Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* inftrees.c -- generate Huffman trees for efficient decoding |
4f3865fb RP |
2 | * Copyright (C) 1995-2005 Mark Adler |
3 | * For conditions of distribution and use, see copyright notice in zlib.h | |
1da177e4 LT |
4 | */ |
5 | ||
6 | #include <linux/zutil.h> | |
7 | #include "inftrees.h" | |
1da177e4 | 8 | |
4f3865fb RP |
9 | #define MAXBITS 15 |
10 | ||
1da177e4 | 11 | /* |
4f3865fb RP |
12 | Build a set of tables to decode the provided canonical Huffman code. |
13 | The code lengths are lens[0..codes-1]. The result starts at *table, | |
14 | whose indices are 0..2^bits-1. work is a writable array of at least | |
15 | lens shorts, which is used as a work area. type is the type of code | |
16 | to be generated, CODES, LENS, or DISTS. On return, zero is success, | |
17 | -1 is an invalid code, and +1 means that ENOUGH isn't enough. table | |
18 | on return points to the next available entry's address. bits is the | |
19 | requested root table index bits, and on return it is the actual root | |
20 | table index bits. It will differ if the request is greater than the | |
21 | longest code or if it is less than the shortest code. | |
1da177e4 | 22 | */ |
4f3865fb RP |
23 | int zlib_inflate_table(type, lens, codes, table, bits, work) |
24 | codetype type; | |
25 | unsigned short *lens; | |
26 | unsigned codes; | |
27 | code **table; | |
28 | unsigned *bits; | |
29 | unsigned short *work; | |
1da177e4 | 30 | { |
4f3865fb RP |
31 | unsigned len; /* a code's length in bits */ |
32 | unsigned sym; /* index of code symbols */ | |
33 | unsigned min, max; /* minimum and maximum code lengths */ | |
34 | unsigned root; /* number of index bits for root table */ | |
35 | unsigned curr; /* number of index bits for current table */ | |
36 | unsigned drop; /* code bits to drop for sub-table */ | |
37 | int left; /* number of prefix codes available */ | |
38 | unsigned used; /* code entries in table used */ | |
39 | unsigned huff; /* Huffman code */ | |
40 | unsigned incr; /* for incrementing code, index */ | |
41 | unsigned fill; /* index for replicating entries */ | |
42 | unsigned low; /* low bits for current root entry */ | |
43 | unsigned mask; /* mask for low root bits */ | |
44 | code this; /* table entry for duplication */ | |
45 | code *next; /* next available space in table */ | |
46 | const unsigned short *base; /* base value table to use */ | |
47 | const unsigned short *extra; /* extra bits table to use */ | |
48 | int end; /* use base and extra for symbol > end */ | |
49 | unsigned short count[MAXBITS+1]; /* number of codes of each length */ | |
50 | unsigned short offs[MAXBITS+1]; /* offsets in table for each length */ | |
51 | static const unsigned short lbase[31] = { /* Length codes 257..285 base */ | |
52 | 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31, | |
53 | 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0}; | |
54 | static const unsigned short lext[31] = { /* Length codes 257..285 extra */ | |
55 | 16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18, | |
56 | 19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 201, 196}; | |
57 | static const unsigned short dbase[32] = { /* Distance codes 0..29 base */ | |
58 | 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193, | |
59 | 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145, | |
60 | 8193, 12289, 16385, 24577, 0, 0}; | |
61 | static const unsigned short dext[32] = { /* Distance codes 0..29 extra */ | |
62 | 16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22, | |
63 | 23, 23, 24, 24, 25, 25, 26, 26, 27, 27, | |
64 | 28, 28, 29, 29, 64, 64}; | |
65 | ||
66 | /* | |
67 | Process a set of code lengths to create a canonical Huffman code. The | |
68 | code lengths are lens[0..codes-1]. Each length corresponds to the | |
69 | symbols 0..codes-1. The Huffman code is generated by first sorting the | |
70 | symbols by length from short to long, and retaining the symbol order | |
71 | for codes with equal lengths. Then the code starts with all zero bits | |
72 | for the first code of the shortest length, and the codes are integer | |
