2d027bae0b9cc2350ea69337ea3412dd5727d79d
[lttng-tools.git] / src / common / hashtable / utils.c
1 /*
2 * Copyright (C) 2006 Bob Jenkins
3 * Copyright (C) 2011 David Goulet <david.goulet@polymtl.ca>
4 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
5 *
6 * SPDX-License-Identifier: GPL-2.0-only
7 *
8 */
9
10 /*
11 * These are functions for producing 32-bit hashes for hash table lookup.
12 * hashword(), hashlittle(), hashlittle2(), hashbig(), mix(), and final() are
13 * externally useful functions. Routines to test the hash are included if
14 * SELF_TEST is defined. You can use this free for any purpose. It's in the
15 * public domain. It has no warranty.
16 *
17 * You probably want to use hashlittle(). hashlittle() and hashbig() hash byte
18 * arrays. hashlittle() is is faster than hashbig() on little-endian machines.
19 * Intel and AMD are little-endian machines. On second thought, you probably
20 * want hashlittle2(), which is identical to hashlittle() except it returns two
21 * 32-bit hashes for the price of one. You could implement hashbig2() if you
22 * wanted but I haven't bothered here.
23 *
24 * If you want to find a hash of, say, exactly 7 integers, do
25 * a = i1; b = i2; c = i3;
26 * mix(a,b,c);
27 * a += i4; b += i5; c += i6;
28 * mix(a,b,c);
29 * a += i7;
30 * final(a,b,c);
31 * then use c as the hash value. If you have a variable length array of
32 * 4-byte integers to hash, use hashword(). If you have a byte array (like
33 * a character string), use hashlittle(). If you have several byte arrays, or
34 * a mix of things, see the comments above hashlittle().
35 *
36 * Why is this so big? I read 12 bytes at a time into 3 4-byte integers, then
37 * mix those integers. This is fast (you can do a lot more thorough mixing
38 * with 12*3 instructions on 3 integers than you can with 3 instructions on 1
39 * byte), but shoehorning those bytes into integers efficiently is messy.
40 */
41
42 #define _LGPL_SOURCE
43 #include <assert.h>
44 #include <stdint.h> /* defines uint32_t etc */
45 #include <stdio.h> /* defines printf for tests */
46 #include <string.h>
47 #include <sys/param.h> /* attempt to define endianness */
48 #include <time.h> /* defines time_t for timings in the test */
49 #include <urcu/compiler.h>
50
51 #include "utils.h"
52 #include <common/compat/endian.h> /* attempt to define endianness */
53 #include <common/common.h>
54 #include <common/hashtable/hashtable.h>
55
56 /*
57 * My best guess at if you are big-endian or little-endian. This may
58 * need adjustment.
59 */
60 #if (defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && \
61 __BYTE_ORDER == __LITTLE_ENDIAN) || \
62 (defined(i386) || defined(__i386__) || defined(__i486__) || \
63 defined(__i586__) || defined(__i686__) || defined(vax) || defined(MIPSEL))
64 # define HASH_LITTLE_ENDIAN 1
65 # define HASH_BIG_ENDIAN 0
66 #elif (defined(__BYTE_ORDER) && defined(__BIG_ENDIAN) && \
67 __BYTE_ORDER == __BIG_ENDIAN) || \
68 (defined(sparc) || defined(POWERPC) || defined(mc68000) || defined(sel))
69 # define HASH_LITTLE_ENDIAN 0
70 # define HASH_BIG_ENDIAN 1
71 #else
72 # define HASH_LITTLE_ENDIAN 0
73 # define HASH_BIG_ENDIAN 0
74 #endif
75
76 #define hashsize(n) ((uint32_t)1<<(n))
77 #define hashmask(n) (hashsize(n)-1)
78 #define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
79
80 /*
81 * mix -- mix 3 32-bit values reversibly.
82 *
83 * This is reversible, so any information in (a,b,c) before mix() is
84 * still in (a,b,c) after mix().
85 *
86 * If four pairs of (a,b,c) inputs are run through mix(), or through
87 * mix() in reverse, there are at least 32 bits of the output that
88 * are sometimes the same for one pair and different for another pair.
89 * This was tested for:
90 * * pairs that differed by one bit, by two bits, in any combination
91 * of top bits of (a,b,c), or in any combination of bottom bits of
92 * (a,b,c).
93 * * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
94 * the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
95 * is commonly produced by subtraction) look like a single 1-bit
96 * difference.
97 * * the base values were pseudorandom, all zero but one bit set, or
98 * all zero plus a counter that starts at zero.
99 *
100 * Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
101 * satisfy this are
102 * 4 6 8 16 19 4
103 * 9 15 3 18 27 15
104 * 14 9 3 7 17 3
105 * Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
106 * for "differ" defined as + with a one-bit base and a two-bit delta. I
107 * used http://burtleburtle.net/bob/hash/avalanche.html to choose
108 * the operations, constants, and arrangements of the variables.
