Line data Source code
1 : /******************************************************************************
2 : * Nickel - a library for hierarchical maps and .ini files
3 : * One of the Bohr Game Libraries (see chaoslizard.org/devel/bohr)
4 : * Copyright (C) 2008 Charles Lindsay. Some rights reserved; see COPYING.
5 : * $Id: hash.c 344 2008-01-19 04:45:41Z chaz $
6 : ******************************************************************************/
7 :
8 :
9 : #include "internal.h"
10 :
11 :
12 : // This is a trivially modified public domain hash library, lookup3.c, from:
13 : // http://burtleburtle.net/bob/hash/
14 :
15 :
16 : /*
17 : -------------------------------------------------------------------------------
18 : lookup3.c, by Bob Jenkins, May 2006, Public Domain.
19 :
20 : These are functions for producing 32-bit hashes for hash table lookup.
21 : hashword(), hashlittle(), hashlittle2(), hashbig(), mix(), and final()
22 : are externally useful functions. Routines to test the hash are included
23 : if SELF_TEST is defined. You can use this free for any purpose. It's in
24 : the public domain. It has no warranty.
25 :
26 : You probably want to use hashlittle(). hashlittle() and hashbig()
27 : hash byte arrays. hashlittle() is is faster than hashbig() on
28 : little-endian machines. Intel and AMD are little-endian machines.
29 : On second thought, you probably want hashlittle2(), which is identical to
30 : hashlittle() except it returns two 32-bit hashes for the price of one.
31 : You could implement hashbig2() if you wanted but I haven't bothered here.
32 :
33 : If you want to find a hash of, say, exactly 7 integers, do
34 : a = i1; b = i2; c = i3;
35 : mix(a,b,c);
36 : a += i4; b += i5; c += i6;
37 : mix(a,b,c);
38 : a += i7;
39 : final(a,b,c);
40 : then use c as the hash value. If you have a variable length array of
41 : 4-byte integers to hash, use hashword(). If you have a byte array (like
42 : a character string), use hashlittle(). If you have several byte arrays, or
43 : a mix of things, see the comments above hashlittle().
44 :
45 : Why is this so big? I read 12 bytes at a time into 3 4-byte integers,
46 : then mix those integers. This is fast (you can do a lot more thorough
47 : mixing with 12*3 instructions on 3 integers than you can with 3 instructions
48 : on 1 byte), but shoehorning those bytes into integers efficiently is messy.
49 : -------------------------------------------------------------------------------
50 : */
51 : //#define SELF_TEST 1
52 :
53 : #include <stdint.h> /* defines uint32_t etc */
54 : #include <stdio.h> /* defines printf for tests */
55 : #include <sys/param.h> /* attempt to define endianness */
56 : #include <time.h> /* defines time_t for timings in the test */
57 : #ifdef linux
58 : #include <endian.h> /* attempt to define endianness */
59 : #endif
60 :
61 : #include <kdbconfig.h> // for no sanitize macros
62 :
63 : /*
64 : * My best guess at if you are big-endian or little-endian. This may
65 : * need adjustment.
66 : */
67 : #if (defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && __BYTE_ORDER == __LITTLE_ENDIAN) || \
68 : (defined(i386) || defined(__i386__) || defined(__i486__) || defined(__i586__) || defined(__i686__) || defined(vax) || \
69 : defined(MIPSEL))
70 : #define HASH_LITTLE_ENDIAN 1
71 : #define HASH_BIG_ENDIAN 0
72 : #elif (defined(__BYTE_ORDER) && defined(__BIG_ENDIAN) && __BYTE_ORDER == __BIG_ENDIAN) || \
73 : (defined(sparc) || defined(POWERPC) || defined(mc68000) || defined(sel))
74 : #define HASH_LITTLE_ENDIAN 0
75 : #define HASH_BIG_ENDIAN 1
76 : #else
77 : #define HASH_LITTLE_ENDIAN 0
78 : #define HASH_BIG_ENDIAN 0
79 : #endif
80 :
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))))
84 :
85 : #ifndef __has_feature
86 : #define __has_feature(something) 0
87 : #endif
88 :
89 : /*
90 : -------------------------------------------------------------------------------
91 : mix -- mix 3 32-bit values reversibly.
92 :
93 : This is reversible, so any information in (a,b,c) before mix() is
94 : still in (a,b,c) after mix().
95 :
96 : If four pairs of (a,b,c) inputs are run through mix(), or through
97 : mix() in reverse, there are at least 32 bits of the output that
98 : are sometimes the same for one pair and different for another pair.
99 : This was tested for:
100 : * pairs that differed by one bit, by two bits, in any combination
101 : of top bits of (a,b,c), or in any combination of bottom bits of
102 : (a,b,c).
103 : * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
104 : the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
105 : is commonly produced by subtraction) look like a single 1-bit
106 : difference.
107 : * the base values were pseudorandom, all zero but one bit set, or
108 : all zero plus a counter that starts at zero.
109 :
110 : Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
111 : satisfy this are
112 : 4 6 8 16 19 4
113 : 9 15 3 18 27 15
114 : 14 9 3 7 17 3
115 : Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
116 : for "differ" defined as + with an one-bit base and a two-bit delta. I
117 : used http://burtleburtle.net/bob/hash/avalanche.html to choose
118 : the operations, constants, and arrangements of the variables.
119 :
120 : This does not achieve avalanche. There are input bits of (a,b,c)
121 : that fail to affect some output bits of (a,b,c), especially of a. The
122 : most thoroughly mixed value is c, but it doesn't really even achieve
123 : avalanche in c.
124 :
125 : This allows some parallelism. Read-after-writes are good at doubling
126 : the number of bits affected, so the goal of mixing pulls in the opposite
127 : direction as the goal of parallelism. I did what I could. Rotates
128 : seem to cost as much as shifts on every machine I could lay my hands
129 : on, and rotates are much kinder to the top and bottom bits, so I used
130 : rotates.
