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 hash_word(), 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 hash_word(). 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 -------------------------------------------------------------------------------
39 #include <stdio.h> /* defines printf for tests */
40 #include <time.h> /* defines time_t for timings in the test */
41 #include <stdint.h> /* defines uint32_t etc */
42 #include <sys/param.h> /* attempt to define endianness */
45 # include <endian.h> /* attempt to define endianness */
49 * My best guess at if you are big-endian or little-endian. This may
52 #if (defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && \
53 __BYTE_ORDER == __LITTLE_ENDIAN) || \
54 (defined(i386) || defined(__i386__) || defined(__i486__) || \
55 defined(__i586__) || defined(__i686__) || defined(__x86_64) || \
56 defined(vax) || defined(MIPSEL))
57 # define HASH_LITTLE_ENDIAN 1
58 # define HASH_BIG_ENDIAN 0
59 #elif (defined(__BYTE_ORDER) && defined(__BIG_ENDIAN) && \
60 __BYTE_ORDER == __BIG_ENDIAN) || \
61 (defined(sparc) || defined(POWERPC) || defined(mc68000) || defined(sel))
62 # define HASH_LITTLE_ENDIAN 0
63 # define HASH_BIG_ENDIAN 1
65 # error Unknown endian
67 #endif /* old hash.c headers. */
71 #if HAVE_LITTLE_ENDIAN
72 #define HASH_LITTLE_ENDIAN 1
73 #define HASH_BIG_ENDIAN 0
75 #define HASH_LITTLE_ENDIAN 0
76 #define HASH_BIG_ENDIAN 1
81 #define hashsize(n) ((uint32_t)1<<(n))
82 #define hashmask(n) (hashsize(n)-1)
83 #define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
86 -------------------------------------------------------------------------------
87 mix -- mix 3 32-bit values reversibly.
89 This is reversible, so any information in (a,b,c) before mix() is
90 still in (a,b,c) after mix().
92 If four pairs of (a,b,c) inputs are run through mix(), or through
93 mix() in reverse, there are at least 32 bits of the output that
94 are sometimes the same for one pair and different for another pair.
96 * pairs that differed by one bit, by two bits, in any combination
97 of top bits of (a,b,c), or in any combination of bottom bits of
99 * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
100 the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
101 is commonly produced by subtraction) look like a single 1-bit
103 * the base values were pseudorandom, all zero but one bit set, or
104 all zero plus a counter that starts at zero.
106 Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
111 Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
112 for "differ" defined as + with a one-bit base and a two-bit delta. I
113 used http://burtleburtle.net/bob/hash/avalanche.html to choose
114 the operations, constants, and arrangements of the variables.
116 This does not achieve avalanche. There are input bits of (a,b,c)
117 that fail to affect some output bits of (a,b,c), especially of a. The
118 most thoroughly mixed value is c, but it doesn't really even achieve
121 This allows some parallelism. Read-after-writes are good at doubling
122 the number of bits affected, so the goal of mixing pulls in the opposite
123 direction as the goal of parallelism. I did what I could. Rotates
124 seem to cost as much as shifts on every machine I could lay my hands
125 on, and rotates are much kinder to the top and bottom bits, so I used
127 -------------------------------------------------------------------------------
131 a -= c; a ^= rot(c, 4); c += b; \
132 b -= a; b ^= rot(a, 6); a += c; \
133 c -= b; c ^= rot(b, 8); b += a; \
134 a -= c; a ^= rot(c,16); c += b; \
135 b -= a; b ^= rot(a,19); a += c; \
136 c -= b; c ^= rot(b, 4); b += a; \
140 -------------------------------------------------------------------------------
141 final -- final mixing of 3 32-bit values (a,b,c) into c
143 Pairs of (a,b,c) values differing in only a few bits will usually
144 produce values of c that look totally different. This was tested for
145 * pairs that differed by one bit, by two bits, in any combination
146 of top bits of (a,b,c), or in any combination of bottom bits of
148 * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
149 the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
150 is commonly produced by subtraction) look like a single 1-bit
152 * the base values were pseudorandom, all zero but one bit set, or
153 all zero plus a counter that starts at zero.
