1 /* -------------------------------------------------------------------- */
2 /*
3 * lookup3.c, by Bob Jenkins, May 2006, Public Domain.
4 *
5 * These are functions for producing 32-bit hashes for hash table lookup.
6 * jlu32w(), jlu32l(), jlu32lpair(), jlu32b(), _JLU3_MIX(), and _JLU3_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.
10 *
11 * You probably want to use jlu32l(). jlu32l() and jlu32b()
12 * hash byte arrays. jlu32l() is is faster than jlu32b() on
13 * little-endian machines. Intel and AMD are little-endian machines.
14 * On second thought, you probably want jlu32lpair(), which is identical to
15 * jlu32l() except it returns two 32-bit hashes for the price of one.
16 * You could implement jlu32bpair() if you wanted but I haven't bothered here.
17 *
18 * If you want to find a hash of, say, exactly 7 integers, do
19 * a = i1; b = i2; c = i3;
20 * _JLU3_MIX(a,b,c);
21 * a += i4; b += i5; c += i6;
22 * _JLU3_MIX(a,b,c);
23 * a += i7;
24 * _JLU3_FINAL(a,b,c);
25 * then use c as the hash value. If you have a variable size array of
26 * 4-byte integers to hash, use jlu32w(). If you have a byte array (like
27 * a character string), use jlu32l(). If you have several byte arrays, or
28 * a mix of things, see the comments above jlu32l().
29 *
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 */
35 /* -------------------------------------------------------------------- */
36
37 #include <stdint.h>
38
39 #if defined(_JLU3_SELFTEST)
40 # define _JLU3_jlu32w 1
41 # define _JLU3_jlu32l 1
42 # define _JLU3_jlu32lpair 1
43 # define _JLU3_jlu32b 1
44 #endif
45
46 static const union _dbswap {
47 const uint32_t ui;
48 const unsigned char uc[4];
49 } endian = { .ui = 0x11223344 };
50 # define HASH_LITTLE_ENDIAN (endian.uc[0] == (unsigned char) 0x44)
51 # define HASH_BIG_ENDIAN (endian.uc[0] == (unsigned char) 0x11)
52
53 #ifndef ROTL32
54 # define ROTL32(x, s) (((x) << (s)) | ((x) >> (32 - (s))))
55 #endif
56
57 /* NOTE: The _size parameter should be in bytes. */
58 #define _JLU3_INIT(_h, _size) (0xdeadbeef + ((uint32_t)(_size)) + (_h))
59
60 /* -------------------------------------------------------------------- */
61 /*
62 * _JLU3_MIX -- mix 3 32-bit values reversibly.
63 *
64 * This is reversible, so any information in (a,b,c) before _JLU3_MIX() is
65 * still in (a,b,c) after _JLU3_MIX().
66 *
67 * If four pairs of (a,b,c) inputs are run through _JLU3_MIX(), or through
68 * _JLU3_MIX() in reverse, there are at least 32 bits of the output that
69 * are sometimes the same for one pair and different for another pair.
70 * This was tested for:
71 * * pairs that differed by one bit, by two bits, in any combination
72 * of top bits of (a,b,c), or in any combination of bottom bits of
73 * (a,b,c).
74 * * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
75 * the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
76 * is commonly produced by subtraction) look like a single 1-bit
77 * difference.
78 * * the base values were pseudorandom, all zero but one bit set, or
79 * all zero plus a counter that starts at zero.
80 *
81 * Some k values for my "a-=c; a^=ROTL32(c,k); c+=b;" arrangement that
82 * satisfy this are
83 * 4 6 8 16 19 4
84 * 9 15 3 18 27 15
85 * 14 9 3 7 17 3
86 * Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
87 * for "differ" defined as + with a one-bit base and a two-bit delta. I
88 * used http://burtleburtle.net/bob/hash/avalanche.html to choose
89 * the operations, constants, and arrangements of the variables.
90 *
91 * This does not achieve avalanche. There are input bits of (a,b,c)
92 * that fail to affect some output bits of (a,b,c), especially of a. The
93 * most thoroughly mixed value is c, but it doesn't really even achieve
94 * avalanche in c.
95 *
96 * This allows some parallelism. Read-after-writes are good at doubling
97 * the number of bits affected, so the goal of mixing pulls in the opposite
98 * direction as the goal of parallelism. I did what I could. Rotates
99 * seem to cost as much as shifts on every machine I could lay my hands
100 * on, and rotates are much kinder to the top and bottom bits, so I used
101 * rotates.
102 */
103 /* -------------------------------------------------------------------- */
104 #define _JLU3_MIX(a,b,c) \
105 { \
106 a -= c; a ^= ROTL32(c, 4); c += b; \
107 b -= a; b ^= ROTL32(a, 6); a += c; \
108 c -= b; c ^= ROTL32(b, 8); b += a; \
109 a -= c; a ^= ROTL32(c,16); c += b; \
110 b -= a; b ^= ROTL32(a,19); a += c; \
111 c -= b; c ^= ROTL32(b, 4); b += a; \
112 }
113
114 /* -------------------------------------------------------------------- */
115 /**
116 * _JLU3_FINAL -- final mixing of 3 32-bit values (a,b,c) into c
117 *
118 * Pairs of (a,b,c) values differing in only a few bits will usually
119 * produce values of c that look totally different. This was tested for
120 * * pairs that differed by one bit, by two bits, in any combination
121 * of top bits of (a,b,c), or in any combination of bottom bits of
122 * (a,b,c).
