2 -------------------------------------------------------------------------------
3 lookup3.c, by Bob Jenkins, May 2006, Public Domain.
5 These are functions for producing 32-bit hashes for hash table lookup.
6 hashword(), hashlittle(), hashlittle2(), hashbig(), mix(), and final()
7 are externally useful functions. Routines to test the hash are included
8 if SELF_TEST is defined. You can use this free for any purpose. It's in
9 the public domain. It has no warranty.
11 You probably want to use hashlittle(). hashlittle() and hashbig()
12 hash byte arrays. hashlittle() is is faster than hashbig() on
13 little-endian machines. Intel and AMD are little-endian machines.
14 On second thought, you probably want hashlittle2(), which is identical to
15 hashlittle() except it returns two 32-bit hashes for the price of one.
16 You could implement hashbig2() if you 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,
31 then mix those integers. This is fast (you can do a lot more thorough
32 mixing with 12*3 instructions on 3 integers than you can with 3 instructions
33 on 1 byte), but shoehorning those bytes into integers efficiently is messy.
34 -------------------------------------------------------------------------------
38 #include <stdio.h> /* defines printf for tests */
39 #include <time.h> /* defines time_t for timings in the test */
40 #include <stdint.h> /* defines uint32_t etc */
41 #include <sys/param.h> /* attempt to define endianness */
43 # include <endian.h> /* attempt to define endianness */
47 * My best guess at if you are big-endian or little-endian. This may
50 #if (defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && \
51 __BYTE_ORDER == __LITTLE_ENDIAN) || \
52 (defined(i386) || defined(__i386__) || defined(__i486__) || \
53 defined(__i586__) || defined(__i686__) || defined(vax) || defined(MIPSEL))
54 # define HASH_LITTLE_ENDIAN 1
55 # define HASH_BIG_ENDIAN 0
56 #elif (defined(__BYTE_ORDER) && defined(__BIG_ENDIAN) && \
57 __BYTE_ORDER == __BIG_ENDIAN) || \
58 (defined(sparc) || defined(POWERPC) || defined(mc68000) || defined(sel))
59 # define HASH_LITTLE_ENDIAN 0
60 # define HASH_BIG_ENDIAN 1
62 # define HASH_LITTLE_ENDIAN 0
63 # define HASH_BIG_ENDIAN 0
66 #define hashsize(n) ((uint32_t)1<<(n))
67 #define hashmask(n) (hashsize(n)-1)
68 #define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
71 -------------------------------------------------------------------------------
72 mix -- mix 3 32-bit values reversibly.
74 This is reversible, so any information in (a,b,c) before mix() is
75 still in (a,b,c) after mix().
77 If four pairs of (a,b,c) inputs are run through mix(), or through
78 mix() in reverse, there are at least 32 bits of the output that
79 are sometimes the same for one pair and different for another pair.
81 * pairs that differed by one bit, by two bits, in any combination
82 of top bits of (a,b,c), or in any combination of bottom bits of
84 * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
85 the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
86 is commonly produced by subtraction) look like a single 1-bit
88 * the base values were pseudorandom, all zero but one bit set, or
89 all zero plus a counter that starts at zero.
91 Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
96 Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
97 for "differ" defined as + with a one-bit base and a two-bit delta. I
98 used http://burtleburtle.net/bob/hash/avalanche.html to choose
99 the operations, constants, and arrangements of the variables.
101 This does not achieve avalanche. There are input bits of (a,b,c)
102 that fail to affect some output bits of (a,b,c), especially of a. The
103 most thoroughly mixed value is c, but it doesn't really even achieve
106 This allows some parallelism. Read-after-writes are good at doubling
107 the number of bits affected, so the goal of mixing pulls in the opposite
108 direction as the goal of parallelism. I did what I could. Rotates
109 seem to cost as much as shifts on every machine I could lay my hands
110 on, and rotates are much kinder to the top and bottom bits, so I used
112 -------------------------------------------------------------------------------
116 a -= c; a ^= rot(c, 4); c += b; \
117 b -= a; b ^= rot(a, 6); a += c; \
118 c -= b; c ^= rot(b, 8); b += a; \
119 a -= c; a ^= rot(c,16); c += b; \
120 b -= a; b ^= rot(a,19); a += c; \
121 c -= b; c ^= rot(b, 4); b += a; \
125 -------------------------------------------------------------------------------
126 final -- final mixing of 3 32-bit values (a,b,c) into c
128 Pairs of (a,b,c) values differing in only a few bits will usually
129 produce values of c that look totally different. This was tested for
130 * pairs that differed by one bit, by two bits, in any combination
131 of top bits of (a,b,c), or in any combination of bottom bits of
133 * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
134 the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
135 is commonly produced by subtraction) look like a single 1-bit
137 * the base values were pseudorandom, all zero but one bit set, or
138 all zero plus a counter that starts at zero.
