9e052883 |
1 | /* inftree9.c -- generate Huffman trees for efficient decoding |
648db22b |
2 | * Copyright (C) 1995-2024 Mark Adler |
b24e7fce |
3 | * For conditions of distribution and use, see copyright notice in zlib.h |
4 | */ |
5 | |
6 | #include "zutil.h" |
9e052883 |
7 | #include "inftree9.h" |
b24e7fce |
8 | |
9 | #define MAXBITS 15 |
10 | |
9e052883 |
11 | const char inflate9_copyright[] = |
648db22b |
12 | " inflate9 1.3.1 Copyright 1995-2024 Mark Adler "; |
b24e7fce |
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 | */ |
648db22b |
32 | int inflate_table9(codetype type, unsigned short FAR *lens, unsigned codes, |
33 | code FAR * FAR *table, unsigned FAR *bits, |
34 | unsigned short FAR *work) { |
b24e7fce |
35 | unsigned len; /* a code's length in bits */ |
36 | unsigned sym; /* index of code symbols */ |
37 | unsigned min, max; /* minimum and maximum code lengths */ |
38 | unsigned root; /* number of index bits for root table */ |
39 | unsigned curr; /* number of index bits for current table */ |
40 | unsigned drop; /* code bits to drop for sub-table */ |
41 | int left; /* number of prefix codes available */ |
42 | unsigned used; /* code entries in table used */ |
43 | unsigned huff; /* Huffman code */ |
44 | unsigned incr; /* for incrementing code, index */ |
45 | unsigned fill; /* index for replicating entries */ |
46 | unsigned low; /* low bits for current root entry */ |
47 | unsigned mask; /* mask for low root bits */ |
9e052883 |
48 | code this; /* table entry for duplication */ |
b24e7fce |
49 | code FAR *next; /* next available space in table */ |
50 | const unsigned short FAR *base; /* base value table to use */ |
51 | const unsigned short FAR *extra; /* extra bits table to use */ |
9e052883 |
52 | int end; /* use base and extra for symbol > end */ |
b24e7fce |
53 | unsigned short count[MAXBITS+1]; /* number of codes of each length */ |
54 | unsigned short offs[MAXBITS+1]; /* offsets in table for each length */ |
55 | static const unsigned short lbase[31] = { /* Length codes 257..285 base */ |
9e052883 |
56 | 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, |
57 | 19, 23, 27, 31, 35, 43, 51, 59, 67, 83, 99, 115, |
58 | 131, 163, 195, 227, 3, 0, 0}; |
b24e7fce |
59 | static const unsigned short lext[31] = { /* Length codes 257..285 extra */ |
9e052883 |
60 | 128, 128, 128, 128, 128, 128, 128, 128, 129, 129, 129, 129, |
61 | 130, 130, 130, 130, 131, 131, 131, 131, 132, 132, 132, 132, |
648db22b |
62 | 133, 133, 133, 133, 144, 203, 77}; |
9e052883 |
63 | static const unsigned short dbase[32] = { /* Distance codes 0..31 base */ |
64 | 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, |
65 | 65, 97, 129, 193, 257, 385, 513, 769, 1025, 1537, 2049, 3073, |
66 | 4097, 6145, 8193, 12289, 16385, 24577, 32769, 49153}; |
67 | static const unsigned short dext[32] = { /* Distance codes 0..31 extra */ |
68 | 128, 128, 128, 128, 129, 129, 130, 130, 131, 131, 132, 132, |
69 | 133, 133, 134, 134, 135, 135, 136, 136, 137, 137, 138, 138, |
70 | 139, 139, 140, 140, 141, 141, 142, 142}; |
b24e7fce |
71 | |
72 | /* |
73 | Process a set of code lengths to create a canonical Huffman code. The |
74 | code lengths are lens[0..codes-1]. Each length corresponds to the |
75 | symbols 0..codes-1. The Huffman code is generated by first sorting the |
76 | symbols by length from short to long, and retaining the symbol order |
77 | for codes with equal lengths. Then the code starts with all zero bits |
78 | for the first code of the shortest length, and the codes are integer |
79 | increments for the same length, and zeros are appended as the length |
80 | increases. For the deflate format, these bits are stored backwards |
81 | from their more natural integer increment ordering, and so when the |
82 | decoding tables are built in the large loop below, the integer codes |
83 | are incremented backwards. |
84 | |
85 | This routine assumes, but does not check, that all of the entries in |
86 | lens[] are in the range 0..MAXBITS. The caller must assure this. |
87 | 1..MAXBITS is interpreted as that code length. zero means that that |
88 | symbol does not occur in this code. |
89 | |
90 | The codes are sorted by computing a count of codes for each length, |
91 | creating from that a table of starting indices for each length in the |
