00001 /* inftrees.c -- generate Huffman trees for efficient decoding 00002 * Copyright (C) 1995-2010 Mark Adler 00003 * For conditions of distribution and use, see copyright notice in zlib.h 00004 */ 00005 00006 #include "zutil.h" 00007 #include "inftrees.h" 00008 00009 #define MAXBITS 15 00010 00011 const char inflate_copyright[] = 00012 " inflate 1.2.5 Copyright 1995-2010 Mark Adler "; 00013 /* 00014 If you use the zlib library in a product, an acknowledgment is welcome 00015 in the documentation of your product. If for some reason you cannot 00016 include such an acknowledgment, I would appreciate that you keep this 00017 copyright string in the executable of your product. 00018 */ 00019 00020 /* 00021 Build a set of tables to decode the provided canonical Huffman code. 00022 The code lengths are lens[0..codes-1]. The result starts at *table, 00023 whose indices are 0..2^bits-1. work is a writable array of at least 00024 lens shorts, which is used as a work area. type is the type of code 00025 to be generated, CODES, LENS, or DISTS. On return, zero is success, 00026 -1 is an invalid code, and +1 means that ENOUGH isn't enough. table 00027 on return points to the next available entry's address. bits is the 00028 requested root table index bits, and on return it is the actual root 00029 table index bits. It will differ if the request is greater than the 00030 longest code or if it is less than the shortest code. 00031 */ 00032 int ZLIB_INTERNAL inflate_table(type, lens, codes, table, bits, work) 00033 codetype type; 00034 unsigned short FAR *lens; 00035 unsigned codes; 00036 code FAR * FAR *table; 00037 unsigned FAR *bits; 00038 unsigned short FAR *work; 00039 { 00040 unsigned len; /* a code's length in bits */ 00041 unsigned sym; /* index of code symbols */ 00042 unsigned min, max; /* minimum and maximum code lengths */ 00043 unsigned root; /* number of index bits for root table */ 00044 unsigned curr; /* number of index bits for current table */ 00045 unsigned drop; /* code bits to drop for sub-table */ 00046 int left; /* number of prefix codes available */ 00047 unsigned used; /* code entries in table used */ 00048 unsigned huff; /* Huffman code */ 00049 unsigned incr; /* for incrementing code, index */ 00050 unsigned fill; /* index for replicating entries */ 00051 unsigned low; /* low bits for current root entry */ 00052 unsigned mask; /* mask for low root bits */ 00053 code here; /* table entry for duplication */ 00054 code FAR *next; /* next available space in table */ 00055 const unsigned short FAR *base; /* base value table to use */ 00056 const unsigned short FAR *extra; /* extra bits table to use */ 00057 int end; /* use base and extra for symbol > end */ 00058 unsigned short count[MAXBITS+1]; /* number of codes of each length */ 00059 unsigned short offs[MAXBITS+1]; /* offsets in table for each length */ 00060 static const unsigned short lbase[31] = { /* Length codes 257..285 base */ 00061 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31, 00062 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0}; 00063 static const unsigned short lext[31] = { /* Length codes 257..285 extra */ 00064 16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18, 00065 19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 73, 195}; 00066 static const unsigned short dbase[32] = { /* Distance codes 0..29 base */ 00067 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193, 00068 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145, 00069 8193, 12289, 16385, 24577, 0, 0}; 00070 static const unsigned short dext[32] = { /* Distance codes 0..29 extra */ 00071 16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22, 00072 23, 23, 24, 24, 25, 25, 26, 26, 27, 27, 00073 28, 28, 29, 29, 64, 64}; 00074 00075 /* 00076 Process a set of code lengths to create a canonical Huffman code. The 00077 code lengths are lens[0..codes-1]. Each length corresponds to the 00078 symbols 0..codes-1. The Huffman code is generated by first sorting the 00079 symbols by length from short to long, and retaining the symbol order 00080 for codes with equal lengths. Then the code starts with all zero bits 00081 for the first code of the shortest length, and the codes are integer 00082 increments for the same length, and zeros are appended as the length 00083 increases. For the deflate format, these bits are stored backwards 00084 from their more natural integer increment ordering, and so when the 00085 decoding tables are built in the large loop below, the integer codes 00086 are incremented backwards. 