2 * Copyright (C) - Bob Jenkins, May 2006
3 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
4 * Copyright (C) 2011 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
21 * These are functions for producing 32-bit hashes for hash table lookup.
22 * hashword(), hashlittle(), hashlittle2(), hashbig(), mix(), and final() are
23 * externally useful functions. Routines to test the hash are included if
24 * SELF_TEST is defined. You can use this free for any purpose. It's in the
25 * public domain. It has no warranty.
27 * You probably want to use hashlittle(). hashlittle() and hashbig() hash byte
28 * arrays. hashlittle() is is faster than hashbig() on little-endian machines.
29 * Intel and AMD are little-endian machines. On second thought, you probably
30 * want hashlittle2(), which is identical to hashlittle() except it returns two
31 * 32-bit hashes for the price of one. You could implement hashbig2() if you
32 * wanted but I haven't bothered here.
34 * If you want to find a hash of, say, exactly 7 integers, do
35 * a = i1; b = i2; c = i3;
37 * a += i4; b += i5; c += i6;
41 * then use c as the hash value. If you have a variable length array of
42 * 4-byte integers to hash, use hashword(). If you have a byte array (like
43 * a character string), use hashlittle(). If you have several byte arrays, or
44 * a mix of things, see the comments above hashlittle().
46 * Why is this so big? I read 12 bytes at a time into 3 4-byte integers, then
47 * mix those integers. This is fast (you can do a lot more thorough mixing
48 * with 12*3 instructions on 3 integers than you can with 3 instructions on 1
49 * byte), but shoehorning those bytes into integers efficiently is messy.
54 #include <stdint.h> /* defines uint32_t etc */
55 #include <stdio.h> /* defines printf for tests */
57 #include <sys/param.h> /* attempt to define endianness */
58 #include <time.h> /* defines time_t for timings in the test */
59 #include <urcu/compiler.h>
62 #include <common/compat/endian.h> /* attempt to define endianness */
63 #include <common/common.h>
64 #include <common/hashtable/hashtable.h>
67 * My best guess at if you are big-endian or little-endian. This may
70 #if (defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && \
71 __BYTE_ORDER == __LITTLE_ENDIAN) || \
72 (defined(i386) || defined(__i386__) || defined(__i486__) || \
73 defined(__i586__) || defined(__i686__) || defined(vax) || defined(MIPSEL))
74 # define HASH_LITTLE_ENDIAN 1
75 # define HASH_BIG_ENDIAN 0
76 #elif (defined(__BYTE_ORDER) && defined(__BIG_ENDIAN) && \
77 __BYTE_ORDER == __BIG_ENDIAN) || \
78 (defined(sparc) || defined(POWERPC) || defined(mc68000) || defined(sel))
79 # define HASH_LITTLE_ENDIAN 0
80 # define HASH_BIG_ENDIAN 1
82 # define HASH_LITTLE_ENDIAN 0
83 # define HASH_BIG_ENDIAN 0
86 #define hashsize(n) ((uint32_t)1<<(n))
87 #define hashmask(n) (hashsize(n)-1)
88 #define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
91 * mix -- mix 3 32-bit values reversibly.
93 * This is reversible, so any information in (a,b,c) before mix() is
94 * still in (a,b,c) after mix().
96 * If four pairs of (a,b,c) inputs are run through mix(), or through
97 * mix() in reverse, there are at least 32 bits of the output that
98 * are sometimes the same for one pair and different for another pair.
99 * This was tested for:
100 * * pairs that differed by one bit, by two bits, in any combination
101 * of top bits of (a,b,c), or in any combination of bottom bits of
103 * * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
104 * the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
105 * is commonly produced by subtraction) look like a single 1-bit
107 * * the base values were pseudorandom, all zero but one bit set, or
108 * all zero plus a counter that starts at zero.
110 * Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
115 * Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
116 * for "differ" defined as + with a one-bit base and a two-bit delta. I
117 * used http://burtleburtle.net/bob/hash/avalanche.html to choose
118 * the operations, constants, and arrangements of the variables.