155 These constants passed:
158 and these came close:
162 -------------------------------------------------------------------------------
164 #define final(a,b,c) \
166 c ^= b; c -= rot(b,14); \
167 a ^= c; a -= rot(c,11); \
168 b ^= a; b -= rot(a,25); \
169 c ^= b; c -= rot(b,16); \
170 a ^= c; a -= rot(c,4); \
171 b ^= a; b -= rot(a,14); \
172 c ^= b; c -= rot(b,24); \
176 --------------------------------------------------------------------
177 This works on all machines. To be useful, it requires
178 -- that the key be an array of uint32_t's, and
179 -- that the length be the number of uint32_t's in the key
181 The function hash_word() is identical to hashlittle() on little-endian
182 machines, and identical to hashbig() on big-endian machines,
183 except that the length has to be measured in uint32_ts rather than in
184 bytes. hashlittle() is more complicated than hash_word() only because
185 hashlittle() has to dance around fitting the key bytes into registers.
186 --------------------------------------------------------------------
189 const uint32_t *k, /* the key, an array of uint32_t values */
190 size_t length, /* the length of the key, in uint32_ts */
191 uint32_t initval) /* the previous hash, or an arbitrary value */
195 /* Set up the internal state */
196 a = b = c = 0xdeadbeef + (((uint32_t)length)<<2) + initval;
198 /*------------------------------------------------- handle most of the key */
209 /*------------------------------------------- handle the last 3 uint32_t's */
210 switch(length) /* all the case statements fall through */
216 case 0: /* case 0: nothing left to add */
219 /*------------------------------------------------------ report the result */
224 -------------------------------------------------------------------------------
225 hashlittle() -- hash a variable-length key into a 32-bit value
226 k : the key (the unaligned variable-length array of bytes)
227 length : the length of the key, counting by bytes
228 val2 : IN: can be any 4-byte value OUT: second 32 bit hash.
229 Returns a 32-bit value. Every bit of the key affects every bit of
230 the return value. Two keys differing by one or two bits will have
231 totally different hash values. Note that the return value is better
232 mixed than val2, so use that first.
234 The best hash table sizes are powers of 2. There is no need to do
235 mod a prime (mod is sooo slow!). If you need less than 32 bits,
236 use a bitmask. For example, if you need only 10 bits, do
237 h = (h & hashmask(10));
238 In which case, the hash table should have hashsize(10) elements.
240 If you are hashing n strings (uint8_t **)k, do it like this:
241 for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h);
243 By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this
244 code any way you wish, private, educational, or commercial. It's free.
246 Use for hash table lookup, or anything where one collision in 2^^32 is
247 acceptable. Do NOT use for cryptographic purposes.
248 -------------------------------------------------------------------------------
251 static uint32_t hashlittle( const void *key, size_t length, uint32_t *val2 )
253 uint32_t a,b,c; /* internal state */
254 union { const void *ptr; size_t i; } u; /* needed for Mac Powerbook G4 */
256 /* Set up the internal state */
257 a = b = c = 0xdeadbeef + ((uint32_t)length) + *val2;
260 if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
261 const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */
266 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
277 /*----------------------------- handle the last (probably partial) block */
279 * "k[2]&0xffffff" actually reads beyond the end of the string, but
280 * then masks off the part it's not allowed to read. Because the
281 * string is aligned, the masked-off tail is in the same word as the
282 * rest of the string. Every machine with memory protection I've seen
283 * does it on word boundaries, so is OK with this. But VALGRIND will
284 * still catch it and complain. The masking trick does make the hash
285 * noticably faster for short strings (like English words).
