1 #include "tdb_private.h"
3 /* This is based on the hash algorithm from gdbm */
4 unsigned int tdb_old_hash(TDB_DATA *key)
6 uint32_t value; /* Used to compute the hash value. */
7 uint32_t i; /* Used to cycle through random values. */
9 /* Set the initial value from the key size. */
10 for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
11 value = (value + (key->dptr[i] << (i*5 % 24)));
13 return (1103515243 * value + 12345);
16 #if HAVE_LITTLE_ENDIAN
17 # define HASH_LITTLE_ENDIAN 1
18 # define HASH_BIG_ENDIAN 0
20 # define HASH_LITTLE_ENDIAN 0
21 # define HASH_BIG_ENDIAN 1
23 # error Unknown endian
27 -------------------------------------------------------------------------------
28 lookup3.c, by Bob Jenkins, May 2006, Public Domain.
30 These are functions for producing 32-bit hashes for hash table lookup.
31 hash_word(), hashlittle(), hashlittle2(), hashbig(), mix(), and final()
32 are externally useful functions. Routines to test the hash are included
33 if SELF_TEST is defined. You can use this free for any purpose. It's in
34 the public domain. It has no warranty.
36 You probably want to use hashlittle(). hashlittle() and hashbig()
37 hash byte arrays. hashlittle() is is faster than hashbig() on
38 little-endian machines. Intel and AMD are little-endian machines.
39 On second thought, you probably want hashlittle2(), which is identical to
40 hashlittle() except it returns two 32-bit hashes for the price of one.
41 You could implement hashbig2() if you wanted but I haven't bothered here.
43 If you want to find a hash of, say, exactly 7 integers, do
44 a = i1; b = i2; c = i3;
46 a += i4; b += i5; c += i6;
50 then use c as the hash value. If you have a variable length array of
51 4-byte integers to hash, use hash_word(). If you have a byte array (like
52 a character string), use hashlittle(). If you have several byte arrays, or
53 a mix of things, see the comments above hashlittle().
55 Why is this so big? I read 12 bytes at a time into 3 4-byte integers,
56 then mix those integers. This is fast (you can do a lot more thorough
57 mixing with 12*3 instructions on 3 integers than you can with 3 instructions
58 on 1 byte), but shoehorning those bytes into integers efficiently is messy.
61 #define hashsize(n) ((uint32_t)1<<(n))
62 #define hashmask(n) (hashsize(n)-1)
63 #define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
66 -------------------------------------------------------------------------------
67 mix -- mix 3 32-bit values reversibly.
69 This is reversible, so any information in (a,b,c) before mix() is
70 still in (a,b,c) after mix().
72 If four pairs of (a,b,c) inputs are run through mix(), or through
73 mix() in reverse, there are at least 32 bits of the output that
74 are sometimes the same for one pair and different for another pair.
76 * pairs that differed by one bit, by two bits, in any combination
77 of top bits of (a,b,c), or in any combination of bottom bits of
79 * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
80 the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
81 is commonly produced by subtraction) look like a single 1-bit
83 * the base values were pseudorandom, all zero but one bit set, or
84 all zero plus a counter that starts at zero.
86 Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
91 Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
92 for "differ" defined as + with a one-bit base and a two-bit delta. I
93 used http://burtleburtle.net/bob/hash/avalanche.html to choose
94 the operations, constants, and arrangements of the variables.
96 This does not achieve avalanche. There are input bits of (a,b,c)
97 that fail to affect some output bits of (a,b,c), especially of a. The
98 most thoroughly mixed value is c, but it doesn't really even achieve
101 This allows some parallelism. Read-after-writes are good at doubling
102 the number of bits affected, so the goal of mixing pulls in the opposite
103 direction as the goal of parallelism. I did what I could. Rotates
104 seem to cost as much as shifts on every machine I could lay my hands
105 on, and rotates are much kinder to the top and bottom bits, so I used
107 -------------------------------------------------------------------------------
111 a -= c; a ^= rot(c, 4); c += b; \
112 b -= a; b ^= rot(a, 6); a += c; \
113 c -= b; c ^= rot(b, 8); b += a; \
114 a -= c; a ^= rot(c,16); c += b; \
115 b -= a; b ^= rot(a,19); a += c; \
116 c -= b; c ^= rot(b, 4); b += a; \
120 -------------------------------------------------------------------------------
121 final -- final mixing of 3 32-bit values (a,b,c) into c
123 Pairs of (a,b,c) values differing in only a few bits will usually
124 produce values of c that look totally different. This was tested for
125 * pairs that differed by one bit, by two bits, in any combination
126 of top bits of (a,b,c), or in any combination of bottom bits of
128 * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
129 the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
130 is commonly produced by subtraction) look like a single 1-bit
132 * the base values were pseudorandom, all zero but one bit set, or
133 all zero plus a counter that starts at zero.
135 These constants passed:
138 and these came close:
142 -------------------------------------------------------------------------------
144 #define final(a,b,c) \
146 c ^= b; c -= rot(b,14); \
147 a ^= c; a -= rot(c,11); \
148 b ^= a; b -= rot(a,25); \
149 c ^= b; c -= rot(b,16); \
150 a ^= c; a -= rot(c,4); \
151 b ^= a; b -= rot(a,14); \
152 c ^= b; c -= rot(b,24); \
157 -------------------------------------------------------------------------------
158 hashlittle() -- hash a variable-length key into a 32-bit value
159 k : the key (the unaligned variable-length array of bytes)
160 length : the length of the key, counting by bytes
161 val2 : IN: can be any 4-byte value OUT: second 32 bit hash.
162 Returns a 32-bit value. Every bit of the key affects every bit of
163 the return value. Two keys differing by one or two bits will have
164 totally different hash values. Note that the return value is better
165 mixed than val2, so use that first.