131 : -------------------------------------------------------------------------------
132 : */
133 : #define mix(a, b, c) \
134 : { \
135 : a -= c; \
136 : a ^= rot (c, 4); \
137 : c += b; \
138 : b -= a; \
139 : b ^= rot (a, 6); \
140 : a += c; \
141 : c -= b; \
142 : c ^= rot (b, 8); \
143 : b += a; \
144 : a -= c; \
145 : a ^= rot (c, 16); \
146 : c += b; \
147 : b -= a; \
148 : b ^= rot (a, 19); \
149 : a += c; \
150 : c -= b; \
151 : c ^= rot (b, 4); \
152 : b += a; \
153 : }
154 :
155 : /*
156 : -------------------------------------------------------------------------------
157 : final -- final mixing of 3 32-bit values (a,b,c) into c
158 :
159 : Pairs of (a,b,c) values differing in only a few bits will usually
160 : produce values of c that look totally different. This was tested for
161 : * pairs that differed by one bit, by two bits, in any combination
162 : of top bits of (a,b,c), or in any combination of bottom bits of
163 : (a,b,c).
164 : * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
165 : the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
166 : is commonly produced by subtraction) look like a single 1-bit
167 : difference.
168 : * the base values were pseudorandom, all zero but one bit set, or
169 : all zero plus a counter that starts at zero.
170 :
171 : These constants passed:
172 : 14 11 25 16 4 14 24
173 : 12 14 25 16 4 14 24
174 : and these came close:
175 : 4 8 15 26 3 22 24
176 : 10 8 15 26 3 22 24
177 : 11 8 15 26 3 22 24
178 : -------------------------------------------------------------------------------
179 : */
180 : #define final(a, b, c) \
181 : { \
182 : c ^= b; \
183 : c -= rot (b, 14); \
184 : a ^= c; \
185 : a -= rot (c, 11); \
186 : b ^= a; \
187 : b -= rot (a, 25); \
188 : c ^= b; \
189 : c -= rot (b, 16); \
190 : a ^= c; \
191 : a -= rot (c, 4); \
192 : b ^= a; \
193 : b -= rot (a, 14); \
194 : c ^= b; \
195 : c -= rot (b, 24); \
196 : }
197 :
198 : /*
199 : --------------------------------------------------------------------
200 : This works on all machines. To be useful, it requires
201 : -- that the key be an array of uint32_t's, and
202 : -- that the length be the number of uint32_t's in the key
203 :
204 : The function hashword() is identical to hashlittle() on little-endian
205 : machines, and identical to hashbig() on big-endian machines,
206 : except that the length has to be measured in uint32_ts rather than in
207 : bytes. hashlittle() is more complicated than hashword() only because
208 : hashlittle() has to dance around fitting the key bytes into registers.
209 : --------------------------------------------------------------------
210 : */
211 : #if (0) // not used here
212 : static uint32_t hashword (const uint32_t * restrict k, /* the key, an array of uint32_t values */
213 : size_t length, /* the length of the key, in uint32_ts */
214 : uint32_t initval) /* the previous hash, or an arbitrary value */
215 : {
216 : uint32_t a, b, c;
217 :
218 : /* Set up the internal state */
219 : a = b = c = 0xdeadbeef + (((uint32_t) length) << 2) + initval;
220 :
221 : /*------------------------------------------------- handle most of the key */
222 : while (length > 3)
223 : {
224 : a += k[0];
225 : b += k[1];
226 : c += k[2];
227 : mix (a, b, c);
228 : length -= 3;
229 : k += 3;
230 : }
231 :
232 : /*------------------------------------------- handle the last 3 uint32_t's */
233 : switch (length) /* all the case statements fall through */
234 : {
235 : case 3:
236 : c += k[2];
237 : case 2:
238 : b += k[1];
239 : case 1:
240 : a += k[0];
241 : final (a, b, c);
242 : case 0: /* case 0: nothing left to add */
243 : break;
244 : }
245 : /*------------------------------------------------------ report the result */
246 : return c;
247 : }
248 : #endif
249 :
250 :
251 : /*
252 : --------------------------------------------------------------------
253 : hashword2() -- same as hashword(), but take two seeds and return two
254 : 32-bit values. pc and pb must both be nonnull, and *pc and *pb must
255 : both be initialized with seeds. If you pass in (*pb)==0, the output
256 : (*pc) will be the same as the return value from hashword().
257 : --------------------------------------------------------------------
258 : */
259 : #if (0) // not used here
260 : static void hashword2 (const uint32_t * restrict k, /* the key, an array of uint32_t values */
261 : size_t length, /* the length of the key, in uint32_ts */
262 : uint32_t * restrict pc, /* IN: seed OUT: primary hash value */
263 : uint32_t * restrict pb) /* IN: more seed OUT: secondary hash value */
264 : {
265 : uint32_t a, b, c;
266 :
267 : /* Set up the internal state */
268 : a = b = c = 0xdeadbeef + ((uint32_t) (length << 2)) + *pc;
269 : c += *pb;
270 :
271 : /*------------------------------------------------- handle most of the key */
272 : while (length > 3)
273 : {
274 : a += k[0];
275 : b += k[1];
276 : c += k[2];
277 : mix (a, b, c);
278 : length -= 3;
279 : k += 3;
280 : }
281 :
282 : /*------------------------------------------- handle the last 3 uint32_t's */
283 : switch (length) /* all the case statements fall through */
284 : {
285 : case 3:
286 : c += k[2];
287 : case 2:
288 : b += k[1];
289 : case 1:
290 : a += k[0];
291 : final (a, b, c);
292 : case 0: /* case 0: nothing left to add */
293 : break;
294 : }
295 : /*------------------------------------------------------ report the result */
296 : *pc = c;
297 : *pb = b;
298 : }
299 : #endif
300 :
301 :
302 : /*
303 : -------------------------------------------------------------------------------
304 : hashlittle() -- hash a variable-length key into a 32-bit value
305 : k : the key (the unaligned variable-length array of bytes)
306 : length : the length of the key, counting by bytes
307 : initval : can be any 4-byte value
308 : Returns a 32-bit value. Every bit of the key affects every bit of
309 : the return value. Two keys differing by one or two bits will have
310 : totally different hash values.
311 :
312 : The best hash table sizes are powers of 2. There is no need to do
313 : mod a prime (mod is sooo slow!). If you need less than 32 bits,
314 : use a bitmask. For example, if you need only 10 bits, do
315 : h = (h & hashmask(10));
316 : In which case, the hash table should have hashsize(10) elements.
317 :
318 : If you are hashing n strings (uint8_t **)k, do it like this:
319 : for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h);
320 :
321 : By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this
322 : code any way you wish, private, educational, or commercial. It's free.
323 :
324 : Use for hash table lookup, or anything where one collision in 2^^32 is
325 : acceptable. Do NOT use for cryptographic purposes.
