1 /* Slightly modified by Lennart Poettering, to avoid name clashes, and
2 * unexport a few functions. */
7 -------------------------------------------------------------------------------
8 lookup3.c, by Bob Jenkins, May 2006, Public Domain.
10 These are functions for producing 32-bit hashes for hash table lookup.
11 hashword(), hashlittle(), hashlittle2(), hashbig(), mix(), and final()
12 are externally useful functions. Routines to test the hash are included
13 if SELF_TEST is defined. You can use this free for any purpose. It's in
14 the public domain. It has no warranty.
16 You probably want to use hashlittle(). hashlittle() and hashbig()
17 hash byte arrays. hashlittle() is is faster than hashbig() on
18 little-endian machines. Intel and AMD are little-endian machines.
19 On second thought, you probably want hashlittle2(), which is identical to
20 hashlittle() except it returns two 32-bit hashes for the price of one.
21 You could implement hashbig2() if you wanted but I haven't bothered here.
23 If you want to find a hash of, say, exactly 7 integers, do
24 a = i1; b = i2; c = i3;
26 a += i4; b += i5; c += i6;
30 then use c as the hash value. If you have a variable length array of
31 4-byte integers to hash, use hashword(). If you have a byte array (like
32 a character string), use hashlittle(). If you have several byte arrays, or
33 a mix of things, see the comments above hashlittle().
35 Why is this so big? I read 12 bytes at a time into 3 4-byte integers,
36 then mix those integers. This is fast (you can do a lot more thorough
37 mixing with 12*3 instructions on 3 integers than you can with 3 instructions
38 on 1 byte), but shoehorning those bytes into integers efficiently is messy.
39 -------------------------------------------------------------------------------
41 /* #define SELF_TEST 1 */
43 #include <stdio.h> /* defines printf for tests */
44 #include <time.h> /* defines time_t for timings in the test */
45 #include <stdint.h> /* defines uint32_t etc */
46 #include <sys/param.h> /* attempt to define endianness */
48 # include <endian.h> /* attempt to define endianness */
52 * My best guess at if you are big-endian or little-endian. This may
55 #if (defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && \
56 __BYTE_ORDER == __LITTLE_ENDIAN) || \
57 (defined(i386) || defined(__i386__) || defined(__i486__) || \
58 defined(__i586__) || defined(__i686__) || defined(vax) || defined(MIPSEL))
59 # define HASH_LITTLE_ENDIAN 1
60 # define HASH_BIG_ENDIAN 0
61 #elif (defined(__BYTE_ORDER) && defined(__BIG_ENDIAN) && \
62 __BYTE_ORDER == __BIG_ENDIAN) || \
63 (defined(sparc) || defined(POWERPC) || defined(mc68000) || defined(sel))
64 # define HASH_LITTLE_ENDIAN 0
65 # define HASH_BIG_ENDIAN 1
67 # define HASH_LITTLE_ENDIAN 0
68 # define HASH_BIG_ENDIAN 0
71 #define hashsize(n) ((uint32_t)1<<(n))
72 #define hashmask(n) (hashsize(n)-1)
73 #define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
76 -------------------------------------------------------------------------------
77 mix -- mix 3 32-bit values reversibly.
79 This is reversible, so any information in (a,b,c) before mix() is
80 still in (a,b,c) after mix().
82 If four pairs of (a,b,c) inputs are run through mix(), or through
83 mix() in reverse, there are at least 32 bits of the output that
84 are sometimes the same for one pair and different for another pair.
86 * pairs that differed by one bit, by two bits, in any combination
87 of top bits of (a,b,c), or in any combination of bottom bits of
89 * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
90 the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
91 is commonly produced by subtraction) look like a single 1-bit
93 * the base values were pseudorandom, all zero but one bit set, or
94 all zero plus a counter that starts at zero.
96 Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
101 Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
102 for "differ" defined as + with a one-bit base and a two-bit delta. I
103 used http://burtleburtle.net/bob/hash/avalanche.html to choose
104 the operations, constants, and arrangements of the variables.
