| 21 |
andreas |
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 |
* hashword(), hashlittle(), hashlittle2(), hashbig(), mix(), and final()
|
|
|
7 |
* are externally useful functions. Routines to test the hash are included
|
|
|
8 |
* if SELF_TEST is defined. You can use this free for any purpose. It's in
|
|
|
9 |
* the public domain. It has no warranty.
|
|
|
10 |
*
|
|
|
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.
|
|
|
17 |
*
|
|
|
18 |
* If you want to find a hash of, say, exactly 7 integers, do
|
|
|
19 |
* a = i1; b = i2; c = i3;
|
|
|
20 |
* mix(a,b,c);
|
|
|
21 |
* a += i4; b += i5; c += i6;
|
|
|
22 |
* mix(a,b,c);
|
|
|
23 |
* a += i7;
|
|
|
24 |
* final(a,b,c);
|
|
|
25 |
* then use c as the hash value. If you have a variable length array of
|
|
|
26 |
* 4-byte integers to hash, use hashword(). If you have a byte array (like
|
|
|
27 |
* a character string), use hashlittle(). If you have several byte arrays, or
|
|
|
28 |
* a mix of things, see the comments above hashlittle().
|
|
|
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 |
/* #define SELF_TEST 1 */
|
|
|
37 |
|
|
|
38 |
#include <stdio.h> /* defines printf for tests */
|
|
|
39 |
#include <time.h> /* defines time_t for timings in the test */
|
|
|
40 |
#include <stdint.h> /* defines uint32_t etc */
|
|
|
41 |
#include <sys/param.h> /* attempt to define endianness */
|
|
|
42 |
#ifdef linux
|
|
|
43 |
# include <endian.h> /* attempt to define endianness */
|
|
|
44 |
#endif
|
|
|
45 |
|
|
|
46 |
/*
|
|
|
47 |
* My best guess at if you are big-endian or little-endian. This may
|
|
|
48 |
* need adjustment.
|
|
|
49 |
*/
|
|
|
50 |
#if (defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && \
|
|
|
51 |
__BYTE_ORDER == __LITTLE_ENDIAN) || \
|
|
|
52 |
(defined(i386) || defined(__i386__) || defined(__i486__) || \
|
|
|
53 |
defined(__i586__) || defined(__i686__) || defined(vax) || defined(MIPSEL))
|
|
|
54 |
# define HASH_LITTLE_ENDIAN 1
|
|
|
55 |
# define HASH_BIG_ENDIAN 0
|
|
|
56 |
#elif (defined(__BYTE_ORDER) && defined(__BIG_ENDIAN) && \
|
|
|
57 |
__BYTE_ORDER == __BIG_ENDIAN) || \
|
|
|
58 |
(defined(sparc) || defined(POWERPC) || defined(mc68000) || defined(sel))
|
|
|
59 |
# define HASH_LITTLE_ENDIAN 0
|
|
|
60 |
# define HASH_BIG_ENDIAN 1
|
|
|
61 |
#else
|
|
|
62 |
# define HASH_LITTLE_ENDIAN 0
|
|
|
63 |
# define HASH_BIG_ENDIAN 0
|
|
|
64 |
#endif
|
|
|
65 |
|
|
|
66 |
#define hashsize(n) ((uint32_t)1<<(n))
|
|
|
67 |
#define hashmask(n) (hashsize(n)-1)
|
|
|
68 |
#define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
|
|
|
69 |
|
|
|
70 |
/*
|
|
|
71 |
* -------------------------------------------------------------------------------
|
|
|
72 |
* mix -- mix 3 32-bit values reversibly.
|
|
|
73 |
*
|
|
|
74 |
* This is reversible, so any information in (a,b,c) before mix() is
|
|
|
75 |
* still in (a,b,c) after mix().
|
|
|
76 |
*
|
|
|
77 |
* If four pairs of (a,b,c) inputs are run through mix(), or through
|
|
|
78 |
* mix() in reverse, there are at least 32 bits of the output that
|
|
|
79 |
* are sometimes the same for one pair and different for another pair.
|
|
|
80 |
* This was tested for:
|
|
|
81 |
* pairs that differed by one bit, by two bits, in any combination
|
|
|
82 |
* of top bits of (a,b,c), or in any combination of bottom bits of
|
|
|
83 |
* (a,b,c).
|
|
|
84 |
* "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
|
|
|
85 |
* the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
|
|
|
86 |
* is commonly produced by subtraction) look like a single 1-bit
|
|
|
87 |
* difference.
|
|
|
88 |
* the base values were pseudorandom, all zero but one bit set, or
|
|
|
89 |
* all zero plus a counter that starts at zero.
|
|
|
90 |
*
|
|
|
91 |
* Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
|
|
|
92 |
* satisfy this are
|
|
|
93 |
* 4 6 8 16 19 4
|
|
|
94 |
* 9 15 3 18 27 15
|
|
|
95 |
* 14 9 3 7 17 3
|
|
|
96 |
* Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
|
|
|
97 |
* for "differ" defined as + with a one-bit base and a two-bit delta. I
|
|
|
98 |
* used http://burtleburtle.net/bob/hash/avalanche.html to choose
|
|
|
99 |
* the operations, constants, and arrangements of the variables.
|
|
|
100 |
*
|
|
|
101 |
* This does not achieve avalanche. There are input bits of (a,b,c)
|
|
|
102 |
* that fail to affect some output bits of (a,b,c), especially of a. The
|
|
|
103 |
* most thoroughly mixed value is c, but it doesn't really even achieve
|
|
|
104 |
* avalanche in c.
|
|
|
105 |
*
|
|
|
106 |
* This allows some parallelism. Read-after-writes are good at doubling
|
|
|
107 |
* the number of bits affected, so the goal of mixing pulls in the opposite
|
|
|
108 |
* direction as the goal of parallelism. I did what I could. Rotates
|
|
|
109 |
* seem to cost as much as shifts on every machine I could lay my hands
|
|
|
110 |
* on, and rotates are much kinder to the top and bottom bits, so I used
|
|
|
111 |
* rotates.
|
|
|
112 |
* -------------------------------------------------------------------------------
|
|
|
113 |
*/
|
|
|
114 |
#define mix(a,b,c) \
|
|
|
115 |
{ \
|
|
|
116 |
a -= c; a ^= rot(c, 4); c += b; \
|
|
|
117 |
b -= a; b ^= rot(a, 6); a += c; \
|
|
|
118 |
c -= b; c ^= rot(b, 8); b += a; \
|
|
|
119 |
a -= c; a ^= rot(c,16); c += b; \
|
|
|
120 |
b -= a; b ^= rot(a,19); a += c; \
|
|
|
121 |
c -= b; c ^= rot(b, 4); b += a; \
|
|
|
122 |
}
|
|
|
123 |
|
|
|
124 |
/*
|
|
|
125 |
* -------------------------------------------------------------------------------
|
|
|
126 |
* final -- final mixing of 3 32-bit values (a,b,c) into c
|
|
|
127 |
*
|
|
|
128 |
* Pairs of (a,b,c) values differing in only a few bits will usually
|
|
|
129 |
* produce values of c that look totally different. This was tested for
|
|
|
130 |
* pairs that differed by one bit, by two bits, in any combination
|
|
|
131 |
* of top bits of (a,b,c), or in any combination of bottom bits of
|
|
|
132 |
* (a,b,c).
