Line data Source code
1 : /* inftree9.c -- generate Huffman trees for efficient decoding
2 : * Copyright (C) 1995-2022 Mark Adler
3 : * For conditions of distribution and use, see copyright notice in zlib.h
4 : */
5 :
6 : #include "minified_zutil.h"
7 : #include "inftree9.h"
8 :
9 : #define MAXBITS 15
10 :
11 : #define inflate_table9 gdal_inflate_table9
12 :
13 : const char inflate9_copyright[] =
14 : " inflate9 1.2.13 Copyright 1995-2022 Mark Adler ";
15 : /*
16 : If you use the zlib library in a product, an acknowledgment is welcome
17 : in the documentation of your product. If for some reason you cannot
18 : include such an acknowledgment, I would appreciate that you keep this
19 : copyright string in the executable of your product.
20 : */
21 :
22 : /*
23 : Build a set of tables to decode the provided canonical Huffman code.
24 : The code lengths are lens[0..codes-1]. The result starts at *table,
25 : whose indices are 0..2^bits-1. work is a writable array of at least
26 : lens shorts, which is used as a work area. type is the type of code
27 : to be generated, CODES, LENS, or DISTS. On return, zero is success,
28 : -1 is an invalid code, and +1 means that ENOUGH isn't enough. table
29 : on return points to the next available entry's address. bits is the
30 : requested root table index bits, and on return it is the actual root
31 : table index bits. It will differ if the request is greater than the
32 : longest code or if it is less than the shortest code.
33 : */
34 387 : int inflate_table9(codetype type, unsigned short FAR *lens,
35 : unsigned codes, code FAR * FAR *table,
36 : unsigned FAR *bits, unsigned short FAR *work)
37 : {
38 : unsigned len; /* a code's length in bits */
39 : unsigned sym; /* index of code symbols */
40 : unsigned min, max; /* minimum and maximum code lengths */
41 : unsigned root; /* number of index bits for root table */
42 : unsigned curr; /* number of index bits for current table */
43 : unsigned drop; /* code bits to drop for sub-table */
44 : int left; /* number of prefix codes available */
45 : unsigned used; /* code entries in table used */
46 : unsigned huff; /* Huffman code */
47 : unsigned incr; /* for incrementing code, index */
48 : unsigned fill; /* index for replicating entries */
49 : unsigned low; /* low bits for current root entry */
50 : unsigned mask; /* mask for low root bits */
51 : code this; /* table entry for duplication */
52 : code FAR *next; /* next available space in table */
53 : const unsigned short FAR *base; /* base value table to use */
54 : const unsigned short FAR *extra; /* extra bits table to use */
55 : int end; /* use base and extra for symbol > end */
56 : unsigned short count[MAXBITS+1]; /* number of codes of each length */
57 : unsigned short offs[MAXBITS+1]; /* offsets in table for each length */
58 : static const unsigned short lbase[31] = { /* Length codes 257..285 base */
59 : 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17,
60 : 19, 23, 27, 31, 35, 43, 51, 59, 67, 83, 99, 115,
61 : 131, 163, 195, 227, 3, 0, 0};
62 : static const unsigned short lext[31] = { /* Length codes 257..285 extra */
63 : 128, 128, 128, 128, 128, 128, 128, 128, 129, 129, 129, 129,
64 : 130, 130, 130, 130, 131, 131, 131, 131, 132, 132, 132, 132,
65 : 133, 133, 133, 133, 144, 194, 65};
66 : static const unsigned short dbase[32] = { /* Distance codes 0..31 base */
67 : 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49,
68 : 65, 97, 129, 193, 257, 385, 513, 769, 1025, 1537, 2049, 3073,
69 : 4097, 6145, 8193, 12289, 16385, 24577, 32769, 49153};
70 : static const unsigned short dext[32] = { /* Distance codes 0..31 extra */
71 : 128, 128, 128, 128, 129, 129, 130, 130, 131, 131, 132, 132,
72 : 133, 133, 134, 134, 135, 135, 136, 136, 137, 137, 138, 138,
73 : 139, 139, 140, 140, 141, 141, 142, 142};
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 6579 : for (len = 0; len <= MAXBITS; len++)
108 6192 : count[len] = 0;
109 40647 : for (sym = 0; sym < codes; sym++)
110 40260 : count[lens[sym]]++;
111 :
112 : /* bound code lengths, force root to be within code lengths */
113 387 : root = *bits;
114 4067 : for (max = MAXBITS; max >= 1; max--)
115 4067 : if (count[max] != 0) break;
116 387 : if (root > max) root = max;
117 387 : if (max == 0) return -1; /* no codes! */
118 633 : for (min = 1; min <= MAXBITS; min++)
119 633 : if (count[min] != 0) break;
120 387 : if (root < min) root = min;
121 :
122 : /* check for an over-subscribed or incomplete set of lengths */
123 387 : left = 1;
124 6192 : for (len = 1; len <= MAXBITS; len++) {
125 5805 : left <<= 1;
126 5805 : left -= count[len];
127 5805 : if (left < 0) return -1; /* over-subscribed */
128 : }
129 387 : if (left > 0 && (type == CODES || max != 1))
130 0 : return -1; /* incomplete set */
131 :
132 : /* generate offsets into symbol table for each length for sorting */
133 387 : offs[1] = 0;
134 5805 : for (len = 1; len < MAXBITS; len++)
135 5418 : offs[len + 1] = offs[len] + count[len];
136 :
137 : /* sort symbols by length, by symbol order within each length */
138 40647 : for (sym = 0; sym < codes; sym++)
139 40260 : 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 for LENS and DIST tables against
162 : the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in
163 : the initial root table size constants. See the comments in inftree9.h
164 : for more information.