123 * * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
124 * the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
125 * is commonly produced by subtraction) look like a single 1-bit
126 * difference.
127 * * the base values were pseudorandom, all zero but one bit set, or
128 * all zero plus a counter that starts at zero.
129 *
130 * These constants passed:
131 * 14 11 25 16 4 14 24
132 * 12 14 25 16 4 14 24
133 * and these came close:
134 * 4 8 15 26 3 22 24
135 * 10 8 15 26 3 22 24
136 * 11 8 15 26 3 22 24
137 */
138 /* -------------------------------------------------------------------- */
139 #define _JLU3_FINAL(a,b,c) \
140 { \
141 c ^= b; c -= ROTL32(b,14); \
142 a ^= c; a -= ROTL32(c,11); \
143 b ^= a; b -= ROTL32(a,25); \
144 c ^= b; c -= ROTL32(b,16); \
145 a ^= c; a -= ROTL32(c,4); \
146 b ^= a; b -= ROTL32(a,14); \
147 c ^= b; c -= ROTL32(b,24); \
148 }
149
150 #if defined(_JLU3_jlu32w)
151 uint32_t jlu32w(uint32_t h, const uint32_t *k, size_t size);
152 /* -------------------------------------------------------------------- */
153 /**
154 * This works on all machines. To be useful, it requires
155 * -- that the key be an array of uint32_t's, and
156 * -- that the size be the number of uint32_t's in the key
157 *
158 * The function jlu32w() is identical to jlu32l() on little-endian
159 * machines, and identical to jlu32b() on big-endian machines,
160 * except that the size has to be measured in uint32_ts rather than in
161 * bytes. jlu32l() is more complicated than jlu32w() only because
162 * jlu32l() has to dance around fitting the key bytes into registers.
163 *
164 * @param h the previous hash, or an arbitrary value
165 * @param *k the key, an array of uint32_t values
166 * @param size the size of the key, in uint32_ts
167 * @return the lookup3 hash
168 */
169 /* -------------------------------------------------------------------- */
jlu32w(uint32_t h,const uint32_t * k,size_t size)170 uint32_t jlu32w(uint32_t h, const uint32_t *k, size_t size)
171 {
172 uint32_t a = _JLU3_INIT(h, (size * sizeof(*k)));
173 uint32_t b = a;
174 uint32_t c = a;
175
176 if (k == NULL)
177 goto exit;
178
179 /*----------------------------------------------- handle most of the key */
180 while (size > 3) {
181 a += k[0];
182 b += k[1];
183 c += k[2];
184 _JLU3_MIX(a,b,c);
185 size -= 3;
186 k += 3;
187 }
188
189 /*----------------------------------------- handle the last 3 uint32_t's */
190 switch (size) {
191 case 3 : c+=k[2];
192 case 2 : b+=k[1];
193 case 1 : a+=k[0];
194 _JLU3_FINAL(a,b,c);
195 /* fallthrough */
196 case 0:
197 break;
198 }
199 /*---------------------------------------------------- report the result */
200 exit:
201 return c;
202 }
203 #endif /* defined(_JLU3_jlu32w) */
204
205 #if defined(_JLU3_jlu32l)
206 uint32_t jlu32l(uint32_t h, const void *key, size_t size);
207 /* -------------------------------------------------------------------- */
208 /*
209 * jlu32l() -- hash a variable-length key into a 32-bit value
210 * h : can be any 4-byte value
211 * k : the key (the unaligned variable-length array of bytes)
212 * size : the size of the key, counting by bytes
213 * Returns a 32-bit value. Every bit of the key affects every bit of
214 * the return value. Two keys differing by one or two bits will have
215 * totally different hash values.
216 *
217 * The best hash table sizes are powers of 2. There is no need to do
218 * mod a prime (mod is sooo slow!). If you need less than 32 bits,
219 * use a bitmask. For example, if you need only 10 bits, do
220 * h = (h & hashmask(10));
221 * In which case, the hash table should have hashsize(10) elements.
222 *
223 * If you are hashing n strings (uint8_t **)k, do it like this:
224 * for (i=0, h=0; i<n; ++i) h = jlu32l(h, k[i], len[i]);
225 *
226 * By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this
227 * code any way you wish, private, educational, or commercial. It's free.
228 *
229 * Use for hash table lookup, or anything where one collision in 2^^32 is
230 * acceptable. Do NOT use for cryptographic purposes.