|
|
|
133 |
* "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
|
|
|
134 |
* the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
|
|
|
135 |
* is commonly produced by subtraction) look like a single 1-bit
|
|
|
136 |
* difference.
|
|
|
137 |
* the base values were pseudorandom, all zero but one bit set, or
|
|
|
138 |
* all zero plus a counter that starts at zero.
|
|
|
139 |
*
|
|
|
140 |
* These constants passed:
|
|
|
141 |
* 14 11 25 16 4 14 24
|
|
|
142 |
* 12 14 25 16 4 14 24
|
|
|
143 |
* and these came close:
|
|
|
144 |
* 4 8 15 26 3 22 24
|
|
|
145 |
* 10 8 15 26 3 22 24
|
|
|
146 |
* 11 8 15 26 3 22 24
|
|
|
147 |
* -------------------------------------------------------------------------------
|
|
|
148 |
*/
|
|
|
149 |
#define final(a,b,c) \
|
|
|
150 |
{ \
|
|
|
151 |
c ^= b; c -= rot(b,14); \
|
|
|
152 |
a ^= c; a -= rot(c,11); \
|
|
|
153 |
b ^= a; b -= rot(a,25); \
|
|
|
154 |
c ^= b; c -= rot(b,16); \
|
|
|
155 |
a ^= c; a -= rot(c,4); \
|
|
|
156 |
b ^= a; b -= rot(a,14); \
|
|
|
157 |
c ^= b; c -= rot(b,24); \
|
|
|
158 |
}
|
|
|
159 |
|
|
|
160 |
/*
|
|
|
161 |
* --------------------------------------------------------------------
|
|
|
162 |
* This works on all machines. To be useful, it requires
|
|
|
163 |
* -- that the key be an array of uint32_t's, and
|
|
|
164 |
* -- that the length be the number of uint32_t's in the key
|
|
|
165 |
*
|
|
|
166 |
* The function hashword() is identical to hashlittle() on little-endian
|
|
|
167 |
* machines, and identical to hashbig() on big-endian machines,
|
|
|
168 |
* except that the length has to be measured in uint32_ts rather than in
|
|
|
169 |
* bytes. hashlittle() is more complicated than hashword() only because
|
|
|
170 |
* hashlittle() has to dance around fitting the key bytes into registers.
|
|
|
171 |
* --------------------------------------------------------------------
|
|
|
172 |
*/
|
|
|
173 |
uint32_t hashword(
|
|
|
174 |
const uint32_t *k, /* the key, an array of uint32_t values */
|
|
|
175 |
size_t length, /* the length of the key, in uint32_ts */
|
|
|
176 |
uint32_t initval) /* the previous hash, or an arbitrary value */
|
|
|
177 |
{
|
|
|
178 |
uint32_t a,b,c;
|
|
|
179 |
|
|
|
180 |
/* Set up the internal state */
|
|
|
181 |
a = b = c = 0xdeadbeef + (((uint32_t)length)<<2) + initval;
|
|
|
182 |
|
|
|
183 |
/*------------------------------------------------- handle most of the key */
|
|
|
184 |
while (length > 3)
|
|
|
185 |
{
|
|
|
186 |
a += k[0];
|
|
|
187 |
b += k[1];
|
|
|
188 |
c += k[2];
|
|
|
189 |
mix(a,b,c);
|
|
|
190 |
length -= 3;
|
|
|
191 |
k += 3;
|
|
|
192 |
}
|
|
|
193 |
|
|
|
194 |
/*------------------------------------------- handle the last 3 uint32_t's */
|
|
|
195 |
switch(length) /* all the case statements fall through */
|
|
|
196 |
{
|
|
|
197 |
case 3 : c+=k[2];
|
|
|
198 |
case 2 : b+=k[1];
|
|
|
199 |
case 1 : a+=k[0];
|
|
|
200 |
final(a,b,c);
|
|
|
201 |
case 0: /* case 0: nothing left to add */
|
|
|
202 |
break;
|
|
|
203 |
}
|
|
|
204 |
/*------------------------------------------------------ report the result */
|
|
|
205 |
return c;
|
|
|
206 |
}
|
|
|
207 |
|
|
|
208 |
/*
|
|
|
209 |
* --------------------------------------------------------------------
|
|
|
210 |
* hashword2() -- same as hashword(), but take two seeds and return two
|
|
|
211 |
* 32-bit values. pc and pb must both be nonnull, and *pc and *pb must
|
|
|
212 |
* both be initialized with seeds. If you pass in (*pb)==0, the output
|
|
|
213 |
* (*pc) will be the same as the return value from hashword().
|
|
|
214 |
* --------------------------------------------------------------------
|
|
|
215 |
*/
|
|
|
216 |
void hashword2 (
|
|
|
217 |
const uint32_t *k, /* the key, an array of uint32_t values */
|
|
|
218 |
size_t length, /* the length of the key, in uint32_ts */
|
|
|
219 |
uint32_t *pc, /* IN: seed OUT: primary hash value */
|
|
|
220 |
uint32_t *pb) /* IN: more seed OUT: secondary hash value */
|
|
|
221 |
{
|
|
|
222 |
uint32_t a,b,c;
|
|
|
223 |
|
|
|
224 |
/* Set up the internal state */
|
|
|
225 |
a = b = c = 0xdeadbeef + ((uint32_t)(length<<2)) + *pc;
|
|
|
226 |
c += *pb;
|
|
|
227 |
|
|
|
228 |
/*------------------------------------------------- handle most of the key */
|
|
|
229 |
while (length > 3)
|
|
|
230 |
{
|
|
|
231 |
a += k[0];
|
|
|
232 |
b += k[1];
|
|
|
233 |
c += k[2];
|
|
|
234 |
mix(a,b,c);
|
|
|
235 |
length -= 3;
|
|
|
236 |
k += 3;
|
|
|
237 |
}
|
|
|
238 |
|
|
|
239 |
/*------------------------------------------- handle the last 3 uint32_t's */
|
|
|
240 |
switch(length) /* all the case statements fall through */
|
|
|
241 |
{
|
|
|
242 |
case 3: c+=k[2];
|
|
|
243 |
case 2: b+=k[1];
|
|
|
244 |
case 1: a+=k[0];
|
|
|
245 |
final(a,b,c);
|
|
|
246 |
case 0: /* case 0: nothing left to add */
|
|
|
247 |
break;
|
|
|
248 |
}
|
|
|
249 |
/*------------------------------------------------------ report the result */
|
|
|
250 |
|
|
|
251 |
*pc=c;
|
|
|
252 |
*pb=b;
|
|
|
253 |
}
|
|
|
254 |
|
|
|
255 |
/*
|
|
|
256 |
* -------------------------------------------------------------------------------
|
|
|
257 |
* hashlittle() -- hash a variable-length key into a 32-bit value
|
|
|
258 |
* k : the key (the unaligned variable-length array of bytes)
|
|
|
259 |
* length : the length of the key, counting by bytes
|
|
|
260 |
* initval : can be any 4-byte value
|
|
|
261 |
* Returns a 32-bit value. Every bit of the key affects every bit of
|
|
|
262 |
* the return value. Two keys differing by one or two bits will have
|
|
|
263 |
* totally different hash values.
|
|
|
264 |
*
|
|
|
265 |
* The best hash table sizes are powers of 2. There is no need to do
|
|
|
266 |
* mod a prime (mod is sooo slow!). If you need less than 32 bits,
|
|
|
267 |
* use a bitmask. For example, if you need only 10 bits, do
|
|
|
268 |
* h = (h & hashmask(10));
|
|
|
269 |
* In which case, the hash table should have hashsize(10) elements.
