Commit | Line | Data |
---|---|---|
f7e7a1b8 MD |
1 | #ifndef _JHASH_H |
2 | #define _JHASH_H | |
3 | ||
4 | /* | |
5 | * jhash.h | |
6 | * | |
7 | * Example hash function. | |
8 | * | |
9 | * Copyright 2009-2012 - Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca> | |
10 | * | |
11 | * THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED | |
12 | * OR IMPLIED. ANY USE IS AT YOUR OWN RISK. | |
13 | * | |
14 | * Permission is hereby granted to use or copy this program for any | |
15 | * purpose, provided the above notices are retained on all copies. | |
16 | * Permission to modify the code and to distribute modified code is | |
17 | * granted, provided the above notices are retained, and a notice that | |
18 | * the code was modified is included with the above copyright notice. | |
19 | */ | |
20 | ||
21 | /* | |
22 | * Hash function | |
23 | * Source: http://burtleburtle.net/bob/c/lookup3.c | |
24 | * Originally Public Domain | |
25 | */ | |
26 | ||
27 | #define rot(x, k) (((x) << (k)) | ((x) >> (32 - (k)))) | |
28 | ||
29 | #define mix(a, b, c) \ | |
30 | do { \ | |
31 | a -= c; a ^= rot(c, 4); c += b; \ | |
32 | b -= a; b ^= rot(a, 6); a += c; \ | |
33 | c -= b; c ^= rot(b, 8); b += a; \ | |
34 | a -= c; a ^= rot(c, 16); c += b; \ | |
35 | b -= a; b ^= rot(a, 19); a += c; \ | |
36 | c -= b; c ^= rot(b, 4); b += a; \ | |
37 | } while (0) | |
38 | ||
39 | #define final(a, b, c) \ | |
40 | { \ | |
41 | c ^= b; c -= rot(b, 14); \ | |
42 | a ^= c; a -= rot(c, 11); \ | |
43 | b ^= a; b -= rot(a, 25); \ | |
44 | c ^= b; c -= rot(b, 16); \ | |
45 | a ^= c; a -= rot(c, 4); \ | |
46 | b ^= a; b -= rot(a, 14); \ | |
47 | c ^= b; c -= rot(b, 24); \ | |
48 | } | |
49 | ||
50 | #if (BYTE_ORDER == LITTLE_ENDIAN) | |
51 | #define HASH_LITTLE_ENDIAN 1 | |
52 | #else | |
53 | #define HASH_LITTLE_ENDIAN 0 | |
54 | #endif | |
55 | ||
56 | /* | |
57 | * | |
58 | * hashlittle() -- hash a variable-length key into a 32-bit value | |
59 | * k : the key (the unaligned variable-length array of bytes) | |
60 | * length : the length of the key, counting by bytes | |
61 | * initval : can be any 4-byte value | |
62 | * Returns a 32-bit value. Every bit of the key affects every bit of | |
63 | * the return value. Two keys differing by one or two bits will have | |
64 | * totally different hash values. | |
65 | * | |
66 | * The best hash table sizes are powers of 2. There is no need to do | |
67 | * mod a prime (mod is sooo slow!). If you need less than 32 bits, | |
68 | * use a bitmask. For example, if you need only 10 bits, do | |
69 | * h = (h & hashmask(10)); | |
70 | * In which case, the hash table should have hashsize(10) elements. | |
71 | * | |
72 | * If you are hashing n strings (uint8_t **)k, do it like this: | |
73 | * for (i = 0, h = 0; i < n; ++i) h = hashlittle(k[i], len[i], h); | |
74 | * | |
75 | * By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this | |
76 | * code any way you wish, private, educational, or commercial. It's free. | |
77 | * | |
78 | * Use for hash table lookup, or anything where one collision in 2^^32 is | |
79 | * acceptable. Do NOT use for cryptographic purposes. | |
80 | */ | |
81 | static | |
82 | uint32_t hashlittle(const void *key, size_t length, uint32_t initval) | |
83 | { | |
84 | uint32_t a, b, c; /* internal state */ | |