326 : -------------------------------------------------------------------------------
327 : */
328 : #if (!HASH_BIG_ENDIAN) // only if not big-endian
329 : ELEKTRA_NO_SANITIZE_UNSIGNED_INTEGER_OVERFLOW
330 : ELEKTRA_NO_SANITIZE_UNDEFINED
331 2357 : static uint32_t hashlittle (const void * restrict key, size_t length, uint32_t initval)
332 : {
333 : uint32_t a, b, c; /* internal state */
334 : union
335 : {
336 : const void * ptr;
337 : size_t i;
338 : } u; /* needed for Mac Powerbook G4 */
339 :
340 : /* Set up the internal state */
341 2357 : a = b = c = (uint32_t) (0xdeadbeef + length + initval);
342 :
343 2357 : u.ptr = key;
344 2357 : if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0))
345 : {
346 : const uint32_t * k = (const uint32_t *) key; /* read 32-bit chunks */
347 : #if defined(VALGRIND) || defined(__SANITIZE_ADDRESS__) || defined(__has_feature) && __has_feature(address_sanitizer)
348 : const uint8_t * k8;
349 : #endif
350 :
351 : /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
352 2583 : while (length > 12)
353 : {
354 516 : a += k[0];
355 516 : b += k[1];
356 516 : c += k[2];
357 516 : mix (a, b, c);
358 516 : length -= 12;
359 516 : k += 3;
360 : }
361 :
362 : /*----------------------------- handle the last (probably partial) block */
363 : /*
364 : * "k[2]&0xffffff" actually reads beyond the end of the string, but
365 : * then masks off the part it's not allowed to read. Because the
366 : * string is aligned, the masked-off tail is in the same word as the
367 : * rest of the string. Every machine with memory protection I've seen
368 : * does it on word boundaries, so is OK with this. But VALGRIND will
369 : * still catch it and complain. The masking trick does make the hash
370 : * noticeably faster for short strings (like English words).
371 : */
372 : #if !defined(VALGRIND) && !defined(__SANITIZE_ADDRESS__) && !(defined(__has_feature) && __has_feature(address_sanitizer))
373 :
374 2067 : switch (length)
375 : {
376 : case 12:
377 66 : c += k[2];
378 66 : b += k[1];
379 66 : a += k[0];
380 66 : break;
381 : case 11:
382 1 : c += k[2] & 0xffffff;
383 1 : b += k[1];
384 1 : a += k[0];
385 1 : break;
386 : case 10:
387 159 : c += k[2] & 0xffff;
388 159 : b += k[1];
389 159 : a += k[0];
390 159 : break;
391 : case 9:
392 7 : c += k[2] & 0xff;
393 7 : b += k[1];
394 7 : a += k[0];
395 7 : break;
396 : case 8:
397 175 : b += k[1];
398 175 : a += k[0];
399 175 : break;
400 : case 7:
401 429 : b += k[1] & 0xffffff;
402 429 : a += k[0];
403 429 : break;
404 : case 6:
405 68 : b += k[1] & 0xffff;
406 68 : a += k[0];
407 68 : break;
408 : case 5:
409 316 : b += k[1] & 0xff;
410 316 : a += k[0];
411 316 : break;
412 : case 4:
413 399 : a += k[0];
414 399 : break;
415 : case 3:
416 82 : a += k[0] & 0xffffff;
417 82 : break;
418 : case 2:
419 70 : a += k[0] & 0xffff;
420 70 : break;
421 : case 1:
422 184 : a += k[0] & 0xff;
423 184 : break;
424 : case 0:
425 : return c; /* zero length strings require no mixing */
426 : }
427 :
428 : #else /* make valgrind happy */
429 :
430 : k8 = (const uint8_t *) k;
431 : switch (length)
432 : {
433 : case 12:
434 : c += k[2];
435 : b += k[1];
436 : a += k[0];
437 : break;
438 : case 11:
439 : c += ((uint32_t) k8[10]) << 16; /* fall through */
440 : case 10:
441 : c += ((uint32_t) k8[9]) << 8; /* fall through */
442 : case 9:
443 : c += k8[8]; /* fall through */
444 : case 8:
445 : b += k[1];
446 : a += k[0];
447 : break;
448 : case 7:
449 : b += ((uint32_t) k8[6]) << 16; /* fall through */
450 : case 6:
451 : b += ((uint32_t) k8[5]) << 8; /* fall through */
452 : case 5:
453 : b += k8[4]; /* fall through */
454 : case 4:
455 : a += k[0];
456 : break;
457 : case 3:
458 : a += ((uint32_t) k8[2]) << 16; /* fall through */
459 : case 2:
460 : a += ((uint32_t) k8[1]) << 8; /* fall through */
461 : case 1:
462 : a += k8[0];
463 : break;
464 : case 0:
465 : return c;
466 : }
467 :
468 : #endif /* !valgrind */
469 : }
470 290 : else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0))
471 : {
472 : const uint16_t * k = (const uint16_t *) key; /* read 16-bit chunks */
473 : const uint8_t * k8;
474 :
475 : /*--------------- all but last block: aligned reads and different mixing */
476 64 : while (length > 12)
477 : {
478 0 : a += k[0] + (((uint32_t) k[1]) << 16);
479 0 : b += k[2] + (((uint32_t) k[3]) << 16);
480 0 : c += k[4] + (((uint32_t) k[5]) << 16);
481 0 : mix (a, b, c);
482 0 : length -= 12;
483 0 : k += 6;
484 : }
485 :
486 : /*----------------------------- handle the last (probably partial) block */
487 64 : k8 = (const uint8_t *) k;
488 64 : switch (length)
489 : {
490 : case 12:
491 0 : c += k[4] + (((uint32_t) k[5]) << 16);
492 0 : b += k[2] + (((uint32_t) k[3]) << 16);
493 0 : a += k[0] + (((uint32_t) k[1]) << 16);
494 0 : break;
495 : case 11:
496 0 : c += ((uint32_t) k8[10]) << 16; /* fall through */