120 * This does not achieve avalanche. There are input bits of (a,b,c)
121 * that fail to affect some output bits of (a,b,c), especially of a. The
122 * most thoroughly mixed value is c, but it doesn't really even achieve
125 * This allows some parallelism. Read-after-writes are good at doubling
126 * the number of bits affected, so the goal of mixing pulls in the opposite
127 * direction as the goal of parallelism. I did what I could. Rotates
128 * seem to cost as much as shifts on every machine I could lay my hands
129 * on, and rotates are much kinder to the top and bottom bits, so I used
134 a -= c; a ^= rot(c, 4); c += b; \
135 b -= a; b ^= rot(a, 6); a += c; \
136 c -= b; c ^= rot(b, 8); b += a; \
137 a -= c; a ^= rot(c,16); c += b; \
138 b -= a; b ^= rot(a,19); a += c; \
139 c -= b; c ^= rot(b, 4); b += a; \
143 * final -- final mixing of 3 32-bit values (a,b,c) into c
145 * Pairs of (a,b,c) values differing in only a few bits will usually
146 * produce values of c that look totally different. This was tested for
147 * * pairs that differed by one bit, by two bits, in any combination
148 * of top bits of (a,b,c), or in any combination of bottom bits of
150 * * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
151 * the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
152 * is commonly produced by subtraction) look like a single 1-bit
154 * * the base values were pseudorandom, all zero but one bit set, or
155 * all zero plus a counter that starts at zero.
157 * These constants passed:
158 * 14 11 25 16 4 14 24
159 * 12 14 25 16 4 14 24
160 * and these came close:
165 #define final(a,b,c) \
167 c ^= b; c -= rot(b,14); \
168 a ^= c; a -= rot(c,11); \
169 b ^= a; b -= rot(a,25); \
170 c ^= b; c -= rot(b,16); \
171 a ^= c; a -= rot(c,4); \
172 b ^= a; b -= rot(a,14); \
173 c ^= b; c -= rot(b,24); \
177 * k - the key, an array of uint32_t values
178 * length - the length of the key, in uint32_ts
179 * initval - the previous hash, or an arbitrary value
181 static uint32_t __attribute__((unused
)) hashword(const uint32_t *k
,
182 size_t length
, uint32_t initval
)
186 /* Set up the internal state */
187 a
= b
= c
= 0xdeadbeef + (((uint32_t) length
) << 2) + initval
;
189 /*----------------------------------------- handle most of the key */
199 /*----------------------------------- handle the last 3 uint32_t's */
200 switch (length
) { /* all the case statements fall through */
205 case 0: /* case 0: nothing left to add */
208 /*---------------------------------------------- report the result */
214 * hashword2() -- same as hashword(), but take two seeds and return two 32-bit
215 * values. pc and pb must both be nonnull, and *pc and *pb must both be
216 * initialized with seeds. If you pass in (*pb)==0, the output (*pc) will be
217 * the same as the return value from hashword().
219 static void __attribute__((unused
)) hashword2(const uint32_t *k
, size_t length
,
220 uint32_t *pc
, uint32_t *pb
)
224 /* Set up the internal state */
225 a
= b
= c
= 0xdeadbeef + ((uint32_t) (length
<< 2)) + *pc
;
245 case 0: /* case 0: nothing left to add */
254 * hashlittle() -- hash a variable-length key into a 32-bit value
255 * k : the key (the unaligned variable-length array of bytes)
256 * length : the length of the key, counting by bytes
257 * initval : can be any 4-byte value
258 * Returns a 32-bit value. Every bit of the key affects every bit of
259 * the return value. Two keys differing by one or two bits will have
260 * totally different hash values.
262 * The best hash table sizes are powers of 2. There is no need to do
263 * mod a prime (mod is sooo slow!). If you need less than 32 bits,
264 * use a bitmask. For example, if you need only 10 bits, do
265 * h = (h & hashmask(10));
266 * In which case, the hash table should have hashsize(10) elements.
268 * If you are hashing n strings (uint8_t **)k, do it like this:
269 * for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h);
271 * By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this
272 * code any way you wish, private, educational, or commercial. It's free.
274 * Use for hash table lookup, or anything where one collision in 2^^32 is
275 * acceptable. Do NOT use for cryptographic purposes.
