[120] | 1 | /** |
---|
| 2 | * Copyright 2007 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.util.hash; |
---|
| 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 | @SuppressWarnings("fallthrough") |
---|
| 82 | public int hash(byte[] key, int nbytes, int initval) { |
---|
| 83 | int length = nbytes; |
---|
| 84 | long a, b, c; // We use longs because we don't have unsigned ints |
---|
| 85 | a = b = c = (0x00000000deadbeefL + length + initval) & INT_MASK; |
---|
| 86 | int offset = 0; |
---|
| 87 | for (; length > 12; offset += 12, length -= 12) { |
---|
| 88 | a = (a + (key[offset + 0] & BYTE_MASK)) & INT_MASK; |
---|
| 89 | a = (a + (((key[offset + 1] & BYTE_MASK) << 8) & INT_MASK)) & INT_MASK; |
---|
| 90 | a = (a + (((key[offset + 2] & BYTE_MASK) << 16) & INT_MASK)) & INT_MASK; |
---|
| 91 | a = (a + (((key[offset + 3] & BYTE_MASK) << 24) & INT_MASK)) & INT_MASK; |
---|
| 92 | b = (b + (key[offset + 4] & BYTE_MASK)) & INT_MASK; |
---|
| 93 | b = (b + (((key[offset + 5] & BYTE_MASK) << 8) & INT_MASK)) & INT_MASK; |
---|
| 94 | b = (b + (((key[offset + 6] & BYTE_MASK) << 16) & INT_MASK)) & INT_MASK; |
---|
| 95 | b = (b + (((key[offset + 7] & BYTE_MASK) << 24) & INT_MASK)) & INT_MASK; |
---|
| 96 | c = (c + (key[offset + 8] & BYTE_MASK)) & INT_MASK; |
---|
| 97 | c = (c + (((key[offset + 9] & BYTE_MASK) << 8) & INT_MASK)) & INT_MASK; |
---|
| 98 | c = (c + (((key[offset + 10] & BYTE_MASK) << 16) & INT_MASK)) & INT_MASK; |
---|
| 99 | c = (c + (((key[offset + 11] & BYTE_MASK) << 24) & INT_MASK)) & INT_MASK; |
---|
| 100 | |
---|
| 101 | /* |
---|
| 102 | * mix -- mix 3 32-bit values reversibly. |
---|
| 103 | * This is reversible, so any information in (a,b,c) before mix() is |
---|
| 104 | * still in (a,b,c) after mix(). |
---|
| 105 | * |
---|
| 106 | * If four pairs of (a,b,c) inputs are run through mix(), or through |
---|
| 107 | * mix() in reverse, there are at least 32 bits of the output that |
---|
| 108 | * are sometimes the same for one pair and different for another pair. |
---|
| 109 | * |
---|
| 110 | * This was tested for: |
---|
| 111 | * - pairs that differed by one bit, by two bits, in any combination |
---|
| 112 | * of top bits of (a,b,c), or in any combination of bottom bits of |
---|
| 113 | * (a,b,c). |
---|
| 114 | * - "differ" is defined as +, -, ^, or ~^. For + and -, I transformed |
---|
| 115 | * the output delta to a Gray code (a^(a>>1)) so a string of 1's (as |
---|
| 116 | * is commonly produced by subtraction) look like a single 1-bit |
---|
| 117 | * difference. |
---|
| 118 | * - the base values were pseudorandom, all zero but one bit set, or |
---|
| 119 | * all zero plus a counter that starts at zero. |
---|
| 120 | * |
---|
| 121 | * Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that |
---|
| 122 | * satisfy this are |
---|
| 123 | * 4 6 8 16 19 4 |
---|
| 124 | * 9 15 3 18 27 15 |
---|
| 125 | * 14 9 3 7 17 3 |
---|
| 126 | * Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing for |
---|
| 127 | * "differ" defined as + with a one-bit base and a two-bit delta. I |
---|
| 128 | * used http://burtleburtle.net/bob/hash/avalanche.html to choose |
---|