85 | union { | |
86 | const void *ptr; | |
87 | size_t i; | |
88 | } u; | |
89 | ||
90 | /* Set up the internal state */ | |
91 | a = b = c = 0xdeadbeef + ((uint32_t)length) + initval; | |
92 | ||
93 | u.ptr = key; | |
94 | if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) { | |
95 | const uint32_t *k = (const uint32_t *) key; /* read 32-bit chunks */ | |
96 | ||
97 | /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */ | |
98 | while (length > 12) { | |
99 | a += k[0]; | |
100 | b += k[1]; | |
101 | c += k[2]; | |
102 | mix(a, b, c); | |
103 | length -= 12; | |
104 | k += 3; | |
105 | } | |
106 | ||
107 | /*----------------------------- handle the last (probably partial) block */ | |
108 | /* | |
109 | * "k[2]&0xffffff" actually reads beyond the end of the string, but | |
110 | * then masks off the part it's not allowed to read. Because the | |
111 | * string is aligned, the masked-off tail is in the same word as the | |
112 | * rest of the string. Every machine with memory protection I've seen | |
113 | * does it on word boundaries, so is OK with this. But VALGRIND will | |
114 | * still catch it and complain. The masking trick does make the hash | |
115 | * noticably faster for short strings (like English words). | |
116 | */ | |
117 | #ifndef VALGRIND | |
118 | ||
119 | switch (length) { | |
120 | case 12: c+=k[2]; b+=k[1]; a+=k[0]; break; | |
121 | case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break; | |
122 | case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break; | |
123 | case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break; | |
124 | case 8 : b+=k[1]; a+=k[0]; break; | |
125 | case 7 : b+=k[1]&0xffffff; a+=k[0]; break; | |
126 | case 6 : b+=k[1]&0xffff; a+=k[0]; break; | |
127 | case 5 : b+=k[1]&0xff; a+=k[0]; break; | |
128 | case 4 : a+=k[0]; break; | |
129 | case 3 : a+=k[0]&0xffffff; break; | |
130 | case 2 : a+=k[0]&0xffff; break; | |
131 | case 1 : a+=k[0]&0xff; break; | |
132 | case 0 : return c; /* zero length strings require no mixing */ | |
133 | } | |
134 | ||
135 | #else /* make valgrind happy */ | |
136 | { | |
137 | const uint8_t *k8; | |
138 | ||
139 | k8 = (const uint8_t *) k; | |
140 | switch (length) { | |
141 | case 12: c+=k[2]; b+=k[1]; a+=k[0]; break; | |
142 | case 11: c+=((uint32_t) k8[10])<<16; /* fall through */ | |
143 | case 10: c+=((uint32_t) k8[9])<<8; /* fall through */ | |
144 | case 9 : c+=k8[8]; /* fall through */ | |
145 | case 8 : b+=k[1]; a+=k[0]; break; | |
146 | case 7 : b+=((uint32_t) k8[6])<<16; /* fall through */ | |
147 | case 6 : b+=((uint32_t) k8[5])<<8; /* fall through */ | |
148 | case 5 : b+=k8[4]; /* fall through */ | |
149 | case 4 : a+=k[0]; break; | |
150 | case 3 : a+=((uint32_t) k8[2])<<16; /* fall through */ | |
151 | case 2 : a+=((uint32_t) k8[1])<<8; /* fall through */ | |
152 | case 1 : a+=k8[0]; break; | |
153 | case 0 : return c; | |
154 | } | |
155 | } | |
156 | #endif /* !valgrind */ | |
157 | ||
158 | } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) { | |
159 | const uint16_t *k = (const uint16_t *) key; /* read 16-bit chunks */ | |
160 | const uint8_t *k8; | |
161 | ||
162 | /*--------------- all but last block: aligned reads and different mixing */ | |
163 | while (length > 12) | |
164 | { | |
165 | a += k[0] + (((uint32_t) k[1])<<16); | |