106 This does not achieve avalanche. There are input bits of (a,b,c)
107 that fail to affect some output bits of (a,b,c), especially of a. The
108 most thoroughly mixed value is c, but it doesn't really even achieve
111 This allows some parallelism. Read-after-writes are good at doubling
112 the number of bits affected, so the goal of mixing pulls in the opposite
113 direction as the goal of parallelism. I did what I could. Rotates
114 seem to cost as much as shifts on every machine I could lay my hands
115 on, and rotates are much kinder to the top and bottom bits, so I used
117 -------------------------------------------------------------------------------
121 a -= c; a ^= rot(c, 4); c += b; \
122 b -= a; b ^= rot(a, 6); a += c; \
123 c -= b; c ^= rot(b, 8); b += a; \
124 a -= c; a ^= rot(c,16); c += b; \
125 b -= a; b ^= rot(a,19); a += c; \
126 c -= b; c ^= rot(b, 4); b += a; \
130 -------------------------------------------------------------------------------
131 final -- final mixing of 3 32-bit values (a,b,c) into c
133 Pairs of (a,b,c) values differing in only a few bits will usually
134 produce values of c that look totally different. This was tested for
135 * pairs that differed by one bit, by two bits, in any combination
136 of top bits of (a,b,c), or in any combination of bottom bits of
138 * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
139 the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
140 is commonly produced by subtraction) look like a single 1-bit
142 * the base values were pseudorandom, all zero but one bit set, or
143 all zero plus a counter that starts at zero.
145 These constants passed:
148 and these came close:
152 -------------------------------------------------------------------------------
154 #define final(a,b,c) \
156 c ^= b; c -= rot(b,14); \
157 a ^= c; a -= rot(c,11); \
158 b ^= a; b -= rot(a,25); \
159 c ^= b; c -= rot(b,16); \
160 a ^= c; a -= rot(c,4); \
161 b ^= a; b -= rot(a,14); \
162 c ^= b; c -= rot(b,24); \
166 --------------------------------------------------------------------
167 This works on all machines. To be useful, it requires
168 -- that the key be an array of uint32_t's, and
169 -- that the length be the number of uint32_t's in the key
171 The function hashword() is identical to hashlittle() on little-endian
172 machines, and identical to hashbig() on big-endian machines,
173 except that the length has to be measured in uint32_ts rather than in
174 bytes. hashlittle() is more complicated than hashword() only because
175 hashlittle() has to dance around fitting the key bytes into registers.
176 --------------------------------------------------------------------
178 uint32_t jenkins_hashword(
179 const uint32_t *k, /* the key, an array of uint32_t values */
180 size_t length, /* the length of the key, in uint32_ts */
181 uint32_t initval) /* the previous hash, or an arbitrary value */
185 /* Set up the internal state */
186 a = b = c = 0xdeadbeef + (((uint32_t)length)<<2) + initval;
188 /*------------------------------------------------- handle most of the key */
199 /*------------------------------------------- handle the last 3 uint32_t's */
200 switch(length) /* all the case statements fall through */
206 case 0: /* case 0: nothing left to add */
209 /*------------------------------------------------------ report the result */
215 --------------------------------------------------------------------
216 hashword2() -- same as hashword(), but take two seeds and return two
217 32-bit values. pc and pb must both be nonnull, and *pc and *pb must
218 both be initialized with seeds. If you pass in (*pb)==0, the output
219 (*pc) will be the same as the return value from hashword().
