Annotation of sys/lib/libz/inftrees.c, Revision 1.1.1.1
1.1 nbrk 1: /* $OpenBSD: inftrees.c,v 1.14 2005/07/20 15:56:46 millert Exp $ */
2: /* inftrees.c -- generate Huffman trees for efficient decoding
3: * Copyright (C) 1995-2005 Mark Adler
4: * For conditions of distribution and use, see copyright notice in zlib.h
5: */
6:
7: #include "zutil.h"
8: #include "inftrees.h"
9:
10: #define MAXBITS 15
11:
12: const char inflate_copyright[] =
13: " inflate 1.2.3 Copyright 1995-2005 Mark Adler ";
14: /*
15: If you use the zlib library in a product, an acknowledgment is welcome
16: in the documentation of your product. If for some reason you cannot
17: include such an acknowledgment, I would appreciate that you keep this
18: copyright string in the executable of your product.
19: */
20:
21: /*
22: Build a set of tables to decode the provided canonical Huffman code.
23: The code lengths are lens[0..codes-1]. The result starts at *table,
24: whose indices are 0..2^bits-1. work is a writable array of at least
25: lens shorts, which is used as a work area. type is the type of code
26: to be generated, CODES, LENS, or DISTS. On return, zero is success,
27: -1 is an invalid code, and +1 means that ENOUGH isn't enough. table
28: on return points to the next available entry's address. bits is the
29: requested root table index bits, and on return it is the actual root
30: table index bits. It will differ if the request is greater than the
31: longest code or if it is less than the shortest code.
32: */
33: int inflate_table(type, lens, codes, table, bits, work)
34: codetype type;
35: unsigned short FAR *lens;
36: unsigned codes;
37: code FAR * FAR *table;
38: unsigned FAR *bits;
39: unsigned short FAR *work;
40: {
41: unsigned len; /* a code's length in bits */
42: unsigned sym; /* index of code symbols */
43: unsigned min, max; /* minimum and maximum code lengths */
44: unsigned root; /* number of index bits for root table */
45: unsigned curr; /* number of index bits for current table */
46: unsigned drop; /* code bits to drop for sub-table */
47: int left; /* number of prefix codes available */
48: unsigned used; /* code entries in table used */
49: unsigned huff; /* Huffman code */
50: unsigned incr; /* for incrementing code, index */
51: unsigned fill; /* index for replicating entries */
52: unsigned low; /* low bits for current root entry */
53: unsigned mask; /* mask for low root bits */
54: code this; /* table entry for duplication */
55: code FAR *next; /* next available space in table */
56: const unsigned short FAR *base; /* base value table to use */
57: const unsigned short FAR *extra; /* extra bits table to use */
58: int end; /* use base and extra for symbol > end */
59: unsigned short count[MAXBITS+1]; /* number of codes of each length */
60: unsigned short offs[MAXBITS+1]; /* offsets in table for each length */
61: static const unsigned short lbase[31] = { /* Length codes 257..285 base */
62: 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31,
63: 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0};
64: static const unsigned short lext[31] = { /* Length codes 257..285 extra */
65: 16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18,
66: 19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 201, 196};
67: static const unsigned short dbase[32] = { /* Distance codes 0..29 base */
68: 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,
69: 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145,
70: 8193, 12289, 16385, 24577, 0, 0};
71: static const unsigned short dext[32] = { /* Distance codes 0..29 extra */
72: 16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22,
73: 23, 23, 24, 24, 25, 25, 26, 26, 27, 27,
74: 28, 28, 29, 29, 64, 64};
75:
76: /*
77: Process a set of code lengths to create a canonical Huffman code. The
78: code lengths are lens[0..codes-1]. Each length corresponds to the
79: symbols 0..codes-1. The Huffman code is generated by first sorting the
80: symbols by length from short to long, and retaining the symbol order
81: for codes with equal lengths. Then the code starts with all zero bits
82: for the first code of the shortest length, and the codes are integer
83: increments for the same length, and zeros are appended as the length
84: increases. For the deflate format, these bits are stored backwards
85: from their more natural integer increment ordering, and so when the
86: decoding tables are built in the large loop below, the integer codes
87: are incremented backwards.
