| 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.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 {
|
| 42 |
private static long INT_MASK = 0x00000000ffffffffL;
|
| 43 |
private static long BYTE_MASK = 0x00000000000000ffL;
|
| 44 |
|
| 45 |
private static long rot(long val, int pos) {
|
| 46 |
return ((Integer.rotateLeft(
|
| 47 |
(int)(val & INT_MASK), pos)) & INT_MASK);
|
| 48 |
}
|
| 49 |
|
| 50 |
/**
|
| 51 |
* Alternate form for hashing an entire byte array
|
| 52 |
*
|
| 53 |
* @param bytes
|
| 54 |
* @return hash value
|
| 55 |
*/
|
| 56 |
public static int hash(byte[] bytes) {
|
| 57 |
return hash(bytes, bytes.length, -1);
|
| 58 |
}
|
| 59 |
|
| 60 |
/**
|
| 61 |
* Alternate form for hashing an entire byte array
|
| 62 |
*
|
| 63 |
* @param bytes
|
| 64 |
* @param initval
|
| 65 |
* @return hash value
|
| 66 |
*/
|
| 67 |
public static int hash(byte[] bytes, int initval) {
|
| 68 |
return hash(bytes, bytes.length, initval);
|
| 69 |
}
|
| 70 |
|
| 71 |
/**
|
| 72 |
* taken from hashlittle() -- hash a variable-length key into a 32-bit value
|
| 73 |
*
|
| 74 |
* @param key the key (the unaligned variable-length array of bytes)
|
| 75 |
* @param nbytes number of bytes to include in hash
|
| 76 |
* @param initval can be any integer value
|
| 77 |
* @return a 32-bit value. Every bit of the key affects every bit of the
|
| 78 |
* return value. Two keys differing by one or two bits will have totally
|
| 79 |
* different hash values.
|
| 80 |
*
|
| 81 |
* <p>The best hash table sizes are powers of 2. There is no need to do mod
|
| 82 |
* a prime (mod is sooo slow!). If you need less than 32 bits, use a bitmask.
|
| 83 |
* For example, if you need only 10 bits, do
|
| 84 |
* <code>h = (h & hashmask(10));</code>
|
| 85 |
* In which case, the hash table should have hashsize(10) elements.
|
| 86 |
*
|
| 87 |
* <p>If you are hashing n strings byte[][] k, do it like this:
|
| 88 |
* for (int i = 0, h = 0; i < n; ++i) h = hash( k[i], h);
|
| 89 |
*
|
| 90 |
* <p>By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this
|
| 91 |
* code any way you wish, private, educational, or commercial. It's free.
|
| 92 |
*
|
| 93 |
* <p>Use for hash table lookup, or anything where one collision in 2^^32 is
|
| 94 |
* acceptable. Do NOT use for cryptographic purposes.
|
| 95 |
*/
|
| 96 |
@SuppressWarnings("fallthrough")
|
| 97 |
public static int hash(byte[] key, int nbytes, int initval) {
|
| 98 |
int length = nbytes;
|
| 99 |
long a, b, c; // We use longs because we don't have unsigned ints
|
| 100 |
a = b = c = (0x00000000deadbeefL + length + initval) & INT_MASK;
|
| 101 |
int offset = 0;
|
| 102 |
for (; length > 12; offset += 12, length -= 12) {
|
| 103 |
a = (a + (key[offset + 0] & BYTE_MASK)) & INT_MASK;
|
| 104 |
a = (a + (((key[offset + 1] & BYTE_MASK) << 8) & INT_MASK)) & INT_MASK;
|
| 105 |
a = (a + (((key[offset + 2] & BYTE_MASK) << 16) & INT_MASK)) & INT_MASK;
|
| 106 |
a = (a + (((key[offset + 3] & BYTE_MASK) << 24) & INT_MASK)) & INT_MASK;
|
| 107 |
b = (b + (key[offset + 4] & BYTE_MASK)) & INT_MASK;
|
| 108 |
b = (b + (((key[offset + 5] & BYTE_MASK) << 8) & INT_MASK)) & INT_MASK;
|
| 109 |
b = (b + (((key[offset + 6] & BYTE_MASK) << 16) & INT_MASK)) & INT_MASK;
|
| 110 |
b = (b + (((key[offset + 7] & BYTE_MASK) << 24) & INT_MASK)) & INT_MASK;
|
| 111 |
c = (c + (key[offset + 8] & BYTE_MASK)) & INT_MASK;
|
| 112 |
c = (c + (((key[offset + 9] & BYTE_MASK) << 8) & INT_MASK)) & INT_MASK;
|
| 113 |
c = (c + (((key[offset + 10] & BYTE_MASK) << 16) & INT_MASK)) & INT_MASK;
|
| 114 |
c = (c + (((key[offset + 11] & BYTE_MASK) << 24) & INT_MASK)) & INT_MASK;
|
| 115 |
|
| 116 |
/*
|
| 117 |
* mix -- mix 3 32-bit values reversibly.