277 static uint32_t __attribute__((unused
)) hashlittle(const void *key
,
278 size_t length
, uint32_t initval
)
284 } u
; /* needed for Mac Powerbook G4 */
286 /* Set up the internal state */
287 a
= b
= c
= 0xdeadbeef + ((uint32_t)length
) + initval
;
290 if (HASH_LITTLE_ENDIAN
&& ((u
.i
& 0x3) == 0)) {
291 const uint32_t *k
= (const uint32_t *)key
; /* read 32-bit chunks */
293 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
294 while (length
> 12) {
304 * "k[2]&0xffffff" actually reads beyond the end of the string, but
305 * then masks off the part it's not allowed to read. Because the
306 * string is aligned, the masked-off tail is in the same word as the
307 * rest of the string. Every machine with memory protection I've seen
308 * does it on word boundaries, so is OK with this. But VALGRIND will
309 * still catch it and complain. The masking trick does make the hash
310 * noticably faster for short strings (like English words).
315 case 12: c
+=k
[2]; b
+=k
[1]; a
+=k
[0]; break;
316 case 11: c
+=k
[2]&0xffffff; b
+=k
[1]; a
+=k
[0]; break;
317 case 10: c
+=k
[2]&0xffff; b
+=k
[1]; a
+=k
[0]; break;
318 case 9 : c
+=k
[2]&0xff; b
+=k
[1]; a
+=k
[0]; break;
319 case 8 : b
+=k
[1]; a
+=k
[0]; break;
320 case 7 : b
+=k
[1]&0xffffff; a
+=k
[0]; break;
321 case 6 : b
+=k
[1]&0xffff; a
+=k
[0]; break;
322 case 5 : b
+=k
[1]&0xff; a
+=k
[0]; break;
323 case 4 : a
+=k
[0]; break;
324 case 3 : a
+=k
[0]&0xffffff; break;
325 case 2 : a
+=k
[0]&0xffff; break;
326 case 1 : a
+=k
[0]&0xff; break;
327 case 0 : return c
; /* zero length strings require no mixing */
329 #else /* make valgrind happy */
332 k8
= (const uint8_t *)k
;
334 case 12: c
+=k
[2]; b
+=k
[1]; a
+=k
[0]; break;
335 case 11: c
+=((uint32_t)k8
[10])<<16; /* fall through */
336 case 10: c
+=((uint32_t)k8
[9])<<8; /* fall through */
337 case 9 : c
+=k8
[8]; /* fall through */
338 case 8 : b
+=k
[1]; a
+=k
[0]; break;
339 case 7 : b
+=((uint32_t)k8
[6])<<16; /* fall through */
340 case 6 : b
+=((uint32_t)k8
[5])<<8; /* fall through */
341 case 5 : b
+=k8
[4]; /* fall through */
342 case 4 : a
+=k
[0]; break;
343 case 3 : a
+=((uint32_t)k8
[2])<<16; /* fall through */
344 case 2 : a
+=((uint32_t)k8
[1])<<8; /* fall through */
345 case 1 : a
+=k8
[0]; break;
348 #endif /* !valgrind */
349 } else if (HASH_LITTLE_ENDIAN
&& ((u
.i
& 0x1) == 0)) {
350 const uint16_t *k
= (const uint16_t *)key
; /* read 16-bit chunks */
353 /*--------------- all but last block: aligned reads and different mixing */
354 while (length
> 12) {
355 a
+= k
[0] + (((uint32_t)k
[1])<<16);
356 b
+= k
[2] + (((uint32_t)k
[3])<<16);
357 c
+= k
[4] + (((uint32_t)k
[5])<<16);
363 k8
= (const uint8_t *)k
;
366 c
+=k
[4]+(((uint32_t)k
[5])<<16);
367 b
+=k
[2]+(((uint32_t)k
[3])<<16);
368 a
+=k
[0]+(((uint32_t)k
[1])<<16);
371 c
+=((uint32_t)k8
[10])<<16; /* fall through */
374 b
+=k
[2]+(((uint32_t)k
[3])<<16);
375 a
+=k
[0]+(((uint32_t)k
[1])<<16);
378 c
+=k8
[8]; /* fall through */
380 b
+=k
[2]+(((uint32_t)k