231 *
232 * @param h the previous hash, or an arbitrary value
233 * @param *k the key, an array of uint8_t values
234 * @param size the size of the key
235 * @return the lookup3 hash
236 */
237 /* -------------------------------------------------------------------- */
jlu32l(uint32_t h,const void * key,size_t size)238 uint32_t jlu32l(uint32_t h, const void *key, size_t size)
239 {
240 union { const void *ptr; size_t i; } u;
241 uint32_t a = _JLU3_INIT(h, size);
242 uint32_t b = a;
243 uint32_t c = a;
244
245 if (key == NULL)
246 goto exit;
247
248 u.ptr = key;
249 if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
250 const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */
251 #ifdef VALGRIND
252 const uint8_t *k8;
253 #endif
254
255 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
256 while (size > 12) {
257 a += k[0];
258 b += k[1];
259 c += k[2];
260 _JLU3_MIX(a,b,c);
261 size -= 12;
262 k += 3;
263 }
264
265 /*------------------------- handle the last (probably partial) block */
266 /*
267 * "k[2]&0xffffff" actually reads beyond the end of the string, but
268 * then masks off the part it's not allowed to read. Because the
269 * string is aligned, the masked-off tail is in the same word as the
270 * rest of the string. Every machine with memory protection I've seen
271 * does it on word boundaries, so is OK with this. But VALGRIND will
272 * still catch it and complain. The masking trick does make the hash
273 * noticably faster for short strings (like English words).
274 */
275 #ifndef VALGRIND
276
277 switch (size) {
278 case 12: c += k[2]; b+=k[1]; a+=k[0]; break;
279 case 11: c += k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
280 case 10: c += k[2]&0xffff; b+=k[1]; a+=k[0]; break;
281 case 9: c += k[2]&0xff; b+=k[1]; a+=k[0]; break;
282 case 8: b += k[1]; a+=k[0]; break;
283 case 7: b += k[1]&0xffffff; a+=k[0]; break;
284 case 6: b += k[1]&0xffff; a+=k[0]; break;
285 case 5: b += k[1]&0xff; a+=k[0]; break;
286 case 4: a += k[0]; break;
287 case 3: a += k[0]&0xffffff; break;
288 case 2: a += k[0]&0xffff; break;
289 case 1: a += k[0]&0xff; break;
290 case 0: goto exit;
291 }
292
293 #else /* make valgrind happy */
294
295 k8 = (const uint8_t *)k;
296 switch (size) {
297 case 12: c += k[2]; b+=k[1]; a+=k[0] break;
298 case 11: c += ((uint32_t)k8[10])<<16; /* fallthrough */
299 case 10: c += ((uint32_t)k8[9])<<8; /* fallthrough */
300 case 9: c += k8[8]; /* fallthrough */
301 case 8: b += k[1]; a+=k[0]; break;
302 case 7: b += ((uint32_t)k8[6])<<16; /* fallthrough */
303 case 6: b += ((uint32_t)k8[5])<<8; /* fallthrough */
304 case 5: b += k8[4]; /* fallthrough */
305 case 4: a += k[0]; break;
306 case 3: a += ((uint32_t)k8[2])<<16; /* fallthrough */
307 case 2: a += ((uint32_t)k8[1])<<8; /* fallthrough */
308 case 1: a += k8[0]; break;
309 case 0: goto exit;
310 }
311
312 #endif /* !valgrind */
313
314 } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
315 const uint16_t *k = (const uint16_t *)key; /* read 16-bit chunks */
316 const uint8_t *k8;
317
318 /*----------- all but last block: aligned reads and different mixing */
319 while (size > 12) {
320 a += k[0] + (((uint32_t)k[1])<<16);
321 b += k[2] + (((uint32_t)k[3])<<16);
322 c += k[4] + (((uint32_t)k[5])<<16);
323 _JLU3_MIX(a,b,c);
324 size -= 12;
325 k += 6;
326 }
327
328 /*------------------------- handle the last (probably partial) block */
329 k8 = (const uint8_t *)k;
330 switch (size) {
331 case 12:
332 c += k[4]+(((uint32_t)k[5])<<16);
333 b += k[2]+(((uint32_t)k[3])<<16);
334 a += k[0]+(((uint32_t)k[1])<<16);
335 break;
336 case 11:
337 c += ((uint32_t)k8[10])<<16;
338 /* fallthrough */
339 case 10:
340 c += (uint32_t)k[4];
341 b += k[2]+(((uint32_t)k[3])<<16);
342 a += k[0]+(((uint32_t)k[1])<<16);
343 break;
344 case 9:
345 c += (uint32_t)k8[8];
346 /* fallthrough */
347 case 8:
348 b += k[2]+(((uint32_t)k[3])<<16);
349 a += k[0]+(((uint32_t)k[1])<<16);
350 break;
351 case 7:
352 b += ((uint32_t)k8[6])<<16;
353 /* fallthrough */
354 case 6:
355 b += (uint32_t)k[2];