00087 00088 This routine assumes, but does not check, that all of the entries in 00089 lens[] are in the range 0..MAXBITS. The caller must assure this. 00090 1..MAXBITS is interpreted as that code length. zero means that that 00091 symbol does not occur in this code. 00092 00093 The codes are sorted by computing a count of codes for each length, 00094 creating from that a table of starting indices for each length in the 00095 sorted table, and then entering the symbols in order in the sorted 00096 table. The sorted table is work[], with that space being provided by 00097 the caller. 00098 00099 The length counts are used for other purposes as well, i.e. finding 00100 the minimum and maximum length codes, determining if there are any 00101 codes at all, checking for a valid set of lengths, and looking ahead 00102 at length counts to determine sub-table sizes when building the 00103 decoding tables. 00104 */ 00105 00106 /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */ 00107 for (len = 0; len <= MAXBITS; len++) 00108 count[len] = 0; 00109 for (sym = 0; sym < codes; sym++) 00110 count[lens[sym]]++; 00111 00112 /* bound code lengths, force root to be within code lengths */ 00113 root = *bits; 00114 for (max = MAXBITS; max >= 1; max--) 00115 if (count[max] != 0) break; 00116 if (root > max) root = max; 00117 if (max == 0) { /* no symbols to code at all */ 00118 here.op = (unsigned char)64; /* invalid code marker */ 00119 here.bits = (unsigned char)1; 00120 here.val = (unsigned short)0; 00121 *(*table)++ = here; /* make a table to force an error */ 00122 *(*table)++ = here; 00123 *bits = 1; 00124 return 0; /* no symbols, but wait for decoding to report error */ 00125 } 00126 for (min = 1; min < max; min++) 00127 if (count[min] != 0) break; 00128 if (root < min) root = min; 00129 00130 /* check for an over-subscribed or incomplete set of lengths */ 00131 left = 1; 00132 for (len = 1; len <= MAXBITS; len++) { 00133 left <<= 1; 00134 left -= count[len]; 00135 if (left < 0) return -1; /* over-subscribed */ 00136 } 00137 if (left > 0 && (type == CODES || max != 1)) 00138 return -1; /* incomplete set */ 00139 00140 /* generate offsets into symbol table for each length for sorting */ 00141 offs[1] = 0; 00142 for (len = 1; len < MAXBITS; len++) 00143 offs[len + 1] = offs[len] + count[len]; 00144 00145 /* sort symbols by length, by symbol order within each length */ 00146 for (sym = 0; sym < codes; sym++) 00147 if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym; 00148 00149 /* 00150 Create and fill in decoding tables. In this loop, the table being 00151 filled is at next and has curr index bits. The code being used is huff 00152 with length len. That code is converted to an index by dropping drop 00153 bits off of the bottom. For codes where len is less than drop + curr, 00154 those top drop + curr - len bits are incremented through all values to 00155 fill the table with replicated entries. 00156 00157 root is the number of index bits for the root table. When len exceeds 00158 root, sub-tables are created pointed to by the root entry with an index 00159 of the low root bits of huff. This is saved in low to check for when a 00160 new sub-table should be started. drop is zero when the root table is 00161 being filled, and drop is root when sub-tables are being filled. 00162 00163 When a new sub-table is needed, it is necessary to look ahead in the 00164 code lengths to determine what size sub-table is needed. The length 00165 counts are used for this, and so count[] is decremented as codes are 00166 entered in the tables. 00167 00168 used keeps track of how many table entries have been allocated from the 00169 provided *table space. It is checked for LENS and DIST tables against 00170 the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in 00171 the initial root table size constants. See the comments in inftrees.h 00172 for more information. 00173 00174 sym increments through all symbols, and the loop terminates when 00175 all codes of length max, i.e. all codes, have been processed. This 00176 routine permits incomplete codes, so another loop after this one fills 00177 in the rest of the decoding tables with invalid code markers. 