220 --------------------------------------------------------------------
222 void jenkins_hashword2 (
223 const uint32_t *k, /* the key, an array of uint32_t values */
224 size_t length, /* the length of the key, in uint32_ts */
225 uint32_t *pc, /* IN: seed OUT: primary hash value */
226 uint32_t *pb) /* IN: more seed OUT: secondary hash value */
230 /* Set up the internal state */
231 a = b = c = 0xdeadbeef + ((uint32_t)(length<<2)) + *pc;
234 /*------------------------------------------------- handle most of the key */
245 /*------------------------------------------- handle the last 3 uint32_t's */
246 switch(length) /* all the case statements fall through */
252 case 0: /* case 0: nothing left to add */
255 /*------------------------------------------------------ report the result */
261 -------------------------------------------------------------------------------
262 hashlittle() -- hash a variable-length key into a 32-bit value
263 k : the key (the unaligned variable-length array of bytes)
264 length : the length of the key, counting by bytes
265 initval : can be any 4-byte value
266 Returns a 32-bit value. Every bit of the key affects every bit of
267 the return value. Two keys differing by one or two bits will have
268 totally different hash values.
270 The best hash table sizes are powers of 2. There is no need to do
271 mod a prime (mod is sooo slow!). If you need less than 32 bits,
272 use a bitmask. For example, if you need only 10 bits, do
273 h = (h & hashmask(10));
274 In which case, the hash table should have hashsize(10) elements.
276 If you are hashing n strings (uint8_t **)k, do it like this:
277 for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h);
279 By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this
280 code any way you wish, private, educational, or commercial. It's free.
282 Use for hash table lookup, or anything where one collision in 2^^32 is
283 acceptable. Do NOT use for cryptographic purposes.
284 -------------------------------------------------------------------------------
287 uint32_t jenkins_hashlittle( const void *key, size_t length, uint32_t initval)
289 uint32_t a,b,c; /* internal state */
290 union { const void *ptr; size_t i; } u; /* needed for Mac Powerbook G4 */
292 /* Set up the internal state */
293 a = b = c = 0xdeadbeef + ((uint32_t)length) + initval;
296 if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
297 const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */
299 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
310 /*----------------------------- handle the last (probably partial) block */
312 * "k[2]&0xffffff" actually reads beyond the end of the string, but
313 * then masks off the part it's not allowed to read. Because the
314 * string is aligned, the masked-off tail is in the same word as the
315 * rest of the string. Every machine with memory protection I've seen
316 * does it on word boundaries, so is OK with this. But VALGRIND will
317 * still catch it and complain. The masking trick does make the hash
318 * noticably faster for short strings (like English words).
324 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
325 case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
326 case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break;
327 case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break;
328 case 8 : b+=k[1]; a+=k[0]; break;
329 case 7 : b+=k[1]&0xffffff; a+=k[0]; break;
330 case 6 : b+=k[1]&0xffff; a+=k[0]; break;
331 case 5 : b+=k[1]&0xff; a+=k[0]; break;
332 case 4 : a+=k[0]; break;
333 case 3 : a+=k[0]&0xffffff; break;
334 case 2 : a+=k[0]&0xffff; break;
335 case 1 : a+=k[0]&0xff; break;
336 case 0 : return c; /* zero length strings require no mixing */
339 #else /* make valgrind happy */
341 k8 = (const uint8_t *)k;
344 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
345 case 11: c+=((uint32_t)k8[10])<<16; /* fall through */
346 case 10: c+=((uint32_t)k8[9])<<8; /* fall through */
347 case 9 : c+=k8[8]; /* fall through */
348 case 8 : b+=k[1]; a+=k[0]; break;