92 | sorted table, and then entering the symbols in order in the sorted |
93 | table. The sorted table is work[], with that space being provided by |
94 | the caller. |
95 | |
96 | The length counts are used for other purposes as well, i.e. finding |
97 | the minimum and maximum length codes, determining if there are any |
98 | codes at all, checking for a valid set of lengths, and looking ahead |
99 | at length counts to determine sub-table sizes when building the |
100 | decoding tables. |
101 | */ |
102 | |
103 | /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */ |
104 | for (len = 0; len <= MAXBITS; len++) |
105 | count[len] = 0; |
106 | for (sym = 0; sym < codes; sym++) |
107 | count[lens[sym]]++; |
108 | |
109 | /* bound code lengths, force root to be within code lengths */ |
110 | root = *bits; |
111 | for (max = MAXBITS; max >= 1; max--) |
112 | if (count[max] != 0) break; |
113 | if (root > max) root = max; |
9e052883 |
114 | if (max == 0) return -1; /* no codes! */ |
115 | for (min = 1; min <= MAXBITS; min++) |
b24e7fce |
116 | if (count[min] != 0) break; |
117 | if (root < min) root = min; |
118 | |
119 | /* check for an over-subscribed or incomplete set of lengths */ |
120 | left = 1; |
121 | for (len = 1; len <= MAXBITS; len++) { |
122 | left <<= 1; |
123 | left -= count[len]; |
124 | if (left < 0) return -1; /* over-subscribed */ |
125 | } |
126 | if (left > 0 && (type == CODES || max != 1)) |
127 | return -1; /* incomplete set */ |
128 | |
129 | /* generate offsets into symbol table for each length for sorting */ |
130 | offs[1] = 0; |
131 | for (len = 1; len < MAXBITS; len++) |
132 | offs[len + 1] = offs[len] + count[len]; |
133 | |
134 | /* sort symbols by length, by symbol order within each length */ |
135 | for (sym = 0; sym < codes; sym++) |
136 | if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym; |
137 | |
138 | /* |
139 | Create and fill in decoding tables. In this loop, the table being |
140 | filled is at next and has curr index bits. The code being used is huff |
141 | with length len. That code is converted to an index by dropping drop |
142 | bits off of the bottom. For codes where len is less than drop + curr, |
143 | those top drop + curr - len bits are incremented through all values to |
144 | fill the table with replicated entries. |
145 | |
146 | root is the number of index bits for the root table. When len exceeds |
147 | root, sub-tables are created pointed to by the root entry with an index |
148 | of the low root bits of huff. This is saved in low to check for when a |
149 | new sub-table should be started. drop is zero when the root table is |
150 | being filled, and drop is root when sub-tables are being filled. |
151 | |
152 | When a new sub-table is needed, it is necessary to look ahead in the |
153 | code lengths to determine what size sub-table is needed. The length |
154 | counts are used for this, and so count[] is decremented as codes are |
155 | entered in the tables. |
156 | |
157 | used keeps track of how many table entries have been allocated from the |
158 | provided *table space. It is checked for LENS and DIST tables against |
159 | the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in |
9e052883 |
160 | the initial root table size constants. See the comments in inftree9.h |
b24e7fce |
161 | for more information. |
162 | |
163 | sym increments through all symbols, and the loop terminates when |
164 | all codes of length max, i.e. all codes, have been processed. This |
165 | routine permits incomplete codes, so another loop after this one fills |
166 | in the rest of the decoding tables with invalid code markers. |
167 | */ |
168 | |
169 | /* set up for code type */ |
170 | switch (type) { |
171 | case CODES: |
172 | base = extra = work; /* dummy value--not used */ |
9e052883 |
173 | end = 19; |
b24e7fce |
174 | break; |
175 | case LENS: |
176 | base = lbase; |
9e052883 |
177 | base -= 257; |
b24e7fce |
178 | extra = lext; |
9e052883 |
179 | extra -= 257; |
180 | end = 256; |
b24e7fce |
181 | break; |
9e052883 |
182 | default: /* DISTS */ |
b24e7fce |
183 | base = dbase; |
184 | extra = dext; |
9e052883 |
185 | end = -1; |
b24e7fce |
186 | } |
187 | |
188 | /* initialize state for loop */ |
189 | huff = 0; /* starting code */ |
190 | sym = 0; /* starting code symbol */ |