|
|
|
270 |
*
|
|
|
271 |
* If you are hashing n strings (uint8_t **)k, do it like this:
|
|
|
272 |
* for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h);
|
|
|
273 |
*
|
|
|
274 |
* By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this
|
|
|
275 |
* code any way you wish, private, educational, or commercial. It's free.
|
|
|
276 |
*
|
|
|
277 |
* Use for hash table lookup, or anything where one collision in 2^^32 is
|
|
|
278 |
* acceptable. Do NOT use for cryptographic purposes.
|
|
|
279 |
* -------------------------------------------------------------------------------
|
|
|
280 |
*/
|
|
|
281 |
|
|
|
282 |
uint32_t hashlittle( const void *key, size_t length, uint32_t initval)
|
|
|
283 |
{
|
|
|
284 |
uint32_t a,b,c; /* internal state */
|
|
|
285 |
union { const void *ptr; size_t i; } u; /* needed for Mac Powerbook G4 */
|
|
|
286 |
|
|
|
287 |
/* Set up the internal state */
|
|
|
288 |
a = b = c = 0xdeadbeef + ((uint32_t)length) + initval;
|
|
|
289 |
u.ptr = key;
|
|
|
290 |
|
|
|
291 |
if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0))
|
|
|
292 |
{
|
|
|
293 |
const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */
|
|
|
294 |
const uint8_t *k8;
|
|
|
295 |
|
|
|
296 |
/*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
|
|
|
297 |
while (length > 12)
|
|
|
298 |
{
|
|
|
299 |
a += k[0];
|
|
|
300 |
b += k[1];
|
|
|
301 |
c += k[2];
|
|
|
302 |
mix(a,b,c);
|
|
|
303 |
length -= 12;
|
|
|
304 |
k += 3;
|
|
|
305 |
}
|
|
|
306 |
|
|
|
307 |
/*----------------------------- handle the last (probably partial) block */
|
|
|
308 |
/*
|
|
|
309 |
* "k[2]&0xffffff" actually reads beyond the end of the string, but
|
|
|
310 |
* then masks off the part it's not allowed to read. Because the
|
|
|
311 |
* string is aligned, the masked-off tail is in the same word as the
|
|
|
312 |
* rest of the string. Every machine with memory protection I've seen
|
|
|
313 |
* does it on word boundaries, so is OK with this. But VALGRIND will
|
|
|
314 |
* still catch it and complain. The masking trick does make the hash
|
|
|
315 |
* noticably faster for short strings (like English words).
|
|
|
316 |
*/
|
|
|
317 |
#ifndef VALGRIND
|
|
|
318 |
|
|
|
319 |
switch(length)
|
|
|
320 |
{
|
|
|
321 |
case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
|
|
|
322 |
case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
|
|
|
323 |
case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break;
|
|
|
324 |
case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break;
|
|
|
325 |
case 8 : b+=k[1]; a+=k[0]; break;
|
|
|
326 |
case 7 : b+=k[1]&0xffffff; a+=k[0]; break;
|
|
|
327 |
case 6 : b+=k[1]&0xffff; a+=k[0]; break;
|
|
|
328 |
case 5 : b+=k[1]&0xff; a+=k[0]; break;
|
|
|
329 |
case 4 : a+=k[0]; break;
|
|
|
330 |
case 3 : a+=k[0]&0xffffff; break;
|
|
|
331 |
case 2 : a+=k[0]&0xffff; break;
|
|
|
332 |
case 1 : a+=k[0]&0xff; break;
|
|
|
333 |
case 0 : return c; /* zero length strings require no mixing */
|
|
|
334 |
}
|
|
|
335 |
#else /* make valgrind happy */
|
|
|
336 |
k8 = (const uint8_t *)k;
|
|
|
337 |
switch(length)
|
|
|
338 |
{
|
|
|
339 |
case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
|
|
|
340 |
case 11: c+=((uint32_t)k8[10])<<16; /* fall through */
|
|
|
341 |
case 10: c+=((uint32_t)k8[9])<<8; /* fall through */
|
|
|
342 |
case 9 : c+=k8[8]; /* fall through */
|
|
|
343 |
case 8 : b+=k[1]; a+=k[0]; break;
|
|
|
344 |
case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */
|
|
|
345 |
case 6 : b+=((uint32_t)k8[5])<<8; /* fall through */
|
|
|
346 |
case 5 : b+=k8[4]; /* fall through */
|
|
|
347 |
case 4 : a+=k[0]; break;
|
|
|
348 |
case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */
|
|
|
349 |
case 2 : a+=((uint32_t)k8[1])<<8; /* fall through */
|
|
|
350 |
case 1 : a+=k8[0]; break;
|
|
|
351 |
case 0 : return c;
|
|
|
352 |
}
|
|
|
353 |
#endif /* !valgrind */
|
|
|
354 |
} else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0))
|
|
|
355 |
{
|
|
|
356 |
const uint16_t *k = (const uint16_t *)key; /* read 16-bit chunks */
|
|
|
357 |
const uint8_t *k8;
|
|
|
358 |
|
|
|
359 |
/*--------------- all but last block: aligned reads and different mixing */
|
|
|
360 |
while (length > 12)
|
|
|
361 |
{
|
|
|
362 |
a += k[0] + (((uint32_t)k[1])<<16);
|
|
|
363 |
b += k[2] + (((uint32_t)k[3])<<16);
|
|
|
364 |
c += k[4] + (((uint32_t)k[5])<<16);
|
|
|
365 |
mix(a,b,c);
|
|
|
366 |
length -= 12;
|
|
|
367 |
k += 6;
|
|
|
368 |
}
|
|
|
369 |
/*----------------------------- handle the last (probably partial) block */
|
|
|
370 |
k8 = (const uint8_t *)k;
|
|
|
371 |
|
|
|
372 |
switch(length)
|
|
|
373 |
{
|
|
|
374 |
case 12:
|
|
|
375 |
c+=k[4]+(((uint32_t)k[5])<<16);
|
|
|
376 |
b+=k[2]+(((uint32_t)k[3])<<16);
|
|
|
377 |
a+=k[0]+(((uint32_t)k[1])<<16);
|
|
|
378 |
break;
|
|
|
379 |
|
|
|
380 |
case 11: c+=((uint32_t)k8[10])<<16; /* fall through */