349 case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */
350 case 6 : b+=((uint32_t)k8[5])<<8; /* fall through */
351 case 5 : b+=k8[4]; /* fall through */
352 case 4 : a+=k[0]; break;
353 case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */
354 case 2 : a+=((uint32_t)k8[1])<<8; /* fall through */
355 case 1 : a+=k8[0]; break;
359 #endif /* !valgrind */
361 } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
362 const uint16_t *k = (const uint16_t *)key; /* read 16-bit chunks */
365 /*--------------- all but last block: aligned reads and different mixing */
368 a += k[0] + (((uint32_t)k[1])<<16);
369 b += k[2] + (((uint32_t)k[3])<<16);
370 c += k[4] + (((uint32_t)k[5])<<16);
376 /*----------------------------- handle the last (probably partial) block */
377 k8 = (const uint8_t *)k;
380 case 12: c+=k[4]+(((uint32_t)k[5])<<16);
381 b+=k[2]+(((uint32_t)k[3])<<16);
382 a+=k[0]+(((uint32_t)k[1])<<16);
384 case 11: c+=((uint32_t)k8[10])<<16; /* fall through */
386 b+=k[2]+(((uint32_t)k[3])<<16);
387 a+=k[0]+(((uint32_t)k[1])<<16);
389 case 9 : c+=k8[8]; /* fall through */
390 case 8 : b+=k[2]+(((uint32_t)k[3])<<16);
391 a+=k[0]+(((uint32_t)k[1])<<16);
393 case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */
395 a+=k[0]+(((uint32_t)k[1])<<16);
397 case 5 : b+=k8[4]; /* fall through */
398 case 4 : a+=k[0]+(((uint32_t)k[1])<<16);
400 case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */
405 case 0 : return c; /* zero length requires no mixing */
408 } else { /* need to read the key one byte at a time */
409 const uint8_t *k = (const uint8_t *)key;
411 /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
415 a += ((uint32_t)k[1])<<8;
416 a += ((uint32_t)k[2])<<16;
417 a += ((uint32_t)k[3])<<24;
419 b += ((uint32_t)k[5])<<8;
420 b += ((uint32_t)k[6])<<16;
421 b += ((uint32_t)k[7])<<24;
423 c += ((uint32_t)k[9])<<8;
424 c += ((uint32_t)k[10])<<16;
425 c += ((uint32_t)k[11])<<24;
431 /*-------------------------------- last block: affect all 32 bits of (c) */
432 switch(length) /* all the case statements fall through */
434 case 12: c+=((uint32_t)k[11])<<24;
435 case 11: c+=((uint32_t)k[10])<<16;
436 case 10: c+=((uint32_t)k[9])<<8;
438 case 8 : b+=((uint32_t)k[7])<<24;
439 case 7 : b+=((uint32_t)k[6])<<16;
440 case 6 : b+=((uint32_t)k[5])<<8;
442 case 4 : a+=((uint32_t)k[3])<<24;
443 case 3 : a+=((uint32_t)k[2])<<16;
444 case 2 : a+=((uint32_t)k[1])<<8;
457 * hashlittle2: return 2 32-bit hash values
459 * This is identical to hashlittle(), except it returns two 32-bit hash
460 * values instead of just one. This is good enough for hash table
461 * lookup with 2^^64 buckets, or if you want a second hash if you're not
462 * happy with the first, or if you want a probably-unique 64-bit ID for
463 * the key. *pc is better mixed than *pb, so use *pc first. If you want
464 * a 64-bit value do something like "*pc + (((uint64_t)*pb)<<32)".
466 void jenkins_hashlittle2(
467 const void *key, /* the key to hash */
468 size_t length, /* length of the key */
469 uint32_t *pc, /* IN: primary initval, OUT: primary hash */
470 uint32_t *pb) /* IN: secondary initval, OUT: secondary hash */
472 uint32_t a,b,c; /* internal state */
473 union { const void *ptr; size_t i; } u; /* needed for Mac Powerbook G4 */
475 /* Set up the internal state */
476 a = b = c = 0xdeadbeef + ((uint32_t)length) + *pc;
480 if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
481 const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */
483 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
494 /*----------------------------- handle the last (probably partial) block */
496 * "k[2]&0xffffff" actually reads beyond the end of the string, but
497 * then masks off the part it's not allowed to read. Because the
498 * string is aligned, the masked-off tail is in the same word as the
499 * rest of the string. Every machine with memory protection I've seen
500 * does it on word boundaries, so is OK with this. But VALGRIND will
501 * still catch it and complain. The masking trick does make the hash
502 * noticably faster for short strings (like English words).
