4941b945a1f96b32e3f19034f414a79413cdb062
[lttng-ust.git] / src / common / jhash.h
1 /*
2 * SPDX-License-Identifier: LGPL-2.1-only
3 *
4 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
5 */
6
7 #ifndef _UST_COMMON_JHASH_H
8 #define _UST_COMMON_JHASH_H
9
10 #include <stddef.h>
11 #include <stdint.h>
12 #include <urcu/compiler.h>
13 #include <lttng/ust-endian.h>
14
15 /*
16 * Hash function
17 * Source: http://burtleburtle.net/bob/c/lookup3.c
18 * Originally Public Domain
19 */
20
21 #define rot(x, k) (((x) << (k)) | ((x) >> (32 - (k))))
22
23 #define mix(a, b, c) \
24 do { \
25 a -= c; a ^= rot(c, 4); c += b; \
26 b -= a; b ^= rot(a, 6); a += c; \
27 c -= b; c ^= rot(b, 8); b += a; \
28 a -= c; a ^= rot(c, 16); c += b; \
29 b -= a; b ^= rot(a, 19); a += c; \
30 c -= b; c ^= rot(b, 4); b += a; \
31 } while (0)
32
33 #define final(a, b, c) \
34 { \
35 c ^= b; c -= rot(b, 14); \
36 a ^= c; a -= rot(c, 11); \
37 b ^= a; b -= rot(a, 25); \
38 c ^= b; c -= rot(b, 16); \
39 a ^= c; a -= rot(c, 4);\
40 b ^= a; b -= rot(a, 14); \
41 c ^= b; c -= rot(b, 24); \
42 }
43
44 #if (BYTE_ORDER == LITTLE_ENDIAN)
45 #define HASH_LITTLE_ENDIAN 1
46 #else
47 #define HASH_LITTLE_ENDIAN 0
48 #endif
49
50 /*
51 *
52 * hashlittle() -- hash a variable-length key into a 32-bit value
53 * k : the key (the unaligned variable-length array of bytes)
54 * length : the length of the key, counting by bytes
55 * initval : can be any 4-byte value
56 * Returns a 32-bit value. Every bit of the key affects every bit of
57 * the return value. Two keys differing by one or two bits will have
58 * totally different hash values.
59 *
60 * The best hash table sizes are powers of 2. There is no need to do
61 * mod a prime (mod is sooo slow!). If you need less than 32 bits,
62 * use a bitmask. For example, if you need only 10 bits, do
63 * h = (h & hashmask(10));
64 * In which case, the hash table should have hashsize(10) elements.
65 *
66 * If you are hashing n strings (uint8_t **)k, do it like this:
67 * for (i = 0, h = 0; i < n; ++i) h = hashlittle(k[i], len[i], h);
68 *
69 * By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this
70 * code any way you wish, private, educational, or commercial. It's free.
71 *
72 * Use for hash table lookup, or anything where one collision in 2^^32 is
73 * acceptable. Do NOT use for cryptographic purposes.
74 */
75 static
76 uint32_t hashlittle(const void *key, size_t length, uint32_t initval)
77 {
78 uint32_t a, b, c; /* internal state */
79 union {
80 const void *ptr;
81 size_t i;
82 } u;
83
84 /* Set up the internal state */
85 a = b = c = 0xdeadbeef + ((uint32_t)length) + initval;
86
87 u.ptr = key;
88 if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
89 const uint32_t *k = (const uint32_t *) key; /* read 32-bit chunks */
90
91 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
92 while (length > 12) {
93 a += k[0];
94 b += k[1];
95 c += k[2];
96 mix(a, b, c);
97 length -= 12;
98 k += 3;
99 }
100
101 /*----------------------------- handle the last (probably partial) block */
102 /*
103 * The original jhash.h reads beyond the end of string, and implements
104 * a special code path for VALGRIND. It seems to make ASan unhappy too
105 * though, so considering that hashing event names is not a fast-path
106 * in lttng-ust, remove the "fast" code entirely and use the slower
107 * but verifiable VALGRIND version of the code which does not issue
108 * out-of-bound reads.
