canonical Huffman code. The 77 code lengths are lens[0..codes-1]. Each length corresponds to the 78 symbols 0..codes-1. The Huffman code is generated by first sorting the 79 symbols by length from short to long, and retaining the symbol order 80 for codes with equal lengths. Then the code starts with all zero bits 81 for the first code of the shortest length, and the codes are integer 82 increments for the same length, and zeros are appended as the length 83 increases. For the deflate format, these bits are stored backwards 84 from their more natural integer increment ordering, and so when the 85 decoding tables are built in the large loop below, the integer codes 86 are incremented backwards. 87 88 This routine assumes, but does not check, that all of the entries in 89 lens[] are in the range 0..MAXBITS. The caller must assure this. 90 1..MAXBITS is interpreted as that code length. zero means that that 91 symbol does not occur in this code. 92 93 The codes are sorted by computing a count of codes for each length, 94 creating from that a table of starting indices for each length in the 95 sorted table, and then entering the symbols in order in the sorted 96 table. The sorted table is work[], with that space being provided by 97 the caller. 98 99 The length counts are used for other purposes as well, i.e. finding 100 the minimum and maximum length codes, determining if there are any 101 codes at all, checking for a valid set of lengths, and looking ahead 102 at length counts to determine sub-table sizes when building the 103 decoding tables. 104 */ 105 106 /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */ 107for(len = 0; len <= MAXBITS; len++) 108 count[len] = 0; 109for(sym = 0; sym < codes; sym++) 110 count[lens[sym]]++; 111 112 /* bound code lengths, force root to be within code lengths */ 113 root = *bits; 114for(max = MAXBITS; max >= 1; max--) 115if(count[max] != 0)break; 116if(root > max) root = max; 117if(max == 0) { /* no symbols to code at all */ 118 here.op = (unsignedchar)64; /* invalid code marker */ 119 here.bits = (unsignedchar)1; 120 here.val = (unsignedshort)0; 121 *(*table)++ = here; /* make a table to force an error */ 122 *(*table)++ = here; 123 *bits = 1; 124return0; /* no symbols, but wait for decoding to report error */ 125 } 126for(min = 1; min < max; min++) 127if(count[min] != 0)break; 128if(root < min) root = min; 129 130 /* check for an over-subscribed or incomplete set of lengths */ 131 left = 1; 132for(len = 1; len <= MAXBITS; len++) { 133 left <<= 1; 134 left -= count[len]; 135if(left < 0)return-1; /* over-subscribed */ 136 } 137if(left > 0 && (type == CODES || max != 1)) 138return-1; /* incomplete set */ 139 140 /* generate offsets into symbol table for each length for sorting */ 141 offs[1] = 0; 142for(len = 1; len < MAXBITS; len++) 143 offs[len + 1] = offs[len] + count[len]; 144 145 /* sort symbols by length, by symbol order within each length */ 146for(sym = 0; sym < codes; sym++) 147if(lens[sym] != 0) work[offs[lens[sym]]++] = (unsignedshort)sym; 148 149 /* 150 Create and fill in decoding tables. In this loop, the table being 151 filled is at next and has curr index bits. The code being used is huff 152 with length len. That code is converted to an index by dropping drop 153 bits off of the bottom. For codes where len is less than drop + curr, 154 those top drop + curr - len bits are incremented through all values to 155 fill the table with replicated entries. 156 157 root is the number of index bits for the root table. When len exceeds 158 root, sub-tables are created pointed to by the root entry with an index 159 of the low root bits of huff. This is saved in low to check for when a 160 new sub-table should be started. drop is zero when the root table is 161 being filled, and drop is root when sub-tables are being filled. 162 163 When a new sub-table is needed, it is necessary to look ahead in the 164 code lengths to determine what size sub-table is needed. The length 165 counts are used for this, and so count[] is decremented as codes are 166 entered in the tables. 