191 | len = min; /* starting code length */ |
192 | next = *table; /* current table to fill in */ |
193 | curr = root; /* current table index bits */ |
194 | drop = 0; /* current bits to drop from code for index */ |
195 | low = (unsigned)(-1); /* trigger new sub-table when len > root */ |
196 | used = 1U << root; /* use root table entries */ |
197 | mask = used - 1; /* mask for comparing low */ |
198 | |
199 | /* check available table space */ |
9e052883 |
200 | if ((type == LENS && used >= ENOUGH_LENS) || |
201 | (type == DISTS && used >= ENOUGH_DISTS)) |
b24e7fce |
202 | return 1; |
203 | |
204 | /* process all codes and make table entries */ |
205 | for (;;) { |
206 | /* create table entry */ |
9e052883 |
207 | this.bits = (unsigned char)(len - drop); |
208 | if ((int)(work[sym]) < end) { |
209 | this.op = (unsigned char)0; |
210 | this.val = work[sym]; |
b24e7fce |
211 | } |
9e052883 |
212 | else if ((int)(work[sym]) > end) { |
213 | this.op = (unsigned char)(extra[work[sym]]); |
214 | this.val = base[work[sym]]; |
b24e7fce |
215 | } |
216 | else { |
9e052883 |
217 | this.op = (unsigned char)(32 + 64); /* end of block */ |
218 | this.val = 0; |
b24e7fce |
219 | } |
220 | |
221 | /* replicate for those indices with low len bits equal to huff */ |
222 | incr = 1U << (len - drop); |
223 | fill = 1U << curr; |
b24e7fce |
224 | do { |
225 | fill -= incr; |
9e052883 |
226 | next[(huff >> drop) + fill] = this; |
b24e7fce |
227 | } while (fill != 0); |
228 | |
229 | /* backwards increment the len-bit code huff */ |
230 | incr = 1U << (len - 1); |
231 | while (huff & incr) |
232 | incr >>= 1; |
233 | if (incr != 0) { |
234 | huff &= incr - 1; |
235 | huff += incr; |
236 | } |
237 | else |
238 | huff = 0; |
239 | |
240 | /* go to next symbol, update count, len */ |
241 | sym++; |
242 | if (--(count[len]) == 0) { |
243 | if (len == max) break; |
244 | len = lens[work[sym]]; |
245 | } |
246 | |
247 | /* create new sub-table if needed */ |
248 | if (len > root && (huff & mask) != low) { |
249 | /* if first time, transition to sub-tables */ |
250 | if (drop == 0) |
251 | drop = root; |
252 | |
253 | /* increment past last table */ |
9e052883 |
254 | next += 1U << curr; |
b24e7fce |
255 | |
256 | /* determine length of next table */ |
257 | curr = len - drop; |
258 | left = (int)(1 << curr); |
259 | while (curr + drop < max) { |
260 | left -= count[curr + drop]; |
261 | if (left <= 0) break; |
262 | curr++; |
263 | left <<= 1; |
264 | } |
265 | |
266 | /* check for enough space */ |
267 | used += 1U << curr; |
9e052883 |
268 | if ((type == LENS && used >= ENOUGH_LENS) || |
269 | (type == DISTS && used >= ENOUGH_DISTS)) |
b24e7fce |
270 | return 1; |
271 | |
272 | /* point entry in root table to sub-table */ |
273 | low = huff & mask; |
274 | (*table)[low].op = (unsigned char)curr; |
275 | (*table)[low].bits = (unsigned char)root; |
276 | (*table)[low].val = (unsigned short)(next - *table); |
277 | } |
278 | } |
279 | |
9e052883 |
280 | /* |
281 | Fill in rest of table for incomplete codes. This loop is similar to the |
282 | loop above in incrementing huff for table indices. It is assumed that |
283 | len is equal to curr + drop, so there is no loop needed to increment |
284 | through high index bits. When the current sub-table is filled, the loop |
285 | drops back to the root table to fill in any remaining entries there. |
286 | */ |
287 | this.op = (unsigned char)64; /* invalid code marker */ |
288 | this.bits = (unsigned char)(len - drop); |
289 | this.val = (unsigned short)0; |
290 | while (huff != 0) { |
291 | /* when done with sub-table, drop back to root table */ |
292 | if (drop != 0 && (huff & mask) != low) { |
293 | drop = 0; |
294 | len = root; |
295 | next = *table; |
296 | curr = root; |
297 | this.bits = (unsigned char)len; |
298 | } |
299 | |
300 | /* put invalid code marker in table */ |
301 | next[huff >> drop] = this; |
302 | |
303 | /* backwards increment the len-bit code huff */ |
304 | incr = 1U << (len - 1); |
305 | while (huff & incr) |
306 | incr >>= 1; |
307 | if (incr != 0) { |
308 | huff &= incr - 1; |
309 | huff += incr; |
310 | } |
311 | else |
312 | huff = 0; |
b24e7fce |
313 | } |
314 | |
315 | /* set return parameters */ |
316 | *table += used; |
317 | *bits = root; |
318 | return 0; |
319 | } |