]>
Commit | Line | Data |
---|---|---|
7eb75bcc DM |
1 | /* inftrees.c -- generate Huffman trees for efficient decoding\r |
2 | * Copyright (C) 1995-2013 Mark Adler\r | |
3 | * For conditions of distribution and use, see copyright notice in zlib.h\r | |
4 | */\r | |
5 | \r | |
6 | #include "zutil.h"\r | |
7 | #include "inftrees.h"\r | |
8 | \r | |
9 | #define MAXBITS 15\r | |
10 | \r | |
11 | const char inflate_copyright[] =\r | |
12 | " inflate 1.2.8 Copyright 1995-2013 Mark Adler ";\r | |
13 | /*\r | |
14 | If you use the zlib library in a product, an acknowledgment is welcome\r | |
15 | in the documentation of your product. If for some reason you cannot\r | |
16 | include such an acknowledgment, I would appreciate that you keep this\r | |
17 | copyright string in the executable of your product.\r | |
18 | */\r | |
19 | \r | |
20 | /*\r | |
21 | Build a set of tables to decode the provided canonical Huffman code.\r | |
22 | The code lengths are lens[0..codes-1]. The result starts at *table,\r | |
23 | whose indices are 0..2^bits-1. work is a writable array of at least\r | |
24 | lens shorts, which is used as a work area. type is the type of code\r | |
25 | to be generated, CODES, LENS, or DISTS. On return, zero is success,\r | |
26 | -1 is an invalid code, and +1 means that ENOUGH isn't enough. table\r | |
27 | on return points to the next available entry's address. bits is the\r | |
28 | requested root table index bits, and on return it is the actual root\r | |
29 | table index bits. It will differ if the request is greater than the\r | |
30 | longest code or if it is less than the shortest code.\r | |
31 | */\r | |
32 | int ZLIB_INTERNAL inflate_table(type, lens, codes, table, bits, work)\r | |
33 | codetype type;\r | |
34 | unsigned short FAR *lens;\r | |
35 | unsigned codes;\r | |
36 | code FAR * FAR *table;\r | |
37 | unsigned FAR *bits;\r | |
38 | unsigned short FAR *work;\r | |
39 | {\r | |
40 | unsigned len; /* a code's length in bits */\r | |
41 | unsigned sym; /* index of code symbols */\r | |
42 | unsigned min, max; /* minimum and maximum code lengths */\r | |
43 | unsigned root; /* number of index bits for root table */\r | |
44 | unsigned curr; /* number of index bits for current table */\r | |
45 | unsigned drop; /* code bits to drop for sub-table */\r | |
46 | int left; /* number of prefix codes available */\r | |
47 | unsigned used; /* code entries in table used */\r | |
48 | unsigned huff; /* Huffman code */\r | |
49 | unsigned incr; /* for incrementing code, index */\r | |
50 | unsigned fill; /* index for replicating entries */\r | |
51 | unsigned low; /* low bits for current root entry */\r | |
52 | unsigned mask; /* mask for low root bits */\r | |
53 | code here; /* table entry for duplication */\r | |
54 | code FAR *next; /* next available space in table */\r | |
55 | const unsigned short FAR *base; /* base value table to use */\r | |
56 | const unsigned short FAR *extra; /* extra bits table to use */\r | |
57 | int end; /* use base and extra for symbol > end */\r | |
58 | unsigned short count[MAXBITS+1]; /* number of codes of each length */\r | |
59 | unsigned short offs[MAXBITS+1]; /* offsets in table for each length */\r | |
60 | static const unsigned short lbase[31] = { /* Length codes 257..285 base */\r | |
61 | 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31,\r | |
62 | 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0};\r | |
63 | static const unsigned short lext[31] = { /* Length codes 257..285 extra */\r | |
64 | 16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18,\r | |