356 a += k[0]+(((uint32_t)k[1])<<16);
357 break;
358 case 5:
359 b += (uint32_t)k8[4];
360 /* fallthrough */
361 case 4:
362 a += k[0]+(((uint32_t)k[1])<<16);
363 break;
364 case 3:
365 a += ((uint32_t)k8[2])<<16;
366 /* fallthrough */
367 case 2:
368 a += (uint32_t)k[0];
369 break;
370 case 1:
371 a += (uint32_t)k8[0];
372 break;
373 case 0:
374 goto exit;
375 }
376
377 } else { /* need to read the key one byte at a time */
378 const uint8_t *k = (const uint8_t *)key;
379
380 /*----------- all but the last block: affect some 32 bits of (a,b,c) */
381 while (size > 12) {
382 a += (uint32_t)k[0];
383 a += ((uint32_t)k[1])<<8;
384 a += ((uint32_t)k[2])<<16;
385 a += ((uint32_t)k[3])<<24;
386 b += (uint32_t)k[4];
387 b += ((uint32_t)k[5])<<8;
388 b += ((uint32_t)k[6])<<16;
389 b += ((uint32_t)k[7])<<24;
390 c += (uint32_t)k[8];
391 c += ((uint32_t)k[9])<<8;
392 c += ((uint32_t)k[10])<<16;
393 c += ((uint32_t)k[11])<<24;
394 _JLU3_MIX(a,b,c);
395 size -= 12;
396 k += 12;
397 }
398
399 /*---------------------------- last block: affect all 32 bits of (c) */
400 switch (size) {
401 case 12: c += ((uint32_t)k[11])<<24; /* fallthrough */
402 case 11: c += ((uint32_t)k[10])<<16; /* fallthrough */
403 case 10: c += ((uint32_t)k[9])<<8; /* fallthrough */
404 case 9: c += (uint32_t)k[8]; /* fallthrough */
405 case 8: b += ((uint32_t)k[7])<<24; /* fallthrough */
406 case 7: b += ((uint32_t)k[6])<<16; /* fallthrough */
407 case 6: b += ((uint32_t)k[5])<<8; /* fallthrough */
408 case 5: b += (uint32_t)k[4]; /* fallthrough */
409 case 4: a += ((uint32_t)k[3])<<24; /* fallthrough */
410 case 3: a += ((uint32_t)k[2])<<16; /* fallthrough */
411 case 2: a += ((uint32_t)k[1])<<8; /* fallthrough */
412 case 1: a += (uint32_t)k[0];
413 break;
414 case 0:
415 goto exit;
416 }
417 }
418
419 _JLU3_FINAL(a,b,c);
420
421 exit:
422 return c;
423 }
424 #endif /* defined(_JLU3_jlu32l) */
425
426 #if defined(_JLU3_jlu32lpair)
427 /**
428 * jlu32lpair: return 2 32-bit hash values.
429 *
430 * This is identical to jlu32l(), except it returns two 32-bit hash
431 * values instead of just one. This is good enough for hash table
432 * lookup with 2^^64 buckets, or if you want a second hash if you're not
433 * happy with the first, or if you want a probably-unique 64-bit ID for
434 * the key. *pc is better mixed than *pb, so use *pc first. If you want
435 * a 64-bit value do something like "*pc + (((uint64_t)*pb)<<32)".
436 *
437 * @param h the previous hash, or an arbitrary value
438 * @param *key the key, an array of uint8_t values
439 * @param size the size of the key in bytes
440 * @retval *pc, IN: primary initval, OUT: primary hash
441 * *retval *pb IN: secondary initval, OUT: secondary hash
442 */
jlu32lpair(const void * key,size_t size,uint32_t * pc,uint32_t * pb)443 void jlu32lpair(const void *key, size_t size, uint32_t *pc, uint32_t *pb)
444 {
445 union { const void *ptr; size_t i; } u;
446 uint32_t a = _JLU3_INIT(*pc, size);
447 uint32_t b = a;
448 uint32_t c = a;
449
450 if (key == NULL)
451 goto exit;
452
453 c += *pb; /* Add the secondary hash. */
454
455 u.ptr = key;
456 if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
457 const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */
458 #ifdef VALGRIND
459 const uint8_t *k8;
460 #endif
461
462 /*-- all but last block: aligned reads and affect 32 bits of (a,b,c) */
463 while (size > (size_t)12) {
464 a += k[0];
465 b += k[1];
466 c += k[2];
467 _JLU3_MIX(a,b,c);
468 size -= 12;
469 k += 3;
470 }
471 /*------------------------- handle the last (probably partial) block */
472 /*
473 * "k[2]&0xffffff" actually reads beyond the end of the string, but
474 * then masks off the part it's not allowed to read. Because the
475 * string is aligned, the masked-off tail is in the same word as the
476 * rest of the string. Every machine with memory protection I've seen
477 * does it on word boundaries, so is OK with this. But VALGRIND will
478 * still catch it and complain. The masking trick does make the hash
479 * noticably faster for short strings (like English words).