|
|
|
381 |
|
|
|
382 |
case 10:
|
|
|
383 |
c+=k[4];
|
|
|
384 |
b+=k[2]+(((uint32_t)k[3])<<16);
|
|
|
385 |
a+=k[0]+(((uint32_t)k[1])<<16);
|
|
|
386 |
break;
|
|
|
387 |
|
|
|
388 |
case 9 : c+=k8[8]; /* fall through */
|
|
|
389 |
|
|
|
390 |
case 8 :
|
|
|
391 |
b+=k[2]+(((uint32_t)k[3])<<16);
|
|
|
392 |
a+=k[0]+(((uint32_t)k[1])<<16);
|
|
|
393 |
break;
|
|
|
394 |
|
|
|
395 |
case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */
|
|
|
396 |
|
|
|
397 |
case 6 :
|
|
|
398 |
b+=k[2];
|
|
|
399 |
a+=k[0]+(((uint32_t)k[1])<<16);
|
|
|
400 |
break;
|
|
|
401 |
|
|
|
402 |
case 5 : b+=k8[4]; /* fall through */
|
|
|
403 |
case 4 : a+=k[0]+(((uint32_t)k[1])<<16); break;
|
|
|
404 |
case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */
|
|
|
405 |
case 2 : a+=k[0]; break;
|
|
|
406 |
case 1 : a+=k8[0]; break;
|
|
|
407 |
case 0 : return c; /* zero length requires no mixing */
|
|
|
408 |
}
|
|
|
409 |
}
|
|
|
410 |
else /* need to read the key one byte at a time */
|
|
|
411 |
{
|
|
|
412 |
const uint8_t *k = (const uint8_t *)key;
|
|
|
413 |
|
|
|
414 |
/*--------------- all but the last block: affect some 32 bits of (a,b,c) */
|
|
|
415 |
while (length > 12)
|
|
|
416 |
{
|
|
|
417 |
a += k[0];
|
|
|
418 |
a += ((uint32_t)k[1])<<8;
|
|
|
419 |
a += ((uint32_t)k[2])<<16;
|
|
|
420 |
a += ((uint32_t)k[3])<<24;
|
|
|
421 |
b += k[4];
|
|
|
422 |
b += ((uint32_t)k[5])<<8;
|
|
|
423 |
b += ((uint32_t)k[6])<<16;
|
|
|
424 |
b += ((uint32_t)k[7])<<24;
|
|
|
425 |
c += k[8];
|
|
|
426 |
c += ((uint32_t)k[9])<<8;
|
|
|
427 |
c += ((uint32_t)k[10])<<16;
|
|
|
428 |
c += ((uint32_t)k[11])<<24;
|
|
|
429 |
mix(a,b,c);
|
|
|
430 |
length -= 12;
|
|
|
431 |
k += 12;
|
|
|
432 |
}
|
|
|
433 |
|
|
|
434 |
/*-------------------------------- last block: affect all 32 bits of (c) */
|
|
|
435 |
switch(length) /* all the case statements fall through */
|
|
|
436 |
{
|
|
|
437 |
case 12: c+=((uint32_t)k[11])<<24;
|
|
|
438 |
case 11: c+=((uint32_t)k[10])<<16;
|
|
|
439 |
case 10: c+=((uint32_t)k[9])<<8;
|
|
|
440 |
case 9 : c+=k[8];
|
|
|
441 |
case 8 : b+=((uint32_t)k[7])<<24;
|
|
|
442 |
case 7 : b+=((uint32_t)k[6])<<16;
|
|
|
443 |
case 6 : b+=((uint32_t)k[5])<<8;
|
|
|
444 |
case 5 : b+=k[4];
|
|
|
445 |
case 4 : a+=((uint32_t)k[3])<<24;
|
|
|
446 |
case 3 : a+=((uint32_t)k[2])<<16;
|
|
|
447 |
case 2 : a+=((uint32_t)k[1])<<8;
|
|
|
448 |
case 1 : a+=k[0];
|
|
|
449 |
break;
|
|
|
450 |
case 0 : return c;
|
|
|
451 |
}
|
|
|
452 |
}
|
|
|
453 |
|
|
|
454 |
final(a,b,c);
|
|
|
455 |
return c;
|
|
|
456 |
}
|
|
|
457 |
|
|
|
458 |
/*
|
|
|
459 |
* hashlittle2: return 2 32-bit hash values
|
|
|
460 |
*
|
|
|
461 |
* This is identical to hashlittle(), except it returns two 32-bit hash
|
|
|
462 |
* values instead of just one. This is good enough for hash table
|
|
|
463 |
* lookup with 2^^64 buckets, or if you want a second hash if you're not
|
|
|
464 |
* happy with the first, or if you want a probably-unique 64-bit ID for
|
|
|
465 |
* the key. *pc is better mixed than *pb, so use *pc first. If you want
|
|
|
466 |
* a 64-bit value do something like "*pc + (((uint64_t)*pb)<<32)".
|
|
|
467 |
*/
|
|
|
468 |
void hashlittle2(
|
|
|
469 |
const void *key, /* the key to hash */
|
|
|
470 |
size_t length, /* length of the key */
|
|
|
471 |
uint32_t *pc, /* IN: primary initval, OUT: primary hash */
|
|
|
472 |
uint32_t *pb) /* IN: secondary initval, OUT: secondary hash */
|
|
|
473 |
{
|
|
|
474 |
uint32_t a,b,c; /* internal state */
|
|
|
475 |
union { const void *ptr; size_t i; } u; /* needed for Mac Powerbook G4 */
|
|
|
476 |
|
|
|
477 |
/* Set up the internal state */
|
|
|
478 |
a = b = c = 0xdeadbeef + ((uint32_t)length) + *pc;
|
|
|
479 |
c += *pb;
|
|
|
480 |
|
|
|
481 |
u.ptr = key;
|
|
|
482 |
|
|
|
483 |
if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0))
|
|
|
484 |
{
|
|
|
485 |
const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */
|
|
|
486 |
const uint8_t *k8;
|
|
|
487 |
|
|
|
488 |
/*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
|
|
|
489 |
while (length > 12)
|
|
|
490 |
{
|
|
|
491 |
a += k[0];
|
|
|
492 |
b += k[1];
|
|
|
493 |
c += k[2];
|
|
|
494 |
mix(a,b,c);
|
|
|
495 |
length -= 12;
|
|
|
496 |
k += 3;
|
|
|
497 |
}
|
|
|
498 |
|
|
|
499 |
/*----------------------------- handle the last (probably partial) block */
|
|
|
500 |
/*
|
|
|
501 |
* "k[2]&0xffffff" actually reads beyond the end of the string, but
|
|
|
502 |
* then masks off the part it's not allowed to read. Because the
|
|
|
503 |
* string is aligned, the masked-off tail is in the same word as the
|
|
|
504 |
* rest of the string. Every machine with memory protection I've seen
|
|
|
505 |
* does it on word boundaries, so is OK with this. But VALGRIND will
|
|
|
506 |
* still catch it and complain. The masking trick does make the hash
|
|
|
507 |
* noticably faster for short strings (like English words).