65 | 19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 72, 78};\r | |
66 | static const unsigned short dbase[32] = { /* Distance codes 0..29 base */\r | |
67 | 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,\r | |
68 | 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145,\r | |
69 | 8193, 12289, 16385, 24577, 0, 0};\r | |
70 | static const unsigned short dext[32] = { /* Distance codes 0..29 extra */\r | |
71 | 16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22,\r | |
72 | 23, 23, 24, 24, 25, 25, 26, 26, 27, 27,\r | |
73 | 28, 28, 29, 29, 64, 64};\r | |
74 | \r | |
75 | /*\r | |
76 | Process a set of code lengths to create a canonical Huffman code. The\r | |
77 | code lengths are lens[0..codes-1]. Each length corresponds to the\r | |
78 | symbols 0..codes-1. The Huffman code is generated by first sorting the\r | |
79 | symbols by length from short to long, and retaining the symbol order\r | |
80 | for codes with equal lengths. Then the code starts with all zero bits\r | |
81 | for the first code of the shortest length, and the codes are integer\r | |
82 | increments for the same length, and zeros are appended as the length\r | |
83 | increases. For the deflate format, these bits are stored backwards\r | |
84 | from their more natural integer increment ordering, and so when the\r | |
85 | decoding tables are built in the large loop below, the integer codes\r | |
86 | are incremented backwards.\r | |
87 | \r | |
88 | This routine assumes, but does not check, that all of the entries in\r | |
89 | lens[] are in the range 0..MAXBITS. The caller must assure this.\r | |
90 | 1..MAXBITS is interpreted as that code length. zero means that that\r | |
91 | symbol does not occur in this code.\r | |
92 | \r | |
93 | The codes are sorted by computing a count of codes for each length,\r | |
94 | creating from that a table of starting indices for each length in the\r | |
95 | sorted table, and then entering the symbols in order in the sorted\r | |
96 | table. The sorted table is work[], with that space being provided by\r | |
97 | the caller.\r | |
98 | \r | |
99 | The length counts are used for other purposes as well, i.e. finding\r | |
100 | the minimum and maximum length codes, determining if there are any\r | |
101 | codes at all, checking for a valid set of lengths, and looking ahead\r | |
102 | at length counts to determine sub-table sizes when building the\r | |
103 | decoding tables.\r | |
104 | */\r | |
105 | \r | |
106 | /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */\r | |
107 | for (len = 0; len <= MAXBITS; len++)\r | |
108 | count[len] = 0;\r | |
109 | for (sym = 0; sym < codes; sym++)\r | |
110 | count[lens[sym]]++;\r | |
111 | \r | |
112 | /* bound code lengths, force root to be within code lengths */\r | |
113 | root = *bits;\r | |
114 | for (max = MAXBITS; max >= 1; max--)\r | |
115 | if (count[max] != 0) break;\r | |
116 | if (root > max) root = max;\r | |
117 | if (max == 0) { /* no symbols to code at all */\r | |
118 | here.op = (unsigned char)64; /* invalid code marker */\r | |
119 | here.bits = (unsigned char)1;\r | |
120 | here.val = (unsigned short)0;\r | |
121 | *(*table)++ = here; /* make a table to force an error */\r | |
122 | *(*table)++ = here;\r | |
123 | *bits = 1;\r | |
124 | return 0; /* no symbols, but wait for decoding to report error */\r | |
125 | }\r | |
126 | for (min = 1; min < max; min++)\r | |
127 | if (count[min] != 0) break;\r | |
128 | if (root < min) root = min;\r | |
129 | \r | |
130 | /* check for an over-subscribed or incomplete set of lengths */\r | |
