a3a3152dcd1af2f03cc7b227867723dc4202d70b
2 * Copyright (C) - Bob Jenkins, May 2006, Public Domain.
3 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
5 * These are functions for producing 32-bit hashes for hash table lookup.
6 * hashword(), hashlittle(), hashlittle2(), hashbig(), mix(), and final() are
7 * externally useful functions. Routines to test the hash are included if
8 * SELF_TEST is defined. You can use this free for any purpose. It's in the
9 * public domain. It has no warranty.
11 * You probably want to use hashlittle(). hashlittle() and hashbig() hash byte
12 * arrays. hashlittle() is is faster than hashbig() on little-endian machines.
13 * Intel and AMD are little-endian machines. On second thought, you probably
14 * want hashlittle2(), which is identical to hashlittle() except it returns two
15 * 32-bit hashes for the price of one. You could implement hashbig2() if you
16 * wanted but I haven't bothered here.
18 * If you want to find a hash of, say, exactly 7 integers, do
19 * a = i1; b = i2; c = i3;
21 * a += i4; b += i5; c += i6;
25 * then use c as the hash value. If you have a variable length array of
26 * 4-byte integers to hash, use hashword(). If you have a byte array (like
27 * a character string), use hashlittle(). If you have several byte arrays, or
28 * a mix of things, see the comments above hashlittle().
30 * Why is this so big? I read 12 bytes at a time into 3 4-byte integers, then
31 * mix those integers. This is fast (you can do a lot more thorough mixing
32 * with 12*3 instructions on 3 integers than you can with 3 instructions on 1
33 * byte), but shoehorning those bytes into integers efficiently is messy.
36 #include <stdio.h> /* defines printf for tests */
37 #include <time.h> /* defines time_t for timings in the test */
38 #include <stdint.h> /* defines uint32_t etc */
39 #include <sys/param.h> /* attempt to define endianness */
40 #include <endian.h> /* attempt to define endianness */
45 * My best guess at if you are big-endian or little-endian. This may
48 #if (defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && \
49 __BYTE_ORDER == __LITTLE_ENDIAN) || \
50 (defined(i386) || defined(__i386__) || defined(__i486__) || \
51 defined(__i586__) || defined(__i686__) || defined(vax) || defined(MIPSEL))
52 # define HASH_LITTLE_ENDIAN 1
53 # define HASH_BIG_ENDIAN 0
54 #elif (defined(__BYTE_ORDER) && defined(__BIG_ENDIAN) && \
55 __BYTE_ORDER == __BIG_ENDIAN) || \
56 (defined(sparc) || defined(POWERPC) || defined(mc68000) || defined(sel))
57 # define HASH_LITTLE_ENDIAN 0
58 # define HASH_BIG_ENDIAN 1
60 # define HASH_LITTLE_ENDIAN 0
61 # define HASH_BIG_ENDIAN 0
64 #define hashsize(n) ((uint32_t)1<<(n))
65 #define hashmask(n) (hashsize(n)-1)
66 #define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
69 * mix -- mix 3 32-bit values reversibly.
71 * This is reversible, so any information in (a,b,c) before mix() is
72 * still in (a,b,c) after mix().
74 * If four pairs of (a,b,c) inputs are run through mix(), or through
75 * mix() in reverse, there are at least 32 bits of the output that
76 * are sometimes the same for one pair and different for another pair.
77 * This was tested for:
78 * * pairs that differed by one bit, by two bits, in any combination
79 * of top bits of (a,b,c), or in any combination of bottom bits of
81 * * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
82 * the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
83 * is commonly produced by subtraction) look like a single 1-bit
85 * * the base values were pseudorandom, all zero but one bit set, or
86 * all zero plus a counter that starts at zero.
88 * Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
93 * Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
94 * for "differ" defined as + with a one-bit base and a two-bit delta. I
95 * used http://burtleburtle.net/bob/hash/avalanche.html to choose
96 * the operations, constants, and arrangements of the variables.