109 *
110 * This does not achieve avalanche. There are input bits of (a,b,c)
111 * that fail to affect some output bits of (a,b,c), especially of a. The
112 * most thoroughly mixed value is c, but it doesn't really even achieve
113 * avalanche in c.
114 *
115 * This allows some parallelism. Read-after-writes are good at doubling
116 * the number of bits affected, so the goal of mixing pulls in the opposite
117 * direction as the goal of parallelism. I did what I could. Rotates
118 * seem to cost as much as shifts on every machine I could lay my hands
119 * on, and rotates are much kinder to the top and bottom bits, so I used
120 * rotates.
121 */
122 #define mix(a,b,c) \
123 { \
124 a -= c; a ^= rot(c, 4); c += b; \
125 b -= a; b ^= rot(a, 6); a += c; \
126 c -= b; c ^= rot(b, 8); b += a; \
127 a -= c; a ^= rot(c,16); c += b; \
128 b -= a; b ^= rot(a,19); a += c; \
129 c -= b; c ^= rot(b, 4); b += a; \
130 }
131
132 /*
133 * final -- final mixing of 3 32-bit values (a,b,c) into c
134 *
135 * Pairs of (a,b,c) values differing in only a few bits will usually
136 * produce values of c that look totally different. This was tested for
137 * * pairs that differed by one bit, by two bits, in any combination
138 * of top bits of (a,b,c), or in any combination of bottom bits of
139 * (a,b,c).
140 * * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
141 * the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
142 * is commonly produced by subtraction) look like a single 1-bit
143 * difference.
144 * * the base values were pseudorandom, all zero but one bit set, or
145 * all zero plus a counter that starts at zero.
146 *
147 * These constants passed:
148 * 14 11 25 16 4 14 24
149 * 12 14 25 16 4 14 24
150 * and these came close:
151 * 4 8 15 26 3 22 24
152 * 10 8 15 26 3 22 24
153 * 11 8 15 26 3 22 24
154 */
155 #define final(a,b,c) \
156 { \
157 c ^= b; c -= rot(b,14); \
158 a ^= c; a -= rot(c,11); \
159 b ^= a; b -= rot(a,25); \
160 c ^= b; c -= rot(b,16); \
161 a ^= c; a -= rot(c,4); \
162 b ^= a; b -= rot(a,14); \
163 c ^= b; c -= rot(b,24); \
164 }
165
166 /*
167 * k - the key, an array of uint32_t values
168 * length - the length of the key, in uint32_ts
169 * initval - the previous hash, or an arbitrary value
170 */
171 static uint32_t __attribute__((unused)) hashword(const uint32_t *k,
172 size_t length, uint32_t initval)
173 {
174 uint32_t a, b, c;
175
176 /* Set up the internal state */
177 a = b = c = 0xdeadbeef + (((uint32_t) length) << 2) + initval;
178
179 /*----------------------------------------- handle most of the key */
180 while (length > 3) {
181 a += k[0];
182 b += k[1];
183 c += k[2];
184 mix(a, b, c);
185 length -= 3;
186 k += 3;
187 }
188
189 /*----------------------------------- handle the last 3 uint32_t's */
190 switch (length) { /* all the case statements fall through */
191 case 3: c += k[2];
192 case 2: b += k[1];
193 case 1: a += k[0];
194 final(a, b, c);
195 case 0: /* case 0: nothing left to add */
196 break;
197 }
198 /*---------------------------------------------- report the result */
199 return c;
200 }
201
202
203 /*
204 * hashword2() -- same as hashword(), but take two seeds and return two 32-bit
205 * values. pc and pb must both be nonnull, and *pc and *pb must both be
206 * initialized with seeds. If you pass in (*pb)==0, the output (*pc) will be
207 * the same as the return value from hashword().
208 */
209 static void __attribute__((unused)) hashword2(const uint32_t *k, size_t length,
210 uint32_t *pc, uint32_t *pb)
211 {
212 uint32_t a, b, c;
213
214 /* Set up the internal state */
215 a = b = c = 0xdeadbeef + ((uint32_t) (length << 2)) + *pc;
216 c += *pb;
217
218 while (length > 3) {
219 a += k[0];
220 b += k[1];
221 c += k[2];
222 mix(a, b, c);
223 length -= 3;
224 k += 3;
225 }
226
227 switch (length) {
228 case 3 :
229 c += k[2];
230 case 2 :
231 b += k[1];
232 case 1 :
233 a += k[0];
234 final(a, b, c);
235 case 0: /* case 0: nothing left to add */
236 break;
237 }
238
239 *pc = c;
240 *pb = b;
241 }
242
243 /*
244 * hashlittle() -- hash a variable-length key into a 32-bit value
245 * k : the key (the unaligned variable-length array of bytes)
246 * length : the length of the key, counting by bytes
247 * initval : can be any 4-byte value
248 * Returns a 32-bit value. Every bit of the key affects every bit of
249 * the return value. Two keys differing by one or two bits will have
250 * totally different hash values.