131 | left = 1;\r | |
132 | for (len = 1; len <= MAXBITS; len++) {\r | |
133 | left <<= 1;\r | |
134 | left -= count[len];\r | |
135 | if (left < 0) return -1; /* over-subscribed */\r | |
136 | }\r | |
137 | if (left > 0 && (type == CODES || max != 1))\r | |
138 | return -1; /* incomplete set */\r | |
139 | \r | |
140 | /* generate offsets into symbol table for each length for sorting */\r | |
141 | offs[1] = 0;\r | |
142 | for (len = 1; len < MAXBITS; len++)\r | |
143 | offs[len + 1] = offs[len] + count[len];\r | |
144 | \r | |
145 | /* sort symbols by length, by symbol order within each length */\r | |
146 | for (sym = 0; sym < codes; sym++)\r | |
147 | if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym;\r | |
148 | \r | |
149 | /*\r | |
150 | Create and fill in decoding tables. In this loop, the table being\r | |
151 | filled is at next and has curr index bits. The code being used is huff\r | |
152 | with length len. That code is converted to an index by dropping drop\r | |
153 | bits off of the bottom. For codes where len is less than drop + curr,\r | |
154 | those top drop + curr - len bits are incremented through all values to\r | |
155 | fill the table with replicated entries.\r | |
156 | \r | |
157 | root is the number of index bits for the root table. When len exceeds\r | |
158 | root, sub-tables are created pointed to by the root entry with an index\r | |
159 | of the low root bits of huff. This is saved in low to check for when a\r | |
160 | new sub-table should be started. drop is zero when the root table is\r | |
161 | being filled, and drop is root when sub-tables are being filled.\r | |
162 | \r | |
163 | When a new sub-table is needed, it is necessary to look ahead in the\r | |
164 | code lengths to determine what size sub-table is needed. The length\r | |
165 | counts are used for this, and so count[] is decremented as codes are\r | |
166 | entered in the tables.\r | |
167 | \r | |
168 | used keeps track of how many table entries have been allocated from the\r | |
169 | provided *table space. It is checked for LENS and DIST tables against\r | |
170 | the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in\r | |
171 | the initial root table size constants. See the comments in inftrees.h\r | |
172 | for more information.\r | |
173 | \r | |
174 | sym increments through all symbols, and the loop terminates when\r | |
175 | all codes of length max, i.e. all codes, have been processed. This\r | |
176 | routine permits incomplete codes, so another loop after this one fills\r | |
177 | in the rest of the decoding tables with invalid code markers.\r | |
178 | */\r | |
179 | \r | |
180 | /* set up for code type */\r | |
181 | switch (type) {\r | |
182 | case CODES:\r | |
183 | base = extra = work; /* dummy value--not used */\r | |
184 | end = 19;\r | |
185 | break;\r | |
186 | case LENS:\r | |
187 | base = lbase;\r | |
188 | base -= 257;\r | |
189 | extra = lext;\r | |
190 | extra -= 257;\r | |
191 | end = 256;\r | |
192 | break;\r | |
193 | default: /* DISTS */\r | |
194 | base = dbase;\r | |
195 | extra = dext;\r | |
196 | end = -1;\r | |
197 | }\r | |
198 | \r | |
199 | /* initialize state for loop */\r | |
200 | huff = 0; /* starting code */\r | |
201 | sym = 0; /* starting code symbol */\r | |
202 | len = min; /* starting code length */\r | |
203 | next = *table; /* current table to fill in */\r | |
204 | curr = root; /* current table index bits */\r | |
205 | drop = 0; /* current bits to drop from code for index */\r | |
206 | low = (unsigned)(-1); /* trigger new sub-table when len > root */\r | |