167 168 used keeps track of how many table entries have been allocated from the 169 provided *table space. It is checked for LENS and DIST tables against 170 the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in 171 the initial root table size constants. See the comments in inftrees.h 172 for more information. 173 174 sym increments through all symbols, and the loop terminates when 175 all codes of length max, i.e. all codes, have been processed. This 176 routine permits incomplete codes, so another loop after this one fills 177 in the rest of the decoding tables with invalid code markers. 178 */ 179 180 /* set up for code type */ 181switch(type) { 182caseCODES: 183 base = extra = work; /* dummy value--not used */ 184 end = 19; 185break; 186caseLENS: 187 base = lbase; 188 base -= 257; 189 extra = lext; 190 extra -= 257; 191 end = 256; 192break; 193default: /* DISTS */ 194 base = dbase; 195 extra = dext; 196 end = -1; 197 } 198 199 /* initialize state for loop */ 200 huff = 0; /* starting code */ 201 sym = 0; /* starting code symbol */ 202 len = min; /* starting code length */ 203 next = *table; /* current table to fill in */ 204 curr = root; /* current table index bits */ 205 drop = 0; /* current bits to drop from code for index */ 206 low = (unsigned)(-1); /* trigger new sub-table when len > root */ 207 used = 1U << root; /* use root table entries */ 208 mask = used - 1; /* mask for comparing low */ 209 210 /* check available table space */ 211if((type == LENS && used > ENOUGH_LENS) || 212 (type == DISTS && used > ENOUGH_DISTS)) 213return1; 214 215 /* process all codes and make table entries */ 216for(;;) { 217 /* create table entry */ 218 here.bits = (unsignedchar)(len - drop); 219if((int)(work[sym]) < end) { 220 here.op = (unsignedchar)0; 221 here.val = work[sym]; 222 } 223elseif((int)(work[sym]) > end) { 224 here.op = (unsignedchar)(extra[work[sym]]); 225 here.val = base[work[sym]]; 226 } 227else{ 228 here.op = (unsignedchar)(32 + 64); /* end of block */ 229 here.val = 0; 230 } 231 232 /* replicate for those indices with low len bits equal to huff */ 233 incr = 1U << (len - drop); 234 fill = 1U << curr; 235 min = fill; /* save offset to next table */ 236do{ 237 fill -= incr; 238 next[(huff >> drop) + fill] = here; 239 }while(fill != 0); 240 241 /* backwards increment the len-bit code huff */ 242 incr = 1U << (len - 1); 243while(huff & incr) 244 incr >>= 1; 245if(incr != 0) { 246 huff &= incr - 1; 247 huff += incr; 248 } 249else250 huff = 0; 251 252 /* go to next symbol, update count, len */ 253 sym++; 254if(--(count[len]) == 0) { 255if(len == max)break; 256 len = lens[work[sym]]; 257 } 258 259 /* create new sub-table if needed */ 260if(len > root && (huff & mask) != low) { 261 /* if first time, transition to sub-tables */ 262if(drop == 0) 263 drop = root; 264 265 /* increment past last table */ 266 next += min; /* here min is 1 << curr */ 267 268 /* determine length of next table */ 269 curr = len - drop; 270 left = (int)(1 << curr); 271while(curr + drop < max) { 272 left -= count[curr + drop]; 273if(left <= 0)break; 274 curr++; 275 left <<= 1; 276 } 277 278 /* check for enough space */ 279 used += 1U << curr; 280if((type == LENS && used > ENOUGH_LENS) || 281 (type == DISTS && used > ENOUGH_DISTS)) 282return1; 283 284 /* point entry in root table to sub-table */ 285 low = huff & mask; 286 (*table)[low].op = (unsignedchar)curr; 287 (*table)[low].bits = (unsignedchar)root; 288 (*table)[low].val = (unsignedshort)(next - *table); 289 } 290 } 291 292 /* fill in remaining table entry if code is incomplete (guaranteed to have 293 at most one remaining entry, since if the code is incomplete, the 294 maximum code length that was allowed to get this far is one bit) */ 295if(huff != 0) { 296 here.op = (unsignedchar)64; /* invalid code marker */ 297 here.bits = (unsignedchar)(len - drop); 298 here.val = (unsignedshort)0; 299 next[huff] = here; 300 } 301 302 /* set return parameters */ 303 *table += used; 304 *bits = root; 305return0; 306} 307