]>
Commit | Line | Data |
---|---|---|
1 | /* inftrees.c -- generate Huffman trees for efficient decoding | |
2 | * Copyright (C) 1995-2003 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 | #define MAXBITS 15 | |
10 | ||
11 | const char inflate_copyright[] = | |
12 | " inflate 1.2.1 Copyright 1995-2003 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 inflate_table(type, lens, codes, table, bits, work) | |
33 | codetype type; | |
34 | unsigned short FAR *lens; | |
35 | unsigned codes; | |
36 | code FAR * FAR *table; | |
37 | unsigned FAR *bits; | |
38 | unsigned short FAR *work; | |
39 | { | |
40 | unsigned len; /* a code's length in bits */ | |
41 | unsigned sym; /* index of code symbols */ | |
42 | unsigned min, max; /* minimum and maximum code lengths */ | |
43 | unsigned root; /* number of index bits for root table */ | |
44 | unsigned curr; /* number of index bits for current table */ | |
45 | unsigned drop; /* code bits to drop for sub-table */ | |
46 | int left; /* number of prefix codes available */ | |
47 | unsigned used; /* code entries in table used */ | |
48 | unsigned huff; /* Huffman code */ | |
49 | unsigned incr; /* for incrementing code, index */ | |
50 | unsigned fill; /* index for replicating entries */ | |
51 | unsigned low; /* low bits for current root entry */ | |
52 | unsigned mask; /* mask for low root bits */ | |
53 | code this; /* table entry for duplication */ | |
54 | code FAR *next; /* next available space in table */ | |
55 | const unsigned short FAR *base; /* base value table to use */ | |
56 | const unsigned short FAR *extra; /* extra bits table to use */ | |
57 | int end; /* use base and extra for symbol > end */ | |
58 | unsigned short count[MAXBITS+1]; /* number of codes of each length */ | |
59 | unsigned short offs[MAXBITS+1]; /* offsets in table for each length */ | |
60 | static const unsigned short lbase[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}; | |
63 | static const unsigned short lext[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, 76, 66}; | |
66 | static const unsigned short dbase[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}; | |
70 | static const unsigned short dext[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) */ | |
107 | for (len = 0; len <= MAXBITS; len++) | |
108 | count[len] = 0; | |
109 | for (sym = 0; sym < codes; sym++) | |
110 | count[lens[sym]]++; | |
111 | ||
112 | /* bound code lengths, force root to be within code lengths */ | |
113 | root = *bits; | |
114 | for (max = MAXBITS; max >= 1; max--) | |
115 | if (count[max] != 0) break; | |
116 | if (root > max) root = max; | |
117 | if (max == 0) return -1; /* no codes! */ | |
118 | for (min = 1; min <= MAXBITS; min++) | |
119 | if (count[min] != 0) break; | |
120 | if (root < min) root = min; | |
121 | ||
122 | /* check for an over-subscribed or incomplete set of lengths */ | |
123 | left = 1; | |
124 | for (len = 1; len <= MAXBITS; len++) { | |
125 | left <<= 1; | |
126 | left -= count[len]; | |
127 | if (left < 0) return -1; /* over-subscribed */ | |
128 | } | |
129 | if (left > 0 && (type == CODES || (codes - count[0] != 1))) | |
130 | return -1; /* incomplete set */ | |
131 | ||
132 | /* generate offsets into symbol table for each length for sorting */ | |
133 | offs[1] = 0; | |
134 | for (len = 1; len < MAXBITS; len++) | |
135 | offs[len + 1] = offs[len] + count[len]; | |
136 | ||
137 | /* sort symbols by length, by symbol order within each length */ | |
138 | for (sym = 0; sym < codes; sym++) | |
139 | if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym; | |
140 | ||
141 | /* | |
142 | Create and fill in decoding tables. In this loop, the table being | |
143 | filled is at next and has curr index bits. The code being used is huff | |
144 | with length len. That code is converted to an index by dropping drop | |
145 | bits off of the bottom. For codes where len is less than drop + curr, | |
146 | those top drop + curr - len bits are incremented through all values to | |
147 | fill the table with replicated entries. | |
148 | ||
149 | root is the number of index bits for the root table. When len exceeds | |
150 | root, sub-tables are created pointed to by the root entry with an index | |
151 | of the low root bits of huff. This is saved in low to check for when a | |
152 | new sub-table should be started. drop is zero when the root table is | |
153 | being filled, and drop is root when sub-tables are being filled. | |
154 | ||
155 | When a new sub-table is needed, it is necessary to look ahead in the | |
156 | code lengths to determine what size sub-table is needed. The length | |
157 | counts are used for this, and so count[] is decremented as codes are | |
158 | entered in the tables. | |
159 | ||
160 | used keeps track of how many table entries have been allocated from the | |
161 | provided *table space. It is checked when a LENS table is being made | |
162 | against the space in *table, ENOUGH, minus the maximum space needed by | |
163 | the worst case distance code, MAXD. This should never happen, but the | |
164 | sufficiency of ENOUGH has not been proven exhaustively, hence the check. | |
165 | This assumes that when type == LENS, bits == 9. | |