140 These constants passed:
143 and these came close:
147 -------------------------------------------------------------------------------
149 #define final(a,b,c) \
151 c ^= b; c -= rot(b,14); \
152 a ^= c; a -= rot(c,11); \
153 b ^= a; b -= rot(a,25); \
154 c ^= b; c -= rot(b,16); \
155 a ^= c; a -= rot(c,4); \
156 b ^= a; b -= rot(a,14); \
157 c ^= b; c -= rot(b,24); \
161 --------------------------------------------------------------------
162 This works on all machines. To be useful, it requires
163 -- that the key be an array of uint32_t's, and
164 -- that the length be the number of uint32_t's in the key
166 The function hashword() is identical to hashlittle() on little-endian
167 machines, and identical to hashbig() on big-endian machines,
168 except that the length has to be measured in uint32_ts rather than in
169 bytes. hashlittle() is more complicated than hashword() only because
170 hashlittle() has to dance around fitting the key bytes into registers.
171 --------------------------------------------------------------------
174 const uint32_t *k
, /* the key, an array of uint32_t values */
175 size_t length
, /* the length of the key, in uint32_ts */
176 uint32_t initval
) /* the previous hash, or an arbitrary value */
180 /* Set up the internal state */
181 a
= b
= c
= 0xdeadbeef + (((uint32_t)length
)<<2) + initval
;
183 /*------------------------------------------------- handle most of the key */
194 /*------------------------------------------- handle the last 3 uint32_t's */
195 switch(length
) /* all the case statements fall through */
201 case 0: /* case 0: nothing left to add */
204 /*------------------------------------------------------ report the result */
210 --------------------------------------------------------------------
211 hashword2() -- same as hashword(), but take two seeds and return two
212 32-bit values. pc and pb must both be nonnull, and *pc and *pb must
213 both be initialized with seeds. If you pass in (*pb)==0, the output
214 (*pc) will be the same as the return value from hashword().
215 --------------------------------------------------------------------
218 const uint32_t *k
, /* the key, an array of uint32_t values */
219 size_t length
, /* the length of the key, in uint32_ts */
220 uint32_t *pc
, /* IN: seed OUT: primary hash value */
221 uint32_t *pb
) /* IN: more seed OUT: secondary hash value */
225 /* Set up the internal state */
226 a
= b
= c
= 0xdeadbeef + ((uint32_t)(length
<<2)) + *pc
;
229 /*------------------------------------------------- handle most of the key */
240 /*------------------------------------------- handle the last 3 uint32_t's */
241 switch(length
) /* all the case statements fall through */
247 case 0: /* case 0: nothing left to add */
250 /*------------------------------------------------------ report the result */
256 -------------------------------------------------------------------------------
257 hashlittle() -- hash a variable-length key into a 32-bit value
258 k : the key (the unaligned variable-length array of bytes)
259 length : the length of the key, counting by bytes
260 initval : can be any 4-byte value
261 Returns a 32-bit value. Every bit of the key affects every bit of
262 the return value. Two keys differing by one or two bits will have
263 totally different hash values.
265 The best hash table sizes are powers of 2. There is no need to do
266 mod a prime (mod is sooo slow!). If you need less than 32 bits,
267 use a bitmask. For example, if you need only 10 bits, do
268 h = (h & hashmask(10));
269 In which case, the hash table should have hashsize(10) elements.
271 If you are hashing n strings (uint8_t **)k, do it like this:
272 for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h);
274 By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this
275 code any way you wish, private, educational, or commercial. It's free.
277 Use for hash table lookup, or anything where one collision in 2^^32 is
278 acceptable. Do NOT use for cryptographic purposes.
279 -------------------------------------------------------------------------------
282 uint32_t hashlittle( const void *key
, size_t length
, uint32_t initval
)
284 uint32_t a
,b
,c
; /* internal state */
285 union { const void *ptr
; size_t i
; } u
; /* needed for Mac Powerbook G4 */
287 /* Set up the internal state */
288 a
= b
= c
= 0xdeadbeef + ((uint32_t)length
) + initval
;
291 if (HASH_LITTLE_ENDIAN
&& ((u
.i
& 0x3) == 0)) {
292 const uint32_t *k
= (const uint32_t *)key
; /* read 32-bit chunks */
295 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
306 /*----------------------------- handle the last (probably partial) block */
308 * "k[2]&0xffffff" actually reads beyond the end of the string, but
309 * then masks off the part it's not allowed to read. Because the
310 * string is aligned, the masked-off tail is in the same word as the
311 * rest of the string. Every machine with memory protection I've seen
312 * does it on word boundaries, so is OK with this. But VALGRIND will
313 * still catch it and complain. The masking trick does make the hash
314 * noticably faster for short strings (like English words).
