cannam@128
|
1 /* inftrees.c -- generate Huffman trees for efficient decoding
|
cannam@128
|
2 * Copyright (C) 1995-2013 Mark Adler
|
cannam@128
|
3 * For conditions of distribution and use, see copyright notice in zlib.h
|
cannam@128
|
4 */
|
cannam@128
|
5
|
cannam@128
|
6 #include "zutil.h"
|
cannam@128
|
7 #include "inftrees.h"
|
cannam@128
|
8
|
cannam@128
|
9 #define MAXBITS 15
|
cannam@128
|
10
|
cannam@128
|
11 const char inflate_copyright[] =
|
cannam@128
|
12 " inflate 1.2.8 Copyright 1995-2013 Mark Adler ";
|
cannam@128
|
13 /*
|
cannam@128
|
14 If you use the zlib library in a product, an acknowledgment is welcome
|
cannam@128
|
15 in the documentation of your product. If for some reason you cannot
|
cannam@128
|
16 include such an acknowledgment, I would appreciate that you keep this
|
cannam@128
|
17 copyright string in the executable of your product.
|
cannam@128
|
18 */
|
cannam@128
|
19
|
cannam@128
|
20 /*
|
cannam@128
|
21 Build a set of tables to decode the provided canonical Huffman code.
|
cannam@128
|
22 The code lengths are lens[0..codes-1]. The result starts at *table,
|
cannam@128
|
23 whose indices are 0..2^bits-1. work is a writable array of at least
|
cannam@128
|
24 lens shorts, which is used as a work area. type is the type of code
|
cannam@128
|
25 to be generated, CODES, LENS, or DISTS. On return, zero is success,
|
cannam@128
|
26 -1 is an invalid code, and +1 means that ENOUGH isn't enough. table
|
cannam@128
|
27 on return points to the next available entry's address. bits is the
|
cannam@128
|
28 requested root table index bits, and on return it is the actual root
|
cannam@128
|
29 table index bits. It will differ if the request is greater than the
|
cannam@128
|
30 longest code or if it is less than the shortest code.
|
cannam@128
|
31 */
|
cannam@128
|
32 int ZLIB_INTERNAL inflate_table(type, lens, codes, table, bits, work)
|
cannam@128
|
33 codetype type;
|
cannam@128
|
34 unsigned short FAR *lens;
|
cannam@128
|
35 unsigned codes;
|
cannam@128
|
36 code FAR * FAR *table;
|
cannam@128
|
37 unsigned FAR *bits;
|
cannam@128
|
38 unsigned short FAR *work;
|
cannam@128
|
39 {
|
cannam@128
|
40 unsigned len; /* a code's length in bits */
|
cannam@128
|
41 unsigned sym; /* index of code symbols */
|
cannam@128
|
42 unsigned min, max; /* minimum and maximum code lengths */
|
cannam@128
|
43 unsigned root; /* number of index bits for root table */
|
cannam@128
|
44 unsigned curr; /* number of index bits for current table */
|
cannam@128
|
45 unsigned drop; /* code bits to drop for sub-table */
|
cannam@128
|
46 int left; /* number of prefix codes available */
|
cannam@128
|
47 unsigned used; /* code entries in table used */
|
cannam@128
|
48 unsigned huff; /* Huffman code */
|
cannam@128
|
49 unsigned incr; /* for incrementing code, index */
|
cannam@128
|
50 unsigned fill; /* index for replicating entries */
|
cannam@128
|
51 unsigned low; /* low bits for current root entry */
|
cannam@128
|
52 unsigned mask; /* mask for low root bits */
|
cannam@128
|
53 code here; /* table entry for duplication */
|
cannam@128
|
54 code FAR *next; /* next available space in table */
|
cannam@128
|
55 const unsigned short FAR *base; /* base value table to use */
|
cannam@128
|
56 const unsigned short FAR *extra; /* extra bits table to use */
|
cannam@128
|
57 int end; /* use base and extra for symbol > end */
|
cannam@128
|
58 unsigned short count[MAXBITS+1]; /* number of codes of each length */
|
cannam@128
|
59 unsigned short offs[MAXBITS+1]; /* offsets in table for each length */
|
cannam@128
|
60 static const unsigned short lbase[31] = { /* Length codes 257..285 base */
|
cannam@128
|
61 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31,
|
cannam@128
|
62 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0};
|
cannam@128
|
63 static const unsigned short lext[31] = { /* Length codes 257..285 extra */
|
cannam@128
|
64 16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18,
|
cannam@128
|
65 19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 72, 78};
|
cannam@128
|
66 static const unsigned short dbase[32] = { /* Distance codes 0..29 base */
|
cannam@128
|
67 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,
|
cannam@128
|
68 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145,
|
cannam@128
|
69 8193, 12289, 16385, 24577, 0, 0};
|
cannam@128
|
70 static const unsigned short dext[32] = { /* Distance codes 0..29 extra */
|
cannam@128
|
71 16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22,
|
cannam@128
|
72 23, 23, 24, 24, 25, 25, 26, 26, 27, 27,
|
cannam@128
|
73 28, 28, 29, 29, 64, 64};
|
cannam@128
|
74
|
cannam@128
|
75 /*
|
cannam@128
|
76 Process a set of code lengths to create a canonical Huffman code. The
|
cannam@128
|
77 code lengths are lens[0..codes-1]. Each length corresponds to the
|
cannam@128
|
78 symbols 0..codes-1. The Huffman code is generated by first sorting the
|
cannam@128
|
79 symbols by length from short to long, and retaining the symbol order
|
cannam@128
|
80 for codes with equal lengths. Then the code starts with all zero bits
|
cannam@128
|
81 for the first code of the shortest length, and the codes are integer
|
cannam@128
|
82 increments for the same length, and zeros are appended as the length
|
cannam@128
|
83 increases. For the deflate format, these bits are stored backwards
|
cannam@128
|
84 from their more natural integer increment ordering, and so when the
|
cannam@128
|
85 decoding tables are built in the large loop below, the integer codes
|
cannam@128
|
86 are incremented backwards.