| 129 | * the operations, constants, and arrangements of the variables. |
---|
| 130 | * |
---|
| 131 | * This does not achieve avalanche. There are input bits of (a,b,c) |
---|
| 132 | * that fail to affect some output bits of (a,b,c), especially of a. |
---|
| 133 | * The most thoroughly mixed value is c, but it doesn't really even |
---|
| 134 | * achieve avalanche in c. |
---|
| 135 | * |
---|
| 136 | * This allows some parallelism. Read-after-writes are good at doubling |
---|
| 137 | * the number of bits affected, so the goal of mixing pulls in the |
---|
| 138 | * opposite direction as the goal of parallelism. I did what I could. |
---|
| 139 | * Rotates seem to cost as much as shifts on every machine I could lay |
---|
| 140 | * my hands on, and rotates are much kinder to the top and bottom bits, |
---|
| 141 | * so I used rotates. |
---|
| 142 | * |
---|
| 143 | * #define mix(a,b,c) \ |
---|
| 144 | * { \ |
---|
| 145 | * a -= c; a ^= rot(c, 4); c += b; \ |
---|
| 146 | * b -= a; b ^= rot(a, 6); a += c; \ |
---|
| 147 | * c -= b; c ^= rot(b, 8); b += a; \ |
---|
| 148 | * a -= c; a ^= rot(c,16); c += b; \ |
---|
| 149 | * b -= a; b ^= rot(a,19); a += c; \ |
---|
| 150 | * c -= b; c ^= rot(b, 4); b += a; \ |
---|
| 151 | * } |
---|
| 152 | * |
---|
| 153 | * mix(a,b,c); |
---|
| 154 | */ |
---|
| 155 | a = (a - c) & INT_MASK; a ^= rot(c, 4); c = (c + b) & INT_MASK; |
---|
| 156 | b = (b - a) & INT_MASK; b ^= rot(a, 6); a = (a + c) & INT_MASK; |
---|
| 157 | c = (c - b) & INT_MASK; c ^= rot(b, 8); b = (b + a) & INT_MASK; |
---|
| 158 | a = (a - c) & INT_MASK; a ^= rot(c,16); c = (c + b) & INT_MASK; |
---|
| 159 | b = (b - a) & INT_MASK; b ^= rot(a,19); a = (a + c) & INT_MASK; |
---|
| 160 | c = (c - b) & INT_MASK; c ^= rot(b, 4); b = (b + a) & INT_MASK; |
---|
| 161 | } |
---|
| 162 | |
---|
| 163 | //-------------------------------- last block: affect all 32 bits of (c) |
---|
| 164 | switch (length) { // all the case statements fall through |
---|
| 165 | case 12: |
---|
| 166 | c = (c + (((key[offset + 11] & BYTE_MASK) << 24) & INT_MASK)) & INT_MASK; |
---|
| 167 | case 11: |
---|
| 168 | c = (c + (((key[offset + 10] & BYTE_MASK) << 16) & INT_MASK)) & INT_MASK; |
---|
| 169 | case 10: |
---|
| 170 | c = (c + (((key[offset + 9] & BYTE_MASK) << 8) & INT_MASK)) & INT_MASK; |
---|
| 171 | case 9: |
---|
| 172 | c = (c + (key[offset + 8] & BYTE_MASK)) & INT_MASK; |
---|
| 173 | case 8: |
---|
| 174 | b = (b + (((key[offset + 7] & BYTE_MASK) << 24) & INT_MASK)) & INT_MASK; |
---|
| 175 | case 7: |
---|
| 176 | b = (b + (((key[offset + 6] & BYTE_MASK) << 16) & INT_MASK)) & INT_MASK; |
---|
| 177 | case 6: |
---|
| 178 | b = (b + (((key[offset + 5] & BYTE_MASK) << 8) & INT_MASK)) & INT_MASK; |
---|
| 179 | case 5: |
---|
| 180 | b = (b + (key[offset + 4] & BYTE_MASK)) & INT_MASK; |
---|
| 181 | case 4: |
---|
| 182 | a = (a + (((key[offset + 3] & BYTE_MASK) << 24) & INT_MASK)) & INT_MASK; |
---|
| 183 | case 3: |
---|
| 184 | a = (a + (((key[offset + 2] & BYTE_MASK) << 16) & INT_MASK)) & INT_MASK; |
---|
| 185 | case 2: |
---|
| 186 | a = (a + (((key[offset + 1] & BYTE_MASK) << 8) & INT_MASK)) & INT_MASK; |