320 case 12: c
+=k
[2]; b
+=k
[1]; a
+=k
[0]; break;
321 case 11: c
+=k
[2]&0xffffff; b
+=k
[1]; a
+=k
[0]; break;
322 case 10: c
+=k
[2]&0xffff; b
+=k
[1]; a
+=k
[0]; break;
323 case 9 : c
+=k
[2]&0xff; b
+=k
[1]; a
+=k
[0]; break;
324 case 8 : b
+=k
[1]; a
+=k
[0]; break;
325 case 7 : b
+=k
[1]&0xffffff; a
+=k
[0]; break;
326 case 6 : b
+=k
[1]&0xffff; a
+=k
[0]; break;
327 case 5 : b
+=k
[1]&0xff; a
+=k
[0]; break;
328 case 4 : a
+=k
[0]; break;
329 case 3 : a
+=k
[0]&0xffffff; break;
330 case 2 : a
+=k
[0]&0xffff; break;
331 case 1 : a
+=k
[0]&0xff; break;
332 case 0 : return c
; /* zero length strings require no mixing */
335 #else /* make valgrind happy */
337 k8
= (const uint8_t *)k
;
340 case 12: c
+=k
[2]; b
+=k
[1]; a
+=k
[0]; break;
341 case 11: c
+=((uint32_t)k8
[10])<<16; /* fall through */
342 case 10: c
+=((uint32_t)k8
[9])<<8; /* fall through */
343 case 9 : c
+=k8
[8]; /* fall through */
344 case 8 : b
+=k
[1]; a
+=k
[0]; break;
345 case 7 : b
+=((uint32_t)k8
[6])<<16; /* fall through */
346 case 6 : b
+=((uint32_t)k8
[5])<<8; /* fall through */
347 case 5 : b
+=k8
[4]; /* fall through */
348 case 4 : a
+=k
[0]; break;
349 case 3 : a
+=((uint32_t)k8
[2])<<16; /* fall through */
350 case 2 : a
+=((uint32_t)k8
[1])<<8; /* fall through */
351 case 1 : a
+=k8
[0]; break;
355 #endif /* !valgrind */
357 } else if (HASH_LITTLE_ENDIAN
&& ((u
.i
& 0x1) == 0)) {
358 const uint16_t *k
= (const uint16_t *)key
; /* read 16-bit chunks */
361 /*--------------- all but last block: aligned reads and different mixing */
364 a
+= k
[0] + (((uint32_t)k
[1])<<16);
365 b
+= k
[2] + (((uint32_t)k
[3])<<16);
366 c
+= k
[4] + (((uint32_t)k
[5])<<16);
372 /*----------------------------- handle the last (probably partial) block */
373 k8
= (const uint8_t *)k
;
376 case 12: c
+=k
[4]+(((uint32_t)k
[5])<<16);
377 b
+=k
[2]+(((uint32_t)k
[3])<<16);
378 a
+=k
[0]+(((uint32_t)k
[1])<<16);
380 case 11: c
+=((uint32_t)k8
[10])<<16; /* fall through */
382 b
+=k
[2]+(((uint32_t)k
[3])<<16);
383 a
+=k
[0]+(((uint32_t)k
[1])<<16);
385 case 9 : c
+=k8
[8]; /* fall through */
386 case 8 : b
+=k
[2]+(((uint32_t)k
[3])<<16);
387 a
+=k
[0]+(((uint32_t)k
[1])<<16);
389 case 7 : b
+=((uint32_t)k8
[6])<<16; /* fall through */
391 a
+=k
[0]+(((uint32_t)k
[1])<<16);
393 case 5 : b
+=k8
[4]; /* fall through */
394 case 4 : a
+=k
[0]+(((uint32_t)k
[1])<<16);
396 case 3 : a
+=((uint32_t)k8
[2])<<16; /* fall through */
401 case 0 : return c
; /* zero length requires no mixing */
404 } else { /* need to read the key one byte at a time */
405 const uint8_t *k
= (const uint8_t *)key
;
407 /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
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 /*-------------------------------- last block: affect all 32 bits of (c) */
428 switch(length
) /* all the case statements fall through */
430 case 12: c
+=((uint32_t)k
[11])<<24;
431 case 11: c
+=((uint32_t)k
[10])<<16;
432 case 10: c
+=((uint32_t)k
[9])<<8;
434 case 8 : b
+=((uint32_t)k
[7])<<24;
435 case 7 : b
+=((uint32_t)k
[6])<<16;
436 case 6 : b
+=((uint32_t)k
[5])<<8;
438 case 4 : a
+=((uint32_t)k
[3])<<24;
439 case 3 : a
+=((uint32_t)k
[2])<<16;
440 case 2 : a
+=((uint32_t)k
[1])<<8;
453 * hashlittle2: return 2 32-bit hash values
455 * This is identical to hashlittle(), except it returns two 32-bit hash
456 * values instead of just one. This is good enough for hash table
457 * lookup with 2^^64 buckets, or if you want a second hash if you're not
458 * happy with the first, or if you want a probably-unique 64-bit ID for
459 * the key. *pc is better mixed than *pb, so use *pc first. If you want
460 * a 64-bit value do something like "*pc + (((uint64_t)*pb)<<32)".