291 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
292 case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
293 case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break;
294 case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break;
295 case 8 : b+=k[1]; a+=k[0]; break;
296 case 7 : b+=k[1]&0xffffff; a+=k[0]; break;
297 case 6 : b+=k[1]&0xffff; a+=k[0]; break;
298 case 5 : b+=k[1]&0xff; a+=k[0]; break;
299 case 4 : a+=k[0]; break;
300 case 3 : a+=k[0]&0xffffff; break;
301 case 2 : a+=k[0]&0xffff; break;
302 case 1 : a+=k[0]&0xff; break;
303 case 0 : return c; /* zero length strings require no mixing */
306 #else /* make valgrind happy */
308 k8 = (const uint8_t *)k;
311 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
312 case 11: c+=((uint32_t)k8[10])<<16; /* fall through */
313 case 10: c+=((uint32_t)k8[9])<<8; /* fall through */
314 case 9 : c+=k8[8]; /* fall through */
315 case 8 : b+=k[1]; a+=k[0]; break;
316 case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */
317 case 6 : b+=((uint32_t)k8[5])<<8; /* fall through */
318 case 5 : b+=k8[4]; /* fall through */
319 case 4 : a+=k[0]; break;
320 case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */
321 case 2 : a+=((uint32_t)k8[1])<<8; /* fall through */
322 case 1 : a+=k8[0]; break;
326 #endif /* !valgrind */
328 } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
329 const uint16_t *k = (const uint16_t *)key; /* read 16-bit chunks */
332 /*--------------- all but last block: aligned reads and different mixing */
335 a += k[0] + (((uint32_t)k[1])<<16);
336 b += k[2] + (((uint32_t)k[3])<<16);
337 c += k[4] + (((uint32_t)k[5])<<16);
343 /*----------------------------- handle the last (probably partial) block */
344 k8 = (const uint8_t *)k;
347 case 12: c+=k[4]+(((uint32_t)k[5])<<16);
348 b+=k[2]+(((uint32_t)k[3])<<16);
349 a+=k[0]+(((uint32_t)k[1])<<16);
351 case 11: c+=((uint32_t)k8[10])<<16; /* fall through */
353 b+=k[2]+(((uint32_t)k[3])<<16);
354 a+=k[0]+(((uint32_t)k[1])<<16);
356 case 9 : c+=k8[8]; /* fall through */
357 case 8 : b+=k[2]+(((uint32_t)k[3])<<16);
358 a+=k[0]+(((uint32_t)k[1])<<16);
360 case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */
362 a+=k[0]+(((uint32_t)k[1])<<16);
364 case 5 : b+=k8[4]; /* fall through */
365 case 4 : a+=k[0]+(((uint32_t)k[1])<<16);
367 case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */
372 case 0 : return c; /* zero length requires no mixing */
375 } else { /* need to read the key one byte at a time */
376 const uint8_t *k = (const uint8_t *)key;
378 /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
382 a += ((uint32_t)k[1])<<8;
383 a += ((uint32_t)k[2])<<16;
384 a += ((uint32_t)k[3])<<24;
386 b += ((uint32_t)k[5])<<8;
387 b += ((uint32_t)k[6])<<16;
388 b += ((uint32_t)k[7])<<24;
390 c += ((uint32_t)k[9])<<8;
391 c += ((uint32_t)k[10])<<16;
392 c += ((uint32_t)k[11])<<24;
398 /*-------------------------------- last block: affect all 32 bits of (c) */
399 switch(length) /* all the case statements fall through */
401 case 12: c+=((uint32_t)k[11])<<24;
402 case 11: c+=((uint32_t)k[10])<<16;
403 case 10: c+=((uint32_t)k[9])<<8;
405 case 8 : b+=((uint32_t)k[7])<<24;
406 case 7 : b+=((uint32_t)k[6])<<16;
407 case 6 : b+=((uint32_t)k[5])<<8;
409 case 4 : a+=((uint32_t)k[3])<<24;
410 case 3 : a+=((uint32_t)k[2])<<16;
411 case 2 : a+=((uint32_t)k[1])<<8;
425 * This is the same as hash_word() on big-endian machines. It is different
426 * from hashlittle() on all machines. hashbig() takes advantage of
427 * big-endian byte ordering.
429 static uint32_t hashbig( const void *key, size_t length, uint32_t *val2)
432 union { const void *ptr; size_t i; } u; /* to cast key to (size_t) happily */
434 /* Set up the internal state */
435 a = b = c = 0xdeadbeef + ((uint32_t)length) + *val2;
438 if (HASH_BIG_ENDIAN && ((u.i & 0x3) == 0)) {
439 const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */
444 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
455 /*----------------------------- handle the last (probably partial) block */
457 * "k[2]<<8" actually reads beyond the end of the string, but
458 * then shifts out the part it's not allowed to read. Because the
459 * string is aligned, the illegal read is in the same word as the
460 * rest of the string. Every machine with memory protection I've seen
461 * does it on word boundaries, so is OK with this. But VALGRIND will
462 * still catch it and complain. The masking trick does make the hash
463 * noticably faster for short strings (like English words).