166 | ||
167 | sym increments through all symbols, and the loop terminates when | |
168 | all codes of length max, i.e. all codes, have been processed. This | |
169 | routine permits incomplete codes, so another loop after this one fills | |
170 | in the rest of the decoding tables with invalid code markers. | |
171 | */ | |
172 | ||
173 | /* set up for code type */ | |
174 | switch (type) { | |
175 | case CODES: | |
176 | base = extra = work; /* dummy value--not used */ | |
177 | end = 19; | |
178 | break; | |
179 | case LENS: | |
180 | base = lbase; | |
181 | base -= 257; | |
182 | extra = lext; | |
183 | extra -= 257; | |
184 | end = 256; | |
185 | break; | |
186 | default: /* DISTS */ | |
187 | base = dbase; | |
188 | extra = dext; | |
189 | end = -1; | |
190 | } | |
191 | ||
192 | /* initialize state for loop */ | |
193 | huff = 0; /* starting code */ | |
194 | sym = 0; /* starting code symbol */ | |
195 | len = min; /* starting code length */ | |
196 | next = *table; /* current table to fill in */ | |
197 | curr = root; /* current table index bits */ | |
198 | drop = 0; /* current bits to drop from code for index */ | |
199 | low = (unsigned)(-1); /* trigger new sub-table when len > root */ | |
200 | used = 1U << root; /* use root table entries */ | |
201 | mask = used - 1; /* mask for comparing low */ | |
202 | ||
203 | /* check available table space */ | |
204 | if (type == LENS && used >= ENOUGH - MAXD) | |
205 | return 1; | |
206 | ||
207 | /* process all codes and make table entries */ | |
208 | for (;;) { | |
209 | /* create table entry */ | |
210 | this.bits = (unsigned char)(len - drop); | |
211 | if ((int)(work[sym]) < end) { | |
212 | this.op = (unsigned char)0; | |
213 | this.val = work[sym]; | |
214 | } | |
215 | else if ((int)(work[sym]) > end) { | |
216 | this.op = (unsigned char)(extra[work[sym]]); | |
217 | this.val = base[work[sym]]; | |
218 | } | |
219 | else { | |
220 | this.op = (unsigned char)(32 + 64); /* end of block */ | |
221 | this.val = 0; | |
222 | } | |
223 | ||
224 | /* replicate for those indices with low len bits equal to huff */ | |
225 | incr = 1U << (len - drop); | |
226 | fill = 1U << curr; | |
227 | do { | |
228 | fill -= incr; | |
229 | next[(huff >> drop) + fill] = this; | |
230 | } while (fill != 0); | |
231 | ||
232 | /* backwards increment the len-bit code huff */ | |
233 | incr = 1U << (len - 1); | |
234 | while (huff & incr) | |
235 | incr >>= 1; | |
236 | if (incr != 0) { | |
237 | huff &= incr - 1; | |
238 | huff += incr; | |
239 | } | |
240 | else | |
241 | huff = 0; | |
242 | ||
243 | /* go to next symbol, update count, len */ | |
244 | sym++; | |
245 | if (--(count[len]) == 0) { | |
246 | if (len == max) break; | |
247 | len = lens[work[sym]]; | |
248 | } | |
249 | ||
250 | /* create new sub-table if needed */ | |
251 | if (len > root && (huff & mask) != low) { | |
252 | /* if first time, transition to sub-tables */ | |
253 | if (drop == 0) | |
254 | drop = root; | |
255 | ||
256 | /* increment past last table */ | |
257 | next += 1U << curr; | |
258 | ||
259 | /* determine length of next table */ | |
260 | curr = len - drop; | |
261 | left = (int)(1 << curr); | |
262 | while (curr + drop < max) { | |
263 | left -= count[curr + drop]; | |
264 | if (left <= 0) break; | |
265 | curr++; | |
266 | left <<= 1; | |
267 | } | |
268 | ||
269 | /* check for enough space */ | |
270 | used += 1U << curr; | |
271 | if (type == LENS && used >= ENOUGH - MAXD) | |
272 | return 1; | |
273 | ||
274 | /* point entry in root table to sub-table */ | |
275 | low = huff & mask; | |
276 | (*table)[low].op = (unsigned char)curr; | |
277 | (*table)[low].bits = (unsigned char)root; | |
278 | (*table)[low].val = (unsigned short)(next - *table); | |
279 | } | |
280 | } | |
281 | ||
282 | /* | |
283 | Fill in rest of table for incomplete codes. This loop is similar to the | |
284 | loop above in incrementing huff for table indices. It is assumed that | |
285 | len is equal to curr + drop, so there is no loop needed to increment | |
286 | through high index bits. When the current sub-table is filled, the loop | |
287 | drops back to the root table to fill in any remaining entries there. | |
288 | */ | |
289 | this.op = (unsigned char)64; /* invalid code marker */ | |
290 | this.bits = (unsigned char)(len - drop); | |
291 | this.val = (unsigned short)0; | |
292 | while (huff != 0) { | |
293 | /* when done with sub-table, drop back to root table */ | |
294 | if (drop != 0 && (huff & mask) != low) { | |
295 | drop = 0; | |
296 | len = root; | |
297 | next = *table; | |
298 | curr = root; | |
299 | this.bits = (unsigned char)len; | |
300 | } | |
301 | ||
302 | /* put invalid code marker in table */ | |
303 | next[huff >> drop] = this; | |
304 | ||
305 | /* backwards increment the len-bit code huff */ | |
306 | incr = 1U << (len - 1); | |
307 | while (huff & incr) | |
308 | incr >>= 1; | |
309 | if (incr != 0) { | |
310 | huff &= incr - 1; | |
311 | huff += incr; | |
312 | } | |
313 | else | |
314 | huff = 0; | |
315 | } | |
316 | ||
317 | /* set return parameters */ | |
318 | *table += used; | |
319 | *bits = root; | |
320 | return 0; | |
321 | } |