508 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
509 case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
510 case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break;
511 case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break;
512 case 8 : b+=k[1]; a+=k[0]; break;
513 case 7 : b+=k[1]&0xffffff; a+=k[0]; break;
514 case 6 : b+=k[1]&0xffff; a+=k[0]; break;
515 case 5 : b+=k[1]&0xff; a+=k[0]; break;
516 case 4 : a+=k[0]; break;
517 case 3 : a+=k[0]&0xffffff; break;
518 case 2 : a+=k[0]&0xffff; break;
519 case 1 : a+=k[0]&0xff; break;
520 case 0 : *pc=c; *pb=b; return; /* zero length strings require no mixing */
523 #else /* make valgrind happy */
525 k8 = (const uint8_t *)k;
528 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
529 case 11: c+=((uint32_t)k8[10])<<16; /* fall through */
530 case 10: c+=((uint32_t)k8[9])<<8; /* fall through */
531 case 9 : c+=k8[8]; /* fall through */
532 case 8 : b+=k[1]; a+=k[0]; break;
533 case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */
534 case 6 : b+=((uint32_t)k8[5])<<8; /* fall through */
535 case 5 : b+=k8[4]; /* fall through */
536 case 4 : a+=k[0]; break;
537 case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */
538 case 2 : a+=((uint32_t)k8[1])<<8; /* fall through */
539 case 1 : a+=k8[0]; break;
540 case 0 : *pc=c; *pb=b; return; /* zero length strings require no mixing */
543 #endif /* !valgrind */
545 } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
546 const uint16_t *k = (const uint16_t *)key; /* read 16-bit chunks */
549 /*--------------- all but last block: aligned reads and different mixing */
552 a += k[0] + (((uint32_t)k[1])<<16);
553 b += k[2] + (((uint32_t)k[3])<<16);
554 c += k[4] + (((uint32_t)k[5])<<16);
560 /*----------------------------- handle the last (probably partial) block */
561 k8 = (const uint8_t *)k;
564 case 12: c+=k[4]+(((uint32_t)k[5])<<16);
565 b+=k[2]+(((uint32_t)k[3])<<16);
566 a+=k[0]+(((uint32_t)k[1])<<16);
568 case 11: c+=((uint32_t)k8[10])<<16; /* fall through */
570 b+=k[2]+(((uint32_t)k[3])<<16);
571 a+=k[0]+(((uint32_t)k[1])<<16);
573 case 9 : c+=k8[8]; /* fall through */
574 case 8 : b+=k[2]+(((uint32_t)k[3])<<16);
575 a+=k[0]+(((uint32_t)k[1])<<16);
577 case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */
579 a+=k[0]+(((uint32_t)k[1])<<16);
581 case 5 : b+=k8[4]; /* fall through */
582 case 4 : a+=k[0]+(((uint32_t)k[1])<<16);
584 case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */
589 case 0 : *pc=c; *pb=b; return; /* zero length strings require no mixing */
592 } else { /* need to read the key one byte at a time */
593 const uint8_t *k = (const uint8_t *)key;
595 /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
599 a += ((uint32_t)k[1])<<8;
600 a += ((uint32_t)k[2])<<16;
601 a += ((uint32_t)k[3])<<24;
603 b += ((uint32_t)k[5])<<8;
604 b += ((uint32_t)k[6])<<16;
605 b += ((uint32_t)k[7])<<24;
607 c += ((uint32_t)k[9])<<8;
608 c += ((uint32_t)k[10])<<16;
609 c += ((uint32_t)k[11])<<24;
615 /*-------------------------------- last block: affect all 32 bits of (c) */
616 switch(length) /* all the case statements fall through */
618 case 12: c+=((uint32_t)k[11])<<24;
619 case 11: c+=((uint32_t)k[10])<<16;
620 case 10: c+=((uint32_t)k[9])<<8;
622 case 8 : b+=((uint32_t)k[7])<<24;
623 case 7 : b+=((uint32_t)k[6])<<16;
624 case 6 : b+=((uint32_t)k[5])<<8;
626 case 4 : a+=((uint32_t)k[3])<<24;
627 case 3 : a+=((uint32_t)k[2])<<16;
628 case 2 : a+=((uint32_t)k[1])<<8;
631 case 0 : *pc=c; *pb=b; return; /* zero length strings require no mixing */
643 * This is the same as hashword() on big-endian machines. It is different
644 * from hashlittle() on all machines. hashbig() takes advantage of
645 * big-endian byte ordering.