469 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
470 case 11: c+=k[2]&0xffffff00; b+=k[1]; a+=k[0]; break;
471 case 10: c+=k[2]&0xffff0000; b+=k[1]; a+=k[0]; break;
472 case 9 : c+=k[2]&0xff000000; b+=k[1]; a+=k[0]; break;
473 case 8 : b+=k[1]; a+=k[0]; break;
474 case 7 : b+=k[1]&0xffffff00; a+=k[0]; break;
475 case 6 : b+=k[1]&0xffff0000; a+=k[0]; break;
476 case 5 : b+=k[1]&0xff000000; a+=k[0]; break;
477 case 4 : a+=k[0]; break;
478 case 3 : a+=k[0]&0xffffff00; break;
479 case 2 : a+=k[0]&0xffff0000; break;
480 case 1 : a+=k[0]&0xff000000; break;
481 case 0 : return c; /* zero length strings require no mixing */
484 #else /* make valgrind happy */
486 k8 = (const uint8_t *)k;
487 switch(length) /* all the case statements fall through */
489 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
490 case 11: c+=((uint32_t)k8[10])<<8; /* fall through */
491 case 10: c+=((uint32_t)k8[9])<<16; /* fall through */
492 case 9 : c+=((uint32_t)k8[8])<<24; /* fall through */
493 case 8 : b+=k[1]; a+=k[0]; break;
494 case 7 : b+=((uint32_t)k8[6])<<8; /* fall through */
495 case 6 : b+=((uint32_t)k8[5])<<16; /* fall through */
496 case 5 : b+=((uint32_t)k8[4])<<24; /* fall through */
497 case 4 : a+=k[0]; break;
498 case 3 : a+=((uint32_t)k8[2])<<8; /* fall through */
499 case 2 : a+=((uint32_t)k8[1])<<16; /* fall through */
500 case 1 : a+=((uint32_t)k8[0])<<24; break;
504 #endif /* !VALGRIND */
506 } else { /* need to read the key one byte at a time */
507 const uint8_t *k = (const uint8_t *)key;
509 /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
512 a += ((uint32_t)k[0])<<24;
513 a += ((uint32_t)k[1])<<16;
514 a += ((uint32_t)k[2])<<8;
515 a += ((uint32_t)k[3]);
516 b += ((uint32_t)k[4])<<24;
517 b += ((uint32_t)k[5])<<16;
518 b += ((uint32_t)k[6])<<8;
519 b += ((uint32_t)k[7]);
520 c += ((uint32_t)k[8])<<24;
521 c += ((uint32_t)k[9])<<16;
522 c += ((uint32_t)k[10])<<8;
523 c += ((uint32_t)k[11]);
529 /*-------------------------------- last block: affect all 32 bits of (c) */
530 switch(length) /* all the case statements fall through */
533 case 11: c+=((uint32_t)k[10])<<8;
534 case 10: c+=((uint32_t)k[9])<<16;
535 case 9 : c+=((uint32_t)k[8])<<24;
537 case 7 : b+=((uint32_t)k[6])<<8;
538 case 6 : b+=((uint32_t)k[5])<<16;
539 case 5 : b+=((uint32_t)k[4])<<24;
541 case 3 : a+=((uint32_t)k[2])<<8;
542 case 2 : a+=((uint32_t)k[1])<<16;
543 case 1 : a+=((uint32_t)k[0])<<24;
554 /* I basically use hashlittle here, but use native endian within each
555 * element. This delivers least-surprise: hash such as "int arr[] = {
556 * 1, 2 }; hash_stable(arr, 2, 0);" will be the same on big and little
557 * endian machines, even though a bytewise hash wouldn't be. */
558 uint64_t hash64_stable_64(const void *key, size_t n, uint64_t base)
560 const uint64_t *k = key;
563 /* Set up the internal state */
564 a = b = c = 0xdeadbeef + ((uint32_t)n*8) + (base >> 32) + base;
568 b += (uint32_t)(k[0] >> 32);
571 a += (uint32_t)(k[1] >> 32);
573 c += (uint32_t)(k[2] >> 32);
581 b += (uint32_t)(k[0] >> 32);
584 a += (uint32_t)(k[1] >> 32);
588 b += (uint32_t)(k[0] >> 32);
594 return ((uint64_t)b << 32) | c;
597 uint64_t hash64_stable_32(const void *key, size_t n, uint64_t base)
599 const uint32_t *k = key;
602 /* Set up the internal state */
603 a = b = c = 0xdeadbeef + ((uint32_t)n*4) + (base >> 32) + base;
624 return ((uint64_t)b << 32) | c;
627 uint64_t hash64_stable_16(const void *key, size_t n, uint64_t base)
629 const uint16_t *k = key;
632 /* Set up the internal state */
633 a = b = c = 0xdeadbeef + ((uint32_t)n*2) + (base >> 32) + base;