480 */
481 #ifndef VALGRIND
482
483 switch (size) {
484 case 12: c += k[2]; b+=k[1]; a+=k[0]; break;
485 case 11: c += k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
486 case 10: c += k[2]&0xffff; b+=k[1]; a+=k[0]; break;
487 case 9: c += k[2]&0xff; b+=k[1]; a+=k[0]; break;
488 case 8: b += k[1]; a+=k[0]; break;
489 case 7: b += k[1]&0xffffff; a+=k[0]; break;
490 case 6: b += k[1]&0xffff; a+=k[0]; break;
491 case 5: b += k[1]&0xff; a+=k[0]; break;
492 case 4: a += k[0]; break;
493 case 3: a += k[0]&0xffffff; break;
494 case 2: a += k[0]&0xffff; break;
495 case 1: a += k[0]&0xff; break;
496 case 0: goto exit;
497 }
498
499 #else /* make valgrind happy */
500
501 k8 = (const uint8_t *)k;
502 switch (size) {
503 case 12: c += k[2]; b+=k[1]; a+=k[0]; break;
504 case 11: c += ((uint32_t)k8[10])<<16; /* fallthrough */
505 case 10: c += ((uint32_t)k8[9])<<8; /* fallthrough */
506 case 9: c += k8[8]; /* fallthrough */
507 case 8: b += k[1]; a+=k[0]; break;
508 case 7: b += ((uint32_t)k8[6])<<16; /* fallthrough */
509 case 6: b += ((uint32_t)k8[5])<<8; /* fallthrough */
510 case 5: b += k8[4]; /* fallthrough */
511 case 4: a += k[0]; break;
512 case 3: a += ((uint32_t)k8[2])<<16; /* fallthrough */
513 case 2: a += ((uint32_t)k8[1])<<8; /* fallthrough */
514 case 1: a += k8[0]; break;
515 case 0: goto exit;
516 }
517
518 #endif /* !valgrind */
519
520 } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
521 const uint16_t *k = (const uint16_t *)key; /* read 16-bit chunks */
522 const uint8_t *k8;
523
524 /*----------- all but last block: aligned reads and different mixing */
525 while (size > (size_t)12) {
526 a += k[0] + (((uint32_t)k[1])<<16);
527 b += k[2] + (((uint32_t)k[3])<<16);
528 c += k[4] + (((uint32_t)k[5])<<16);
529 _JLU3_MIX(a,b,c);
530 size -= 12;
531 k += 6;
532 }
533
534 /*------------------------- handle the last (probably partial) block */
535 k8 = (const uint8_t *)k;
536 switch (size) {
537 case 12:
538 c += k[4]+(((uint32_t)k[5])<<16);
539 b += k[2]+(((uint32_t)k[3])<<16);
540 a += k[0]+(((uint32_t)k[1])<<16);
541 break;
542 case 11:
543 c += ((uint32_t)k8[10])<<16;
544 /* fallthrough */
545 case 10:
546 c += k[4];
547 b += k[2]+(((uint32_t)k[3])<<16);
548 a += k[0]+(((uint32_t)k[1])<<16);
549 break;
550 case 9:
551 c += k8[8];
552 /* fallthrough */
553 case 8:
554 b += k[2]+(((uint32_t)k[3])<<16);
555 a += k[0]+(((uint32_t)k[1])<<16);
556 break;
557 case 7:
558 b += ((uint32_t)k8[6])<<16;
559 /* fallthrough */
560 case 6:
561 b += k[2];
562 a += k[0]+(((uint32_t)k[1])<<16);
563 break;
564 case 5:
565 b += k8[4];
566 /* fallthrough */
567 case 4:
568 a += k[0]+(((uint32_t)k[1])<<16);
569 break;
570 case 3:
571 a += ((uint32_t)k8[2])<<16;
572 /* fallthrough */
573 case 2:
574 a += k[0];
575 break;
576 case 1:
577 a += k8[0];
578 break;
579 case 0:
580 goto exit;
581 }
582
583 } else { /* need to read the key one byte at a time */
584 const uint8_t *k = (const uint8_t *)key;
585
586 /*----------- all but the last block: affect some 32 bits of (a,b,c) */
587 while (size > (size_t)12) {
588 a += k[0];
589 a += ((uint32_t)k[1])<<8;
590 a += ((uint32_t)k[2])<<16;
591 a += ((uint32_t)k[3])<<24;
592 b += k[4];
593 b += ((uint32_t)k[5])<<8;
594 b += ((uint32_t)k[6])<<16;
595 b += ((uint32_t)k[7])<<24;
596 c += k[8];
597 c += ((uint32_t)k[9])<<8;
598 c += ((uint32_t)k[10])<<16;
599 c += ((uint32_t)k[11])<<24;
600 _JLU3_MIX(a,b,c);
601 size -= 12;
602 k += 12;
603 }
604
605 /*---------------------------- last block: affect all 32 bits of (c) */
606 switch (size) {
607 case 12: c += ((uint32_t)k[11])<<24; /* fallthrough */
608 case 11: c += ((uint32_t)k[10])<<16; /* fallthrough */
609 case 10: c += ((uint32_t)k[9])<<8; /* fallthrough */
610 case 9: c += k[8]; /* fallthrough */
611 case 8: b += ((uint32_t)k[7])<<24; /* fallthrough */
612 case 7: b += ((uint32_t)k[6])<<16; /* fallthrough */
613 case 6: b += ((uint32_t)k[5])<<8; /* fallthrough */
614 case 5: b += k[4]; /* fallthrough */
615 case 4: a += ((uint32_t)k[3])<<24; /* fallthrough */
616 case 3: a += ((uint32_t)k[2])<<16; /* fallthrough */
617 case 2: a += ((uint32_t)k[1])<<8; /* fallthrough */
618 case 1: a += k[0];
619 break;
620 case 0:
621 goto exit;
622 }
623 }
624
625 _JLU3_FINAL(a,b,c);
626
627 exit:
628 *pc = c;
629 *pb = b;
630 return;
631 }
632 #endif /* defined(_JLU3_jlu32lpair) */
633
634 #if defined(_JLU3_jlu32b)
635 uint32_t jlu32b(uint32_t h, const void *key, size_t size);
636 /*
637 * jlu32b():
638 * This is the same as jlu32w() on big-endian machines. It is different
639 * from jlu32l() on all machines. jlu32b() takes advantage of
640 * big-endian byte ordering.