166 | b += k[2] + (((uint32_t) k[3])<<16); | |
167 | c += k[4] + (((uint32_t) k[5])<<16); | |
168 | mix(a, b, c); | |
169 | length -= 12; | |
170 | k += 6; | |
171 | } | |
172 | ||
173 | /*----------------------------- handle the last (probably partial) block */ | |
174 | k8 = (const uint8_t *) k; | |
175 | switch(length) | |
176 | { | |
177 | case 12: c+=k[4]+(((uint32_t) k[5])<<16); | |
178 | b+=k[2]+(((uint32_t) k[3])<<16); | |
179 | a+=k[0]+(((uint32_t) k[1])<<16); | |
180 | break; | |
181 | case 11: c+=((uint32_t) k8[10])<<16; /* fall through */ | |
182 | case 10: c+=k[4]; | |
183 | b+=k[2]+(((uint32_t) k[3])<<16); | |
184 | a+=k[0]+(((uint32_t) k[1])<<16); | |
185 | break; | |
186 | case 9 : c+=k8[8]; /* fall through */ | |
187 | case 8 : b+=k[2]+(((uint32_t) k[3])<<16); | |
188 | a+=k[0]+(((uint32_t) k[1])<<16); | |
189 | break; | |
190 | case 7 : b+=((uint32_t) k8[6])<<16; /* fall through */ | |
191 | case 6 : b+=k[2]; | |
192 | a+=k[0]+(((uint32_t) k[1])<<16); | |
193 | break; | |
194 | case 5 : b+=k8[4]; /* fall through */ | |
195 | case 4 : a+=k[0]+(((uint32_t) k[1])<<16); | |
196 | break; | |
197 | case 3 : a+=((uint32_t) k8[2])<<16; /* fall through */ | |
198 | case 2 : a+=k[0]; | |
199 | break; | |
200 | case 1 : a+=k8[0]; | |
201 | break; | |
202 | case 0 : return c; /* zero length requires no mixing */ | |
203 | } | |
204 | ||
205 | } else { /* need to read the key one byte at a time */ | |
206 | const uint8_t *k = (const uint8_t *)key; | |
207 | ||
208 | /*--------------- all but the last block: affect some 32 bits of (a, b, c) */ | |
209 | while (length > 12) { | |
210 | a += k[0]; | |
211 | a += ((uint32_t) k[1])<<8; | |
212 | a += ((uint32_t) k[2])<<16; | |
213 | a += ((uint32_t) k[3])<<24; | |
214 | b += k[4]; | |
215 | b += ((uint32_t) k[5])<<8; | |
216 | b += ((uint32_t) k[6])<<16; | |
217 | b += ((uint32_t) k[7])<<24; | |
218 | c += k[8]; | |
219 | c += ((uint32_t) k[9])<<8; | |
220 | c += ((uint32_t) k[10])<<16; | |
221 | c += ((uint32_t) k[11])<<24; | |
222 | mix(a,b,c); | |
223 | length -= 12; | |
224 | k += 12; | |
225 | } | |
226 | ||
227 | /*-------------------------------- last block: affect all 32 bits of (c) */ | |
228 | switch (length) { /* all the case statements fall through */ | |
447c9339 MJ |
229 | case 12: c+=((uint32_t) k[11])<<24; /* fall through */ |
230 | case 11: c+=((uint32_t) k[10])<<16; /* fall through */ | |
231 | case 10: c+=((uint32_t) k[9])<<8; /* fall through */ | |
232 | case 9 : c+=k[8]; /* fall through */ | |
233 | case 8 : b+=((uint32_t) k[7])<<24; /* fall through */ | |
234 | case 7 : b+=((uint32_t) k[6])<<16; /* fall through */ | |
235 | case 6 : b+=((uint32_t) k[5])<<8; /* fall through */ | |
236 | case 5 : b+=k[4]; /* fall through */ | |
237 | case 4 : a+=((uint32_t) k[3])<<24; /* fall through */ | |
238 | case 3 : a+=((uint32_t) k[2])<<16; /* fall through */ | |
239 | case 2 : a+=((uint32_t) k[1])<<8; /* fall through */ | |
f7e7a1b8 MD |
240 | case 1 : a+=k[0]; |
241 | break; | |
242 | case 0 : return c; | |
243 | } | |
244 | } | |
245 | ||
246 | final(a, b, c); | |
247 | return c; | |
248 | } | |
249 | ||
250 | static inline | |
251 | uint32_t jhash(const void *key, size_t length, uint32_t seed) | |
252 | { | |
253 | return hashlittle(key, length, seed); | |
254 | } | |
255 | ||
256 | #endif /* _JHASH_H */ |