|
cannam@128
|
87
|
cannam@128
|
88 This routine assumes, but does not check, that all of the entries in
|
cannam@128
|
89 lens[] are in the range 0..MAXBITS. The caller must assure this.
|
cannam@128
|
90 1..MAXBITS is interpreted as that code length. zero means that that
|
cannam@128
|
91 symbol does not occur in this code.
|
cannam@128
|
92
|
cannam@128
|
93 The codes are sorted by computing a count of codes for each length,
|
cannam@128
|
94 creating from that a table of starting indices for each length in the
|
cannam@128
|
95 sorted table, and then entering the symbols in order in the sorted
|
cannam@128
|
96 table. The sorted table is work[], with that space being provided by
|
cannam@128
|
97 the caller.
|
cannam@128
|
98
|
cannam@128
|
99 The length counts are used for other purposes as well, i.e. finding
|
cannam@128
|
100 the minimum and maximum length codes, determining if there are any
|
cannam@128
|
101 codes at all, checking for a valid set of lengths, and looking ahead
|
cannam@128
|
102 at length counts to determine sub-table sizes when building the
|
cannam@128
|
103 decoding tables.
|
cannam@128
|
104 */
|
cannam@128
|
105
|
cannam@128
|
106 /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */
|
cannam@128
|
107 for (len = 0; len <= MAXBITS; len++)
|
cannam@128
|
108 count[len] = 0;
|
cannam@128
|
109 for (sym = 0; sym < codes; sym++)
|
cannam@128
|
110 count[lens[sym]]++;
|
cannam@128
|
111
|
cannam@128
|
112 /* bound code lengths, force root to be within code lengths */
|
cannam@128
|
113 root = *bits;
|
cannam@128
|
114 for (max = MAXBITS; max >= 1; max--)
|
cannam@128
|
115 if (count[max] != 0) break;
|
cannam@128
|
116 if (root > max) root = max;
|
cannam@128
|
117 if (max == 0) { /* no symbols to code at all */
|
cannam@128
|
118 here.op = (unsigned char)64; /* invalid code marker */
|
cannam@128
|
119 here.bits = (unsigned char)1;
|
cannam@128
|
120 here.val = (unsigned short)0;
|
cannam@128
|
121 *(*table)++ = here; /* make a table to force an error */
|
cannam@128
|
122 *(*table)++ = here;
|
cannam@128
|
123 *bits = 1;
|
cannam@128
|
124 return 0; /* no symbols, but wait for decoding to report error */
|
cannam@128
|
125 }
|
cannam@128
|
126 for (min = 1; min < max; min++)
|
cannam@128
|
127 if (count[min] != 0) break;
|
cannam@128
|
128 if (root < min) root = min;
|
cannam@128
|
129
|
cannam@128
|
130 /* check for an over-subscribed or incomplete set of lengths */
|
cannam@128
|
131 left = 1;
|
cannam@128
|
132 for (len = 1; len <= MAXBITS; len++) {
|
cannam@128
|
133 left <<= 1;
|
cannam@128
|
134 left -= count[len];
|
cannam@128
|
135 if (left < 0) return -1; /* over-subscribed */
|
cannam@128
|
136 }
|
cannam@128
|
137 if (left > 0 && (type == CODES || max != 1))
|
cannam@128
|
138 return -1; /* incomplete set */
|
cannam@128
|
139
|
cannam@128
|
140 /* generate offsets into symbol table for each length for sorting */
|
cannam@128
|
141 offs[1] = 0;
|
cannam@128
|
142 for (len = 1; len < MAXBITS; len++)
|
cannam@128
|
143 offs[len + 1] = offs[len] + count[len];
|
cannam@128
|
144
|
cannam@128
|
145 /* sort symbols by length, by symbol order within each length */
|
cannam@128
|
146 for (sym = 0; sym < codes; sym++)
|
cannam@128
|
147 if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym;
|
cannam@128
|
148
|
cannam@128
|
149 /*
|
cannam@128
|
150 Create and fill in decoding tables. In this loop, the table being
|
cannam@128
|
151 filled is at next and has curr index bits. The code being used is huff
|
cannam@128
|
152 with length len. That code is converted to an index by dropping drop
|
cannam@128
|
153 bits off of the bottom. For codes where len is less than drop + curr,
|
cannam@128
|
154 those top drop + curr - len bits are incremented through all values to
|
cannam@128
|
155 fill the table with replicated entries.