00178 */ 00179 00180 /* set up for code type */ 00181 switch (type) { 00182 case CODES: 00183 base = extra = work; /* dummy value--not used */ 00184 end = 19; 00185 break; 00186 case LENS: 00187 base = lbase; 00188 base -= 257; 00189 extra = lext; 00190 extra -= 257; 00191 end = 256; 00192 break; 00193 default: /* DISTS */ 00194 base = dbase; 00195 extra = dext; 00196 end = -1; 00197 } 00198 00199 /* initialize state for loop */ 00200 huff = 0; /* starting code */ 00201 sym = 0; /* starting code symbol */ 00202 len = min; /* starting code length */ 00203 next = *table; /* current table to fill in */ 00204 curr = root; /* current table index bits */ 00205 drop = 0; /* current bits to drop from code for index */ 00206 low = (unsigned)(-1); /* trigger new sub-table when len > root */ 00207 used = 1U << root; /* use root table entries */ 00208 mask = used - 1; /* mask for comparing low */ 00209 00210 /* check available table space */ 00211 if ((type == LENS && used >= ENOUGH_LENS) || 00212 (type == DISTS && used >= ENOUGH_DISTS)) 00213 return 1; 00214 00215 /* process all codes and make table entries */ 00216 for (;;) { 00217 /* create table entry */ 00218 here.bits = (unsigned char)(len - drop); 00219 if ((int)(work[sym]) < end) { 00220 here.op = (unsigned char)0; 00221 here.val = work[sym]; 00222 } 00223 else if ((int)(work[sym]) > end) { 00224 here.op = (unsigned char)(extra[work[sym]]); 00225 here.val = base[work[sym]]; 00226 } 00227 else { 00228 here.op = (unsigned char)(32 + 64); /* end of block */ 00229 here.val = 0; 00230 } 00231 00232 /* replicate for those indices with low len bits equal to huff */ 00233 incr = 1U << (len - drop); 00234 fill = 1U << curr; 00235 min = fill; /* save offset to next table */ 00236 do { 00237 fill -= incr; 00238 next[(huff >> drop) + fill] = here; 00239 } while (fill != 0); 00240 00241 /* backwards increment the len-bit code huff */ 00242 incr = 1U << (len - 1); 00243 while (huff & incr) 00244 incr >>= 1; 00245 if (incr != 0) { 00246 huff &= incr - 1; 00247 huff += incr; 00248 } 00249 else 00250 huff = 0; 00251 00252 /* go to next symbol, update count, len */ 00253 sym++; 00254 if (--(count[len]) == 0) { 00255 if (len == max) break; 00256 len = lens[work[sym]]; 00257 } 00258 00259 /* create new sub-table if needed */ 00260 if (len > root && (huff & mask) != low) { 00261 /* if first time, transition to sub-tables */ 00262 if (drop == 0) 00263 drop = root; 00264 00265 /* increment past last table */ 00266 next += min; /* here min is 1 << curr */ 00267 00268 /* determine length of next table */ 00269 curr = len - drop; 00270 left = (int)(1 << curr); 00271 while (curr + drop < max) { 00272 left -= count[curr + drop]; 00273 if (left <= 0) break; 00274 curr++; 00275 left <<= 1; 00276 } 00277 00278 /* check for enough space */ 00279 used += 1U << curr; 00280 if ((type == LENS && used >= ENOUGH_LENS) || 00281 (type == DISTS && used >= ENOUGH_DISTS)) 00282 return 1; 00283 00284 /* point entry in root table to sub-table */ 00285 low = huff & mask; 00286 (*table)[low].op = (unsigned char)curr; 00287 (*table)[low].bits = (unsigned char)root; 00288 (*table)[low].val = (unsigned short)(next - *table); 00289 } 00290 } 00291 00292 /* 00293 Fill in rest of table for incomplete codes. This loop is similar to the 00294 loop above in incrementing huff for table indices. It is assumed that 00295 len is equal to curr + drop, so there is no loop needed to increment 00296 through high index bits. When the current sub-table is filled, the loop 00297 drops back to the root table to fill in any remaining entries there. 00298 */ 00299 here.op = (unsigned char)64; /* invalid code marker */ 00300 here.bits = (unsigned char)(len - drop); 00301 here.val = (unsigned short)0; 00302 while (huff != 0) { 00303 /* when done with sub-table, drop back to root table */ 00304 if (drop != 0 && (huff & mask) != low) { 00305 drop = 0; 00306 len = root; 00307 next = *table; 00308 here.bits = (unsigned char)len; 00309 } 00310 00311 /* put invalid code marker in table */ 00312 next[huff >> drop] = here; 00313 00314 /* backwards increment the len-bit code huff */ 00315 incr = 1U << (len - 1); 00316 while (huff & incr) 00317 incr >>= 1; 00318 if (incr != 0) { 00319 huff &= incr - 1; 00320 huff += incr; 00321 } 00322 else 00323 huff = 0; 00324 } 00325 00326 /* set return parameters */ 00327 *table += used; 00328 *bits = root; 00329 return 0; 00330 }