207 | used = 1U << root; /* use root table entries */\r | |
208 | mask = used - 1; /* mask for comparing low */\r | |
209 | \r | |
210 | /* check available table space */\r | |
211 | if ((type == LENS && used > ENOUGH_LENS) ||\r | |
212 | (type == DISTS && used > ENOUGH_DISTS))\r | |
213 | return 1;\r | |
214 | \r | |
215 | /* process all codes and make table entries */\r | |
216 | for (;;) {\r | |
217 | /* create table entry */\r | |
218 | here.bits = (unsigned char)(len - drop);\r | |
219 | if ((int)(work[sym]) < end) {\r | |
220 | here.op = (unsigned char)0;\r | |
221 | here.val = work[sym];\r | |
222 | }\r | |
223 | else if ((int)(work[sym]) > end) {\r | |
224 | here.op = (unsigned char)(extra[work[sym]]);\r | |
225 | here.val = base[work[sym]];\r | |
226 | }\r | |
227 | else {\r | |
228 | here.op = (unsigned char)(32 + 64); /* end of block */\r | |
229 | here.val = 0;\r | |
230 | }\r | |
231 | \r | |
232 | /* replicate for those indices with low len bits equal to huff */\r | |
233 | incr = 1U << (len - drop);\r | |
234 | fill = 1U << curr;\r | |
235 | min = fill; /* save offset to next table */\r | |
236 | do {\r | |
237 | fill -= incr;\r | |
238 | next[(huff >> drop) + fill] = here;\r | |
239 | } while (fill != 0);\r | |
240 | \r | |
241 | /* backwards increment the len-bit code huff */\r | |
242 | incr = 1U << (len - 1);\r | |
243 | while (huff & incr)\r | |
244 | incr >>= 1;\r | |
245 | if (incr != 0) {\r | |
246 | huff &= incr - 1;\r | |
247 | huff += incr;\r | |
248 | }\r | |
249 | else\r | |
250 | huff = 0;\r | |
251 | \r | |
252 | /* go to next symbol, update count, len */\r | |
253 | sym++;\r | |
254 | if (--(count[len]) == 0) {\r | |
255 | if (len == max) break;\r | |
256 | len = lens[work[sym]];\r | |
257 | }\r | |
258 | \r | |
259 | /* create new sub-table if needed */\r | |
260 | if (len > root && (huff & mask) != low) {\r | |
261 | /* if first time, transition to sub-tables */\r | |
262 | if (drop == 0)\r | |
263 | drop = root;\r | |
264 | \r | |
265 | /* increment past last table */\r | |
266 | next += min; /* here min is 1 << curr */\r | |
267 | \r | |
268 | /* determine length of next table */\r | |
269 | curr = len - drop;\r | |
270 | left = (int)(1 << curr);\r | |
271 | while (curr + drop < max) {\r | |
272 | left -= count[curr + drop];\r | |
273 | if (left <= 0) break;\r | |
274 | curr++;\r | |
275 | left <<= 1;\r | |
276 | }\r | |
277 | \r | |
278 | /* check for enough space */\r | |
279 | used += 1U << curr;\r | |
280 | if ((type == LENS && used > ENOUGH_LENS) ||\r | |
281 | (type == DISTS && used > ENOUGH_DISTS))\r | |
282 | return 1;\r | |
283 | \r | |
284 | /* point entry in root table to sub-table */\r | |
285 | low = huff & mask;\r | |
286 | (*table)[low].op = (unsigned char)curr;\r | |
287 | (*table)[low].bits = (unsigned char)root;\r | |
288 | (*table)[low].val = (unsigned short)(next - *table);\r | |
289 | }\r | |
290 | }\r | |
291 | \r | |
292 | /* fill in remaining table entry if code is incomplete (guaranteed to have\r | |
293 | at most one remaining entry, since if the code is incomplete, the\r | |
294 | maximum code length that was allowed to get this far is one bit) */\r | |
295 | if (huff != 0) {\r | |
296 | here.op = (unsigned char)64; /* invalid code marker */\r | |
297 | here.bits = (unsigned char)(len - drop);\r | |
298 | here.val = (unsigned short)0;\r | |
299 | next[huff] = here;\r | |
300 | }\r | |
301 | \r | |
302 | /* set return parameters */\r | |
303 | *table += used;\r | |
304 | *bits = root;\r | |
305 | return 0;\r | |
306 | }\r |