463 const void *key
, /* the key to hash */
464 size_t length
, /* length of the key */
465 uint32_t *pc
, /* IN: primary initval, OUT: primary hash */
466 uint32_t *pb
) /* IN: secondary initval, OUT: secondary hash */
468 uint32_t a
,b
,c
; /* internal state */
469 union { const void *ptr
; size_t i
; } u
; /* needed for Mac Powerbook G4 */
471 /* Set up the internal state */
472 a
= b
= c
= 0xdeadbeef + ((uint32_t)length
) + *pc
;
476 if (HASH_LITTLE_ENDIAN
&& ((u
.i
& 0x3) == 0)) {
477 const uint32_t *k
= (const uint32_t *)key
; /* read 32-bit chunks */
480 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
491 /*----------------------------- handle the last (probably partial) block */
493 * "k[2]&0xffffff" actually reads beyond the end of the string, but
494 * then masks off the part it's not allowed to read. Because the
495 * string is aligned, the masked-off tail is in the same word as the
496 * rest of the string. Every machine with memory protection I've seen
497 * does it on word boundaries, so is OK with this. But VALGRIND will
498 * still catch it and complain. The masking trick does make the hash
499 * noticably faster for short strings (like English words).
505 case 12: c
+=k
[2]; b
+=k
[1]; a
+=k
[0]; break;
506 case 11: c
+=k
[2]&0xffffff; b
+=k
[1]; a
+=k
[0]; break;
507 case 10: c
+=k
[2]&0xffff; b
+=k
[1]; a
+=k
[0]; break;
508 case 9 : c
+=k
[2]&0xff; b
+=k
[1]; a
+=k
[0]; break;
509 case 8 : b
+=k
[1]; a
+=k
[0]; break;
510 case 7 : b
+=k
[1]&0xffffff; a
+=k
[0]; break;
511 case 6 : b
+=k
[1]&0xffff; a
+=k
[0]; break;
512 case 5 : b
+=k
[1]&0xff; a
+=k
[0]; break;
513 case 4 : a
+=k
[0]; break;
514 case 3 : a
+=k
[0]&0xffffff; break;
515 case 2 : a
+=k
[0]&0xffff; break;
516 case 1 : a
+=k
[0]&0xff; break;
517 case 0 : *pc
=c
; *pb
=b
; return; /* zero length strings require no mixing */
520 #else /* make valgrind happy */
522 k8
= (const uint8_t *)k
;
525 case 12: c
+=k
[2]; b
+=k
[1]; a
+=k
[0]; break;
526 case 11: c
+=((uint32_t)k8
[10])<<16; /* fall through */
527 case 10: c
+=((uint32_t)k8
[9])<<8; /* fall through */
528 case 9 : c
+=k8
[8]; /* fall through */
529 case 8 : b
+=k
[1]; a
+=k
[0]; break;
530 case 7 : b
+=((uint32_t)k8
[6])<<16; /* fall through */
531 case 6 : b
+=((uint32_t)k8
[5])<<8; /* fall through */
532 case 5 : b
+=k8
[4]; /* fall through */
533 case 4 : a
+=k
[0]; break;
534 case 3 : a
+=((uint32_t)k8
[2])<<16; /* fall through */
535 case 2 : a
+=((uint32_t)k8
[1])<<8; /* fall through */
536 case 1 : a
+=k8
[0]; break;
537 case 0 : *pc
=c
; *pb
=b
; return; /* zero length strings require no mixing */
540 #endif /* !valgrind */
542 } else if (HASH_LITTLE_ENDIAN
&& ((u
.i
& 0x1) == 0)) {
543 const uint16_t *k
= (const uint16_t *)key
; /* read 16-bit chunks */
546 /*--------------- all but last block: aligned reads and different mixing */
549 a
+= k
[0] + (((uint32_t)k
[1])<<16);
550 b
+= k
[2] + (((uint32_t)k
[3])<<16);
551 c
+= k
[4] + (((uint32_t)k
[5])<<16);
557 /*----------------------------- handle the last (probably partial) block */