167 The best hash table sizes are powers of 2. There is no need to do
168 mod a prime (mod is sooo slow!). If you need less than 32 bits,
169 use a bitmask. For example, if you need only 10 bits, do
170 h = (h & hashmask(10));
171 In which case, the hash table should have hashsize(10) elements.
173 If you are hashing n strings (uint8_t **)k, do it like this:
174 for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h);
176 By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this
177 code any way you wish, private, educational, or commercial. It's free.
179 Use for hash table lookup, or anything where one collision in 2^^32 is
180 acceptable. Do NOT use for cryptographic purposes.
181 -------------------------------------------------------------------------------
184 static uint32_t hashlittle( const void *key, size_t length )
186 uint32_t a,b,c; /* internal state */
187 union { const void *ptr; size_t i; } u; /* needed for Mac Powerbook G4 */
189 /* Set up the internal state */
190 a = b = c = 0xdeadbeef + ((uint32_t)length);
193 if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
194 const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */
199 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
210 /*----------------------------- handle the last (probably partial) block */
212 * "k[2]&0xffffff" actually reads beyond the end of the string, but
213 * then masks off the part it's not allowed to read. Because the
214 * string is aligned, the masked-off tail is in the same word as the
215 * rest of the string. Every machine with memory protection I've seen
216 * does it on word boundaries, so is OK with this. But VALGRIND will
217 * still catch it and complain. The masking trick does make the hash
218 * noticably faster for short strings (like English words).
224 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
225 case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
226 case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break;
227 case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break;
228 case 8 : b+=k[1]; a+=k[0]; break;
229 case 7 : b+=k[1]&0xffffff; a+=k[0]; break;
230 case 6 : b+=k[1]&0xffff; a+=k[0]; break;
231 case 5 : b+=k[1]&0xff; a+=k[0]; break;
232 case 4 : a+=k[0]; break;
233 case 3 : a+=k[0]&0xffffff; break;
234 case 2 : a+=k[0]&0xffff; break;
235 case 1 : a+=k[0]&0xff; break;
236 case 0 : return c; /* zero length strings require no mixing */
239 #else /* make valgrind happy */
241 k8 = (const uint8_t *)k;
244 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
245 case 11: c+=((uint32_t)k8[10])<<16; /* fall through */
246 case 10: c+=((uint32_t)k8[9])<<8; /* fall through */
247 case 9 : c+=k8[8]; /* fall through */
248 case 8 : b+=k[1]; a+=k[0]; break;
249 case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */
250 case 6 : b+=((uint32_t)k8[5])<<8; /* fall through */
251 case 5 : b+=k8[4]; /* fall through */
252 case 4 : a+=k[0]; break;
253 case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */
254 case 2 : a+=((uint32_t)k8[1])<<8; /* fall through */
255 case 1 : a+=k8[0]; break;
259 #endif /* !valgrind */
261 } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
262 const uint16_t *k = (const uint16_t *)key; /* read 16-bit chunks */
265 /*--------------- all but last block: aligned reads and different mixing */
268 a += k[0] + (((uint32_t)k[1])<<16);
269 b += k[2] + (((uint32_t)k[3])<<16);
270 c += k[4] + (((uint32_t)k[5])<<16);
276 /*----------------------------- handle the last (probably partial) block */
277 k8 = (const uint8_t *)k;
280 case 12: c+=k[4]+(((uint32_t)k[5])<<16);
281 b+=k[2]+(((uint32_t)k[3])<<16);
282 a+=k[0]+(((uint32_t)k[1])<<16);
284 case 11: c+=((uint32_t)k8[10])<<16; /* fall through */
286 b+=k[2]+(((uint32_t)k[3])<<16);
287 a+=k[0]+(((uint32_t)k[1])<<16);
289 case 9 : c+=k8[8]; /* fall through */
290 case 8 : b+=k[2]+(((uint32_t)k[3])<<16);
291 a+=k[0]+(((uint32_t)k[1])<<16);
293 case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */
295 a+=k[0]+(((uint32_t)k[1])<<16);
297 case 5 : b+=k8[4]; /* fall through */
298 case 4 : a+=k[0]+(((uint32_t)k[1])<<16);
300 case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */
305 case 0 : return c; /* zero length requires no mixing */
308 } else { /* need to read the key one byte at a time */
309 const uint8_t *k = (const uint8_t *)key;
311 /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
315 a += ((uint32_t)k[1])<<8;
316 a += ((uint32_t)k[2])<<16;
317 a += ((uint32_t)k[3])<<24;
319 b += ((uint32_t)k[5])<<8;
320 b += ((uint32_t)k[6])<<16;
321 b += ((uint32_t)k[7])<<24;
323 c += ((uint32_t)k[9])<<8;
324 c += ((uint32_t)k[10])<<16;
325 c += ((uint32_t)k[11])<<24;
331 /*-------------------------------- last block: affect all 32 bits of (c) */
332 switch(length) /* all the case statements fall through */
334 case 12: c+=((uint32_t)k[11])<<24;
335 case 11: c+=((uint32_t)k[10])<<16;
336 case 10: c+=((uint32_t)k[9])<<8;
338 case 8 : b+=((uint32_t)k[7])<<24;
339 case 7 : b+=((uint32_t)k[6])<<16;
340 case 6 : b+=((uint32_t)k[5])<<8;
342 case 4 : a+=((uint32_t)k[3])<<24;
343 case 3 : a+=((uint32_t)k[2])<<16;
344 case 2 : a+=((uint32_t)k[1])<<8;
355 unsigned int tdb_jenkins_hash(TDB_DATA *key)
357 return hashlittle(key->dptr, key->dsize);