641 *
642 * @param h the previous hash, or an arbitrary value
643 * @param *k the key, an array of uint8_t values
644 * @param size the size of the key
645 * @return the lookup3 hash
646 */
jlu32b(uint32_t h,const void * key,size_t size)647 uint32_t jlu32b(uint32_t h, const void *key, size_t size)
648 {
649 union { const void *ptr; size_t i; } u;
650 uint32_t a = _JLU3_INIT(h, size);
651 uint32_t b = a;
652 uint32_t c = a;
653
654 if (key == NULL)
655 return h;
656
657 u.ptr = key;
658 if (HASH_BIG_ENDIAN && ((u.i & 0x3) == 0)) {
659 const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */
660 #ifdef VALGRIND
661 const uint8_t *k8;
662 #endif
663
664 /*-- all but last block: aligned reads and affect 32 bits of (a,b,c) */
665 while (size > 12) {
666 a += k[0];
667 b += k[1];
668 c += k[2];
669 _JLU3_MIX(a,b,c);
670 size -= 12;
671 k += 3;
672 }
673
674 /*------------------------- handle the last (probably partial) block */
675 /*
676 * "k[2]<<8" actually reads beyond the end of the string, but
677 * then shifts out the part it's not allowed to read. Because the
678 * string is aligned, the illegal read is in the same word as the
679 * rest of the string. Every machine with memory protection I've seen
680 * does it on word boundaries, so is OK with this. But VALGRIND will
681 * still catch it and complain. The masking trick does make the hash
682 * noticably faster for short strings (like English words).
683 */
684 #ifndef VALGRIND
685
686 switch (size) {
687 case 12: c += k[2]; b+=k[1]; a+=k[0]; break;
688 case 11: c += k[2]&0xffffff00; b+=k[1]; a+=k[0]; break;
689 case 10: c += k[2]&0xffff0000; b+=k[1]; a+=k[0]; break;
690 case 9: c += k[2]&0xff000000; b+=k[1]; a+=k[0]; break;
691 case 8: b += k[1]; a+=k[0]; break;
692 case 7: b += k[1]&0xffffff00; a+=k[0]; break;
693 case 6: b += k[1]&0xffff0000; a+=k[0]; break;
694 case 5: b += k[1]&0xff000000; a+=k[0]; break;
695 case 4: a += k[0]; break;
696 case 3: a += k[0]&0xffffff00; break;
697 case 2: a += k[0]&0xffff0000; break;
698 case 1: a += k[0]&0xff000000; break;
699 case 0: goto exit;
700 }
701
702 #else /* make valgrind happy */
703
704 k8 = (const uint8_t *)k;
705 switch (size) { /* all the case statements fall through */
706 case 12: c += k[2]; b+=k[1]; a+=k[0]; break;
707 case 11: c += ((uint32_t)k8[10])<<8; /* fallthrough */
708 case 10: c += ((uint32_t)k8[9])<<16; /* fallthrough */
709 case 9: c += ((uint32_t)k8[8])<<24; /* fallthrough */
710 case 8: b += k[1]; a+=k[0]; break;
711 case 7: b += ((uint32_t)k8[6])<<8; /* fallthrough */
712 case 6: b += ((uint32_t)k8[5])<<16; /* fallthrough */
713 case 5: b += ((uint32_t)k8[4])<<24; /* fallthrough */
714 case 4: a += k[0]; break;
715 case 3: a += ((uint32_t)k8[2])<<8; /* fallthrough */
716 case 2: a += ((uint32_t)k8[1])<<16; /* fallthrough */
717 case 1: a += ((uint32_t)k8[0])<<24; break;
718 case 0: goto exit;
719 }
720
721 #endif /* !VALGRIND */
722
723 } else { /* need to read the key one byte at a time */
724 const uint8_t *k = (const uint8_t *)key;
725
726 /*----------- all but the last block: affect some 32 bits of (a,b,c) */
727 while (size > 12) {
728 a += ((uint32_t)k[0])<<24;
729 a += ((uint32_t)k[1])<<16;
730 a += ((uint32_t)k[2])<<8;
731 a += ((uint32_t)k[3]);
732 b += ((uint32_t)k[4])<<24;
733 b += ((uint32_t)k[5])<<16;
734 b += ((uint32_t)k[6])<<8;
735 b += ((uint32_t)k[7]);
736 c += ((uint32_t)k[8])<<24;
737 c += ((uint32_t)k[9])<<16;
738 c += ((uint32_t)k[10])<<8;
739 c += ((uint32_t)k[11]);
740 _JLU3_MIX(a,b,c);
741 size -= 12;
742 k += 12;
743 }
744
745 /*---------------------------- last block: affect all 32 bits of (c) */