98 * This does not achieve avalanche. There are input bits of (a,b,c)
99 * that fail to affect some output bits of (a,b,c), especially of a. The
100 * most thoroughly mixed value is c, but it doesn't really even achieve
103 * This allows some parallelism. Read-after-writes are good at doubling
104 * the number of bits affected, so the goal of mixing pulls in the opposite
105 * direction as the goal of parallelism. I did what I could. Rotates
106 * seem to cost as much as shifts on every machine I could lay my hands
107 * on, and rotates are much kinder to the top and bottom bits, so I used
112 a -= c; a ^= rot(c, 4); c += b; \
113 b -= a; b ^= rot(a, 6); a += c; \
114 c -= b; c ^= rot(b, 8); b += a; \
115 a -= c; a ^= rot(c,16); c += b; \
116 b -= a; b ^= rot(a,19); a += c; \
117 c -= b; c ^= rot(b, 4); b += a; \
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:
136 * 14 11 25 16 4 14 24
137 * 12 14 25 16 4 14 24
138 * and these came close:
143 #define final(a,b,c) \
145 c ^= b; c -= rot(b,14); \
146 a ^= c; a -= rot(c,11); \
147 b ^= a; b -= rot(a,25); \
148 c ^= b; c -= rot(b,16); \
149 a ^= c; a -= rot(c,4); \
150 b ^= a; b -= rot(a,14); \
151 c ^= b; c -= rot(b,24); \
155 * hashword2() -- same as hashword(), but take two seeds and return two 32-bit
156 * values. pc and pb must both be nonnull, and *pc and *pb must both be
157 * initialized with seeds. If you pass in (*pb)==0, the output (*pc) will be
158 * the same as the return value from hashword().
160 static void hashword2(const uint32_t *k
, size_t length
,
161 uint32_t *pc
, uint32_t *pb
)
165 /* Set up the internal state */
166 a
= b
= c
= 0xdeadbeef + ((uint32_t) (length
<< 2)) + *pc
;
186 case 0: /* case 0: nothing left to add */
195 * hashlittle() -- hash a variable-length key into a 32-bit value
196 * k : the key (the unaligned variable-length array of bytes)
197 * length : the length of the key, counting by bytes
198 * initval : can be any 4-byte value
199 * Returns a 32-bit value. Every bit of the key affects every bit of
200 * the return value. Two keys differing by one or two bits will have
201 * totally different hash values.
203 * The best hash table sizes are powers of 2. There is no need to do
204 * mod a prime (mod is sooo slow!). If you need less than 32 bits,
205 * use a bitmask. For example, if you need only 10 bits, do
206 * h = (h & hashmask(10));
207 * In which case, the hash table should have hashsize(10) elements.
209 * If you are hashing n strings (uint8_t **)k, do it like this:
210 * for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h);
212 * By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this
213 * code any way you wish, private, educational, or commercial. It's free.
215 * Use for hash table lookup, or anything where one collision in 2^^32 is
216 * acceptable. Do NOT use for cryptographic purposes.
219 static uint32_t hashlittle(const void *key
, size_t length
, uint32_t initval
)
225 } u
; /* needed for Mac Powerbook G4 */
227 /* Set up the internal state */
228 a
= b
= c
= 0xdeadbeef + ((uint32_t)length
) + initval
;
231 if (HASH_LITTLE_ENDIAN
&& ((u
.i
& 0x3) == 0)) {
232 const uint32_t *k
= (const uint32_t *)key
; /* read 32-bit chunks */
234 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
235 while (length
> 12) {
245 * "k[2]&0xffffff" actually reads beyond the end of the string, but
246 * then masks off the part it's not allowed to read. Because the
247 * string is aligned, the masked-off tail is in the same word as the
248 * rest of the string. Every machine with memory protection I've seen
249 * does it on word boundaries, so is OK with this. But VALGRIND will
250 * still catch it and complain. The masking trick does make the hash
251 * noticably faster for short strings (like English words).
256 case 12: c
+=k
[2]; b
+=k
[1]; a
+=k
[0]; break;
257 case 11: c
+=k
[2]&0xffffff; b
+=k
[1]; a
+=k
[0]; break;
258 case 10: c
+=k
[2]&0xffff; b
+=k
[1]; a