88:
89: This routine assumes, but does not check, that all of the entries in
90: lens[] are in the range 0..MAXBITS. The caller must assure this.
91: 1..MAXBITS is interpreted as that code length. zero means that that
92: symbol does not occur in this code.
93:
94: The codes are sorted by computing a count of codes for each length,
95: creating from that a table of starting indices for each length in the
96: sorted table, and then entering the symbols in order in the sorted
97: table. The sorted table is work[], with that space being provided by
98: the caller.
99:
100: The length counts are used for other purposes as well, i.e. finding
101: the minimum and maximum length codes, determining if there are any
102: codes at all, checking for a valid set of lengths, and looking ahead
103: at length counts to determine sub-table sizes when building the
104: decoding tables.
105: */
106:
107: /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */
108: for (len = 0; len <= MAXBITS; len++)
109: count[len] = 0;
110: for (sym = 0; sym < codes; sym++)
111: count[lens[sym]]++;
112:
113: /* bound code lengths, force root to be within code lengths */
114: root = *bits;
115: for (max = MAXBITS; max >= 1; max--)
116: if (count[max] != 0) break;
117: if (root > max) root = max;
118: if (max == 0) { /* no symbols to code at all */
119: this.op = (unsigned char)64; /* invalid code marker */
120: this.bits = (unsigned char)1;
121: this.val = (unsigned short)0;
122: *(*table)++ = this; /* make a table to force an error */
123: *(*table)++ = this;
124: *bits = 1;
125: return 0; /* no symbols, but wait for decoding to report error */
126: }
127: for (min = 1; min <= MAXBITS; min++)
128: if (count[min] != 0) break;
129: if (root < min) root = min;
130:
131: /* check for an over-subscribed or incomplete set of lengths */
132: left = 1;
133: for (len = 1; len <= MAXBITS; len++) {
134: left <<= 1;
135: left -= count[len];
136: if (left < 0) return -1; /* over-subscribed */
137: }
138: if (left > 0 && (type == CODES || max != 1))
139: return -1; /* incomplete set */
140:
141: /* generate offsets into symbol table for each length for sorting */
142: offs[1] = 0;
143: for (len = 1; len < MAXBITS; len++)
144: offs[len + 1] = offs[len] + count[len];
145:
146: /* sort symbols by length, by symbol order within each length */
147: for (sym = 0; sym < codes; sym++)
148: if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym;
149:
150: /*
151: Create and fill in decoding tables. In this loop, the table being
152: filled is at next and has curr index bits. The code being used is huff
153: with length len. That code is converted to an index by dropping drop
154: bits off of the bottom. For codes where len is less than drop + curr,
155: those top drop + curr - len bits are incremented through all values to
156: fill the table with replicated entries.
157:
158: root is the number of index bits for the root table. When len exceeds
159: root, sub-tables are created pointed to by the root entry with an index
160: of the low root bits of huff. This is saved in low to check for when a
161: new sub-table should be started. drop is zero when the root table is
162: being filled, and drop is root when sub-tables are being filled.
163:
164: When a new sub-table is needed, it is necessary to look ahead in the
165: code lengths to determine what size sub-table is needed. The length
166: counts are used for this, and so count[] is decremented as codes are
167: entered in the tables.
168:
169: used keeps track of how many table entries have been allocated from the
170: provided *table space. It is checked when a LENS table is being made
171: against the space in *table, ENOUGH, minus the maximum space needed by
172: the worst case distance code, MAXD. This should never happen, but the
173: sufficiency of ENOUGH has not been proven exhaustively, hence the check.
174: This assumes that when type == LENS, bits == 9.
175:
176: sym increments through all symbols, and the loop terminates when
177: all codes of length max, i.e. all codes, have been processed. This
178: routine permits incomplete codes, so another loop after this one fills
179: in the rest of the decoding tables with invalid code markers.