558 k8
= (const uint8_t *)k
;
561 case 12: c
+=k
[4]+(((uint32_t)k
[5])<<16);
562 b
+=k
[2]+(((uint32_t)k
[3])<<16);
563 a
+=k
[0]+(((uint32_t)k
[1])<<16);
565 case 11: c
+=((uint32_t)k8
[10])<<16; /* fall through */
567 b
+=k
[2]+(((uint32_t)k
[3])<<16);
568 a
+=k
[0]+(((uint32_t)k
[1])<<16);
570 case 9 : c
+=k8
[8]; /* fall through */
571 case 8 : b
+=k
[2]+(((uint32_t)k
[3])<<16);
572 a
+=k
[0]+(((uint32_t)k
[1])<<16);
574 case 7 : b
+=((uint32_t)k8
[6])<<16; /* fall through */
576 a
+=k
[0]+(((uint32_t)k
[1])<<16);
578 case 5 : b
+=k8
[4]; /* fall through */
579 case 4 : a
+=k
[0]+(((uint32_t)k
[1])<<16);
581 case 3 : a
+=((uint32_t)k8
[2])<<16; /* fall through */
586 case 0 : *pc
=c
; *pb
=b
; return; /* zero length strings require no mixing */
589 } else { /* need to read the key one byte at a time */
590 const uint8_t *k
= (const uint8_t *)key
;
592 /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
596 a
+= ((uint32_t)k
[1])<<8;
597 a
+= ((uint32_t)k
[2])<<16;
598 a
+= ((uint32_t)k
[3])<<24;
600 b
+= ((uint32_t)k
[5])<<8;
601 b
+= ((uint32_t)k
[6])<<16;
602 b
+= ((uint32_t)k
[7])<<24;
604 c
+= ((uint32_t)k
[9])<<8;
605 c
+= ((uint32_t)k
[10])<<16;
606 c
+= ((uint32_t)k
[11])<<24;
612 /*-------------------------------- last block: affect all 32 bits of (c) */
613 switch(length
) /* all the case statements fall through */
615 case 12: c
+=((uint32_t)k
[11])<<24;
616 case 11: c
+=((uint32_t)k
[10])<<16;
617 case 10: c
+=((uint32_t)k
[9])<<8;
619 case 8 : b
+=((uint32_t)k
[7])<<24;
620 case 7 : b
+=((uint32_t)k
[6])<<16;
621 case 6 : b
+=((uint32_t)k
[5])<<8;
623 case 4 : a
+=((uint32_t)k
[3])<<24;
624 case 3 : a
+=((uint32_t)k
[2])<<16;
625 case 2 : a
+=((uint32_t)k
[1])<<8;
628 case 0 : *pc
=c
; *pb
=b
; return; /* zero length strings require no mixing */
640 * This is the same as hashword() on big-endian machines. It is different
641 * from hashlittle() on all machines. hashbig() takes advantage of
642 * big-endian byte ordering.
644 uint32_t hashbig( const void *key
, size_t length
, uint32_t initval
)
647 union { const void *ptr
; size_t i
; } u
; /* to cast key to (size_t) happily */
649 /* Set up the internal state */
650 a
= b
= c
= 0xdeadbeef + ((uint32_t)length
) + initval
;
653 if (HASH_BIG_ENDIAN
&& ((u
.i
& 0x3) == 0)) {
654 const uint32_t *k
= (const uint32_t *)key
; /* read 32-bit chunks */
657 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
668 /*----------------------------- handle the last (probably partial) block */
670 * "k[2]<<8" actually reads beyond the end of the string, but
671 * then shifts out the part it's not allowed to read. Because the
672 * string is aligned, the illegal read is in the same word as the
673 * rest of the string. Every machine with memory protection I've seen
674 * does it on word boundaries, so is OK with this. But VALGRIND will
675 * still catch it and complain. The masking trick does make the hash
676 * noticably faster for short strings (like English words).
