65aa1d11c6110269fc46e0bf7ea20e098a45200a
[m6w6/libmemcached] / src / libhashkit / jenkins.cc
1 /* vim:expandtab:shiftwidth=2:tabstop=2:smarttab:
2 *
3 * HashKit library
4 *
5 * Copyright (C) 2011-2012 Data Differential, http://datadifferential.com/
6 * Copyright (C) 2006-2009 Brian Aker All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions are
10 * met:
11 *
12 * * Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 *
15 * * Redistributions in binary form must reproduce the above
16 * copyright notice, this list of conditions and the following disclaimer
17 * in the documentation and/or other materials provided with the
18 * distribution.
19 *
20 * * The names of its contributors may not be used to endorse or
21 * promote products derived from this software without specific prior
22 * written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
25 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
26 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
27 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
28 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
29 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
30 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
31 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
32 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
33 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
34 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 *
36 */
37
38 /*
39 *
40 * By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this
41 * code any way you wish, private, educational, or commercial. It's free.
42 * Use for hash table lookup, or anything where one collision in 2^^32 is
43 * acceptable. Do NOT use for cryptographic purposes.
44 * http://burtleburtle.net/bob/hash/index.html
45 *
46 * Modified by Brian Pontz for libmemcached
47 * TODO:
48 * Add big endian support
49 */
50
51 #include "libhashkit/common.h"
52
53 #define hashsize(n) ((uint32_t)1<<(n))
54 #define hashmask(n) (hashsize(n)-1)
55 #define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
56
57 #define mix(a,b,c) \
58 { \
59 a -= c; a ^= rot(c, 4); c += b; \
60 b -= a; b ^= rot(a, 6); a += c; \
61 c -= b; c ^= rot(b, 8); b += a; \
62 a -= c; a ^= rot(c,16); c += b; \
63 b -= a; b ^= rot(a,19); a += c; \
64 c -= b; c ^= rot(b, 4); b += a; \
65 }
66
67 #define final(a,b,c) \
68 { \
69 c ^= b; c -= rot(b,14); \
70 a ^= c; a -= rot(c,11); \
71 b ^= a; b -= rot(a,25); \
72 c ^= b; c -= rot(b,16); \
73 a ^= c; a -= rot(c,4); \
74 b ^= a; b -= rot(a,14); \
75 c ^= b; c -= rot(b,24); \
76 }
77
78 #define JENKINS_INITVAL 13
79
80 /*
81 jenkins_hash() -- hash a variable-length key into a 32-bit value
82 k : the key (the unaligned variable-length array of bytes)
83 length : the length of the key, counting by bytes
84 initval : can be any 4-byte value
85 Returns a 32-bit value. Every bit of the key affects every bit of
86 the return value. Two keys differing by one or two bits will have
87 totally different hash values.
88
89 The best hash table sizes are powers of 2. There is no need to do
90 mod a prime (mod is sooo slow!). If you need less than 32 bits,
91 use a bitmask. For example, if you need only 10 bits, do
92 h = (h & hashmask(10));
93 In which case, the hash table should have hashsize(10) elements.
94 */
95
96 #if HAVE_ASAN
97 __attribute__((no_sanitize_address,no_sanitize("address")))
98 #endif
99 uint32_t hashkit_jenkins(const char *key, size_t length, void *)
100 {
101 uint32_t a,b,c; /* internal state */
102 #if !WORDS_BIGENDIAN
103 union { const void *ptr; size_t i; } u;
104 u.ptr = key;
105 #endif
106
107 /* Set up the internal state */
108 a = b = c = 0xdeadbeef + ((uint32_t)length) + JENKINS_INITVAL;
109
110 #if !WORDS_BIGENDIAN
111 if ((u.i & 0x3) == 0)
112 {
113 const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */
114
115 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
116 while (length > 12)
117 {
118 a += k[0];
119 b += k[1];
120 c += k[2];
121 mix(a,b,c);
122 length -= 12;
123 k += 3;
124 }
125
126 /*----------------------------- handle the last (probably partial) block */
127 /*
128 * "k[2]&0xffffff" actually reads beyond the end of the string, but
129 * then masks off the part it's not allowed to read. Because the
130 * string is aligned, the masked-off tail is in the same word as the
131 * rest of the string. Every machine with memory protection I've seen
132 * does it on word boundaries, so is OK with this. But VALGRIND will
133 * still catch it and complain. The masking trick does make the hash
134 * noticably faster for short strings (like English words).
