1 /**
2 * Copyright 2010 The Apache Software Foundation
3 *
4 * Licensed to the Apache Software Foundation (ASF) under one
5 * or more contributor license agreements. See the NOTICE file
6 * distributed with this work for additional information
7 * regarding copyright ownership. The ASF licenses this file
8 * to you under the Apache License, Version 2.0 (the
9 * "License"); you may not use this file except in compliance
10 * with the License. You may obtain a copy of the License at
11 *
12 * http://www.apache.org/licenses/LICENSE-2.0
13 *
14 * Unless required by applicable law or agreed to in writing, software
15 * distributed under the License is distributed on an "AS IS" BASIS,
16 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17 * See the License for the specific language governing permissions and
18 * limitations under the License.
19 */
20
21 package org.apache.hadoop.hbase.util;
22
23 import java.io.FileInputStream;
24 import java.io.IOException;
25
26 /**
27 * Produces 32-bit hash for hash table lookup.
28 *
29 * <pre>lookup3.c, by Bob Jenkins, May 2006, Public Domain.
30 *
31 * You can use this free for any purpose. It's in the public domain.
32 * It has no warranty.
33 * </pre>
34 *
35 * @see <a href="http://burtleburtle.net/bob/c/lookup3.c">lookup3.c</a>
36 * @see <a href="http://www.ddj.com/184410284">Hash Functions (and how this
37 * function compares to others such as CRC, MD?, etc</a>
38 * @see <a href="http://burtleburtle.net/bob/hash/doobs.html">Has update on the
39 * Dr. Dobbs Article</a>
40 */
41 public class JenkinsHash extends Hash {
42 private static long INT_MASK = 0x00000000ffffffffL;
43 private static long BYTE_MASK = 0x00000000000000ffL;
44
45 private static JenkinsHash _instance = new JenkinsHash();
46
47 public static Hash getInstance() {
48 return _instance;
49 }
50
51 private static long rot(long val, int pos) {
52 return ((Integer.rotateLeft(
53 (int)(val & INT_MASK), pos)) & INT_MASK);
54 }
55
56 /**
57 * taken from hashlittle() -- hash a variable-length key into a 32-bit value
58 *
59 * @param key the key (the unaligned variable-length array of bytes)
60 * @param nbytes number of bytes to include in hash
61 * @param initval can be any integer value
62 * @return a 32-bit value. Every bit of the key affects every bit of the
63 * return value. Two keys differing by one or two bits will have totally
64 * different hash values.
65 *
66 * <p>The best hash table sizes are powers of 2. There is no need to do mod
67 * a prime (mod is sooo slow!). If you need less than 32 bits, use a bitmask.
68 * For example, if you need only 10 bits, do
69 * <code>h = (h & hashmask(10));</code>
70 * In which case, the hash table should have hashsize(10) elements.
71 *
72 * <p>If you are hashing n strings byte[][] k, do it like this:
73 * for (int i = 0, h = 0; i < n; ++i) h = hash( k[i], h);
74 *
75 * <p>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 * <p>Use for hash table lookup, or anything where one collision in 2^^32 is
79 * acceptable. Do NOT use for cryptographic purposes.
80 */
81 @Override
82 @SuppressWarnings("fallthrough")
83 public int hash(byte[] key, int off, int nbytes, int initval) {
84 int length = nbytes;
85 long a, b, c; // We use longs because we don't have unsigned ints
86 a = b = c = (0x00000000deadbeefL + length + initval) & INT_MASK;
87 int offset = off;
88 for (; length > 12; offset += 12, length -= 12) {
89 //noinspection PointlessArithmeticExpression
90 a = (a + (key[offset + 0] & BYTE_MASK)) & INT_MASK;
91 a = (a + (((key[offset + 1] & BYTE_MASK) << 8) & INT_MASK)) & INT_MASK;
92 a = (a + (((key[offset + 2] & BYTE_MASK) << 16) & INT_MASK)) & INT_MASK;
93 a = (a + (((key[offset + 3] & BYTE_MASK) << 24) & INT_MASK)) & INT_MASK;
94 b = (b + (key[offset + 4] & BYTE_MASK)) & INT_MASK;
95 b = (b + (((key[offset + 5] & BYTE_MASK) << 8) & INT_MASK)) & INT_MASK;
96 b = (b + (((key[offset + 6] & BYTE_MASK) << 16) & INT_MASK)) & INT_MASK;
97 b = (b + (((key[offset + 7] & BYTE_MASK) << 24) & INT_MASK)) & INT_MASK;
98 c = (c + (key[offset + 8] & BYTE_MASK)) & INT_MASK;
99 c = (c + (((key[offset + 9] & BYTE_MASK) << 8) & INT_MASK)) & INT_MASK;
100 c = (c + (((key[offset + 10] & BYTE_MASK) << 16) & INT_MASK)) & INT_MASK;
101 c = (c + (((key[offset + 11] & BYTE_MASK) << 24) & INT_MASK)) & INT_MASK;
102
103 /*
104 * mix -- mix 3 32-bit values reversibly.