73 | increments for the same length, and zeros are appended as the length | |
74 | increases. For the deflate format, these bits are stored backwards | |
75 | from their more natural integer increment ordering, and so when the | |
76 | decoding tables are built in the large loop below, the integer codes | |
77 | are incremented backwards. | |
78 | ||
79 | This routine assumes, but does not check, that all of the entries in | |
80 | lens[] are in the range 0..MAXBITS. The caller must assure this. | |
81 | 1..MAXBITS is interpreted as that code length. zero means that that | |
82 | symbol does not occur in this code. | |
83 | ||
84 | The codes are sorted by computing a count of codes for each length, | |
85 | creating from that a table of starting indices for each length in the | |
86 | sorted table, and then entering the symbols in order in the sorted | |
87 | table. The sorted table is work[], with that space being provided by | |
88 | the caller. | |
89 | ||
90 | The length counts are used for other purposes as well, i.e. finding | |
91 | the minimum and maximum length codes, determining if there are any | |
92 | codes at all, checking for a valid set of lengths, and looking ahead | |
93 | at length counts to determine sub-table sizes when building the | |
94 | decoding tables. | |
95 | */ | |
96 | ||
97 | /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */ | |
98 | for (len = 0; len <= MAXBITS; len++) | |
99 | count[len] = 0; | |
100 | for (sym = 0; sym < codes; sym++) | |
101 | count[lens[sym]]++; | |
102 | ||
103 | /* bound code lengths, force root to be within code lengths */ | |
104 | root = *bits; | |
105 | for (max = MAXBITS; max >= 1; max--) | |
106 | if (count[max] != 0) break; | |
107 | if (root > max) root = max; | |
108 | if (max == 0) { /* no symbols to code at all */ | |
109 | this.op = (unsigned char)64; /* invalid code marker */ | |
110 | this.bits = (unsigned char)1; | |
111 | this.val = (unsigned short)0; | |
112 | *(*table)++ = this; /* make a table to force an error */ | |
113 | *(*table)++ = this; | |
114 | *bits = 1; | |
115 | return 0; /* no symbols, but wait for decoding to report error */ | |
116 | } | |
117 | for (min = 1; min <= MAXBITS; min++) | |
118 | if (count[min] != 0) break; | |
119 | if (root < min) root = min; | |
120 | ||
121 | /* check for an over-subscribed or incomplete set of lengths */ | |
122 | left = 1; | |
123 | for (len = 1; len <= MAXBITS; len++) { | |
124 | left <<= 1; | |
125 | left -= count[len]; | |
126 | if (left < 0) return -1; /* over-subscribed */ | |
127 | } | |
128 | if (left > 0 && (type == CODES || max != 1)) | |
129 | return -1; /* incomplete set */ | |
130 | ||
131 | /* generate offsets into symbol table for each length for sorting */ | |
132 | offs[1] = 0; | |
133 | for (len = 1; len < MAXBITS; len++) | |
134 | offs[len + 1] = offs[len] + count[len]; | |
135 | ||
136 | /* sort symbols by length, by symbol order within each length */ | |
137 | for (sym = 0; sym < codes; sym++) | |
138 | if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym; | |
139 | ||
140 | /* | |
141 | Create and fill in decoding tables. In this loop, the table being | |
142 | filled is at next and has curr index bits. The code being used is huff | |
143 | with length len. That code is converted to an index by dropping drop | |
144 | bits off of the bottom. For codes where len is less than drop + curr, | |
145 | those top drop + curr - len bits are incremented through all values to | |
146 | fill the table with replicated entries. | |
147 | ||
148 | root is the number of index bits for the root table. When len exceeds | |
149 | root, sub-tables are created pointed to by the root entry with an index | |
150 | of the low root bits of huff. This is saved in low to check for when a | |
151 | new sub-table should be started. drop is zero when the root table is | |
152 | being filled, and drop is root when sub-tables are being filled. | |
153 | ||
154 | When a new sub-table is needed, it is necessary to look ahead in the | |
155 | code lengths to determine what size sub-table is needed. The length | |
156 | counts are used for this, and so count[] is decremented as codes are | |