682 case 12: c
+=k
[2]; b
+=k
[1]; a
+=k
[0]; break;
683 case 11: c
+=k
[2]&0xffffff00; b
+=k
[1]; a
+=k
[0]; break;
684 case 10: c
+=k
[2]&0xffff0000; b
+=k
[1]; a
+=k
[0]; break;
685 case 9 : c
+=k
[2]&0xff000000; b
+=k
[1]; a
+=k
[0]; break;
686 case 8 : b
+=k
[1]; a
+=k
[0]; break;
687 case 7 : b
+=k
[1]&0xffffff00; a
+=k
[0]; break;
688 case 6 : b
+=k
[1]&0xffff0000; a
+=k
[0]; break;
689 case 5 : b
+=k
[1]&0xff000000; a
+=k
[0]; break;
690 case 4 : a
+=k
[0]; break;
691 case 3 : a
+=k
[0]&0xffffff00; break;
692 case 2 : a
+=k
[0]&0xffff0000; break;
693 case 1 : a
+=k
[0]&0xff000000; break;
694 case 0 : return c
; /* zero length strings require no mixing */
697 #else /* make valgrind happy */
699 k8
= (const uint8_t *)k
;
700 switch(length
) /* all the case statements fall through */
702 case 12: c
+=k
[2]; b
+=k
[1]; a
+=k
[0]; break;
703 case 11: c
+=((uint32_t)k8
[10])<<8; /* fall through */
704 case 10: c
+=((uint32_t)k8
[9])<<16; /* fall through */
705 case 9 : c
+=((uint32_t)k8
[8])<<24; /* fall through */
706 case 8 : b
+=k
[1]; a
+=k
[0]; break;
707 case 7 : b
+=((uint32_t)k8
[6])<<8; /* fall through */
708 case 6 : b
+=((uint32_t)k8
[5])<<16; /* fall through */
709 case 5 : b
+=((uint32_t)k8
[4])<<24; /* fall through */
710 case 4 : a
+=k
[0]; break;
711 case 3 : a
+=((uint32_t)k8
[2])<<8; /* fall through */
712 case 2 : a
+=((uint32_t)k8
[1])<<16; /* fall through */
713 case 1 : a
+=((uint32_t)k8
[0])<<24; break;
717 #endif /* !VALGRIND */
719 } else { /* need to read the key one byte at a time */
720 const uint8_t *k
= (const uint8_t *)key
;
722 /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
725 a
+= ((uint32_t)k
[0])<<24;
726 a
+= ((uint32_t)k
[1])<<16;
727 a
+= ((uint32_t)k
[2])<<8;
728 a
+= ((uint32_t)k
[3]);
729 b
+= ((uint32_t)k
[4])<<24;
730 b
+= ((uint32_t)k
[5])<<16;
731 b
+= ((uint32_t)k
[6])<<8;
732 b
+= ((uint32_t)k
[7]);
733 c
+= ((uint32_t)k
[8])<<24;
734 c
+= ((uint32_t)k
[9])<<16;
735 c
+= ((uint32_t)k
[10])<<8;
736 c
+= ((uint32_t)k
[11]);
742 /*-------------------------------- last block: affect all 32 bits of (c) */
743 switch(length
) /* all the case statements fall through */
746 case 11: c
+=((uint32_t)k
[10])<<8;
747 case 10: c
+=((uint32_t)k
[9])<<16;
748 case 9 : c
+=((uint32_t)k
[8])<<24;
750 case 7 : b
+=((uint32_t)k
[6])<<8;
751 case 6 : b
+=((uint32_t)k
[5])<<16;
752 case 5 : b
+=((uint32_t)k
[4])<<24;
754 case 3 : a
+=((uint32_t)k
[2])<<8;
755 case 2 : a
+=((uint32_t)k
[1])<<16;
756 case 1 : a
+=((uint32_t)k
[0])<<24;
769 /* used for timings */
778 for (i
=0; i
<256; ++i
) buf
[i
] = 'x';
781 h
= hashlittle(&buf
[0],1,h
);
784 if (z
-a
> 0) printf("time %d %.8x\n", z
-a
, h
);
787 /* check that every input bit changes every output bit half the time */
794 uint8_t qa
[MAXLEN
+1], qb
[MAXLEN
+2], *a
= &qa
[0], *b
= &qb
[1];
795 uint32_t c
[HASHSTATE
], d
[HASHSTATE
], i
=0, j
=0, k
, l
, m
=0, z
;
796 uint32_t e
[HASHSTATE
],f
[HASHSTATE
],g
[HASHSTATE
],h
[HASHSTATE
];
797 uint32_t x
[HASHSTATE
],y
[HASHSTATE
];
800 printf("No more than %d trials should ever be needed \n",MAXPAIR
/2);
801 for (hlen
=0; hlen
< MAXLEN
; ++hlen
)
804 for (i
=0; i
<hlen
; ++i
) /*----------------------- for each input byte, */
806 for (j
=0; j
<8; ++j