746 switch (size) { /* all the case statements fall through */
747 case 12: c += k[11]; /* fallthrough */
748 case 11: c += ((uint32_t)k[10])<<8; /* fallthrough */
749 case 10: c += ((uint32_t)k[9])<<16; /* fallthrough */
750 case 9: c += ((uint32_t)k[8])<<24; /* fallthrough */
751 case 8: b += k[7]; /* fallthrough */
752 case 7: b += ((uint32_t)k[6])<<8; /* fallthrough */
753 case 6: b += ((uint32_t)k[5])<<16; /* fallthrough */
754 case 5: b += ((uint32_t)k[4])<<24; /* fallthrough */
755 case 4: a += k[3]; /* fallthrough */
756 case 3: a += ((uint32_t)k[2])<<8; /* fallthrough */
757 case 2: a += ((uint32_t)k[1])<<16; /* fallthrough */
758 case 1: a += ((uint32_t)k[0])<<24; /* fallthrough */
759 break;
760 case 0:
761 goto exit;
762 }
763 }
764
765 _JLU3_FINAL(a,b,c);
766
767 exit:
768 return c;
769 }
770 #endif /* defined(_JLU3_jlu32b) */
771
772 #if defined(_JLU3_SELFTEST)
773
774 /* used for timings */
driver1(void)775 static void driver1(void)
776 {
777 uint8_t buf[256];
778 uint32_t i;
779 uint32_t h=0;
780 time_t a,z;
781
782 time(&a);
783 for (i=0; i<256; ++i) buf[i] = 'x';
784 for (i=0; i<1; ++i) {
785 h = jlu32l(h, &buf[0], sizeof(buf[0]));
786 }
787 time(&z);
788 if (z-a > 0) printf("time %d %.8x\n", (int)(z-a), h);
789 }
790
791 /* check that every input bit changes every output bit half the time */
792 #define HASHSTATE 1
793 #define HASHLEN 1
794 #define MAXPAIR 60
795 #define MAXLEN 70
driver2(void)796 static void driver2(void)
797 {
798 uint8_t qa[MAXLEN+1], qb[MAXLEN+2], *a = &qa[0], *b = &qb[1];
799 uint32_t c[HASHSTATE], d[HASHSTATE], i=0, j=0, k, l, m=0, z;
800 uint32_t e[HASHSTATE],f[HASHSTATE],g[HASHSTATE],h[HASHSTATE];
801 uint32_t x[HASHSTATE],y[HASHSTATE];
802 uint32_t hlen;
803
804 printf("No more than %d trials should ever be needed \n",MAXPAIR/2);
805 for (hlen=0; hlen < MAXLEN; ++hlen) {
806 z=0;
807 for (i=0; i<hlen; ++i) { /*-------------- for each input byte, */
808 for (j=0; j<8; ++j) { /*--------------- for each input bit, */
809 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);
812
813 /* check that every output bit is affected by that input bit */
814 for (k=0; k<MAXPAIR; k+=2) {
815 uint32_t finished=1;
816 /* keys have one bit different */
817 for (l=0; l<hlen+1; ++l) {a[l] = b[l] = (uint8_t)0;}
818 /* have a and b be two keys differing in only one bit */
819 a[i] ^= (k<<j);
820 a[i] ^= (k>>(8-j));
821 c[0] = jlu32l(m, a, hlen);
822 b[i] ^= ((k+1)<<j);
823 b[i] ^= ((k+1)>>(8-j));
824 d[0] = jlu32l(m, b, hlen);
825 /* check every bit is 1, 0, set, and not set at least once */
826 for (l=0; l<HASHSTATE; ++l) {
827 e[l] &= (c[l]^d[l]);
828 f[l] &= ~(c[l]^d[l]);
829 g[l] &= c[l];
830 h[l] &= ~c[l];
831 x[l] &= d[l];
832 y[l] &= ~d[l];
833 if (e[l]|f[l]|g[l]|h[l]|x[l]|y[l]) finished=0;
834 }
835 if (finished) break;
836 }
837 if (k>z) z=k;
838 if (k == MAXPAIR) {
839 printf("Some bit didn't change: ");
840 printf("%.8x %.8x %.8x %.8x %.8x %.8x ",
841 e[0],f[0],g[0],h[0],x[0],y[0]);
842 printf("i %d j %d m %d len %d\n", i, j, m, hlen);
843 }
844 if (z == MAXPAIR) goto done;
845 }
846 }
847 }
848 done:
849 if (z < MAXPAIR) {
850 printf("Mix success %2d bytes %2d initvals ",i,m);
851 printf("required %d trials\n", z/2);
852 }
853 }
854 printf("\n");
855 }
856
857 /* Check for reading beyond the end of the buffer and alignment problems */
driver3(void)858 static void driver3(void)
859 {
860 uint8_t buf[MAXLEN+20], *b;
861 uint32_t len;
862 uint8_t q[] = "This is the time for all good men to come to the aid of their country...";
863 uint32_t h;
864 uint8_t qq[] = "xThis is the time for all good men to come to the aid of their country...";