157 | entered in the tables. | |
158 | ||
159 | used keeps track of how many table entries have been allocated from the | |
160 | provided *table space. It is checked when a LENS table is being made | |
161 | against the space in *table, ENOUGH, minus the maximum space needed by | |
162 | the worst case distance code, MAXD. This should never happen, but the | |
163 | sufficiency of ENOUGH has not been proven exhaustively, hence the check. | |
164 | This assumes that when type == LENS, bits == 9. | |
165 | ||
166 | sym increments through all symbols, and the loop terminates when | |
167 | all codes of length max, i.e. all codes, have been processed. This | |
168 | routine permits incomplete codes, so another loop after this one fills | |
169 | in the rest of the decoding tables with invalid code markers. | |
170 | */ | |
171 | ||
172 | /* set up for code type */ | |
173 | switch (type) { | |
174 | case CODES: | |
175 | base = extra = work; /* dummy value--not used */ | |
176 | end = 19; | |
177 | break; | |
178 | case LENS: | |
179 | base = lbase; | |
180 | base -= 257; | |
181 | extra = lext; | |
182 | extra -= 257; | |
183 | end = 256; | |
184 | break; | |
185 | default: /* DISTS */ | |
186 | base = dbase; | |
187 | extra = dext; | |
188 | end = -1; | |
189 | } | |
1da177e4 | 190 | |
4f3865fb RP |
191 | /* initialize state for loop */ |
192 | huff = 0; /* starting code */ | |
193 | sym = 0; /* starting code symbol */ | |
194 | len = min; /* starting code length */ | |
195 | next = *table; /* current table to fill in */ | |
196 | curr = root; /* current table index bits */ | |
197 | drop = 0; /* current bits to drop from code for index */ | |
198 | low = (unsigned)(-1); /* trigger new sub-table when len > root */ | |
199 | used = 1U << root; /* use root table entries */ | |
200 | mask = used - 1; /* mask for comparing low */ | |
201 | ||
202 | /* check available table space */ | |
203 | if (type == LENS && used >= ENOUGH - MAXD) | |
204 | return 1; | |
205 | ||
206 | /* process all codes and make table entries */ | |
207 | for (;;) { | |
208 | /* create table entry */ | |
209 | this.bits = (unsigned char)(len - drop); | |
210 | if ((int)(work[sym]) < end) { | |
211 | this.op = (unsigned char)0; | |
212 | this.val = work[sym]; | |
1da177e4 | 213 | } |
4f3865fb RP |
214 | else if ((int)(work[sym]) > end) { |
215 | this.op = (unsigned char)(extra[work[sym]]); | |
216 | this.val = base[work[sym]]; | |
217 | } | |
218 | else { | |
219 | this.op = (unsigned char)(32 + 64); /* end of block */ | |
220 | this.val = 0; | |
1da177e4 | 221 | } |
1da177e4 | 222 | |
4f3865fb RP |
223 | /* replicate for those indices with low len bits equal to huff */ |
224 | incr = 1U << (len - drop); | |
225 | fill = 1U << curr; | |
226 | min = fill; /* save offset to next table */ | |
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; | |
1da177e4 | 242 | |
4f3865fb RP |
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 | } | |
1da177e4 | 249 | |
4f3865fb RP |
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 += min; /* here min is 1 << 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 | } | |
1da177e4 | 268 | |
4f3865fb RP |
269 | /* check for enough space */ |
270 | used += 1U << curr; | |
271 | if (type == LENS && used >= ENOUGH - MAXD) | |
272 | return 1; | |
1da177e4 | 273 | |
4f3865fb RP |
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 | } | |
1da177e4 | 280 | } |
1da177e4 | 281 | |
4f3865fb RP |
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 | this.bits = (unsigned char)len; | |
299 | } | |
1da177e4 | 300 | |
4f3865fb RP |
301 | /* put invalid code marker in table */ |
302 | next[huff >> drop] = this; | |
1da177e4 | 303 | |
4f3865fb RP |
304 | /* backwards increment the len-bit code huff */ |
305 | incr = 1U << (len - 1); | |
306 | while (huff & incr) | |
307 | incr >>= 1; | |
308 | if (incr != 0) { | |
309 | huff &= incr - 1; | |
310 | huff += incr; | |
311 | } | |
312 | else | |
313 | huff = 0; | |
314 | } | |
315 | ||
316 | /* set return parameters */ | |
317 | *table += used; | |
318 | *bits = root; | |
319 | return 0; | |
1da177e4 | 320 | } |