105 * This is reversible, so any information in (a,b,c) before mix() is
106 * still in (a,b,c) after mix().
107 *
108 * If four pairs of (a,b,c) inputs are run through mix(), or through
109 * mix() in reverse, there are at least 32 bits of the output that
110 * are sometimes the same for one pair and different for another pair.
111 *
112 * This was tested for:
113 * - pairs that differed by one bit, by two bits, in any combination
114 * of top bits of (a,b,c), or in any combination of bottom bits of
115 * (a,b,c).
116 * - "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
117 * the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
118 * is commonly produced by subtraction) look like a single 1-bit
119 * difference.
120 * - the base values were pseudorandom, all zero but one bit set, or
121 * all zero plus a counter that starts at zero.
122 *
123 * Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
124 * satisfy this are
125 * 4 6 8 16 19 4
126 * 9 15 3 18 27 15
127 * 14 9 3 7 17 3
128 * Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing for
129 * "differ" defined as + with a one-bit base and a two-bit delta. I
130 * used http://burtleburtle.net/bob/hash/avalanche.html to choose
131 * the operations, constants, and arrangements of the variables.
132 *
133 * This does not achieve avalanche. There are input bits of (a,b,c)
134 * that fail to affect some output bits of (a,b,c), especially of a.
135 * The most thoroughly mixed value is c, but it doesn't really even
136 * achieve avalanche in c.
137 *
138 * This allows some parallelism. Read-after-writes are good at doubling
139 * the number of bits affected, so the goal of mixing pulls in the
140 * opposite direction as the goal of parallelism. I did what I could.
141 * Rotates seem to cost as much as shifts on every machine I could lay
142 * my hands on, and rotates are much kinder to the top and bottom bits,
143 * so I used rotates.
144 *
145 * #define mix(a,b,c) \
146 * { \
147 * a -= c; a ^= rot(c, 4); c += b; \
148 * b -= a; b ^= rot(a, 6); a += c; \
149 * c -= b; c ^= rot(b, 8); b += a; \
150 * a -= c; a ^= rot(c,16); c += b; \
151 * b -= a; b ^= rot(a,19); a += c; \
152 * c -= b; c ^= rot(b, 4); b += a; \
153 * }
154 *
155 * mix(a,b,c);
156 */
157 a = (a - c) & INT_MASK; a ^= rot(c, 4); c = (c + b) & INT_MASK;
158 b = (b - a) & INT_MASK; b ^= rot(a, 6); a = (a + c) & INT_MASK;
159 c = (c - b) & INT_MASK; c ^= rot(b, 8); b = (b + a) & INT_MASK;
160 a = (a - c) & INT_MASK; a ^= rot(c,16); c = (c + b) & INT_MASK;
161 b = (b - a) & INT_MASK; b ^= rot(a,19); a = (a + c) & INT_MASK;
162 c = (c - b) & INT_MASK; c ^= rot(b, 4); b = (b + a) & INT_MASK;
163 }
164
165 //-------------------------------- last block: affect all 32 bits of (c)
166 switch (length) { // all the case statements fall through
167 case 12:
168 c = (c + (((key[offset + 11] & BYTE_MASK) << 24) & INT_MASK)) & INT_MASK;
169 case 11:
170 c = (c + (((key[offset + 10] & BYTE_MASK) << 16) & INT_MASK)) & INT_MASK;
171 case 10:
172 c = (c + (((key[offset + 9] & BYTE_MASK) << 8) & INT_MASK)) & INT_MASK;