865 uint32_t i;
866 uint8_t qqq[] = "xxThis is the time for all good men to come to the aid of their country...";
867 uint32_t j;
868 uint8_t qqqq[] = "xxxThis is the time for all good men to come to the aid of their country...";
869 uint32_t ref,x,y;
870 uint8_t *p;
871 uint32_t m = 13;
872
873 printf("Endianness. These lines should all be the same (for values filled in):\n");
874 printf("%.8x %.8x %.8x\n",
875 jlu32w(m, (const uint32_t *)q, (sizeof(q)-1)/4),
876 jlu32w(m, (const uint32_t *)q, (sizeof(q)-5)/4),
877 jlu32w(m, (const uint32_t *)q, (sizeof(q)-9)/4));
878 p = q;
879 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
880 jlu32l(m, p, sizeof(q)-1), jlu32l(m, p, sizeof(q)-2),
881 jlu32l(m, p, sizeof(q)-3), jlu32l(m, p, sizeof(q)-4),
882 jlu32l(m, p, sizeof(q)-5), jlu32l(m, p, sizeof(q)-6),
883 jlu32l(m, p, sizeof(q)-7), jlu32l(m, p, sizeof(q)-8),
884 jlu32l(m, p, sizeof(q)-9), jlu32l(m, p, sizeof(q)-10),
885 jlu32l(m, p, sizeof(q)-11), jlu32l(m, p, sizeof(q)-12));
886 p = &qq[1];
887 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
888 jlu32l(m, p, sizeof(q)-1), jlu32l(m, p, sizeof(q)-2),
889 jlu32l(m, p, sizeof(q)-3), jlu32l(m, p, sizeof(q)-4),
890 jlu32l(m, p, sizeof(q)-5), jlu32l(m, p, sizeof(q)-6),
891 jlu32l(m, p, sizeof(q)-7), jlu32l(m, p, sizeof(q)-8),
892 jlu32l(m, p, sizeof(q)-9), jlu32l(m, p, sizeof(q)-10),
893 jlu32l(m, p, sizeof(q)-11), jlu32l(m, p, sizeof(q)-12));
894 p = &qqq[2];
895 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
896 jlu32l(m, p, sizeof(q)-1), jlu32l(m, p, sizeof(q)-2),
897 jlu32l(m, p, sizeof(q)-3), jlu32l(m, p, sizeof(q)-4),
898 jlu32l(m, p, sizeof(q)-5), jlu32l(m, p, sizeof(q)-6),
899 jlu32l(m, p, sizeof(q)-7), jlu32l(m, p, sizeof(q)-8),
900 jlu32l(m, p, sizeof(q)-9), jlu32l(m, p, sizeof(q)-10),
901 jlu32l(m, p, sizeof(q)-11), jlu32l(m, p, sizeof(q)-12));
902 p = &qqqq[3];
903 printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
904 jlu32l(m, p, sizeof(q)-1), jlu32l(m, p, sizeof(q)-2),
905 jlu32l(m, p, sizeof(q)-3), jlu32l(m, p, sizeof(q)-4),
906 jlu32l(m, p, sizeof(q)-5), jlu32l(m, p, sizeof(q)-6),
907 jlu32l(m, p, sizeof(q)-7), jlu32l(m, p, sizeof(q)-8),
908 jlu32l(m, p, sizeof(q)-9), jlu32l(m, p, sizeof(q)-10),
909 jlu32l(m, p, sizeof(q)-11), jlu32l(m, p, sizeof(q)-12));
910 printf("\n");
911 for (h=0, b=buf+1; h<8; ++h, ++b) {
912 for (i=0; i<MAXLEN; ++i) {
913 len = i;
914 for (j=0; j<i; ++j)
915 *(b+j)=0;
916
917 /* these should all be equal */
918 m = 1;
919 ref = jlu32l(m, b, len);
920 *(b+i)=(uint8_t)~0;
921 *(b-1)=(uint8_t)~0;
922 x = jlu32l(m, b, len);
923 y = jlu32l(m, b, len);
924 if ((ref != x) || (ref != y))
925 printf("alignment error: %.8x %.8x %.8x %d %d\n",ref,x,y, h, i);
926 }
927 }
928 }
929
930 /* check for problems with nulls */
driver4(void)931 static void driver4(void)
932 {
933 uint8_t buf[1];
934 uint32_t h;
935 uint32_t i;
936 uint32_t state[HASHSTATE];
937
938 buf[0] = ~0;
939 for (i=0; i<HASHSTATE; ++i)
940 state[i] = 1;
941 printf("These should all be different\n");
942 h = 0;
943 for (i=0; i<8; ++i) {
944 h = jlu32l(h, buf, 0);
945 printf("%2ld 0-byte strings, hash is %.8x\n", (long)i, h);
946 }
947 }
948
949
main(int argc,char ** argv)950 int main(int argc, char ** argv)
951 {
952 driver1(); /* test that the key is hashed: used for timings */
953 driver2(); /* test that whole key is hashed thoroughly */
954 driver3(); /* test that nothing but the key is hashed */
955 driver4(); /* test hashing multiple buffers (all buffers are null) */
956 return 1;
957 }
958
959 #endif /* _JLU3_SELFTEST */
960