) /*------------------------ for each input bit, */
808 for (m
=1; m
<8; ++m
) /*------------ for serveral possible initvals, */
810 for (l
=0; l
<HASHSTATE
; ++l
)
811 e
[l
]=f
[l
]=g
[l
]=h
[l
]=x
[l
]=y
[l
]=~((uint32_t)0);
813 /*---- check that every output bit is affected by that input bit */
814 for (k
=0; k
<MAXPAIR
; k
+=2)
817 /* keys have one bit different */
818 for (l
=0; l
<hlen
+1; ++l
) {a
[l
] = b
[l
] = (uint8_t)0;}
819 /* have a and b be two keys differing in only one bit */
822 c
[0] = hashlittle(a
, hlen
, m
);
824 b
[i
] ^= ((k
+1)>>(8-j
));
825 d
[0] = hashlittle(b
, hlen
, m
);
826 /* check every bit is 1, 0, set, and not set at least once */
827 for (l
=0; l
<HASHSTATE
; ++l
)
830 f
[l
] &= ~(c
[l
]^d
[l
]);
835 if (e
[l
]|f
[l
]|g
[l
]|h
[l
]|x
[l
]|y
[l
]) finished
=0;
842 printf("Some bit didn't change: ");
843 printf("%.8x %.8x %.8x %.8x %.8x %.8x ",
844 e
[0],f
[0],g
[0],h
[0],x
[0],y
[0]);
845 printf("i %d j %d m %d len %d\n", i
, j
, m
, hlen
);
847 if (z
==MAXPAIR
) goto done
;
854 printf("Mix success %2d bytes %2d initvals ",i
,m
);
855 printf("required %d trials\n", z
/2);
861 /* Check for reading beyond the end of the buffer and alignment problems */
864 uint8_t buf
[MAXLEN
+20], *b
;
866 uint8_t q
[] = "This is the time for all good men to come to the aid of their country...";
868 uint8_t qq
[] = "xThis is the time for all good men to come to the aid of their country...";
870 uint8_t qqq
[] = "xxThis is the time for all good men to come to the aid of their country...";
872 uint8_t qqqq
[] = "xxxThis is the time for all good men to come to the aid of their country...";
876 printf("Endianness. These lines should all be the same (for values filled in):\n");
877 printf("%.8x %.8x %.8x\n",
878 hashword((const uint32_t *)q
, (sizeof(q
)-1)/4, 13),
879 hashword((const uint32_t *)q
, (sizeof(q
)-5)/4, 13),
880 hashword((const uint32_t *)q
, (sizeof(q
)-9)/4, 13));
882 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
883 hashlittle(p
, sizeof(q
)-1, 13), hashlittle(p
, sizeof(q
)-2, 13),
884 hashlittle(p
, sizeof(q
)-3, 13), hashlittle(p
, sizeof(q
)-4, 13),
885 hashlittle(p
, sizeof(q
)-5, 13), hashlittle(p
, sizeof(q
)-6, 13),
886 hashlittle(p
, sizeof(q
)-7, 13), hashlittle(p
, sizeof(q
)-8, 13),
887 hashlittle(p
, sizeof(q
)-9, 13), hashlittle(p
, sizeof(q
)-10, 13),
888 hashlittle(p
, sizeof(q
)-11, 13), hashlittle(p
, sizeof(q
)-12, 13));
890 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
891 hashlittle(p
, sizeof(q
)-1, 13), hashlittle(p
, sizeof(q
)-2, 13),
892 hashlittle(p
, sizeof(q
)-3, 13), hashlittle(p
, sizeof(q
)-4, 13),
893 hashlittle(p
, sizeof(q
)-5, 13), hashlittle(p
, sizeof(q
)-6, 13),
894 hashlittle(p
, sizeof(q
)-7, 13), hashlittle(p
, sizeof(q
)-8, 13),
895 hashlittle(p
, sizeof(q
)-9, 13), hashlittle(p
, sizeof(q
)-10, 13),
896 hashlittle(p
, sizeof(q
)-11, 13), hashlittle(p
, sizeof(q
)-12, 13));
898 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
899 hashlittle(p
, sizeof(q
)-1, 13), hashlittle(p
, sizeof(q
)-2, 13),
900 hashlittle(p
, sizeof(q
)-3, 13), hashlittle(p
, sizeof(q
)-4, 13),
901 hashlittle(p
, sizeof(q
)-5, 13), hashlittle(p
, sizeof(q
)-6, 13),
902 hashlittle(p
, sizeof(q
)-7, 13), hashlittle(p
, sizeof(q