497 : case 10:
498 0 : c += k[4];
499 0 : b += k[2] + (((uint32_t) k[3]) << 16);
500 0 : a += k[0] + (((uint32_t) k[1]) << 16);
501 0 : break;
502 : case 9:
503 3 : c += k8[8]; /* fall through */
504 : case 8:
505 3 : b += k[2] + (((uint32_t) k[3]) << 16);
506 3 : a += k[0] + (((uint32_t) k[1]) << 16);
507 3 : break;
508 : case 7:
509 4 : b += ((uint32_t) k8[6]) << 16; /* fall through */
510 : case 6:
511 17 : b += k[2];
512 17 : a += k[0] + (((uint32_t) k[1]) << 16);
513 17 : break;
514 : case 5:
515 12 : b += k8[4]; /* fall through */
516 : case 4:
517 12 : a += k[0] + (((uint32_t) k[1]) << 16);
518 12 : break;
519 : case 3:
520 4 : a += ((uint32_t) k8[2]) << 16; /* fall through */
521 : case 2:
522 4 : a += k[0];
523 4 : break;
524 : case 1:
525 14 : a += k8[0];
526 14 : break;
527 : case 0:
528 : return c; /* zero length requires no mixing */
529 : }
530 : }
531 : else
532 : { /* need to read the key one byte at a time */
533 : const uint8_t * k = (const uint8_t *) key;
534 :
535 : /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
536 274 : while (length > 12)
537 : {
538 48 : a += k[0];
539 48 : a += ((uint32_t) k[1]) << 8;
540 48 : a += ((uint32_t) k[2]) << 16;
541 48 : a += ((uint32_t) k[3]) << 24;
542 48 : b += k[4];
543 48 : b += ((uint32_t) k[5]) << 8;
544 48 : b += ((uint32_t) k[6]) << 16;
545 48 : b += ((uint32_t) k[7]) << 24;
546 48 : c += k[8];
547 48 : c += ((uint32_t) k[9]) << 8;
548 48 : c += ((uint32_t) k[10]) << 16;
549 48 : c += ((uint32_t) k[11]) << 24;
550 48 : mix (a, b, c);
551 48 : length -= 12;
552 48 : k += 12;
553 : }
554 :
555 : /*-------------------------------- last block: affect all 32 bits of (c) */
556 226 : switch (length) /* all the case statements fall through */
557 : {
558 : case 12:
559 49 : c += ((uint32_t) k[11]) << 24; // FALLTHROUGH
560 : case 11:
561 49 : c += ((uint32_t) k[10]) << 16; // FALLTHROUGH
562 : case 10:
563 52 : c += ((uint32_t) k[9]) << 8; // FALLTHROUGH
564 : case 9:
565 52 : c += k[8]; // FALLTHROUGH
566 : case 8:
567 92 : b += ((uint32_t) k[7]) << 24; // FALLTHROUGH
568 : case 7:
569 94 : b += ((uint32_t) k[6]) << 16; // FALLTHROUGH
570 : case 6:
571 121 : b += ((uint32_t) k[5]) << 8; // FALLTHROUGH
572 : case 5:
573 179 : b += k[4]; // FALLTHROUGH
574 : case 4:
575 179 : a += ((uint32_t) k[3]) << 24; // FALLTHROUGH
576 : case 3:
577 190 : a += ((uint32_t) k[2]) << 16; // FALLTHROUGH
578 : case 2:
579 210 : a += ((uint32_t) k[1]) << 8; // FALLTHROUGH
580 : case 1:
581 224 : a += k[0];
582 224 : break;
583 : case 0:
584 : return c;
585 : }
586 : }
587 :
588 2230 : final (a, b, c);
589 2230 : return c;
590 : }
591 : #endif
592 :
593 :
594 : /*
595 : * hashlittle2: return 2 32-bit hash values
596 : *
597 : * This is identical to hashlittle(), except it returns two 32-bit hash
598 : * values instead of just one. This is good enough for hash table
599 : * lookup with 2^^64 buckets, or if you want a second hash if you're not
600 : * happy with the first, or if you want a probably-unique 64-bit ID for
601 : * the key. *pc is better mixed than *pb, so use *pc first. If you want
602 : * a 64-bit value do something like "*pc + (((uint64_t)*pb)<<32)".
603 : */
604 : #if (0) // not used here
605 : static void hashlittle2 (const void * restrict key, /* the key to hash */
606 : size_t length, /* length of the key */
607 : uint32_t * restrict pc, /* IN: primary initval, OUT: primary hash */
608 : uint32_t * restrict pb) /* IN: secondary initval, OUT: secondary hash */
609 : {
610 : uint32_t a, b, c; /* internal state */
611 : union
612 : {
613 : const void * ptr;
614 : size_t i;
615 : } u; /* needed for Mac Powerbook G4 */
616 :
617 : /* Set up the internal state */
618 : a = b = c = 0xdeadbeef + ((uint32_t) length) + *pc;
619 : c += *pb;
620 :
621 : u.ptr = key;
622 : if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0))
623 : {
624 : const uint32_t * k = (const uint32_t *) key; /* read 32-bit chunks */
625 : #if defined(VALGRIND) || defined(__SANITIZE_ADDRESS__)
626 : const uint8_t * k8;
627 : #endif
628 :
629 : /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
630 : while (length > 12)
631 : {
632 : a += k[0];
633 : b += k[1];
634 : c += k[2];
635 : mix (a, b, c);
636 : length -= 12;
637 : k += 3;
638 : }
639 :
640 : /*----------------------------- handle the last (probably partial) block */
641 : /*
642 : * "k[2]&0xffffff" actually reads beyond the end of the string, but
643 : * then masks off the part it's not allowed to read. Because the
644 : * string is aligned, the masked-off tail is in the same word as the
645 : * rest of the string. Every machine with memory protection I've seen
646 : * does it on word boundaries, so is OK with this. But VALGRIND will
647 : * still catch it and complain. The masking trick does make the hash
648 : * noticeably faster for short strings (like English words).