636 a += (uint32_t)k[0] + ((uint32_t)k[1] << 16);
637 b += (uint32_t)k[2] + ((uint32_t)k[3] << 16);
638 c += (uint32_t)k[4] + ((uint32_t)k[5] << 16);
649 b += ((uint32_t)k[3] << 16);
653 a += ((uint32_t)k[1] << 16);
661 return ((uint64_t)b << 32) | c;
664 uint64_t hash64_stable_8(const void *key, size_t n, uint64_t base)
666 uint32_t b32 = base + (base >> 32);
667 uint32_t lower = hashlittle(key, n, &b32);
669 return ((uint64_t)b32 << 32) | lower;
672 uint32_t hash_any(const void *key, size_t length, uint32_t base)
675 return hashbig(key, length, &base);
677 return hashlittle(key, length, &base);
680 uint32_t hash_stable_64(const void *key, size_t n, uint32_t base)
682 return hash64_stable_64(key, n, base);
685 uint32_t hash_stable_32(const void *key, size_t n, uint32_t base)
687 return hash64_stable_32(key, n, base);
690 uint32_t hash_stable_16(const void *key, size_t n, uint32_t base)
692 return hash64_stable_16(key, n, base);
695 uint32_t hash_stable_8(const void *key, size_t n, uint32_t base)
697 return hashlittle(key, n, &base);
700 /* Jenkins' lookup8 is a 64 bit hash, but he says it's obsolete. Use
701 * the plain one and recombine into 64 bits. */
702 uint64_t hash64_any(const void *key, size_t length, uint64_t base)
704 uint32_t b32 = base + (base >> 32);
708 lower = hashbig(key, length, &b32);
710 lower = hashlittle(key, length, &b32);
712 return ((uint64_t)b32 << 32) | lower;
717 /* used for timings */
726 for (i=0; i<256; ++i) buf[i] = 'x';
729 h = hashlittle(&buf[0],1,h);
732 if (z-a > 0) printf("time %d %.8x\n", z-a, h);
735 /* check that every input bit changes every output bit half the time */
742 uint8_t qa[MAXLEN+1], qb[MAXLEN+2], *a = &qa[0], *b = &qb[1];
743 uint32_t c[HASHSTATE], d[HASHSTATE], i=0, j=0, k, l, m=0, z;
744 uint32_t e[HASHSTATE],f[HASHSTATE],g[HASHSTATE],h[HASHSTATE];
745 uint32_t x[HASHSTATE],y[HASHSTATE];
748 printf("No more than %d trials should ever be needed \n",MAXPAIR/2);
749 for (hlen=0; hlen < MAXLEN; ++hlen)
752 for (i=0; i<hlen; ++i) /*----------------------- for each input byte, */
754 for (j=0; j<8; ++j) /*------------------------ for each input bit, */
756 for (m=1; m<8; ++m) /*------------ for several possible initvals, */
758 for (l=0; l<HASHSTATE; ++l)
759 e[l]=f[l]=g[l]=h[l]=x[l]=y[l]=~((uint32_t)0);
761 /*---- check that every output bit is affected by that input bit */
762 for (k=0; k<MAXPAIR; k+=2)
765 /* keys have one bit different */
766 for (l=0; l<hlen+1; ++l) {a[l] = b[l] = (uint8_t)0;}
767 /* have a and b be two keys differing in only one bit */
770 c[0] = hashlittle(a, hlen, m);
772 b[i] ^= ((k+1)>>(8-j));
773 d[0] = hashlittle(b, hlen, m);
774 /* check every bit is 1, 0, set, and not set at least once */
775 for (l=0; l<HASHSTATE; ++l)
778 f[l] &= ~(c[l]^d[l]);
783 if (e[l]|f[l]|g[l]|h[l]|x[l]|y[l]) finished=0;
790 printf("Some bit didn't change: ");
791 printf("%.8x %.8x %.8x %.8x %.8x %.8x ",
792 e[0],f[0],g[0],h[0],x[0],y[0]);
793 printf("i %d j %d m %d len %d\n", i, j, m, hlen);
795 if (z==MAXPAIR) goto done;
802 printf("Mix success %2d bytes %2d initvals ",i,m);
803 printf("required %d trials\n", z/2);
809 /* Check for reading beyond the end of the buffer and alignment problems */
812 uint8_t buf[MAXLEN+20], *b;
814 uint8_t q[] = "This is the time for all good men to come to the aid of their country...";
816 uint8_t qq[] = "xThis is the time for all good men to come to the aid of their country...";
818 uint8_t qqq[] = "xxThis is the time for all good men to come to the aid of their country...";