647 uint32_t jenkins_hashbig( const void *key, size_t length, uint32_t initval)
650 union { const void *ptr; size_t i; } u; /* to cast key to (size_t) happily */
652 /* Set up the internal state */
653 a = b = c = 0xdeadbeef + ((uint32_t)length) + initval;
656 if (HASH_BIG_ENDIAN && ((u.i & 0x3) == 0)) {
657 const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */
659 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
670 /*----------------------------- handle the last (probably partial) block */
672 * "k[2]<<8" actually reads beyond the end of the string, but
673 * then shifts out the part it's not allowed to read. Because the
674 * string is aligned, the illegal read is in the same word as the
675 * rest of the string. Every machine with memory protection I've seen
676 * does it on word boundaries, so is OK with this. But VALGRIND will
677 * still catch it and complain. The masking trick does make the hash
678 * noticably faster for short strings (like English words).
684 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
685 case 11: c+=k[2]&0xffffff00; b+=k[1]; a+=k[0]; break;
686 case 10: c+=k[2]&0xffff0000; b+=k[1]; a+=k[0]; break;
687 case 9 : c+=k[2]&0xff000000; b+=k[1]; a+=k[0]; break;
688 case 8 : b+=k[1]; a+=k[0]; break;
689 case 7 : b+=k[1]&0xffffff00; a+=k[0]; break;
690 case 6 : b+=k[1]&0xffff0000; a+=k[0]; break;
691 case 5 : b+=k[1]&0xff000000; a+=k[0]; break;
692 case 4 : a+=k[0]; break;
693 case 3 : a+=k[0]&0xffffff00; break;
694 case 2 : a+=k[0]&0xffff0000; break;
695 case 1 : a+=k[0]&0xff000000; break;
696 case 0 : return c; /* zero length strings require no mixing */
699 #else /* make valgrind happy */
701 k8 = (const uint8_t *)k;
702 switch(length) /* all the case statements fall through */
704 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
705 case 11: c+=((uint32_t)k8[10])<<8; /* fall through */
706 case 10: c+=((uint32_t)k8[9])<<16; /* fall through */
707 case 9 : c+=((uint32_t)k8[8])<<24; /* fall through */
708 case 8 : b+=k[1]; a+=k[0]; break;
709 case 7 : b+=((uint32_t)k8[6])<<8; /* fall through */
710 case 6 : b+=((uint32_t)k8[5])<<16; /* fall through */
711 case 5 : b+=((uint32_t)k8[4])<<24; /* fall through */
712 case 4 : a+=k[0]; break;
713 case 3 : a+=((uint32_t)k8[2])<<8; /* fall through */
714 case 2 : a+=((uint32_t)k8[1])<<16; /* fall through */
715 case 1 : a+=((uint32_t)k8[0])<<24; break;
719 #endif /* !VALGRIND */
721 } else { /* need to read the key one byte at a time */
722 const uint8_t *k = (const uint8_t *)key;
724 /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
727 a += ((uint32_t)k[0])<<24;
728 a += ((uint32_t)k[1])<<16;
729 a += ((uint32_t)k[2])<<8;
730 a += ((uint32_t)k[3]);
731 b += ((uint32_t)k[4])<<24;
732 b += ((uint32_t)k[5])<<16;
733 b += ((uint32_t)k[6])<<8;
734 b += ((uint32_t)k[7]);
735 c += ((uint32_t)k[8])<<24;
736 c += ((uint32_t)k[9])<<16;
737 c += ((uint32_t)k[10])<<8;
738 c += ((uint32_t)k[11]);
744 /*-------------------------------- last block: affect all 32 bits of (c) */
745 switch(length) /* all the case statements fall through */
748 case 11: c+=((uint32_t)k[10])<<8;
749 case 10: c+=((uint32_t)k[9])<<16;
750 case 9 : c+=((uint32_t)k[8])<<24;
752 case 7 : b+=((uint32_t)k[6])<<8;
753 case 6 : b+=((uint32_t)k[5])<<16;
754 case 5 : b+=((uint32_t)k[4])<<24;
756 case 3 : a+=((uint32_t)k[2])<<8;
757 case 2 : a+=((uint32_t)k[1])<<16;