649 : */
650 : #if !defined(VALGRIND) && !defined(__SANITIZE_ADDRESS__)
651 :
652 : switch (length)
653 : {
654 : case 12:
655 : c += k[2];
656 : b += k[1];
657 : a += k[0];
658 : break;
659 : case 11:
660 : c += k[2] & 0xffffff;
661 : b += k[1];
662 : a += k[0];
663 : break;
664 : case 10:
665 : c += k[2] & 0xffff;
666 : b += k[1];
667 : a += k[0];
668 : break;
669 : case 9:
670 : c += k[2] & 0xff;
671 : b += k[1];
672 : a += k[0];
673 : break;
674 : case 8:
675 : b += k[1];
676 : a += k[0];
677 : break;
678 : case 7:
679 : b += k[1] & 0xffffff;
680 : a += k[0];
681 : break;
682 : case 6:
683 : b += k[1] & 0xffff;
684 : a += k[0];
685 : break;
686 : case 5:
687 : b += k[1] & 0xff;
688 : a += k[0];
689 : break;
690 : case 4:
691 : a += k[0];
692 : break;
693 : case 3:
694 : a += k[0] & 0xffffff;
695 : break;
696 : case 2:
697 : a += k[0] & 0xffff;
698 : break;
699 : case 1:
700 : a += k[0] & 0xff;
701 : break;
702 : case 0:
703 : *pc = c;
704 : *pb = b;
705 : return; /* zero length strings require no mixing */
706 : }
707 :
708 : #else /* make valgrind happy */
709 :
710 : k8 = (const uint8_t *) k;
711 : switch (length)
712 : {
713 : case 12:
714 : c += k[2];
715 : b += k[1];
716 : a += k[0];
717 : break;
718 : case 11:
719 : c += ((uint32_t) k8[10]) << 16; /* fall through */
720 : case 10:
721 : c += ((uint32_t) k8[9]) << 8; /* fall through */
722 : case 9:
723 : c += k8[8]; /* fall through */
724 : case 8:
725 : b += k[1];
726 : a += k[0];
727 : break;
728 : case 7:
729 : b += ((uint32_t) k8[6]) << 16; /* fall through */
730 : case 6:
731 : b += ((uint32_t) k8[5]) << 8; /* fall through */
732 : case 5:
733 : b += k8[4]; /* fall through */
734 : case 4:
735 : a += k[0];
736 : break;
737 : case 3:
738 : a += ((uint32_t) k8[2]) << 16; /* fall through */
739 : case 2:
740 : a += ((uint32_t) k8[1]) << 8; /* fall through */
741 : case 1:
742 : a += k8[0];
743 : break;
744 : case 0:
745 : *pc = c;
746 : *pb = b;
747 : return; /* zero length strings require no mixing */
748 : }
749 :
750 : #endif /* !valgrind */
751 : }
752 : else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0))
753 : {
754 : const uint16_t * k = (const uint16_t *) key; /* read 16-bit chunks */
755 : const uint8_t * k8;
756 :
757 : /*--------------- all but last block: aligned reads and different mixing */
758 : while (length > 12)
759 : {
760 : a += k[0] + (((uint32_t) k[1]) << 16);
761 : b += k[2] + (((uint32_t) k[3]) << 16);
762 : c += k[4] + (((uint32_t) k[5]) << 16);
763 : mix (a, b, c);
764 : length -= 12;
765 : k += 6;
766 : }
767 :
768 : /*----------------------------- handle the last (probably partial) block */
769 : k8 = (const uint8_t *) k;
770 : switch (length)
771 : {
772 : case 12:
773 : c += k[4] + (((uint32_t) k[5]) << 16);
774 : b += k[2] + (((uint32_t) k[3]) << 16);
775 : a += k[0] + (((uint32_t) k[1]) << 16);
776 : break;
777 : case 11:
778 : c += ((uint32_t) k8[10]) << 16; /* fall through */
779 : case 10:
780 : c += k[4];
781 : b += k[2] + (((uint32_t) k[3]) << 16);
782 : a += k[0] + (((uint32_t) k[1]) << 16);
783 : break;
784 : case 9:
785 : c += k8[8]; /* fall through */
786 : case 8:
787 : b += k[2] + (((uint32_t) k[3]) << 16);
788 : a += k[0] + (((uint32_t) k[1]) << 16);
789 : break;
790 : case 7:
791 : b += ((uint32_t) k8[6]) << 16; /* fall through */
792 : case 6:
793 : b += k[2];
794 : a += k[0] + (((uint32_t) k[1]) << 16);
795 : break;
796 : case 5:
797 : b += k8[4]; /* fall through */
798 : case 4:
799 : a += k[0] + (((uint32_t) k[1]) << 16);
800 : break;
801 : case 3:
802 : a += ((uint32_t) k8[2]) << 16; /* fall through */
803 : case 2:
804 : a += k[0];
805 : break;
806 : case 1:
807 : a += k8[0];
808 : break;
809 : case 0:
810 : *pc = c;
811 : *pb = b;
812 : return; /* zero length strings require no mixing */
813 : }
814 : }
815 : else
816 : { /* need to read the key one byte at a time */
817 : const uint8_t * k = (const uint8_t *) key;
818 :
819 : /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
820 : while (length > 12)
821 : {
822 : a += k[0];
823 : a += ((uint32_t) k[1]) << 8;
824 : a += ((uint32_t) k[2]) << 16;
825 : a += ((uint32_t) k[3]) << 24;
826 : b += k[4];
827 : b += ((uint32_t) k[5]) << 8;
828 : b += ((uint32_t) k[6]) << 16;
829 : b += ((uint32_t) k[7]) << 24;
830 : c += k[8];
831 : c += ((uint32_t) k[9]) << 8;
832 : c += ((uint32_t) k[10]) << 16;
833 : c += ((uint32_t) k[11]) << 24;
834 : mix (a, b, c);
835 : length -= 12;
836 : k += 12;
837 : }
838 :
839 : /*-------------------------------- last block: affect all 32 bits of (c) */
840 : switch (length) /* all the case statements fall through */
841 : {
842 : case 12:
843 : c += ((uint32_t) k[11]) << 24;
844 : case 11:
845 : c += ((uint32_t) k[10]) << 16;
846 : case 10:
847 : c += ((uint32_t) k[9]) << 8;
848 : case 9:
849 : c += k[8];
850 : case 8:
851 : b += ((uint32_t) k[7]) << 24;
852 : case 7:
853 : b += ((uint32_t) k[6]) << 16;
854 : case 6:
855 : b += ((uint32_t) k[5]) << 8;
856 : case 5:
857 : b += k[4];
858 : case 4:
859 : a += ((uint32_t) k[3]) << 24;
860 : case 3:
861 : a += ((uint32_t) k[2]) << 16;
862 : case 2:
863 : a += ((uint32_t) k[1]) << 8;
864 : case 1:
865 : a += k[0];
866 : break;
867 : case 0:
868 : *pc = c;
869 : *pb = b;
870 : return; /* zero length strings require no mixing */
871 : }
872 : }
873 :
874 : final (a, b, c);
875 : *pc = c;
876 : *pb = b;
877 : }
878 : #endif
879 :
880 :
881 : /*
882 : * hashbig():
883 : * This is the same as hashword() on big-endian machines. It is different
884 : * from hashlittle() on all machines. hashbig() takes advantage of
885 : * big-endian byte ordering.