|
| 118 |
* This is reversible, so any information in (a,b,c) before mix() is
|
| 119 |
* still in (a,b,c) after mix().
|
| 120 |
*
|
| 121 |
* If four pairs of (a,b,c) inputs are run through mix(), or through
|
| 122 |
* mix() in reverse, there are at least 32 bits of the output that
|
| 123 |
* are sometimes the same for one pair and different for another pair.
|
| 124 |
*
|
| 125 |
* This was tested for:
|
| 126 |
* - pairs that differed by one bit, by two bits, in any combination
|
| 127 |
* of top bits of (a,b,c), or in any combination of bottom bits of
|
| 128 |
* (a,b,c).
|
| 129 |
* - "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
|
| 130 |
* the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
|
| 131 |
* is commonly produced by subtraction) look like a single 1-bit
|
| 132 |
* difference.
|
| 133 |
* - the base values were pseudorandom, all zero but one bit set, or
|
| 134 |
* all zero plus a counter that starts at zero.
|
| 135 |
*
|
| 136 |
* Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
|
| 137 |
* satisfy this are
|
| 138 |
* 4 6 8 16 19 4
|
| 139 |
* 9 15 3 18 27 15
|
| 140 |
* 14 9 3 7 17 3
|
| 141 |
* Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing for
|
| 142 |
* "differ" defined as + with a one-bit base and a two-bit delta. I
|
| 143 |
* used http://burtleburtle.net/bob/hash/avalanche.html to choose
|
| 144 |
* the operations, constants, and arrangements of the variables.
|
| 145 |
*
|
| 146 |
* This does not achieve avalanche. There are input bits of (a,b,c)
|
| 147 |
* that fail to affect some output bits of (a,b,c), especially of a.
|
| 148 |
* The most thoroughly mixed value is c, but it doesn't really even
|
| 149 |
* achieve avalanche in c.
|
| 150 |
*
|
| 151 |
* This allows some parallelism. Read-after-writes are good at doubling
|
| 152 |
* the number of bits affected, so the goal of mixing pulls in the
|
| 153 |
* opposite direction as the goal of parallelism. I did what I could.
|
| 154 |
* Rotates seem to cost as much as shifts on every machine I could lay
|
| 155 |
* my hands on, and rotates are much kinder to the top and bottom bits,
|
| 156 |
* so I used rotates.
|
| 157 |
*
|
| 158 |
* #define mix(a,b,c) \
|
| 159 |
* { \
|
| 160 |
* a -= c; a ^= rot(c, 4); c += b; \
|
| 161 |
* b -= a; b ^= rot(a, 6); a += c; \
|
| 162 |
* c -= b; c ^= rot(b, 8); b += a; \
|
| 163 |
* a -= c; a ^= rot(c,16); c += b; \
|
| 164 |
* b -= a; b ^= rot(a,19); a += c; \
|
| 165 |
* c -= b; c ^= rot(b, 4); b += a; \
|
| 166 |
* }
|
| 167 |
*
|
| 168 |
* mix(a,b,c);
|
| 169 |
*/
|
| 170 |
a = (a - c) & INT_MASK; a ^= rot(c, 4); c = (c + b) & INT_MASK;
|
| 171 |
b = (b - a) & INT_MASK; b ^= rot(a, 6); a = (a + c) & INT_MASK;
|
| 172 |
c = (c - b) & INT_MASK; c ^= rot(b, 8); b = (b + a) & INT_MASK;
|
| 173 |
a = (a - c) & INT_MASK; a ^= rot(c,16); c = (c + b) & INT_MASK;
|
| 174 |
b = (b - a) & INT_MASK; b ^= rot(a,19); a = (a + c) & INT_MASK;
|
| 175 |
c = (c - b) & INT_MASK; c ^= rot(b, 4); b = (b + a) & INT_MASK;
|
| 176 |
}
|
| 177 |
|
| 178 |
//-------------------------------- last block: affect all 32 bits of (c)
|
| 179 |
switch (length) { // all the case statements fall through
|
| 180 |
case 12:
|
| 181 |
c = (c + (((key[offset + 11] & BYTE_MASK) << 24) & INT_MASK)) & INT_MASK;
|
| 182 |
case 11:
|
| 183 |
c = (c + (((key[offset + 10] & BYTE_MASK) << 16) & INT_MASK)) & INT_MASK;
|
| 184 |
case 10:
|
| 185 |
c = (c + (((key[offset + 9] & BYTE_MASK) << 8) & INT_MASK)) & INT_MASK;
|
| 186 |
case 9:
|
| 187 |
c = (c + (key[offset + 8] & BYTE_MASK)) & INT_MASK;
|
| 188 |
case 8:
|
| 189 |
b = (b + (((key[offset + 7] & BYTE_MASK) << 24) & INT_MASK)) & INT_MASK;
|
| 190 |
case 7:
|
| 191 |
b = (b + (((key[offset + 6] & BYTE_MASK) << 16) & INT_MASK)) & INT_MASK;
|
| 192 |
case 6:
|
| 193 |
b = (b + (((key[offset + 5] & BYTE_MASK) << 8) & INT_MASK)) & INT_MASK;
|
| 194 |
case 5:
|
| 195 |
b = (b + (key[offset + 4] & BYTE_MASK)) & INT_MASK;
|
| 196 |
case 4:
|
| 197 |
a = (a + (((key[offset + 3] & BYTE_MASK) << 24) & INT_MASK)) & INT_MASK;
|
| 198 |
case 3:
|
| 199 |
a = (a + (((key[offset + 2] & BYTE_MASK) << 16) & INT_MASK)) & INT_MASK;
|
| 200 |
case 2:
|
| 201 |
a = (a + (((key[offset + 1] & BYTE_MASK) << 8) & INT_MASK)) & INT_MASK;
|
| 202 |
case 1:
|
| 203 |
a = (a + (key[offset + 0] & BYTE_MASK)) & INT_MASK;
|
| 204 |
break;
|
| 205 |
case 0:
|
| 206 |
return (int)(c & INT_MASK);
|
| 207 |
}
|
| 208 |
/*
|
| 209 |
* final -- final mixing of 3 32-bit values (a,b,c) into c
|
| 210 |
*
|
| 211 |
* Pairs of (a,b,c) values differing in only a few bits will usually
|
| 212 |
* produce values of c that look totally different. This was tested for
|
| 213 |
* - pairs that differed by one bit, by two bits, in any combination
|
| 214 |
* of top bits of (a,b,c), or in any combination of bottom bits of
|
| 215 |
* (a,b,c).
|
| 216 |
*
|
| 217 |
* - "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
|
| 218 |
* the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
|
| 219 |
* is commonly produced by subtraction) look like a single 1-bit
|
| 220 |
* difference.
|
| 221 |
*
|
| 222 |
* - the base values were pseudorandom, all zero but one bit set, or
|
| 223 |
* all zero plus a counter that starts at zero.
|
| 224 |
*
|
| 225 |
* These constants passed:
|
| 226 |
* 14 11 25 16 4 14 24
|
| 227 |
* 12 14 25 16 4 14 24
|
| 228 |
* and these came close:
|
| 229 |
* 4 8 15 26 3 22 24
|
| 230 |
* 10 8 15 26 3 22 24
|
| 231 |
* 11 8 15 26 3 22 24
|
| 232 |
*
|
| 233 |
* #define final(a,b,c) \
|
| 234 |
* {
|
| 235 |
* c ^= b; c -= rot(b,14); \
|
| 236 |
* a ^= c; a -= rot(c,11); \
|
| 237 |
* b ^= a; b -= rot(a,25); \
|
| 238 |
* c ^= b; c -= rot(b,16); \
|
| 239 |
* a ^= c; a -= rot(c,4); \
|
| 240 |
* b ^= a; b -= rot(a,14); \
|
| 241 |
* c ^= b; c -= rot(b,24); \
|
| 242 |
* }
|
| 243 |
*
|
| 244 |
*/
|
| 245 |
c ^= b; c = (c - rot(b,14)) & INT_MASK;
|
| 246 |
a ^= c; a = (a - rot(c,11)) & INT_MASK;
|
| 247 |
b ^= a; b = (b - rot(a,25)) & INT_MASK;
|
| 248 |
c ^= b; c = (c - rot(b,16)) & INT_MASK;
|
| 249 |
a ^= c; a = (a - rot(c,4)) & INT_MASK;
|
| 250 |
b ^= a; b = (b - rot(a,14)) & INT_MASK;
|
| 251 |
c ^= b; c = (c - rot(b,24)) & INT_MASK;
|
| 252 |
|
| 253 |
return (int)(c & INT_MASK);
|
| 254 |
}
|
| 255 |
|
| 256 |
/**
|
| 257 |
* Compute the hash of the specified file
|
| 258 |
* @param args name of file to compute hash of.
|
| 259 |
* @throws IOException
|
| 260 |
*/
|
| 261 |
public static void main(String[] args) throws IOException {
|
| 262 |
if (args.length != 1) {
|
| 263 |
System.err.println("Usage: JenkinsHash filename");
|
| 264 |
System.exit(-1);
|
| 265 |
}
|
| 266 |
FileInputStream in = new FileInputStream(args[0]);
|
| 267 |
byte[] bytes = new byte[512];
|
| 268 |
int value = 0;
|
| 269 |
for (int length = in.read(bytes); length > 0 ; length = in.read(bytes)) {
|
| 270 |
value = hash(bytes, length, value);
|
| 271 |
}
|
| 272 |
System.out.println(Math.abs(value));
|
| 273 |
}
|
| 274 |
}
|
Parent Directory
| 
Revision Log