|
|
|
508 |
*/
|
|
|
509 |
#ifndef VALGRIND
|
|
|
510 |
switch(length)
|
|
|
511 |
{
|
|
|
512 |
case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
|
|
|
513 |
case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
|
|
|
514 |
case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break;
|
|
|
515 |
case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break;
|
|
|
516 |
case 8 : b+=k[1]; a+=k[0]; break;
|
|
|
517 |
case 7 : b+=k[1]&0xffffff; a+=k[0]; break;
|
|
|
518 |
case 6 : b+=k[1]&0xffff; a+=k[0]; break;
|
|
|
519 |
case 5 : b+=k[1]&0xff; a+=k[0]; break;
|
|
|
520 |
case 4 : a+=k[0]; break;
|
|
|
521 |
case 3 : a+=k[0]&0xffffff; break;
|
|
|
522 |
case 2 : a+=k[0]&0xffff; break;
|
|
|
523 |
case 1 : a+=k[0]&0xff; break;
|
|
|
524 |
case 0 : *pc=c; *pb=b; return; /* zero length strings require no mixing */
|
|
|
525 |
}
|
|
|
526 |
#else /* make valgrind happy */
|
|
|
527 |
k8 = (const uint8_t *)k;
|
|
|
528 |
switch(length)
|
|
|
529 |
{
|
|
|
530 |
case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
|
|
|
531 |
case 11: c+=((uint32_t)k8[10])<<16; /* fall through */
|
|
|
532 |
case 10: c+=((uint32_t)k8[9])<<8; /* fall through */
|
|
|
533 |
case 9 : c+=k8[8]; /* fall through */
|
|
|
534 |
case 8 : b+=k[1]; a+=k[0]; break;
|
|
|
535 |
case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */
|
|
|
536 |
case 6 : b+=((uint32_t)k8[5])<<8; /* fall through */
|
|
|
537 |
case 5 : b+=k8[4]; /* fall through */
|
|
|
538 |
case 4 : a+=k[0]; break;
|
|
|
539 |
case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */
|
|
|
540 |
case 2 : a+=((uint32_t)k8[1])<<8; /* fall through */
|
|
|
541 |
case 1 : a+=k8[0]; break;
|
|
|
542 |
case 0 : *pc=c; *pb=b; return; /* zero length strings require no mixing */
|
|
|
543 |
}
|
|
|
544 |
#endif /* !valgrind */
|
|
|
545 |
|
|
|
546 |
}
|
|
|
547 |
else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0))
|
|
|
548 |
{
|
|
|
549 |
const uint16_t *k = (const uint16_t *)key; /* read 16-bit chunks */
|
|
|
550 |
const uint8_t *k8;
|
|
|
551 |
|
|
|
552 |
/*--------------- all but last block: aligned reads and different mixing */
|
|
|
553 |
while (length > 12)
|
|
|
554 |
{
|
|
|
555 |
a += k[0] + (((uint32_t)k[1])<<16);
|
|
|
556 |
b += k[2] + (((uint32_t)k[3])<<16);
|
|
|
557 |
c += k[4] + (((uint32_t)k[5])<<16);
|
|
|
558 |
mix(a,b,c);
|
|
|
559 |
length -= 12;
|
|
|
560 |
k += 6;
|
|
|
561 |
}
|
|
|
562 |
|
|
|
563 |
/*----------------------------- handle the last (probably partial) block */
|
|
|
564 |
k8 = (const uint8_t *)k;
|
|
|
565 |
switch(length)
|
|
|
566 |
{
|
|
|
567 |
case 12:
|
|
|
568 |
c+=k[4]+(((uint32_t)k[5])<<16);
|
|
|
569 |
b+=k[2]+(((uint32_t)k[3])<<16);
|
|
|
570 |
a+=k[0]+(((uint32_t)k[1])<<16);
|
|
|
571 |
break;
|
|
|
572 |
|
|
|
573 |
case 11: c+=((uint32_t)k8[10])<<16; /* fall through */
|
|
|
574 |
|
|
|
575 |
case 10:
|
|
|
576 |
c+=k[4];
|
|
|
577 |
b+=k[2]+(((uint32_t)k[3])<<16);
|
|
|
578 |
a+=k[0]+(((uint32_t)k[1])<<16);
|
|
|
579 |
break;
|
|
|
580 |
|
|
|
581 |
case 9 : c+=k8[8]; /* fall through */
|
|
|
582 |
|
|
|
583 |
case 8 :
|
|
|
584 |
b+=k[2]+(((uint32_t)k[3])<<16);
|
|
|
585 |
a+=k[0]+(((uint32_t)k[1])<<16);
|
|
|
586 |
break;
|
|
|
587 |
|
|
|
588 |
case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */
|
|
|
589 |
|
|
|
590 |
case 6 :
|
|
|
591 |
b+=k[2];
|
|
|
592 |
a+=k[0]+(((uint32_t)k[1])<<16);
|
|
|
593 |
break;
|
|
|
594 |
|
|
|
595 |
case 5 : b+=k8[4]; /* fall through */
|
|
|
596 |
case 4 : a+=k[0]+(((uint32_t)k[1])<<16); break;
|
|
|
597 |
case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */
|
|
|
598 |
case 2 : a+=k[0]; break;
|
|
|
599 |
case 1 : a+=k8[0]; break;
|
|
|
600 |
case 0 : *pc=c; *pb=b; return; /* zero length strings require no mixing */
|
|
|
601 |
}
|
|
|
602 |
}
|
|
|
603 |
else /* need to read the key one byte at a time */
|
|
|
604 |
{
|
|
|
605 |
const uint8_t *k = (const uint8_t *)key;
|
|
|
606 |
|
|
|
607 |
/*--------------- all but the last block: affect some 32 bits of (a,b,c) */
|
|
|
608 |
while (length > 12)
|
|
|
609 |
{
|
|
|
610 |
a += k[0];
|
|
|
611 |
a += ((uint32_t)k[1])<<8;
|
|
|
612 |
a += ((uint32_t)k[2])<<16;
|
|
|
613 |
a += ((uint32_t)k[3])<<24;
|
|
|
614 |
b += k[4];
|
|
|
615 |
b += ((uint32_t)k[5])<<8;
|
|
|
616 |
b += ((uint32_t)k[6])<<16;
|
|
|
617 |
b += ((uint32_t)k[7])<<24;
|
|
|
618 |
c += k[8];
|
|
|
619 |
c += ((uint32_t)k[9])<<8;
|
|
|
620 |
c += ((uint32_t)k[10])<<16;
|
|
|
621 |
c += ((uint32_t)k[11])<<24;
|
|
|
622 |
mix(a,b,c);
|
|
|
623 |
length -= 12;
|
|
|
624 |
k += 12;
|
|
|
625 |
}
|
|
|
626 |
|
|
|
627 |
/*-------------------------------- last block: affect all 32 bits of (c) */
|
|
|
628 |
switch(length) /* all the case statements fall through */
|
|
|
629 |
{
|
|
|
630 |
case 12: c+=((uint32_t)k[11])<<24;
|
|
|
631 |
case 11: c+=((uint32_t)k[10])<<16;
|
|
|
632 |
case 10: c+=((uint32_t)k[9])<<8;
|
|
|
633 |
case 9 : c+=k[8];
|
|
|
634 |
case 8 : b+=((uint32_t)k[7])<<24;
|
|
|
635 |
case 7 : b+=((uint32_t)k[6])<<16;
|
|
|
636 |
case 6 : b+=((uint32_t)k[5])<<8;
|
|
|
637 |
case 5 : b+=k[4];
|
|
|
638 |
case 4 : a+=((uint32_t)k[3])<<24;
|
|
|
639 |
case 3 : a+=((uint32_t)k[2])<<16;
|
|
|
640 |
case 2 : a+=((uint32_t)k[1])<<8;
|
|
|
641 |
case 1 : a+=k[0]; break;
|
|
|
642 |
case 0 : *pc=c; *pb=b; return; /* zero length strings require no mixing */
|
|
|
643 |
}
|
|
|
644 |
}
|
|
|
645 |
|
|
|
646 |
final(a,b,c);
|
|
|
647 |
*pc=c; *pb=b;
|
|
|
648 |
}
|
|
|
649 |
|
|
|
650 |
/*
|
|
|
651 |
* hashbig():
|
|
|
652 |
* This is the same as hashword() on big-endian machines. It is different
|
|
|
653 |
* from hashlittle() on all machines. hashbig() takes advantage of
|
|
|
654 |
* big-endian byte ordering.