886 : */
887 : #if (HASH_BIG_ENDIAN) // only if big-endian
888 : ELEKTRA_NO_SANITIZE_UNSIGNED_INTEGER_OVERFLOW
889 : ELEKTRA_NO_SANITIZE_UNDEFINED
890 : static uint32_t hashbig (const void * restrict key, size_t length, uint32_t initval)
891 : {
892 : uint32_t a, b, c;
893 : union
894 : {
895 : const void * ptr;
896 : size_t i;
897 : } u; /* to cast key to (size_t) happily */
898 :
899 : /* Set up the internal state */
900 : a = b = c = 0xdeadbeef + ((uint32_t) length) + initval;
901 :
902 : u.ptr = key;
903 : if (HASH_BIG_ENDIAN && ((u.i & 0x3) == 0))
904 : {
905 : const uint32_t * k = (const uint32_t *) key; /* read 32-bit chunks */
906 : #if defined(VALGRIND) || defined(__SANITIZE_ADDRESS__)
907 : const uint8_t * k8;
908 : #endif
909 :
910 : /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
911 : while (length > 12)
912 : {
913 : a += k[0];
914 : b += k[1];
915 : c += k[2];
916 : mix (a, b, c);
917 : length -= 12;
918 : k += 3;
919 : }
920 :
921 : /*----------------------------- handle the last (probably partial) block */
922 : /*
923 : * "k[2]<<8" actually reads beyond the end of the string, but
924 : * then shifts out the part it's not allowed to read. Because the
925 : * string is aligned, the illegal read is in the same word as the
926 : * rest of the string. Every machine with memory protection I've seen
927 : * does it on word boundaries, so is OK with this. But VALGRIND will
928 : * still catch it and complain. The masking trick does make the hash
929 : * noticeably faster for short strings (like English words).
930 : */
931 : #if !defined(VALGRIND) && !defined(__SANITIZE_ADDRESS__)
932 :
933 : switch (length)
934 : {
935 : case 12:
936 : c += k[2];
937 : b += k[1];
938 : a += k[0];
939 : break;
940 : case 11:
941 : c += k[2] & 0xffffff00;
942 : b += k[1];
943 : a += k[0];
944 : break;
945 : case 10:
946 : c += k[2] & 0xffff0000;
947 : b += k[1];
948 : a += k[0];
949 : break;
950 : case 9:
951 : c += k[2] & 0xff000000;
952 : b += k[1];
953 : a += k[0];
954 : break;
955 : case 8:
956 : b += k[1];
957 : a += k[0];
958 : break;
959 : case 7:
960 : b += k[1] & 0xffffff00;
961 : a += k[0];
962 : break;
963 : case 6:
964 : b += k[1] & 0xffff0000;
965 : a += k[0];
966 : break;
967 : case 5:
968 : b += k[1] & 0xff000000;
969 : a += k[0];
970 : break;
971 : case 4:
972 : a += k[0];
973 : break;
974 : case 3:
975 : a += k[0] & 0xffffff00;
976 : break;
977 : case 2:
978 : a += k[0] & 0xffff0000;
979 : break;
980 : case 1:
981 : a += k[0] & 0xff000000;
982 : break;
983 : case 0:
984 : return c; /* zero length strings require no mixing */
985 : }
986 :
987 : #else /* make valgrind happy */
988 :
989 : k8 = (const uint8_t *) k;
990 : switch (length) /* all the case statements fall through */
991 : {
992 : case 12:
993 : c += k[2];
994 : b += k[1];
995 : a += k[0];
996 : break;
997 : case 11:
998 : c += ((uint32_t) k8[10]) << 8; /* fall through */
999 : case 10:
1000 : c += ((uint32_t) k8[9]) << 16; /* fall through */
1001 : case 9:
1002 : c += ((uint32_t) k8[8]) << 24; /* fall through */
1003 : case 8:
1004 : b += k[1];
1005 : a += k[0];
1006 : break;
1007 : case 7:
1008 : b += ((uint32_t) k8[6]) << 8; /* fall through */
1009 : case 6:
1010 : b += ((uint32_t) k8[5]) << 16; /* fall through */
1011 : case 5:
1012 : b += ((uint32_t) k8[4]) << 24; /* fall through */
1013 : case 4:
1014 : a += k[0];
1015 : break;
1016 : case 3:
1017 : a += ((uint32_t) k8[2]) << 8; /* fall through */
1018 : case 2:
1019 : a += ((uint32_t) k8[1]) << 16; /* fall through */
1020 : case 1:
1021 : a += ((uint32_t) k8[0]) << 24;
1022 : break;
1023 : case 0:
1024 : return c;
1025 : }
1026 :
1027 : #endif /* !VALGRIND */
1028 : }
1029 : else
1030 : { /* need to read the key one byte at a time */
1031 : const uint8_t * k = (const uint8_t *) key;
1032 :
1033 : /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
1034 : while (length > 12)
1035 : {
1036 : a += ((uint32_t) k[0]) << 24;
1037 : a += ((uint32_t) k[1]) << 16;
1038 : a += ((uint32_t) k[2]) << 8;
1039 : a += ((uint32_t) k[3]);
1040 : b += ((uint32_t) k[4]) << 24;
1041 : b += ((uint32_t) k[5]) << 16;
1042 : b += ((uint32_t) k[6]) << 8;
1043 : b += ((uint32_t) k[7]);
1044 : c += ((uint32_t) k[8]) << 24;
1045 : c += ((uint32_t) k[9]) << 16;
1046 : c += ((uint32_t) k[10]) << 8;
1047 : c += ((uint32_t) k[11]);
1048 : mix (a, b, c);
1049 : length -= 12;
1050 : k += 12;
1051 : }
1052 :
1053 : /*-------------------------------- last block: affect all 32 bits of (c) */
1054 : switch (length) /* all the case statements fall through */
1055 : {
1056 : case 12:
1057 : c += k[11];
1058 : case 11:
1059 : c += ((uint32_t) k[10]) << 8;
1060 : case 10:
1061 : c += ((uint32_t) k[9]) << 16;
1062 : case 9:
1063 : c += ((uint32_t) k[8]) << 24;
1064 : case 8:
1065 : b += k[7];
1066 : case 7:
1067 : b += ((uint32_t) k[6]) << 8;
1068 : case 6:
1069 : b += ((uint32_t) k[5]) << 16;
1070 : case 5:
1071 : b += ((uint32_t) k[4]) << 24;
1072 : case 4:
1073 : a += k[3];
1074 : case 3:
1075 : a += ((uint32_t) k[2]) << 8;
1076 : case 2:
1077 : a += ((uint32_t) k[1]) << 16;
1078 : case 1:
1079 : a += ((uint32_t) k[0]) << 24;
1080 : break;
1081 : case 0:
1082 : return c;
1083 : }
1084 : }
1085 :
1086 : final (a, b, c);
1087 : return c;
1088 : }
1089 : #endif
1090 :
1091 :
1092 2357 : elektraNi_PRIVATE uint32_t Hash (const void * restrict key, size_t length, uint32_t initval)
1093 : {
1094 : #if (HASH_BIG_ENDIAN)
1095 : return hashbig (key, length, initval);
1096 : #else
1097 2357 : return hashlittle (key, length, initval);
1098 : #endif
1099 : }
1100 :
1101 :
1102 : #ifdef SELF_TEST
1103 :
1104 : /* used for timings */
1105 : void driver1 (void)
1106 : {
1107 : uint8_t buf[256];
1108 : uint32_t i;
1109 : uint32_t h = 0;
1110 : time_t a, z;
1111 :
1112 : time (&a);
1113 : for (i = 0; i < 256; ++i)
1114 : buf[i] = 'x';
1115 : for (i = 0; i < 1; ++i)
1116 : {
1117 : h = hashlittle (&buf[0], 1, h);
1118 : }
1119 : time (&z);
1120 : if (z - a > 0) printf ("time %d %.8x\n", z - a, h);
1121 : }
1122 :
1123 : /* check that every input bit changes every output bit half the time */
1124 : #define HASHSTATE 1
1125 : #define HASHLEN 1
1126 : #define MAXPAIR 60
1127 : #define MAXLEN 70
1128 : void driver2 (void)
1129 : {
1130 : uint8_t qa[MAXLEN + 1], qb[MAXLEN + 2], *a = &qa[0], *b = &qb[1];
1131 : uint32_t c[HASHSTATE], d[HASHSTATE], i = 0, j = 0, k, l, m = 0, z;
1132 : uint32_t e[HASHSTATE], f[HASHSTATE], g[HASHSTATE], h[HASHSTATE];
1133 : uint32_t x[HASHSTATE], y[HASHSTATE];
1134 : uint32_t hlen;
1135 :
1136 : printf ("No more than %d trials should ever be needed \n", MAXPAIR / 2);
1137 : for (hlen = 0; hlen < MAXLEN; ++hlen)
1138 : {
1139 : z = 0;
1140 : for (i = 0; i < hlen; ++i) /*----------------------- for each input byte, */
1141 : {
1142 : for (j = 0; j < 8; ++j) /*------------------------ for each input bit, */
1143 : {
1144 : for (m = 1; m < 8; ++m) /*------------ for several possible initvals, */
1145 : {
1146 : for (l = 0; l < HASHSTATE; ++l)
1147 : e[l] = f[l] = g[l] = h[l] = x[l] = y[l] = ~((uint32_t) 0);
1148 :
1149 : /*---- check that every output bit is affected by that input bit */
1150 : for (k = 0; k < MAXPAIR; k += 2)
1151 : {
1152 : uint32_t finished = 1;
1153 : /* keys have one bit different */
1154 : for (l = 0; l < hlen + 1; ++l)
1155 : {
1156 : a[l] = b[l] = (uint8_t) 0;
1157 : }
1158 : /* have a and b be two keys differing in only one bit */
1159 : a[i] ^= (k << j);
1160 : a[i] ^= (k >> (8 - j));
1161 : c[0] = hashlittle (a, hlen, m);
1162 : b[i] ^= ((k + 1) << j);
1163 : b[i] ^= ((k + 1) >> (8 - j));
1164 : d[0] = hashlittle (b, hlen, m);
1165 : /* check every bit is 1, 0, set, and not set at least once */
1166 : for (l = 0; l < HASHSTATE; ++l)
1167 : {
1168 : e[l] &= (c[l] ^ d[l]);
1169 : f[l] &= ~(c[l] ^ d[l]);
1170 : g[l] &= c[l];
1171 : h[l] &= ~c[l];
1172 : x[l] &= d[l];
1173 : y[l] &= ~d[l];
1174 : if (e[l] | f[l] | g[l] | h[l] | x[l] | y[l]) finished = 0;
1175 : }
1176 : if (finished) break;
1177 : }
1178 : if (k > z) z = k;
1179 : if (k == MAXPAIR)
1180 : {
1181 : printf ("Some bit didn't change: ");
1182 : printf ("%.8x %.8x %.8x %.8x %.8x %.8x ", e[0], f[0], g[0], h[0], x[0], y[0]);
1183 : printf ("i %d j %d m %d len %d\n", i, j, m, hlen);
1184 : }
1185 : if (z == MAXPAIR) goto done;
1186 : }
1187 : }
1188 : }
1189 : done:
1190 : if (z < MAXPAIR)
1191 : {
1192 : printf ("Mix success %2d bytes %2d initvals ", i, m);
1193 : printf ("required %d trials\n", z / 2);
1194 : }
1195 : }
1196 : printf ("\n");
1197 : }
1198 :
1199 : /* Check for reading beyond the end of the buffer and alignment problems */
1200 : void driver3 (void)
1201 : {
1202 : uint8_t buf[MAXLEN + 20], *b;
1203 : uint32_t len;
1204 : uint8_t q[] = "This is the time for all good men to come to the aid of their country...";
1205 : uint32_t h;
1206 : uint8_t qq[] = "xThis is the time for all good men to come to the aid of their country...";
1207 : uint32_t i;
1208 : uint8_t qqq[] = "xxThis is the time for all good men to come to the aid of their country...";
1209 : uint32_t j;
1210 : uint8_t qqqq[] = "xxxThis is the time for all good men to come to the aid of their country...";
1211 : uint32_t ref, x, y;
1212 : uint8_t * p;
1213 :