173 case 9:
174 c = (c + (key[offset + 8] & BYTE_MASK)) & INT_MASK;
175 case 8:
176 b = (b + (((key[offset + 7] & BYTE_MASK) << 24) & INT_MASK)) & INT_MASK;
177 case 7:
178 b = (b + (((key[offset + 6] & BYTE_MASK) << 16) & INT_MASK)) & INT_MASK;
179 case 6:
180 b = (b + (((key[offset + 5] & BYTE_MASK) << 8) & INT_MASK)) & INT_MASK;
181 case 5:
182 b = (b + (key[offset + 4] & BYTE_MASK)) & INT_MASK;
183 case 4:
184 a = (a + (((key[offset + 3] & BYTE_MASK) << 24) & INT_MASK)) & INT_MASK;
185 case 3:
186 a = (a + (((key[offset + 2] & BYTE_MASK) << 16) & INT_MASK)) & INT_MASK;
187 case 2:
188 a = (a + (((key[offset + 1] & BYTE_MASK) << 8) & INT_MASK)) & INT_MASK;
189 case 1:
190 //noinspection PointlessArithmeticExpression
191 a = (a + (key[offset + 0] & BYTE_MASK)) & INT_MASK;
192 break;
193 case 0:
194 return (int)(c & INT_MASK);
195 }
196 /*
197 * final -- final mixing of 3 32-bit values (a,b,c) into c
198 *
199 * Pairs of (a,b,c) values differing in only a few bits will usually
200 * produce values of c that look totally different. This was tested for
201 * - pairs that differed by one bit, by two bits, in any combination
202 * of top bits of (a,b,c), or in any combination of bottom bits of
203 * (a,b,c).
204 *
205 * - "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
206 * the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
207 * is commonly produced by subtraction) look like a single 1-bit
208 * difference.
209 *
210 * - the base values were pseudorandom, all zero but one bit set, or
211 * all zero plus a counter that starts at zero.
212 *
213 * These constants passed:
214 * 14 11 25 16 4 14 24
215 * 12 14 25 16 4 14 24
216 * and these came close:
217 * 4 8 15 26 3 22 24
218 * 10 8 15 26 3 22 24
219 * 11 8 15 26 3 22 24
220 *
221 * #define final(a,b,c) \
222 * {
223 * c ^= b; c -= rot(b,14); \
224 * a ^= c; a -= rot(c,11); \
225 * b ^= a; b -= rot(a,25); \
226 * c ^= b; c -= rot(b,16); \
227 * a ^= c; a -= rot(c,4); \
228 * b ^= a; b -= rot(a,14); \
229 * c ^= b; c -= rot(b,24); \
230 * }
231 *
232 */
233 c ^= b; c = (c - rot(b,14)) & INT_MASK;
234 a ^= c; a = (a - rot(c,11)) & INT_MASK;
235 b ^= a; b = (b - rot(a,25)) & INT_MASK;
236 c ^= b; c = (c - rot(b,16)) & INT_MASK;
237 a ^= c; a = (a - rot(c,4)) & INT_MASK;
238 b ^= a; b = (b - rot(a,14)) & INT_MASK;
239 c ^= b; c = (c - rot(b,24)) & INT_MASK;
240
241 return (int)(c & INT_MASK);
242 }
243
244 /**
245 * Compute the hash of the specified file
246 * @param args name of file to compute hash of.
247 * @throws IOException e
248 */
249 public static void main(String[] args) throws IOException {
250 if (args.length != 1) {
251 System.err.println("Usage: JenkinsHash filename");
252 System.exit(-1);
253 }
254 FileInputStream in = new FileInputStream(args[0]);
255 byte[] bytes = new byte[512];
256 int value = 0;
257 JenkinsHash hash = new JenkinsHash();
258 for (int length = in.read(bytes); length > 0 ; length = in.read(bytes)) {
259 value = hash.hash(bytes, length, value);
260 }
261 System.out.println(Math.abs(value));
262 }
263 }