|
|
|
655 |
*/
|
|
|
656 |
uint32_t hashbig( const void *key, size_t length, uint32_t initval)
|
|
|
657 |
{
|
|
|
658 |
uint32_t a,b,c;
|
|
|
659 |
union { const void *ptr; size_t i; } u; /* to cast key to (size_t) happily */
|
|
|
660 |
|
|
|
661 |
/* Set up the internal state */
|
|
|
662 |
a = b = c = 0xdeadbeef + ((uint32_t)length) + initval;
|
|
|
663 |
u.ptr = key;
|
|
|
664 |
|
|
|
665 |
if (HASH_BIG_ENDIAN && ((u.i & 0x3) == 0))
|
|
|
666 |
{
|
|
|
667 |
const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */
|
|
|
668 |
const uint8_t *k8;
|
|
|
669 |
|
|
|
670 |
/*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
|
|
|
671 |
while (length > 12)
|
|
|
672 |
{
|
|
|
673 |
a += k[0];
|
|
|
674 |
b += k[1];
|
|
|
675 |
c += k[2];
|
|
|
676 |
mix(a,b,c);
|
|
|
677 |
length -= 12;
|
|
|
678 |
k += 3;
|
|
|
679 |
}
|
|
|
680 |
|
|
|
681 |
/*----------------------------- handle the last (probably partial) block */
|
|
|
682 |
/*
|
|
|
683 |
* "k[2]<<8" actually reads beyond the end of the string, but
|
|
|
684 |
* then shifts out the part it's not allowed to read. Because the
|
|
|
685 |
* string is aligned, the illegal read is in the same word as the
|
|
|
686 |
* rest of the string. Every machine with memory protection I've seen
|
|
|
687 |
* does it on word boundaries, so is OK with this. But VALGRIND will
|
|
|
688 |
* still catch it and complain. The masking trick does make the hash
|
|
|
689 |
* noticably faster for short strings (like English words).
|
|
|
690 |
*/
|
|
|
691 |
#ifndef VALGRIND
|
|
|
692 |
switch(length)
|
|
|
693 |
{
|
|
|
694 |
case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
|
|
|
695 |
case 11: c+=k[2]&0xffffff00; b+=k[1]; a+=k[0]; break;
|
|
|
696 |
case 10: c+=k[2]&0xffff0000; b+=k[1]; a+=k[0]; break;
|
|
|
697 |
case 9 : c+=k[2]&0xff000000; b+=k[1]; a+=k[0]; break;
|
|
|
698 |
case 8 : b+=k[1]; a+=k[0]; break;
|
|
|
699 |
case 7 : b+=k[1]&0xffffff00; a+=k[0]; break;
|
|
|
700 |
case 6 : b+=k[1]&0xffff0000; a+=k[0]; break;
|
|
|
701 |
case 5 : b+=k[1]&0xff000000; a+=k[0]; break;
|
|
|
702 |
case 4 : a+=k[0]; break;
|
|
|
703 |
case 3 : a+=k[0]&0xffffff00; break;
|
|
|
704 |
case 2 : a+=k[0]&0xffff0000; break;
|
|
|
705 |
case 1 : a+=k[0]&0xff000000; break;
|
|
|
706 |
case 0 : return c; /* zero length strings require no mixing */
|
|
|
707 |
}
|
|
|
708 |
#else /* make valgrind happy */
|
|
|
709 |
k8 = (const uint8_t *)k;
|
|
|
710 |
|
|
|
711 |
switch(length) /* all the case statements fall through */
|
|
|
712 |
{
|
|
|
713 |
case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
|
|
|
714 |
case 11: c+=((uint32_t)k8[10])<<8; /* fall through */
|
|
|
715 |
case 10: c+=((uint32_t)k8[9])<<16; /* fall through */
|
|
|
716 |
case 9 : c+=((uint32_t)k8[8])<<24; /* fall through */
|
|
|
717 |
case 8 : b+=k[1]; a+=k[0]; break;
|
|
|
718 |
case 7 : b+=((uint32_t)k8[6])<<8; /* fall through */
|
|
|
719 |
case 6 : b+=((uint32_t)k8[5])<<16; /* fall through */
|
|
|
720 |
case 5 : b+=((uint32_t)k8[4])<<24; /* fall through */
|
|
|
721 |
case 4 : a+=k[0]; break;
|
|
|
722 |
case 3 : a+=((uint32_t)k8[2])<<8; /* fall through */
|
|
|
723 |
case 2 : a+=((uint32_t)k8[1])<<16; /* fall through */
|
|
|
724 |
case 1 : a+=((uint32_t)k8[0])<<24; break;
|
|
|
725 |
case 0 : return c;
|
|
|
726 |
}
|
|
|
727 |
#endif /* !VALGRIND */
|
|
|
728 |
|
|
|
729 |
}
|
|
|
730 |
else /* need to read the key one byte at a time */
|
|
|
731 |
{
|
|
|
732 |
const uint8_t *k = (const uint8_t *)key;
|
|
|
733 |
|
|
|
734 |
/*--------------- all but the last block: affect some 32 bits of (a,b,c) */
|
|
|
735 |
while (length > 12)
|
|
|
736 |
{
|
|
|
737 |
a += ((uint32_t)k[0])<<24;
|
|
|
738 |
a += ((uint32_t)k[1])<<16;
|
|
|
739 |
a += ((uint32_t)k[2])<<8;
|
|
|
740 |
a += ((uint32_t)k[3]);
|
|
|
741 |
b += ((uint32_t)k[4])<<24;
|
|
|
742 |
b += ((uint32_t)k[5])<<16;
|
|
|
743 |
b += ((uint32_t)k[6])<<8;
|
|
|
744 |
b += ((uint32_t)k[7]);
|
|
|
745 |
c += ((uint32_t)k[8])<<24;
|
|
|
746 |
c += ((uint32_t)k[9])<<16;
|
|
|
747 |
c += ((uint32_t)k[10])<<8;
|
|
|
748 |
c += ((uint32_t)k[11]);
|
|
|
749 |
mix(a,b,c);
|
|
|
750 |
length -= 12;
|
|
|
751 |
k += 12;
|
|
|
752 |
}
|
|
|
753 |
|
|
|
754 |
/*-------------------------------- last block: affect all 32 bits of (c) */
|
|
|
755 |
switch(length) /* all the case statements fall through */
|
|
|
756 |
{
|
|
|
757 |
case 12: c+=k[11];
|
|
|
758 |
case 11: c+=((uint32_t)k[10])<<8;
|
|
|
759 |
case 10: c+=((uint32_t)k[9])<<16;
|
|
|
760 |
case 9 : c+=((uint32_t)k[8])<<24;
|
|
|
761 |
case 8 : b+=k[7];
|
|
|
762 |
case 7 : b+=((uint32_t)k[6])<<8;
|
|
|
763 |
case 6 : b+=((uint32_t)k[5])<<16;
|
|
|
764 |
case 5 : b+=((uint32_t)k[4])<<24;
|
|
|
765 |
case 4 : a+=k[3];
|
|
|
766 |
case 3 : a+=((uint32_t)k[2])<<8;
|
|
|
767 |
case 2 : a+=((uint32_t)k[1])<<16;
|
|
|
768 |
case 1 : a+=((uint32_t)k[0])<<24; break;
|
|
|
769 |
case 0 : return c;
|
|
|
770 |
}
|
|
|
771 |
}
|
|
|
772 |
|
|
|
773 |
final(a,b,c);
|
|
|
774 |
return c;
|
|
|
775 |
}
|
|
|
776 |
|
|
|
777 |
#ifdef SELF_TEST
|
|
|
778 |
/* used for timings */
|
|
|
779 |