[3])<<16);
381 a
+=k
[0]+(((uint32_t)k
[1])<<16);
384 b
+=((uint32_t)k8
[6])<<16; /* fall through */
387 a
+=k
[0]+(((uint32_t)k
[1])<<16);
390 b
+=k8
[4]; /* fall through */
392 a
+=k
[0]+(((uint32_t)k
[1])<<16);
395 a
+=((uint32_t)k8
[2])<<16; /* fall through */
403 return c
; /* zero length requires no mixing */
406 } else { /* need to read the key one byte at a time */
407 const uint8_t *k
= (const uint8_t *)key
;
409 while (length
> 12) {
411 a
+= ((uint32_t)k
[1])<<8;
412 a
+= ((uint32_t)k
[2])<<16;
413 a
+= ((uint32_t)k
[3])<<24;
415 b
+= ((uint32_t)k
[5])<<8;
416 b
+= ((uint32_t)k
[6])<<16;
417 b
+= ((uint32_t)k
[7])<<24;
419 c
+= ((uint32_t)k
[9])<<8;
420 c
+= ((uint32_t)k
[10])<<16;
421 c
+= ((uint32_t)k
[11])<<24;
427 switch(length
) { /* all the case statements fall through */
428 case 12: c
+=((uint32_t)k
[11])<<24;
429 case 11: c
+=((uint32_t)k
[10])<<16;
430 case 10: c
+=((uint32_t)k
[9])<<8;
432 case 8: b
+=((uint32_t)k
[7])<<24;
433 case 7: b
+=((uint32_t)k
[6])<<16;
434 case 6: b
+=((uint32_t)k
[5])<<8;
436 case 4: a
+=((uint32_t)k
[3])<<24;
437 case 3: a
+=((uint32_t)k
[2])<<16;
438 case 2: a
+=((uint32_t)k
[1])<<8;
452 unsigned long hash_key_u64(void *_key
, unsigned long seed
)
463 v
.v64
= (uint64_t) seed
;
464 key
.v64
= *(uint64_t *) _key
;
465 hashword2(key
.v32
, 2, &v
.v32
[0], &v
.v32
[1]);
469 #if (CAA_BITS_PER_LONG == 64)
471 * Hash function for number value.
474 unsigned long hash_key_ulong(void *_key
, unsigned long seed
)
476 uint64_t __key
= (uint64_t) _key
;
477 return (unsigned long) hash_key_u64(&__key
, seed
);
481 * Hash function for number value.
484 unsigned long hash_key_ulong(void *_key
, unsigned long seed
)
486 uint32_t key
= (uint32_t) _key
;
488 return hashword(&key
, 1, seed
);
490 #endif /* CAA_BITS_PER_LONG */
493 * Hash function for string.
496 unsigned long hash_key_str(void *key
, unsigned long seed
)
498 return hashlittle(key
, strlen((char *) key
), seed
);
502 * Hash function for two uint64_t.
505 unsigned long hash_key_two_u64(void *key
, unsigned long seed
)
507 struct lttng_ht_two_u64
*k
= (struct lttng_ht_two_u64
*) key
;
509 return hash_key_u64(&k
->key1
, seed
) ^ hash_key_u64(&k
->key2
, seed
);
513 * Hash function compare for number value.
516 int hash_match_key_ulong(void *key1
, void *key2
)
526 * Hash function compare for number value.
529 int hash_match_key_u64(void *key1
, void *key2
)
531 if (*(uint64_t *) key1
== *(uint64_t *) key2
) {
539 * Hash compare function for string.
542 int hash_match_key_str(void *key1
, void *key2
)
544 if (strcmp(key1
, key2
) == 0) {
552 * Hash function compare two uint64_t.
555 int hash_match_key_two_u64(void *key1
, void *key2
)
557 struct lttng_ht_two_u64
*k1
= (struct lttng_ht_two_u64
*) key1
;
558 struct lttng_ht_two_u64
*k2
= (struct lttng_ht_two_u64
*) key2
;
560 if (hash_match_key_u64(&k1
->key1
, &k2
->key1
) &&
561 hash_match_key_u64(&k1
->key2
, &k2
->key2
)) {
This page took 0.041783 seconds and 4 git commands to generate.