820 uint8_t qqqq[] = "xxxThis is the time for all good men to come to the aid of their country...";
824 printf("Endianness. These lines should all be the same (for values filled in):\n");
825 printf("%.8x %.8x %.8x\n",
826 hash_word((const uint32_t *)q, (sizeof(q)-1)/4, 13),
827 hash_word((const uint32_t *)q, (sizeof(q)-5)/4, 13),
828 hash_word((const uint32_t *)q, (sizeof(q)-9)/4, 13));
830 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
831 hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13),
832 hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13),
833 hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13),
834 hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13),
835 hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13),
836 hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13));
838 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
839 hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13),
840 hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13),
841 hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13),
842 hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13),
843 hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13),
844 hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13));
846 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
847 hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13),
848 hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13),
849 hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13),
850 hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13),
851 hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13),
852 hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13));
854 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
855 hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13),
856 hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13),
857 hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13),
858 hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13),
859 hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13),
860 hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13));
863 /* check that hashlittle2 and hashlittle produce the same results */
865 hashlittle2(q, sizeof(q), &i, &j);
866 if (hashlittle(q, sizeof(q), 47) != i)
867 printf("hashlittle2 and hashlittle mismatch\n");
869 /* check that hash_word2 and hash_word produce the same results */
872 hash_word2(&len, 1, &i, &j);
873 if (hash_word(&len, 1, 47) != i)
874 printf("hash_word2 and hash_word mismatch %x %x\n",
875 i, hash_word(&len, 1, 47));
877 /* check hashlittle doesn't read before or after the ends of the string */
878 for (h=0, b=buf+1; h<8; ++h, ++b)
880 for (i=0; i<MAXLEN; ++i)
883 for (j=0; j<i; ++j) *(b+j)=0;
885 /* these should all be equal */
886 ref = hashlittle(b, len, (uint32_t)1);
889 x = hashlittle(b, len, (uint32_t)1);
890 y = hashlittle(b, len, (uint32_t)1);
891 if ((ref != x) || (ref != y))
893 printf("alignment error: %.8x %.8x %.8x %d %d\n",ref,x,y,
900 /* check for problems with nulls */
904 uint32_t h,i,state[HASHSTATE];
908 for (i=0; i<HASHSTATE; ++i) state[i] = 1;
909 printf("These should all be different\n");
910 for (i=0, h=0; i<8; ++i)
912 h = hashlittle(buf, 0, h);
913 printf("%2ld 0-byte strings, hash is %.8x\n", i, h);
920 driver1(); /* test that the key is hashed: used for timings */
921 driver2(); /* test that whole key is hashed thoroughly */
922 driver3(); /* test that nothing but the key is hashed */
923 driver4(); /* test hashing multiple buffers (all buffers are null) */
927 #endif /* SELF_TEST */