void driver1()
|
|
|
780 |
{
|
|
|
781 |
uint8_t buf[256];
|
|
|
782 |
uint32_t i;
|
|
|
783 |
uint32_t h=0;
|
|
|
784 |
time_t a,z;
|
|
|
785 |
|
|
|
786 |
time(&a);
|
|
|
787 |
|
|
|
788 |
for (i=0; i<256; ++i)
|
|
|
789 |
buf[i] = 'x';
|
|
|
790 |
|
|
|
791 |
for (i=0; i<1; ++i)
|
|
|
792 |
h = hashlittle(&buf[0],1,h);
|
|
|
793 |
|
|
|
794 |
time(&z);
|
|
|
795 |
|
|
|
796 |
if (z-a > 0)
|
|
|
797 |
printf("time %d %.8x\n", z-a, h);
|
|
|
798 |
}
|
|
|
799 |
|
|
|
800 |
/* check that every input bit changes every output bit half the time */
|
|
|
801 |
#define HASHSTATE 1
|
|
|
802 |
#define HASHLEN 1
|
|
|
803 |
#define MAXPAIR 60
|
|
|
804 |
#define MAXLEN 70
|
|
|
805 |
void driver2()
|
|
|
806 |
{
|
|
|
807 |
uint8_t qa[MAXLEN+1], qb[MAXLEN+2], *a = &qa[0], *b = &qb[1];
|
|
|
808 |
uint32_t c[HASHSTATE], d[HASHSTATE], i=0, j=0, k, l, m=0, z;
|
|
|
809 |
uint32_t e[HASHSTATE],f[HASHSTATE],g[HASHSTATE],h[HASHSTATE];
|
|
|
810 |
uint32_t x[HASHSTATE],y[HASHSTATE];
|
|
|
811 |
uint32_t hlen;
|
|
|
812 |
|
|
|
813 |
printf("No more than %d trials should ever be needed \n",MAXPAIR/2);
|
|
|
814 |
|
|
|
815 |
for (hlen=0; hlen < MAXLEN; ++hlen)
|
|
|
816 |
{
|
|
|
817 |
z=0;
|
|
|
818 |
|
|
|
819 |
for (i=0; i<hlen; ++i) /*----------------------- for each input byte, */
|
|
|
820 |
{
|
|
|
821 |
for (j=0; j<8; ++j) /*------------------------ for each input bit, */
|
|
|
822 |
{
|
|
|
823 |
for (m=1; m<8; ++m) /*------------ for serveral possible initvals, */
|
|
|
824 |
{
|
|
|
825 |
for (l=0; l<HASHSTATE; ++l)
|
|
|
826 |
e[l]=f[l]=g[l]=h[l]=x[l]=y[l]=~((uint32_t)0);
|
|
|
827 |
|
|
|
828 |
/*---- check that every output bit is affected by that input bit */
|
|
|
829 |
for (k=0; k<MAXPAIR; k+=2)
|
|
|
830 |
{
|
|
|
831 |
uint32_t finished=1;
|
|
|
832 |
/* keys have one bit different */
|
|
|
833 |
for (l=0; l<hlen+1; ++l)
|
|
|
834 |
a[l] = b[l] = (uint8_t)0;
|
|
|
835 |
|
|
|
836 |
/* have a and b be two keys differing in only one bit */
|
|
|
837 |
a[i] ^= (k<<j);
|
|
|
838 |
a[i] ^= (k>>(8-j));
|
|
|
839 |
c[0] = hashlittle(a, hlen, m);
|
|
|
840 |
b[i] ^= ((k+1)<<j);
|
|
|
841 |
b[i] ^= ((k+1)>>(8-j));
|
|
|
842 |
d[0] = hashlittle(b, hlen, m);
|
|
|
843 |
/* check every bit is 1, 0, set, and not set at least once */
|
|
|
844 |
for (l=0; l<HASHSTATE; ++l)
|
|
|
845 |
{
|
|
|
846 |
e[l] &= (c[l]^d[l]);
|
|
|
847 |
f[l] &= ~(c[l]^d[l]);
|
|
|
848 |
g[l] &= c[l];
|
|
|
849 |
h[l] &= ~c[l];
|
|
|
850 |
x[l] &= d[l];
|
|
|
851 |
y[l] &= ~d[l];
|
|
|
852 |
if (e[l]|f[l]|g[l]|h[l]|x[l]|y[l]) finished=0;
|
|
|
853 |
}
|
|
|
854 |
|
|
|
855 |
if (finished) break;
|
|
|
856 |
}
|
|
|
857 |
|
|
|
858 |
if (k>z)
|
|
|
859 |
z=k;
|
|
|
860 |
|
|
|
861 |
if (k==MAXPAIR)
|
|
|
862 |
{
|
|
|
863 |
printf("Some bit didn't change: ");
|
|
|
864 |
printf("%.8x %.8x %.8x %.8x %.8x %.8x ", e[0],f[0],g[0],h[0],x[0],y[0]);
|
|
|
865 |
printf("i %d j %d m %d len %d\n", i, j, m, hlen);
|
|
|
866 |
}
|
|
|
867 |
|
|
|
868 |
if (z==MAXPAIR)
|
|
|
869 |
goto done;
|
|
|
870 |
}
|
|
|
871 |
}
|
|
|
872 |
}
|
|
|
873 |
done:
|
|
|
874 |
if (z < MAXPAIR)
|
|
|
875 |
{
|
|
|
876 |
printf("Mix success %2d bytes %2d initvals ",i,m);
|
|
|
877 |
printf("required %d trials\n", z/2);
|
|
|
878 |
}
|
|
|
879 |
}
|
|
|
880 |
|
|
|
881 |
printf("\n");
|
|
|
882 |
}
|
|
|
883 |
|
|
|
884 |
/* Check for reading beyond the end of the buffer and alignment problems */
|
|
|
885 |
void driver3()
|
|
|
886 |
{
|
|
|
887 |
uint8_t buf[MAXLEN+20], *b;
|
|
|
888 |
uint32_t len;
|
|
|
889 |
uint8_t q[] = "This is the time for all good men to come to the aid of their country...";
|
|
|
890 |
uint32_t h;
|
|
|
891 |
uint8_t qq[] = "xThis is the time for all good men to come to the aid of their country...";
|
|
|
892 |
uint32_t i;
|
|
|
893 |
uint8_t qqq[] = "xxThis is the time for all good men to come to the aid of their country...";
|
|
|
894 |
uint32_t j;
|
|
|
895 |
uint8_t qqqq[] = "xxxThis is the time for all good men to come to the aid of their country...";
|
|
|
896 |
uint32_t ref,x,y;
|
|
|
897 |
uint8_t *p;
|
|
|
898 |
|
|
|
899 |
printf("Endianness. These lines should all be the same (for values filled in):\n");
|
|
|
900 |
printf("%.8x %.8x %.8x\n",
|
|
|
901 |
hashword((const uint32_t *)q, (sizeof(q)-1)/4, 13),
|
|
|
902 |
hashword((const uint32_t *)q, (sizeof(q)-5)/4, 13),
|
|
|
903 |
hashword((const uint32_t *)q, (sizeof(q)-9)/4, 13));
|
|
|
904 |
p = q;
|
|
|
905 |
printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
|
|
|
906 |
hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13),
|
|
|
907 |
hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13),
|
|
|
908 |
hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13),
|
|
|
909 |
hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13),
|
|
|
910 |
hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13),
|
|
|
911 |
hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13));
|
|
|
912 |
p = &qq[1];
|
|
|
913 |
printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
|
|
|
914 |
hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13),
|
|
|
915 |
hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13),
|
|
|
916 |
hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13),
|
|
|
917 |
hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13),
|
|
|
918 |
hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13),
|
|
|
919 |
hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13));
|
|
|
920 |
p = &qqq[2];
|
|
|
921 |
printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
|
|
|
922 |
hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13),
|
|
|
923 |
hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13),
|
|
|
924 |
hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13),
|
|
|
925 |
hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13),
|
|
|
926 |
hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13),
|
|
|
927 |
hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13));
|
|
|
928 |
p = &qqqq[3];
|
|
|
929 |
printf("%.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x %.8x\n",
|
|
|
930 |
hashlittle(p, sizeof(q)-1, 13), hashlittle(p, sizeof(q)-2, 13),
|
|
|
931 |
hashlittle(p, sizeof(q)-3, 13), hashlittle(p, sizeof(q)-4, 13),
|
|
|
932 |
hashlittle(p, sizeof(q)-5, 13), hashlittle(p, sizeof(q)-6, 13),
|
|
|
933 |
hashlittle(p, sizeof(q)-7, 13), hashlittle(p, sizeof(q)-8, 13),
|
|
|
934 |
hashlittle(p, sizeof(q)-9, 13), hashlittle(p, sizeof(q)-10, 13),
|
|
|
935 |
hashlittle(p, sizeof(q)-11, 13), hashlittle(p, sizeof(q)-12, 13));
|
|
|
936 |
printf("\n");
|
|
|
937 |
|
|
|
938 |
/* check that hashlittle2 and hashlittle produce the same results */
|
|
|
939 |
i=47; j=0;
|
|
|
940 |
hashlittle2(q, sizeof(q), &i, &j);
|
|
|
941 |
if (hashlittle(q, sizeof(q), 47) != i)
|
|
|
942 |
printf("hashlittle2 and hashlittle mismatch\n");
|
|
|
943 |
|
|
|
944 |
/* check that hashword2 and hashword produce the same results */
|
|
|
945 |
len = 0xdeadbeef;
|
|
|
946 |
i=47, j=0;
|
|
|
947 |
hashword2(&len, 1, &i, &j);
|
|
|
948 |
if (hashword(&len, 1, 47) != i)
|
|
|
949 |
printf("hashword2 and hashword mismatch %x %x\n",
|
|
|
950 |
i, hashword(&len, 1, 47));
|
|
|
951 |
|
|
|
952 |
/* check hashlittle doesn't read before or after the ends of the string */
|
|
|
953 |
for (h=0, b=buf+1; h<8; ++h, ++b)
|
|
|
954 |
{
|
|
|
955 |
for (i=0; i<MAXLEN; ++i)
|
|
|
956 |
{
|
|
|
957 |
len = i;
|
|
|
958 |
for (j=0; j<i; ++j) *(b+j)=0;
|
|
|
959 |
|
|
|
960 |
/* these should all be equal */
|
|
|
961 |
ref = hashlittle(b, len, (uint32_t)1);
|
|
|
962 |
*(b+i)=(uint8_t)~0;
|
|
|
963 |
*(b-1)=(uint8_t)~0;
|
|
|
964 |
x = hashlittle(b, len, (uint32_t)1);
|
|
|
965 |
y = hashlittle(b, len, (uint32_t)1);
|
|
|
966 |
if ((ref != x) || (ref != y))
|
|
|
967 |
{
|
|
|
968 |
printf("alignment error: %.8x %.8x %.8x %d %d\n",ref,x,y,
|
|
|
969 |
h, i);
|
|
|
970 |
}
|
|
|
971 |
}
|
|
|
972 |
}
|
|
|
973 |
}
|
|
|
974 |
|
|
|
975 |
/* check for problems with nulls */
|
|
|
976 |
void driver4()
|
|
|
977 |
{
|
|
|
978 |
uint8_t buf[1];
|
|
|
979 |
uint32_t h,i,state[HASHSTATE];
|
|
|
980 |
|
|
|
981 |
|
|
|
982 |
buf[0] = ~0;
|
|
|
983 |
for (i=0; i<HASHSTATE; ++i) state[i] = 1;
|
|
|
984 |
printf("These should all be different\n");
|
|
|
985 |
for (i=0, h=0; i<8; ++i)
|
|
|
986 |
{
|
|
|
987 |
h = hashlittle(buf, 0, h);
|
|
|
988 |
printf("%2ld 0-byte strings, hash is %.8x\n", i, h);
|
|
|
989 |
}
|
|
|
990 |
}
|
|
|
991 |
|
|
|
992 |
void driver5()
|
|
|
993 |
{
|
|
|
994 |
uint32_t b,c;
|
|
|
995 |
b=0, c=0, hashlittle2("", 0, &c, &b);
|
|
|
996 |
printf("hash is %.8lx %.8lx\n", c, b); /* deadbeef deadbeef */
|
|
|
997 |
b=0xdeadbeef, c=0, hashlittle2("", 0, &c, &b);
|
|
|
998 |
printf("hash is %.8lx %.8lx\n", c, b); /* bd5b7dde deadbeef */
|
|
|
999 |
b=0xdeadbeef, c=0xdeadbeef, hashlittle2("", 0, &c, &b);
|
|
|
1000 |
printf("hash is %.8lx %.8lx\n", c, b); /* 9c093ccd bd5b7dde */
|
|
|
1001 |
b=0, c=0, hashlittle2("Four score and seven years ago", 30, &c, &b);
|
|
|
1002 |
printf("hash is %.8lx %.8lx\n", c, b); /* 17770551 ce7226e6 */
|
|
|
1003 |
b=1, c=0, hashlittle2("Four score and seven years ago", 30, &c, &b);
|
|
|
1004 |
printf("hash is %.8lx %.8lx\n", c, b); /* e3607cae bd371de4 */
|
|
|
1005 |
b=0, c=1, hashlittle2("Four score and seven years ago", 30, &c, &b);
|
|
|
1006 |
printf("hash is %.8lx %.8lx\n", c, b); /* cd628161 6cbea4b3 */
|
|
|
1007 |
c = hashlittle("Four score and seven years ago", 30, 0);
|
|
|
1008 |
printf("hash is %.8lx\n", c); /* 17770551 */
|
|
|
1009 |
c = hashlittle("Four score and seven years ago", 30, 1);
|
|
|
1010 |
printf("hash is %.8lx\n", c); /* cd628161 */
|
|
|
1011 |
}
|
|
|
1012 |
|
|
|
1013 |
|
|
|
1014 |
int main()
|
|
|
1015 |
{
|
|
|
1016 |
driver1(); /* test that the key is hashed: used for timings */
|
|
|
1017 |
driver2(); /* test that whole key is hashed thoroughly */
|
|
|
1018 |
driver3(); /* test that nothing but the key is hashed */
|
|
|
1019 |
driver4(); /* test hashing multiple buffers (all buffers are null) */
|
|
|
1020 |
driver5(); /* test the hash against known vectors */
|
|
|
1021 |
return 1;
|
|
|
1022 |
}
|
|
|
1023 |
|
|
|
1024 |
#endif /* SELF_TEST */
|
|
|
1025 |
|