180: */
181:
182: /* set up for code type */
183: switch (type) {
184: case CODES:
185: base = extra = work; /* dummy value--not used */
186: end = 19;
187: break;
188: case LENS:
189: base = lbase;
190: base -= 257;
191: extra = lext;
192: extra -= 257;
193: end = 256;
194: break;
195: default: /* DISTS */
196: base = dbase;
197: extra = dext;
198: end = -1;
199: }
200:
201: /* initialize state for loop */
202: huff = 0; /* starting code */
203: sym = 0; /* starting code symbol */
204: len = min; /* starting code length */
205: next = *table; /* current table to fill in */
206: curr = root; /* current table index bits */
207: drop = 0; /* current bits to drop from code for index */
208: low = (unsigned)(-1); /* trigger new sub-table when len > root */
209: used = 1U << root; /* use root table entries */
210: mask = used - 1; /* mask for comparing low */
211:
212: /* check available table space */
213: if (type == LENS && used >= ENOUGH - MAXD)
214: return 1;
215:
216: /* process all codes and make table entries */
217: for (;;) {
218: /* create table entry */
219: this.bits = (unsigned char)(len - drop);
220: if ((int)(work[sym]) < end) {
221: this.op = (unsigned char)0;
222: this.val = work[sym];
223: }
224: else if ((int)(work[sym]) > end) {
225: this.op = (unsigned char)(extra[work[sym]]);
226: this.val = base[work[sym]];
227: }
228: else {
229: this.op = (unsigned char)(32 + 64); /* end of block */
230: this.val = 0;
231: }
232:
233: /* replicate for those indices with low len bits equal to huff */
234: incr = 1U << (len - drop);
235: fill = 1U << curr;
236: min = fill; /* save offset to next table */
237: do {
238: fill -= incr;
239: next[(huff >> drop) + fill] = this;
240: } while (fill != 0);
241:
242: /* backwards increment the len-bit code huff */
243: incr = 1U << (len - 1);
244: while (huff & incr)
245: incr >>= 1;
246: if (incr != 0) {
247: huff &= incr - 1;
248: huff += incr;
249: }
250: else
251: huff = 0;
252:
253: /* go to next symbol, update count, len */
254: sym++;
255: if (--(count[len]) == 0) {
256: if (len == max) break;
257: len = lens[work[sym]];
258: }
259:
260: /* create new sub-table if needed */
261: if (len > root && (huff & mask) != low) {
262: /* if first time, transition to sub-tables */
263: if (drop == 0)
264: drop = root;
265:
266: /* increment past last table */
267: next += min; /* here min is 1 << curr */
268:
269: /* determine length of next table */
270: curr = len - drop;
271: left = (int)(1 << curr);
272: while (curr + drop < max) {
273: left -= count[curr + drop];
274: if (left <= 0) break;
275: curr++;
276: left <<= 1;
277: }
278:
279: /* check for enough space */
280: used += 1U << curr;
281: if (type == LENS && used >= ENOUGH - MAXD)
282: return 1;
283:
284: /* point entry in root table to sub-table */
285: low = huff & mask;
286: (*table)[low].op = (unsigned char)curr;
287: (*table)[low].bits = (unsigned char)root;
288: (*table)[low].val = (unsigned short)(next - *table);
289: }
290: }
291:
292: /*
293: Fill in rest of table for incomplete codes. This loop is similar to the
294: loop above in incrementing huff for table indices. It is assumed that
295: len is equal to curr + drop, so there is no loop needed to increment
296: through high index bits. When the current sub-table is filled, the loop
297: drops back to the root table to fill in any remaining entries there.
298: */
299: this.op = (unsigned char)64; /* invalid code marker */
300: this.bits = (unsigned char)(len - drop);
301: this.val = (unsigned short)0;
302: while (huff != 0) {
303: /* when done with sub-table, drop back to root table */
304: if (drop != 0 && (huff & mask) != low) {
305: drop = 0;
306: len = root;
307: next = *table;
308: this.bits = (unsigned char)len;
309: }
310:
311: /* put invalid code marker in table */
312: next[huff >> drop] = this;
313:
314: /* backwards increment the len-bit code huff */
315: incr = 1U << (len - 1);
316: while (huff & incr)
317: incr >>= 1;
318: if (incr != 0) {
319: huff &= incr - 1;
320: huff += incr;
321: }
322: else
323: huff = 0;
324: }
325:
326: /* set return parameters */
327: *table += used;
328: *bits = root;
329: return 0;
330: }
CVSweb