xref: /illumos-gate/usr/src/cmd/spell/huff.c (revision 0d8b5334)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
28 /*	  All Rights Reserved  	*/
29 
30 #pragma ident	"%Z%%M%	%I%	%E% SMI"
31 
32 
33 #include <unistd.h>
34 #include <stdlib.h>
35 #include <stdio.h>
36 
37 #define	BYTE 8
38 #define	QW 1		/* width of bas-q digit in bits */
39 
40 /*
41  * this stuff should be local and hidden; it was made
42  * accessible outside for dirty reasons: 20% faster spell
43  */
44 #include "huff.h"
45 struct huff huffcode;
46 
47 /*
48  * Infinite Huffman code
49  *
50  * Let the messages be exponentially distributed with ratio r:
51  * 	P {message k} = r^k*(1-r),	k = 0, 1, ...
52  * Let the messages be coded in base q, and suppose
53  * 	r^n = 1/q
54  * If each decade(base q) contains n codes, then
55  * the messages assigned to each decade will be q times
56  * as probable as the next. Moreover the code for the tail of
57  * the distribution after truncating one decade should look
58  * just like the original, but longer by one leading digit q-1.
59  * 	q(z+n) = z + (q-1)q^w
60  * where z is first code of decade, w is width of code, in shortest
61  * full decade. Examples, base 2:
62  * 	r^1 = 1/2	r^5 = 1/2
63  * 	0		0110
64  * 	10		0111
65  * 	110		1000
66  * 	1110		1001
67  * 	...		1010
68  * 			10110
69  * 	w = 1, z = 0		w = 4, z = 0110
70  * Rewriting slightly
71  * 	(q-1)z + q*n = (q-1)q^w
72  * whence z is a multiple of q and n is a multiple of q-1. Let
73  * 	z = cq, n = d(q-1)
74  * We pick w to be the least integer such that
75  * 	d = n/(q-1) <= q^(w-1)
76  * Then solve for c
77  * 	c = q^(w-1) - d
78  * If c is not zero, the first decade may be preceded by
79  * even shorter (w-1)-digit codes 0, 1, ..., c-1. Thus
80  * the example code with r^5 = 1/2 becomes
81  * 	000
82  * 	001
83  * 	010
84  * 	0110
85  * 	0111
86  * 	1000
87  * 	1001
88  * 	1010
89  * 	10110
90  * 	...
91  * 	110110
92  * 	...
93  * The expected number of base-q digits in a codeword is then
94  *	w - 1 + r^c/(1-r^n)
95  * The present routines require q to be a power of 2
96  */
97 /*
98  * There is a lot of hanky-panky with left justification against
99  * sign instead of simple left justification because
100  * unsigned long is not available
101  */
102 #define	L (BYTE*(sizeof (long))-1)	/* length of signless long */
103 #define	MASK (~((unsigned long)1<<L))	/* mask out sign */
104 
105 /*
106  * decode the prefix of word y (which is left justified against sign)
107  * place mesage number into place pointed to by kp
108  * return length (in bits) of decoded prefix or 0 if code is out of
109  * range
110  */
111 int
112 decode(long y, long *pk)
113 {
114 	int l;
115 	long v;
116 	if (y < cs) {
117 		*pk = y >> (long)(L+QW-w);
118 		return (w-QW);
119 	}
120 	for (l = w, v = v0; y >= qcs;
121 	    y = ((unsigned long)y << QW) & MASK, v += n)
122 		if ((l += QW) > L)
123 			return (0);
124 	*pk = v + (y>>(long)(L-w));
125 	return (l);
126 }
127 
128 /*
129  * encode message k and put result (right justified) into
130  * place pointed to by py.
131  * return length (in bits) of result,
132  * or 0 if code is too long
133  */
134 
135 int
136 encode(long k, long *py)
137 {
138 	int l;
139 	long y;
140 	if (k < c) {
141 		*py = k;
142 		return (w-QW);
143 	}
144 	for (k -= c, y = 1, l = w; k >= n; k -= n, y <<= QW)
145 		if ((l += QW) > L)
146 			return (0);
147 	*py = ((y-1)<<w) + cq + k;
148 	return (l);
149 }
150 
151 
152 /*
153  * Initialization code, given expected value of k
154  * E(k) = r/(1-r) = a
155  * and given base width b
156  * return expected length of coded messages
157  */
158 static struct qlog {
159 	long p;
160 	double u;
161 } z;
162 
163 static struct qlog
164 qlog(double x, double y, long p, double u)	/* find smallest p so x^p<=y */
165 {
166 
167 	if (u/x <= y) {
168 		z.p = 0;
169 		z.u = 1;
170 	} else {
171 		z = qlog(x, y, p+p, u*u);
172 		if (u*z.u/x > y) {
173 			z.p += p;
174 			z.u *= u;
175 		}
176 	}
177 	return (z);
178 }
179 
180 double
181 huff(float a)
182 {
183 	int i, q;
184 	long d, j;
185 	double r = a/(1.0 + a);
186 	double rc, rq;
187 
188 	for (i = 0, q = 1, rq = r; i < QW; i++, q *= 2, rq *= rq)
189 		continue;
190 	rq /= r;	/* rq = r^(q-1) */
191 	(void) qlog(rq, 1./q, 1L, rq);
192 	d = z.p;
193 	n = d*(q-1);
194 	if (n != d * (q - 1))
195 		abort();	/* time to make n long */
196 	for (w = QW, j = 1; j < d; w += QW, j *= q)
197 		continue;
198 	c = j - d;
199 	cq = c*q;
200 	cs = cq<<(L-w);
201 	qcs = (((long)(q-1)<<w) + cq) << (L-QW-w);
202 	v0 = c - cq;
203 	for (i = 0, rc = 1; i < c; i++, rc *= r)	/* rc = r^c */
204 		continue;
205 	return (w + QW*(rc/(1-z.u) - 1));
206 }
207 
208 void
209 whuff(void)
210 {
211 	(void) fwrite((char *) & huffcode, sizeof (huffcode), 1, stdout);
212 }
213 
214 int
215 rhuff(FILE *f)
216 {
217 	return (read(fileno(f), (char *)&huffcode, sizeof (huffcode)) ==
218 	    sizeof (huffcode));
219 }
220