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 */
173 387 : switch (type) {
174 129 : case CODES:
175 129 : base = extra = work; /* dummy value--not used */
176 129 : end = 19;
177 129 : break;
178 129 : case LENS:
179 129 : base = lbase;
180 129 : base -= 257;
181 129 : extra = lext;
182 129 : extra -= 257;
183 129 : end = 256;
184 129 : break;
185 129 : default: /* DISTS */
186 129 : base = dbase;
187 129 : extra = dext;
188 129 : end = -1;
189 : }
190 :
191 : /* initialize state for loop */
192 387 : huff = 0; /* starting code */
193 387 : sym = 0; /* starting code symbol */
194 387 : len = min; /* starting code length */
195 387 : next = *table; /* current table to fill in */
196 387 : curr = root; /* current table index bits */
197 387 : drop = 0; /* current bits to drop from code for index */
198 387 : low = (unsigned)(-1); /* trigger new sub-table when len > root */
199 387 : used = 1U << root; /* use root table entries */
200 387 : mask = used - 1; /* mask for comparing low */
201 :
202 : /* check available table space */
203 387 : if ((type == LENS && used >= ENOUGH_LENS) ||
204 129 : (type == DISTS && used >= ENOUGH_DISTS))
205 0 : return 1;
206 :
207 : /* process all codes and make table entries */
208 : for (;;) {
209 : /* create table entry */
210 4610 : this.bits = (unsigned char)(len - drop);
211 4610 : if ((int)(work[sym]) < end) {
212 2863 : this.op = (unsigned char)0;
213 2863 : this.val = work[sym];
214 : }
215 1747 : else if ((int)(work[sym]) > end) {
216 1618 : this.op = (unsigned char)(extra[work[sym]]);
217 1618 : this.val = base[work[sym]];
218 : }
219 : else {
220 129 : this.op = (unsigned char)(32 + 64); /* end of block */
221 129 : this.val = 0;
222 : }
223 :
224 : /* replicate for those indices with low len bits equal to huff */
225 4610 : incr = 1U << (len - drop);
226 4610 : fill = 1U << curr;
227 : do {
228 68596 : fill -= incr;
229 68596 : next[(huff >> drop) + fill] = this;
230 68596 : } while (fill != 0);
231 :
232 : /* backwards increment the len-bit code huff */
233 4610 : incr = 1U << (len - 1);
234 8833 : while (huff & incr)
235 4223 : incr >>= 1;
236 4610 : if (incr != 0) {
237 4223 : huff &= incr - 1;
238 4223 : huff += incr;
239 : }
240 : else
241 387 : huff = 0;
242 :
243 : /* go to next symbol, update count, len */
244 4610 : sym++;
245 4610 : if (--(count[len]) == 0) {
246 1375 : if (len == max) break;
247 988 : len = lens[work[sym]];
248 : }
249 :
250 : /* create new sub-table if needed */
251 4223 : if (len > root && (huff & mask) != low) {
252 : /* if first time, transition to sub-tables */
253 336 : if (drop == 0)
254 121 : drop = root;
255 :
256 : /* increment past last table */
257 336 : next += (size_t)(1U << curr);
258 :
259 : /* determine length of next table */
260 336 : curr = len - drop;
261 336 : left = (int)(1 << curr);
262 340 : while (curr + drop < max) {
263 12 : left -= count[curr + drop];
264 12 : if (left <= 0) break;
265 4 : curr++;
266 4 : left <<= 1;
267 : }
268 :
269 : /* check for enough space */
270 336 : used += 1U << curr;
271 336 : if ((type == LENS && used >= ENOUGH_LENS) ||
272 12 : (type == DISTS && used >= ENOUGH_DISTS))
273 0 : return 1;
274 :
275 : /* point entry in root table to sub-table */
276 336 : low = huff & mask;
277 336 : (*table)[low].op = (unsigned char)curr;
278 336 : (*table)[low].bits = (unsigned char)root;
279 336 : (*table)[low].val = (unsigned short)(next - *table);
280 : }
281 : }
282 :
283 : /*
284 : Fill in rest of table for incomplete codes. This loop is similar to the
285 : loop above in incrementing huff for table indices. It is assumed that
286 : len is equal to curr + drop, so there is no loop needed to increment
287 : through high index bits. When the current sub-table is filled, the loop
288 : drops back to the root table to fill in any remaining entries there.
289 : */
290 387 : this.op = (unsigned char)64; /* invalid code marker */
291 387 : this.bits = (unsigned char)(len - drop);
292 387 : this.val = (unsigned short)0;
293 387 : while (huff != 0) {
294 : /* when done with sub-table, drop back to root table */
295 0 : if (drop != 0 && (huff & mask) != low) {
296 0 : drop = 0;
297 0 : len = root;
298 0 : next = *table;
299 : /* curr = root; */
300 0 : this.bits = (unsigned char)len;
301 : }
302 :
303 : /* put invalid code marker in table */
304 0 : next[huff >> drop] = this;
305 :
306 : /* backwards increment the len-bit code huff */
307 0 : incr = 1U << (len - 1);
308 0 : while (huff & incr)
309 0 : incr >>= 1;
310 0 : if (incr != 0) {
311 0 : huff &= incr - 1;
312 0 : huff += incr;
313 : }
314 : else
315 0 : huff = 0;
316 : }
317 :
318 : /* set return parameters */
319 387 : *table += used;
320 387 : *bits = root;
321 387 : return 0;
322 : }
|