758 case 1 : a+=((uint32_t)k[0])<<24;
771 /* used for timings */
780 for (i=0; i<256; ++i) buf[i] = 'x';
783 h = hashlittle(&buf[0],1,h);
786 if (z-a > 0) printf("time %d %.8x\n", z-a, h);
789 /* check that every input bit changes every output bit half the time */
796 uint8_t qa[MAXLEN+1], qb[MAXLEN+2], *a = &qa[0], *b = &qb[1];
797 uint32_t c[HASHSTATE], d[HASHSTATE], i=0, j=0, k, l, m=0, z;
798 uint32_t e[HASHSTATE],f[HASHSTATE],g[HASHSTATE],h[HASHSTATE];
799 uint32_t x[HASHSTATE],y[HASHSTATE];
802 printf("No more than %d trials should ever be needed \n",MAXPAIR/2);
803 for (hlen=0; hlen < MAXLEN; ++hlen)
806 for (i=0; i<hlen; ++i) /*----------------------- for each input byte, */
808 for (j=0; j<8; ++j) /*------------------------ for each input bit, */
810 for (m=1; m<8; ++m) /*------------ for serveral possible initvals, */
812 for (l=0; l<HASHSTATE; ++l)
813 e[l]=f[l]=g[l]=h[l]=x[l]=y[l]=~((uint32_t)0);
815 /*---- check that every output bit is affected by that input bit */
816 for (k=0; k<MAXPAIR; k+=2)
819 /* keys have one bit different */
820 for (l=0; l<hlen+1; ++l) {a[l] = b[l] = (uint8_t)0;}
821 /* have a and b be two keys differing in only one bit */
824 c[0] = hashlittle(a, hlen, m);
826 b[i] ^= ((k+1)>>(8-j));
827 d[0] = hashlittle(b, hlen, m);
828 /* check every bit is 1, 0, set, and not set at least once */
829 for (l=0; l<HASHSTATE; ++l)
832 f[l] &= ~(c[l]^d[l]);
837 if (e[l]|f[l]|g[l]|h[l]|x[l]|y[l]) finished=0;
844 printf("Some bit didn't change: ");
845 printf("%.8x %.8x %.8x %.8x %.8x %.8x ",
846 e[0],f[0],g[0],h[0],x[0],y[0]);
847 printf("i %d j %d m %d len %d\n", i, j, m, hlen);
849 if (z==MAXPAIR) goto done;
856 printf("Mix success %2d bytes %2d initvals ",i,m);
857 printf("required %d trials\n", z/2);
863 /* Check for reading beyond the end of the buffer and alignment problems */
866 uint8_t buf[MAXLEN+20], *b;
868 uint8_t q[] = "This is the time for all good men to come to the aid of their country...";
870 uint8_t qq[] = "xThis is the time for all good men to come to the aid of their country...";
872 uint8_t qqq[] = "xxThis is the time for all good men to come to the aid of their country...";
874 uint8_t qqqq[] = "xxxThis is the time for all good men to come to the aid of their country...";
878 printf("Endianness. These lines should all be the same (for values filled in):\n");
879 printf("%.8x %.8x %.8x\n",
880 hashword((const uint32_t *)q, (sizeof(q)-1)/4, 13),
881 hashword((const uint32_t *)q, (sizeof(q)-5)/4, 13),
882 hashword((const uint32_t *)q, (sizeof(q)-9)/4, 13));
884 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
885 hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13),
886 hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13),
887 hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13),
888 hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13),
889 hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13),
890 hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13));
892 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
893 hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13),
894 hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13),
895 hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13),
896 hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13),
897 hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13),
898 hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13));