+=k
[0]; break;
259 case 9 : c
+=k
[2]&0xff; b
+=k
[1]; a
+=k
[0]; break;
260 case 8 : b
+=k
[1]; a
+=k
[0]; break;
261 case 7 : b
+=k
[1]&0xffffff; a
+=k
[0]; break;
262 case 6 : b
+=k
[1]&0xffff; a
+=k
[0]; break;
263 case 5 : b
+=k
[1]&0xff; a
+=k
[0]; break;
264 case 4 : a
+=k
[0]; break;
265 case 3 : a
+=k
[0]&0xffffff; break;
266 case 2 : a
+=k
[0]&0xffff; break;
267 case 1 : a
+=k
[0]&0xff; break;
268 case 0 : return c
; /* zero length strings require no mixing */
270 #else /* make valgrind happy */
273 k8
= (const uint8_t *)k
;
275 case 12: c
+=k
[2]; b
+=k
[1]; a
+=k
[0]; break;
276 case 11: c
+=((uint32_t)k8
[10])<<16; /* fall through */
277 case 10: c
+=((uint32_t)k8
[9])<<8; /* fall through */
278 case 9 : c
+=k8
[8]; /* fall through */
279 case 8 : b
+=k
[1]; a
+=k
[0]; break;
280 case 7 : b
+=((uint32_t)k8
[6])<<16; /* fall through */
281 case 6 : b
+=((uint32_t)k8
[5])<<8; /* fall through */
282 case 5 : b
+=k8
[4]; /* fall through */
283 case 4 : a
+=k
[0]; break;
284 case 3 : a
+=((uint32_t)k8
[2])<<16; /* fall through */
285 case 2 : a
+=((uint32_t)k8
[1])<<8; /* fall through */
286 case 1 : a
+=k8
[0]; break;
289 #endif /* !valgrind */
290 } else if (HASH_LITTLE_ENDIAN
&& ((u
.i
& 0x1) == 0)) {
291 const uint16_t *k
= (const uint16_t *)key
; /* read 16-bit chunks */
294 /*--------------- all but last block: aligned reads and different mixing */
295 while (length
> 12) {
296 a
+= k
[0] + (((uint32_t)k
[1])<<16);
297 b
+= k
[2] + (((uint32_t)k
[3])<<16);
298 c
+= k
[4] + (((uint32_t)k
[5])<<16);
304 k8
= (const uint8_t *)k
;
307 c
+=k
[4]+(((uint32_t)k
[5])<<16);
308 b
+=k
[2]+(((uint32_t)k
[3])<<16);
309 a
+=k
[0]+(((uint32_t)k
[1])<<16);
312 c
+=((uint32_t)k8
[10])<<16; /* fall through */
315 b
+=k
[2]+(((uint32_t)k
[3])<<16);
316 a
+=k
[0]+(((uint32_t)k
[1])<<16);
319 c
+=k8
[8]; /* fall through */
321 b
+=k
[2]+(((uint32_t)k
[3])<<16);
322 a
+=k
[0]+(((uint32_t)k
[1])<<16);
325 b
+=((uint32_t)k8
[6])<<16; /* fall through */
328 a
+=k
[0]+(((uint32_t)k
[1])<<16);
331 b
+=k8
[4]; /* fall through */
333 a
+=k
[0]+(((uint32_t)k
[1])<<16);
336 a
+=((uint32_t)k8
[2])<<16; /* fall through */
344 return c
; /* zero length requires no mixing */
347 } else { /* need to read the key one byte at a time */
348 const uint8_t *k
= (const uint8_t *)key
;
350 while (length
> 12) {
352 a
+= ((uint32_t)k
[1])<<8;
353 a
+= ((uint32_t)k
[2])<<16;
354 a
+= ((uint32_t)k
[3])<<24;
356 b
+= ((uint32_t)k
[5])<<8;
357 b
+= ((uint32_t)k
[6])<<16;
358 b
+= ((uint32_t)k
[7])<<24;
360 c
+= ((uint32_t)k
[9])<<8;
361 c
+= ((uint32_t)k
[10])<<16;
362 c
+= ((uint32_t)k
[11])<<24;
368 switch(length
) { /* all the case statements fall through */
369 case 12: c
+=((uint32_t)k
[11])<<24;
370 case 11: c
+=((uint32_t)k
[10])<<16;
371 case 10: c
+=((uint32_t)k
[9])<<8;
373 case 8: b
+=((uint32_t)k
[7])<<24;
374 case 7: b
+=((uint32_t)k
[6])<<16;
375 case 6: b
+=((uint32_t)k
[5])<<8;
377 case 4: a
+=((uint32_t)k
[3])<<24;
378 case 3: a
+=((uint32_t)k
[2])<<16;
379 case 2: a
+=((uint32_t)k
[1])<<8;
393 * Hash function for number value.
395 unsigned long hash_key(void *_key
, size_t length
, unsigned long seed
)
406 assert(length
== sizeof(unsigned long));
407 v
.v64
= (uint64_t) seed
;
408 key
.v64
= (uint64_t) _key
;
409 hashword2(key
.v32
, 2, &v
.v32
[0], &v
.v32
[1]);
414 * Hash function for string.
416 unsigned long hash_key_str(void *key
, size_t length
, unsigned long seed
)
418 return hashlittle(key
, length
, seed
);
422 * Hash function compare for number value.
424 unsigned long hash_compare_key(void *key1
, size_t key1_len
,
425 void *key2
, size_t key2_len
)
427 if (key1_len
!= key2_len
) {
439 * Hash compare function for string.
441 unsigned long hash_compare_key_str(void *key1
, size_t key1_len
,
442 void *key2
, size_t key2_len
)
444 if (key1_len
!= key2_len
) {
448 if (strncmp(key1
, key2
, key1_len
) == 0) {
This page took 0.039446 seconds and 4 git commands to generate.