135 */
136 switch(length)
137 {
138 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
139 case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
140 case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break;
141 case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break;
142 case 8 : b+=k[1]; a+=k[0]; break;
143 case 7 : b+=k[1]&0xffffff; a+=k[0]; break;
144 case 6 : b+=k[1]&0xffff; a+=k[0]; break;
145 case 5 : b+=k[1]&0xff; a+=k[0]; break;
146 case 4 : a+=k[0]; break;
147 case 3 : a+=k[0]&0xffffff; break;
148 case 2 : a+=k[0]&0xffff; break;
149 case 1 : a+=k[0]&0xff; break;
150 case 0 : return c; /* zero length strings require no mixing */
151 default: return c;
152 }
153
154 }
155 else if ((u.i & 0x1) == 0)
156 {
157 const uint16_t *k = (const uint16_t *)key; /* read 16-bit chunks */
158 const uint8_t *k8;
159
160 /*--------------- all but last block: aligned reads and different mixing */
161 while (length > 12)
162 {
163 a += k[0] + (((uint32_t)k[1])<<16);
164 b += k[2] + (((uint32_t)k[3])<<16);
165 c += k[4] + (((uint32_t)k[5])<<16);
166 mix(a,b,c);
167 length -= 12;
168 k += 6;
169 }
170
171 /*----------------------------- handle the last (probably partial) block */
172 k8 = (const uint8_t *)k;
173 switch(length)
174 {
175 case 12: c+=k[4]+(((uint32_t)k[5])<<16);
176 b+=k[2]+(((uint32_t)k[3])<<16);
177 a+=k[0]+(((uint32_t)k[1])<<16);
178 break;
179 case 11: c+=((uint32_t)k8[10])<<16;
180 /* fall through */
181 case 10: c+=k[4];
182 b+=k[2]+(((uint32_t)k[3])<<16);
183 a+=k[0]+(((uint32_t)k[1])<<16);
184 break;
185 case 9 : c+=k8[8];
186 /* 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;
191 /* fall through */
192 case 6 : b+=k[2];
193 a+=k[0]+(((uint32_t)k[1])<<16);
194 break;
195 case 5 : b+=k8[4];
196 /* fall through */
197 case 4 : a+=k[0]+(((uint32_t)k[1])<<16);
198 break;
199 case 3 : a+=((uint32_t)k8[2])<<16;
200 /* fall through */
201 case 2 : a+=k[0];
202 break;
203 case 1 : a+=k8[0];
204 break;
205 case 0 : return c; /* zero length requires no mixing */
206 default: return c;
207 }
208
209 }
210 else
211 { /* need to read the key one byte at a time */
212 #endif /* little endian */
213 const uint8_t *k = (const uint8_t *)key;
214
215 /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
216 while (length > 12)
217 {
218 a += k[0];
219 a += ((uint32_t)k[1])<<8;
220 a += ((uint32_t)k[2])<<16;
221 a += ((uint32_t)k[3])<<24;
222 b += k[4];
223 b += ((uint32_t)k[5])<<8;
224 b += ((uint32_t)k[6])<<16;
225 b += ((uint32_t)k[7])<<24;
226 c += k[8];
227 c += ((uint32_t)k[9])<<8;
228 c += ((uint32_t)k[10])<<16;
229 c += ((uint32_t)k[11])<<24;
230 mix(a,b,c);
231 length -= 12;
232 k += 12;
233 }
234
235 /*-------------------------------- last block: affect all 32 bits of (c) */
236 switch(length) /* all the case statements fall through */
237 {
238 case 12: c+=((uint32_t)k[11])<<24;
239 /* fall through */
240 case 11: c+=((uint32_t)k[10])<<16;
241 /* fall through */
242 case 10: c+=((uint32_t)k[9])<<8;
243 /* fall through */
244 case 9 : c+=k[8];
245 /* fall through */
246 case 8 : b+=((uint32_t)k[7])<<24;
247 /* fall through */
248 case 7 : b+=((uint32_t)k[6])<<16;
249 /* fall through */
250 case 6 : b+=((uint32_t)k[5])<<8;
251 /* fall through */
252 case 5 : b+=k[4];
253 /* fall through */
254 case 4 : a+=((uint32_t)k[3])<<24;
255 /* fall through */
256 case 3 : a+=((uint32_t)k[2])<<16;
257 /* fall through */
258 case 2 : a+=((uint32_t)k[1])<<8;
259 /* fall through */
260 case 1 : a+=k[0];
261 break;
262 case 0 : return c;
263 default : return c;
264 }
265 #if !WORDS_BIGENDIAN
266 }
267 #endif
268
269 final(a,b,c);
270 return c;
271 }