1214 : printf ("Endianness. These lines should all be the same (for values filled in):\n");
1215 : printf ("%.8x %.8x %.8x\n",
1216 : hashword ((const uint32_t *) q, (sizeof (q) - 1) / 4, 13), hashword ((const uint32_t *) q, (sizeof (q) - 5) / 4, 13),
1217 : hashword ((const uint32_t *) q, (sizeof (q) - 9) / 4, 13));
1218 : p = q;
1219 : printf ("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", hashlittle (p, sizeof (q) - 1, 13),
1220 : hashlittle (p, sizeof (q) - 2, 13), hashlittle (p, sizeof (q) - 3, 13), hashlittle (p, sizeof (q) - 4, 13),
1221 : hashlittle (p, sizeof (q) - 5, 13), hashlittle (p, sizeof (q) - 6, 13), hashlittle (p, sizeof (q) - 7, 13),
1222 : hashlittle (p, sizeof (q) - 8, 13), hashlittle (p, sizeof (q) - 9, 13), hashlittle (p, sizeof (q) - 10, 13),
1223 : hashlittle (p, sizeof (q) - 11, 13), hashlittle (p, sizeof (q) - 12, 13));
1224 : p = &qq[1];
1225 : printf ("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", hashlittle (p, sizeof (q) - 1, 13),
1226 : hashlittle (p, sizeof (q) - 2, 13), hashlittle (p, sizeof (q) - 3, 13), hashlittle (p, sizeof (q) - 4, 13),
1227 : hashlittle (p, sizeof (q) - 5, 13), hashlittle (p, sizeof (q) - 6, 13), hashlittle (p, sizeof (q) - 7, 13),
1228 : hashlittle (p, sizeof (q) - 8, 13), hashlittle (p, sizeof (q) - 9, 13), hashlittle (p, sizeof (q) - 10, 13),
1229 : hashlittle (p, sizeof (q) - 11, 13), hashlittle (p, sizeof (q) - 12, 13));
1230 : p = &qqq[2];
1231 : printf ("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", hashlittle (p, sizeof (q) - 1, 13),
1232 : hashlittle (p, sizeof (q) - 2, 13), hashlittle (p, sizeof (q) - 3, 13), hashlittle (p, sizeof (q) - 4, 13),
1233 : hashlittle (p, sizeof (q) - 5, 13), hashlittle (p, sizeof (q) - 6, 13), hashlittle (p, sizeof (q) - 7, 13),
1234 : hashlittle (p, sizeof (q) - 8, 13), hashlittle (p, sizeof (q) - 9, 13), hashlittle (p, sizeof (q) - 10, 13),
1235 : hashlittle (p, sizeof (q) - 11, 13), hashlittle (p, sizeof (q) - 12, 13));
1236 : p = &qqqq[3];
1237 : printf ("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n", hashlittle (p, sizeof (q) - 1, 13),
1238 : hashlittle (p, sizeof (q) - 2, 13), hashlittle (p, sizeof (q) - 3, 13), hashlittle (p, sizeof (q) - 4, 13),
1239 : hashlittle (p, sizeof (q) - 5, 13), hashlittle (p, sizeof (q) - 6, 13), hashlittle (p, sizeof (q) - 7, 13),
1240 : hashlittle (p, sizeof (q) - 8, 13), hashlittle (p, sizeof (q) - 9, 13), hashlittle (p, sizeof (q) - 10, 13),
1241 : hashlittle (p, sizeof (q) - 11, 13), hashlittle (p, sizeof (q) - 12, 13));
1242 : printf ("\n");
1243 :
1244 : /* check that hashlittle2 and hashlittle produce the same results */
1245 : i = 47;
1246 : j = 0;
1247 : hashlittle2 (q, sizeof (q), &i, &j);
1248 : if (hashlittle (q, sizeof (q), 47) != i) printf ("hashlittle2 and hashlittle mismatch\n");
1249 :
1250 : /* check that hashword2 and hashword produce the same results */
1251 : len = 0xdeadbeef;
1252 : i = 47, j = 0;
1253 : hashword2 (&len, 1, &i, &j);
1254 : if (hashword (&len, 1, 47) != i) printf ("hashword2 and hashword mismatch %x %x\n", i, hashword (&len, 1, 47));
1255 :
1256 : /* check hashlittle doesn't read before or after the ends of the string */
1257 : for (h = 0, b = buf + 1; h < 8; ++h, ++b)
1258 : {
1259 : for (i = 0; i < MAXLEN; ++i)
1260 : {
1261 : len = i;
1262 : for (j = 0; j < i; ++j)
1263 : *(b + j) = 0;
1264 :
1265 : /* these should all be equal */
1266 : ref = hashlittle (b, len, (uint32_t) 1);
1267 : *(b + i) = (uint8_t) ~0;
1268 : *(b - 1) = (uint8_t) ~0;
1269 : x = hashlittle (b, len, (uint32_t) 1);
1270 : y = hashlittle (b, len, (uint32_t) 1);
1271 : if ((ref != x) || (ref != y))
1272 : {
1273 : printf ("alignment error: %.8x %.8x %.8x %d %d\n", ref, x, y, h, i);
1274 : }
1275 : }
1276 : }
1277 : }
1278 :
1279 : /* check for problems with nulls */
1280 : void driver4 (void)
1281 : {
1282 : uint8_t buf[1];
1283 : uint32_t h, i, state[HASHSTATE];
1284 :
1285 :
1286 : buf[0] = ~0;
1287 : for (i = 0; i < HASHSTATE; ++i)
1288 : state[i] = 1;
1289 : printf ("These should all be different\n");
1290 : for (i = 0, h = 0; i < 8; ++i)
1291 : {
1292 : h = hashlittle (buf, 0, h);
1293 : printf ("%2ld 0-byte strings, hash is %.8x\n", i, h);
1294 : }
1295 : }
1296 :
1297 :
1298 : int main (void)
1299 : {
1300 : driver1 (); /* test that the key is hashed: used for timings */
1301 : driver2 (); /* test that whole key is hashed thoroughly */
1302 : driver3 (); /* test that nothing but the key is hashed */
1303 : driver4 (); /* test hashing multiple buffers (all buffers are null) */
1304 : return 1;
1305 : }
1306 :
1307 : #endif /* SELF_TEST */
|