|
cannam@128
|
156
|
cannam@128
|
157 root is the number of index bits for the root table. When len exceeds
|
cannam@128
|
158 root, sub-tables are created pointed to by the root entry with an index
|
cannam@128
|
159 of the low root bits of huff. This is saved in low to check for when a
|
cannam@128
|
160 new sub-table should be started. drop is zero when the root table is
|
cannam@128
|
161 being filled, and drop is root when sub-tables are being filled.
|
cannam@128
|
162
|
cannam@128
|
163 When a new sub-table is needed, it is necessary to look ahead in the
|
cannam@128
|
164 code lengths to determine what size sub-table is needed. The length
|
cannam@128
|
165 counts are used for this, and so count[] is decremented as codes are
|
cannam@128
|
166 entered in the tables.
|
cannam@128
|
167
|
cannam@128
|
168 used keeps track of how many table entries have been allocated from the
|
cannam@128
|
169 provided *table space. It is checked for LENS and DIST tables against
|
cannam@128
|
170 the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in
|
cannam@128
|
171 the initial root table size constants. See the comments in inftrees.h
|
cannam@128
|
172 for more information.
|
cannam@128
|
173
|
cannam@128
|
174 sym increments through all symbols, and the loop terminates when
|
cannam@128
|
175 all codes of length max, i.e. all codes, have been processed. This
|
cannam@128
|
176 routine permits incomplete codes, so another loop after this one fills
|
cannam@128
|
177 in the rest of the decoding tables with invalid code markers.
|
cannam@128
|
178 */
|
cannam@128
|
179
|
cannam@128
|
180 /* set up for code type */
|
cannam@128
|
181 switch (type) {
|
cannam@128
|
182 case CODES:
|
cannam@128
|
183 base = extra = work; /* dummy value--not used */
|
cannam@128
|
184 end = 19;
|
cannam@128
|
185 break;
|
cannam@128
|
186 case LENS:
|
cannam@128
|
187 base = lbase;
|
cannam@128
|
188 base -= 257;
|
cannam@128
|
189 extra = lext;
|
cannam@128
|
190 extra -= 257;
|
cannam@128
|
191 end = 256;
|
cannam@128
|
192 break;
|
cannam@128
|
193 default: /* DISTS */
|
cannam@128
|
194 base = dbase;
|
cannam@128
|
195 extra = dext;
|
cannam@128
|
196 end = -1;
|
cannam@128
|
197 }
|
cannam@128
|
198
|
cannam@128
|
199 /* initialize state for loop */
|
cannam@128
|
200 huff = 0; /* starting code */
|
cannam@128
|
201 sym = 0; /* starting code symbol */
|
cannam@128
|
202 len = min; /* starting code length */
|
cannam@128
|
203 next = *table; /* current table to fill in */
|
cannam@128
|
204 curr = root; /* current table index bits */
|
cannam@128
|
205 drop = 0; /* current bits to drop from code for index */
|
cannam@128
|
206 low = (unsigned)(-1); /* trigger new sub-table when len > root */
|
cannam@128
|
207 used = 1U << root; /* use root table entries */
|
cannam@128
|
208 mask = used - 1; /* mask for comparing low */
|
cannam@128
|
209
|
cannam@128
|
210 /* check available table space */
|
cannam@128
|
211 if ((type == LENS && used > ENOUGH_LENS) ||
|
cannam@128
|
212 (type == DISTS && used > ENOUGH_DISTS))
|
cannam@128
|
213 return 1;
|
cannam@128
|
214
|
cannam@128
|
215 /* process all codes and make table entries */
|
cannam@128
|
216 for (;;) {
|
cannam@128
|
217 /* create table entry */
|
cannam@128
|
218 here.bits = (unsigned char)(len - drop);
|
cannam@128
|
219 if ((int)(work[sym]) < end) {
|
cannam@128
|
220 here.op = (unsigned char)0;
|
cannam@128
|
221 here.val = work[sym];
|
cannam@128
|
222 }
|
cannam@128
|