)-8, 13),
903 hashlittle(p
, sizeof(q
)-9, 13), hashlittle(p
, sizeof(q
)-10, 13),
904 hashlittle(p
, sizeof(q
)-11, 13), hashlittle(p
, sizeof(q
)-12, 13));
906 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
907 hashlittle(p
, sizeof(q
)-1, 13), hashlittle(p
, sizeof(q
)-2, 13),
908 hashlittle(p
, sizeof(q
)-3, 13), hashlittle(p
, sizeof(q
)-4, 13),
909 hashlittle(p
, sizeof(q
)-5, 13), hashlittle(p
, sizeof(q
)-6, 13),
910 hashlittle(p
, sizeof(q
)-7, 13), hashlittle(p
, sizeof(q
)-8, 13),
911 hashlittle(p
, sizeof(q
)-9, 13), hashlittle(p
, sizeof(q
)-10, 13),
912 hashlittle(p
, sizeof(q
)-11, 13), hashlittle(p
, sizeof(q
)-12, 13));
915 /* check that hashlittle2 and hashlittle produce the same results */
917 hashlittle2(q
, sizeof(q
), &i
, &j
);
918 if (hashlittle(q
, sizeof(q
), 47) != i
)
919 printf("hashlittle2 and hashlittle mismatch\n");
921 /* check that hashword2 and hashword produce the same results */
924 hashword2(&len
, 1, &i
, &j
);
925 if (hashword(&len
, 1, 47) != i
)
926 printf("hashword2 and hashword mismatch %x %x\n",
927 i
, hashword(&len
, 1, 47));
929 /* check hashlittle doesn't read before or after the ends of the string */
930 for (h
=0, b
=buf
+1; h
<8; ++h
, ++b
)
932 for (i
=0; i
<MAXLEN
; ++i
)
935 for (j
=0; j
<i
; ++j
) *(b
+j
)=0;
937 /* these should all be equal */
938 ref
= hashlittle(b
, len
, (uint32_t)1);
941 x
= hashlittle(b
, len
, (uint32_t)1);
942 y
= hashlittle(b
, len
, (uint32_t)1);
943 if ((ref
!= x
) || (ref
!= y
))
945 printf("alignment error: %.8x %.8x %.8x %d %d\n",ref
,x
,y
,
952 /* check for problems with nulls */
956 uint32_t h
,i
,state
[HASHSTATE
];
960 for (i
=0; i
<HASHSTATE
; ++i
) state
[i
] = 1;
961 printf("These should all be different\n");
962 for (i
=0, h
=0; i
<8; ++i
)
964 h
= hashlittle(buf
, 0, h
);
965 printf("%2ld 0-byte strings, hash is %.8x\n", i
, h
);
972 b
=0, c
=0, hashlittle2("", 0, &c
, &b
);
973 printf("hash is %.8lx %.8lx\n", c
, b
); /* deadbeef deadbeef */
974 b
=0xdeadbeef, c
=0, hashlittle2("", 0, &c
, &b
);
975 printf("hash is %.8lx %.8lx\n", c
, b
); /* bd5b7dde deadbeef */
976 b
=0xdeadbeef, c
=0xdeadbeef, hashlittle2("", 0, &c
, &b
);
977 printf("hash is %.8lx %.8lx\n", c
, b
); /* 9c093ccd bd5b7dde */
978 b
=0, c
=0, hashlittle2("Four score and seven years ago", 30, &c
, &b
);
979 printf("hash is %.8lx %.8lx\n", c
, b
); /* 17770551 ce7226e6 */
980 b
=1, c
=0, hashlittle2("Four score and seven years ago", 30, &c
, &b
);
981 printf("hash is %.8lx %.8lx\n", c
, b
); /* e3607cae bd371de4 */
982 b
=0, c
=1, hashlittle2("Four score and seven years ago", 30, &c
, &b
);
983 printf("hash is %.8lx %.8lx\n", c
, b
); /* cd628161 6cbea4b3 */
984 c
= hashlittle("Four score and seven years ago", 30, 0);
985 printf("hash is %.8lx\n", c
); /* 17770551 */
986 c
= hashlittle("Four score and seven years ago", 30, 1);
987 printf("hash is %.8lx\n", c
); /* cd628161 */
993 driver1(); /* test that the key is hashed: used for timings */
994 driver2(); /* test that whole key is hashed thoroughly */
995 driver3(); /* test that nothing but the key is hashed */
996 driver4(); /* test hashing multiple buffers (all buffers are null) */
997 driver5(); /* test the hash against known vectors */
1001 #endif /* SELF_TEST */