900 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
901 hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13),
902 hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13),
903 hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13),
904 hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13),
905 hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13),
906 hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13));
908 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
909 hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13),
910 hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13),
911 hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13),
912 hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13),
913 hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13),
914 hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13));
917 /* check that hashlittle2 and hashlittle produce the same results */
919 hashlittle2(q, sizeof(q), &i, &j);
920 if (hashlittle(q, sizeof(q), 47) != i)
921 printf("hashlittle2 and hashlittle mismatch\n");
923 /* check that hashword2 and hashword produce the same results */
926 hashword2(&len, 1, &i, &j);
927 if (hashword(&len, 1, 47) != i)
928 printf("hashword2 and hashword mismatch %x %x\n",
929 i, hashword(&len, 1, 47));
931 /* check hashlittle doesn't read before or after the ends of the string */
932 for (h=0, b=buf+1; h<8; ++h, ++b)
934 for (i=0; i<MAXLEN; ++i)
937 for (j=0; j<i; ++j) *(b+j)=0;
939 /* these should all be equal */
940 ref = hashlittle(b, len, (uint32_t)1);
943 x = hashlittle(b, len, (uint32_t)1);
944 y = hashlittle(b, len, (uint32_t)1);
945 if ((ref != x) || (ref != y))
947 printf("alignment error: %.8x %.8x %.8x %d %d\n",ref,x,y,
954 /* check for problems with nulls */
958 uint32_t h,i,state[HASHSTATE];
962 for (i=0; i<HASHSTATE; ++i) state[i] = 1;
963 printf("These should all be different\n");
964 for (i=0, h=0; i<8; ++i)
966 h = hashlittle(buf, 0, h);
967 printf("%2ld 0-byte strings, hash is %.8x\n", i, h);
974 b=0, c=0, hashlittle2("", 0, &c, &b);
975 printf("hash is %.8lx %.8lx\n", c, b); /* deadbeef deadbeef */
976 b=0xdeadbeef, c=0, hashlittle2("", 0, &c, &b);
977 printf("hash is %.8lx %.8lx\n", c, b); /* bd5b7dde deadbeef */
978 b=0xdeadbeef, c=0xdeadbeef, hashlittle2("", 0, &c, &b);
979 printf("hash is %.8lx %.8lx\n", c, b); /* 9c093ccd bd5b7dde */
980 b=0, c=0, hashlittle2("Four score and seven years ago", 30, &c, &b);
981 printf("hash is %.8lx %.8lx\n", c, b); /* 17770551 ce7226e6 */
982 b=1, c=0, hashlittle2("Four score and seven years ago", 30, &c, &b);
983 printf("hash is %.8lx %.8lx\n", c, b); /* e3607cae bd371de4 */
984 b=0, c=1, hashlittle2("Four score and seven years ago", 30, &c, &b);
985 printf("hash is %.8lx %.8lx\n", c, b); /* cd628161 6cbea4b3 */
986 c = hashlittle("Four score and seven years ago", 30, 0);
987 printf("hash is %.8lx\n", c); /* 17770551 */
988 c = hashlittle("Four score and seven years ago", 30, 1);
989 printf("hash is %.8lx\n", c); /* cd628161 */
995 driver1(); /* test that the key is hashed: used for timings */
996 driver2(); /* test that whole key is hashed thoroughly */
997 driver3(); /* test that nothing but the key is hashed */
998 driver4(); /* test hashing multiple buffers (all buffers are null) */
999 driver5(); /* test the hash against known vectors */
1003 #endif /* SELF_TEST */