---|
| 187 | case 1: |
---|
| 188 | a = (a + (key[offset + 0] & BYTE_MASK)) & INT_MASK; |
---|
| 189 | break; |
---|
| 190 | case 0: |
---|
| 191 | return (int)(c & INT_MASK); |
---|
| 192 | } |
---|
| 193 | /* |
---|
| 194 | * final -- final mixing of 3 32-bit values (a,b,c) into c |
---|
| 195 | * |
---|
| 196 | * Pairs of (a,b,c) values differing in only a few bits will usually |
---|
| 197 | * produce values of c that look totally different. This was tested for |
---|
| 198 | * - pairs that differed by one bit, by two bits, in any combination |
---|
| 199 | * of top bits of (a,b,c), or in any combination of bottom bits of |
---|
| 200 | * (a,b,c). |
---|
| 201 | * |
---|
| 202 | * - "differ" is defined as +, -, ^, or ~^. For + and -, I transformed |
---|
| 203 | * the output delta to a Gray code (a^(a>>1)) so a string of 1's (as |
---|
| 204 | * is commonly produced by subtraction) look like a single 1-bit |
---|
| 205 | * difference. |
---|
| 206 | * |
---|
| 207 | * - the base values were pseudorandom, all zero but one bit set, or |
---|
| 208 | * all zero plus a counter that starts at zero. |
---|
| 209 | * |
---|
| 210 | * These constants passed: |
---|
| 211 | * 14 11 25 16 4 14 24 |
---|
| 212 | * 12 14 25 16 4 14 24 |
---|
| 213 | * and these came close: |
---|
| 214 | * 4 8 15 26 3 22 24 |
---|
| 215 | * 10 8 15 26 3 22 24 |
---|
| 216 | * 11 8 15 26 3 22 24 |
---|
| 217 | * |
---|
| 218 | * #define final(a,b,c) \ |
---|
| 219 | * { |
---|
| 220 | * c ^= b; c -= rot(b,14); \ |
---|
| 221 | * a ^= c; a -= rot(c,11); \ |
---|
| 222 | * b ^= a; b -= rot(a,25); \ |
---|
| 223 | * c ^= b; c -= rot(b,16); \ |
---|
| 224 | * a ^= c; a -= rot(c,4); \ |
---|
| 225 | * b ^= a; b -= rot(a,14); \ |
---|
| 226 | * c ^= b; c -= rot(b,24); \ |
---|
| 227 | * } |
---|
| 228 | * |
---|
| 229 | */ |
---|
| 230 | c ^= b; c = (c - rot(b,14)) & INT_MASK; |
---|
| 231 | a ^= c; a = (a - rot(c,11)) & INT_MASK; |
---|
| 232 | b ^= a; b = (b - rot(a,25)) & INT_MASK; |
---|
| 233 | c ^= b; c = (c - rot(b,16)) & INT_MASK; |
---|
| 234 | a ^= c; a = (a - rot(c,4)) & INT_MASK; |
---|
| 235 | b ^= a; b = (b - rot(a,14)) & INT_MASK; |
---|
| 236 | c ^= b; c = (c - rot(b,24)) & INT_MASK; |
---|
| 237 | |
---|
| 238 | return (int)(c & INT_MASK); |
---|
| 239 | } |
---|
| 240 | |
---|
| 241 | /** |
---|
| 242 | * Compute the hash of the specified file |
---|
| 243 | * @param args name of file to compute hash of. |
---|
| 244 | * @throws IOException |
---|
| 245 | */ |
---|
| 246 | public static void main(String[] args) throws IOException { |
---|
| 247 | if (args.length != 1) { |
---|
| 248 | System.err.println("Usage: JenkinsHash filename"); |
---|
| 249 | System.exit(-1); |
---|
| 250 | } |
---|
| 251 | FileInputStream in = new FileInputStream(args[0]); |
---|
| 252 | byte[] bytes = new byte[512]; |
---|
| 253 | int value = 0; |
---|
| 254 | JenkinsHash hash = new JenkinsHash(); |
---|
| 255 | for (int length = in.read(bytes); length > 0 ; length = in.read(bytes)) { |
---|
| 256 | value = hash.hash(bytes, length, value); |
---|
| 257 | } |
---|
| 258 | System.out.println(Math.abs(value)); |
---|
| 259 | } |
---|
| 260 | } |
---|