109 */
110 {
111 const uint8_t *k8;
112
113 k8 = (const uint8_t *) k;
114 switch (length) {
115 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
116 case 11: c+=((uint32_t) k8[10])<<16; /* fall through */
117 case 10: c+=((uint32_t) k8[9])<<8; /* fall through */
118 case 9 : c+=k8[8]; /* fall through */
119 case 8 : b+=k[1]; a+=k[0]; break;
120 case 7 : b+=((uint32_t) k8[6])<<16; /* fall through */
121 case 6 : b+=((uint32_t) k8[5])<<8; /* fall through */
122 case 5 : b+=k8[4]; /* fall through */
123 case 4 : a+=k[0]; break;
124 case 3 : a+=((uint32_t) k8[2])<<16; /* fall through */
125 case 2 : a+=((uint32_t) k8[1])<<8; /* fall through */
126 case 1 : a+=k8[0]; break;
127 case 0 : return c;
128 }
129 }
130
131 } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
132 const uint16_t *k = (const uint16_t *) key; /* read 16-bit chunks */
133 const uint8_t *k8;
134
135 /*--------------- all but last block: aligned reads and different mixing */
136 while (length > 12)
137 {
138 a += k[0] + (((uint32_t) k[1])<<16);
139 b += k[2] + (((uint32_t) k[3])<<16);
140 c += k[4] + (((uint32_t) k[5])<<16);
141 mix(a, b, c);
142 length -= 12;
143 k += 6;
144 }
145
146 /*----------------------------- handle the last (probably partial) block */
147 k8 = (const uint8_t *) k;
148 switch(length)
149 {
150 case 12: c+=k[4]+(((uint32_t) k[5])<<16);
151 b+=k[2]+(((uint32_t) k[3])<<16);
152 a+=k[0]+(((uint32_t) k[1])<<16);
153 break;
154 case 11: c+=((uint32_t) k8[10])<<16; /* fall through */
155 case 10: c+=k[4];
156 b+=k[2]+(((uint32_t) k[3])<<16);
157 a+=k[0]+(((uint32_t) k[1])<<16);
158 break;
159 case 9 : c+=k8[8]; /* fall through */
160 case 8 : b+=k[2]+(((uint32_t) k[3])<<16);
161 a+=k[0]+(((uint32_t) k[1])<<16);
162 break;
163 case 7 : b+=((uint32_t) k8[6])<<16; /* fall through */
164 case 6 : b+=k[2];
165 a+=k[0]+(((uint32_t) k[1])<<16);
166 break;
167 case 5 : b+=k8[4]; /* fall through */
168 case 4 : a+=k[0]+(((uint32_t) k[1])<<16);
169 break;
170 case 3 : a+=((uint32_t) k8[2])<<16; /* fall through */
171 case 2 : a+=k[0];
172 break;
173 case 1 : a+=k8[0];
174 break;
175 case 0 : return c; /* zero length requires no mixing */
176 }
177
178 } else { /* need to read the key one byte at a time */
179 const uint8_t *k = (const uint8_t *)key;
180
181 /*--------------- all but the last block: affect some 32 bits of (a, b, c) */
182 while (length > 12) {
183 a += k[0];
184 a += ((uint32_t) k[1])<<8;
185 a += ((uint32_t) k[2])<<16;
186 a += ((uint32_t) k[3])<<24;
187 b += k[4];
188 b += ((uint32_t) k[5])<<8;
189 b += ((uint32_t) k[6])<<16;
190 b += ((uint32_t) k[7])<<24;
191 c += k[8];
192 c += ((uint32_t) k[9])<<8;
193 c += ((uint32_t) k[10])<<16;
194 c += ((uint32_t) k[11])<<24;
195 mix(a,b,c);
196 length -= 12;
197 k += 12;
198 }
199
200 /*-------------------------------- last block: affect all 32 bits of (c) */
201 switch (length) { /* all the case statements fall through */
202 case 12: c+=((uint32_t) k[11])<<24; /* fall through */
203 case 11: c+=((uint32_t) k[10])<<16; /* fall through */
204 case 10: c+=((uint32_t) k[9])<<8; /* fall through */
205 case 9 : c+=k[8]; /* fall through */
206 case 8 : b+=((uint32_t) k[7])<<24; /* fall through */
207 case 7 : b+=((uint32_t) k[6])<<16; /* fall through */
208 case 6 : b+=((uint32_t) k[5])<<8; /* fall through */
209 case 5 : b+=k[4]; /* fall through */
210 case 4 : a+=((uint32_t) k[3])<<24; /* fall through */
211 case 3 : a+=((uint32_t) k[2])<<16; /* fall through */
212 case 2 : a+=((uint32_t) k[1])<<8; /* fall through */
213 case 1 : a+=k[0];
214 break;
215 case 0 : return c;
216 }
217 }
218
219 final(a, b, c);
220 return c;
221 }
222
223 static inline
224 uint32_t jhash(const void *key, size_t length, uint32_t seed)
225 {
226 return hashlittle(key, length, seed);
227 }
228
229 #endif /* _UST_COMMON_JHASH_H */
This page took 0.035256 seconds and 3 git commands to generate.