223 else if ((int)(work[sym]) > end) {
|
cannam@128
|
224 here.op = (unsigned char)(extra[work[sym]]);
|
cannam@128
|
225 here.val = base[work[sym]];
|
cannam@128
|
226 }
|
cannam@128
|
227 else {
|
cannam@128
|
228 here.op = (unsigned char)(32 + 64); /* end of block */
|
cannam@128
|
229 here.val = 0;
|
cannam@128
|
230 }
|
cannam@128
|
231
|
cannam@128
|
232 /* replicate for those indices with low len bits equal to huff */
|
cannam@128
|
233 incr = 1U << (len - drop);
|
cannam@128
|
234 fill = 1U << curr;
|
cannam@128
|
235 min = fill; /* save offset to next table */
|
cannam@128
|
236 do {
|
cannam@128
|
237 fill -= incr;
|
cannam@128
|
238 next[(huff >> drop) + fill] = here;
|
cannam@128
|
239 } while (fill != 0);
|
cannam@128
|
240
|
cannam@128
|
241 /* backwards increment the len-bit code huff */
|
cannam@128
|
242 incr = 1U << (len - 1);
|
cannam@128
|
243 while (huff & incr)
|
cannam@128
|
244 incr >>= 1;
|
cannam@128
|
245 if (incr != 0) {
|
cannam@128
|
246 huff &= incr - 1;
|
cannam@128
|
247 huff += incr;
|
cannam@128
|
248 }
|
cannam@128
|
249 else
|
cannam@128
|
250 huff = 0;
|
cannam@128
|
251
|
cannam@128
|
252 /* go to next symbol, update count, len */
|
cannam@128
|
253 sym++;
|
cannam@128
|
254 if (--(count[len]) == 0) {
|
cannam@128
|
255 if (len == max) break;
|
cannam@128
|
256 len = lens[work[sym]];
|
cannam@128
|
257 }
|
cannam@128
|
258
|
cannam@128
|
259 /* create new sub-table if needed */
|
cannam@128
|
260 if (len > root && (huff & mask) != low) {
|
cannam@128
|
261 /* if first time, transition to sub-tables */
|
cannam@128
|
262 if (drop == 0)
|
cannam@128
|
263 drop = root;
|
cannam@128
|
264
|
cannam@128
|
265 /* increment past last table */
|
cannam@128
|
266 next += min; /* here min is 1 << curr */
|
cannam@128
|
267
|
cannam@128
|
268 /* determine length of next table */
|
cannam@128
|
269 curr = len - drop;
|
cannam@128
|
270 left = (int)(1 << curr);
|
cannam@128
|
271 while (curr + drop < max) {
|
cannam@128
|
272 left -= count[curr + drop];
|
cannam@128
|
273 if (left <= 0) break;
|
cannam@128
|
274 curr++;
|
cannam@128
|
275 left <<= 1;
|
cannam@128
|
276 }
|
cannam@128
|
277
|
cannam@128
|
278 /* check for enough space */
|
cannam@128
|
279 used += 1U << curr;
|
cannam@128
|
280 if ((type == LENS && used > ENOUGH_LENS) ||
|
cannam@128
|
281 (type == DISTS && used > ENOUGH_DISTS))
|
cannam@128
|
282 return 1;
|
cannam@128
|
283
|
cannam@128
|
284 /* point entry in root table to sub-table */
|
cannam@128
|
285 low = huff & mask;
|
cannam@128
|
286 (*table)[low].op = (unsigned char)curr;
|
cannam@128
|
287 (*table)[low].bits = (unsigned char)root;
|
cannam@128
|
288 (*table)[low].val = (unsigned short)(next - *table);
|
cannam@128
|
289 }
|
cannam@128
|
290 }
|
cannam@128
|
291
|
cannam@128
|
292 /* fill in remaining table entry if code is incomplete (guaranteed to have
|
cannam@128
|
293 at most one remaining entry, since if the code is incomplete, the
|
cannam@128
|
294 maximum code length that was allowed to get this far is one bit) */
|
cannam@128
|
295 if (huff != 0) {
|
cannam@128
|
296 here.op = (unsigned char)64; /* invalid code marker */
|
cannam@128
|
297 here.bits = (unsigned char)(len - drop);
|
cannam@128
|
298 here.val = (unsigned short)0;
|
cannam@128
|
299 next[huff] = here;
|
cannam@128
|
300 }
|
cannam@128
|
301
|
cannam@128
|
302 /* set return parameters */
|
cannam@128
|
303 *table += used;
|
cannam@128
|
304 *bits = root;
|
cannam@128
|
305 return 0;
|
cannam@128
|
306 }
|