251 *
252 * The best hash table sizes are powers of 2. There is no need to do
253 * mod a prime (mod is sooo slow!). If you need less than 32 bits,
254 * use a bitmask. For example, if you need only 10 bits, do
255 * h = (h & hashmask(10));
256 * In which case, the hash table should have hashsize(10) elements.
257 *
258 * If you are hashing n strings (uint8_t **)k, do it like this:
259 * for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h);
260 *
261 * By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this
262 * code any way you wish, private, educational, or commercial. It's free.
263 *
264 * Use for hash table lookup, or anything where one collision in 2^^32 is
265 * acceptable. Do NOT use for cryptographic purposes.
266 */
267 static uint32_t __attribute__((unused)) hashlittle(const void *key,
268 size_t length, uint32_t initval)
269 {
270 uint32_t a,b,c;
271 union {
272 const void *ptr;
273 size_t i;
274 } u; /* needed for Mac Powerbook G4 */
275
276 /* Set up the internal state */
277 a = b = c = 0xdeadbeef + ((uint32_t)length) + initval;
278
279 u.ptr = key;
280 if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
281 const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */
282
283 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
284 while (length > 12) {
285 a += k[0];
286 b += k[1];
287 c += k[2];
288 mix(a,b,c);
289 length -= 12;
290 k += 3;
291 }
292
293 /*
294 * "k[2]&0xffffff" actually reads beyond the end of the string, but
295 * then masks off the part it's not allowed to read. Because the
296 * string is aligned, the masked-off tail is in the same word as the
297 * rest of the string. Every machine with memory protection I've seen
298 * does it on word boundaries, so is OK with this. But VALGRIND will
299 * still catch it and complain. The masking trick does make the hash
300 * noticably faster for short strings (like English words).
301 */
302 #ifndef VALGRIND
303
304 switch (length) {
305 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
306 case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
307 case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break;
308 case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break;
309 case 8 : b+=k[1]; a+=k[0]; break;
310 case 7 : b+=k[1]&0xffffff; a+=k[0]; break;
311 case 6 : b+=k[1]&0xffff; a+=k[0]; break;
312 case 5 : b+=k[1]&0xff; a+=k[0]; break;
313 case 4 : a+=k[0]; break;
314 case 3 : a+=k[0]&0xffffff; break;
315 case 2 : a+=k[0]&0xffff; break;
316 case 1 : a+=k[0]&0xff; break;
317 case 0 : return c; /* zero length strings require no mixing */
318 }
319 #else /* make valgrind happy */
320 const uint8_t *k8;
321
322 k8 = (const uint8_t *)k;
323 switch (length) {
324 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
325 case 11: c+=((uint32_t)k8[10])<<16; /* fall through */
326 case 10: c+=((uint32_t)k8[9])<<8; /* fall through */
327 case 9 : c+=k8[8]; /* fall through */
328 case 8 : b+=k[1]; a+=k[0]; break;
329 case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */
330 case 6 : b+=((uint32_t)k8[5])<<8; /* fall through */
331 case 5 : b+=k8[4]; /* fall through */
332 case 4 : a+=k[0]; break;
333 case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */
334 case 2 : a+=((uint32_t)k8[1])<<8; /* fall through */
335 case 1 : a+=k8[0]; break;
336 case 0 : return c;
337 }
338 #endif /* !valgrind */
339 } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
340 const uint16_t *k = (const uint16_t *)key; /* read 16-bit chunks */
341 const uint8_t *k8;
342
343 /*--------------- all but last block: aligned reads and different mixing */
344 while (length > 12) {
345 a += k[0] + (((uint32_t)k[1])<<16);
346 b += k[2] + (((uint32_t)k[3])<<16);
347 c += k[4] + (((uint32_t)k[5])<<16);
348 mix(a,b,c);
349 length -= 12;
350 k += 6;
351 }
352
353 k8 = (const uint8_t *)k;
354 switch (length) {
355 case 12:
356 c+=k[4]+(((uint32_t)k[5])<<16);
357 b+=k[2]+(((uint32_t)k[3])<<16);
358 a+=k[0]+(((uint32_t)k[1])<<16);
359 break;
360 case 11:
361 c+=((uint32_t)k8[10])<<16; /* fall through */
362 case 10:
363 c+=k[4];
364 b+=k[2]+(((uint32_t)k[3])<<16);
365 a+=k[0]+(((uint32_t)k[1])<<16);
366 break;
367 case 9:
368 c+=k8[8]; /* fall through */
369 case 8:
370 b+=k[2]+(((uint32_t)k[3])<<16);
371 a+=k[0]+(((uint32_t)k[1])<<16);
372 break;
373 case 7:
374 b+=((uint32_t)k8[6])<<16; /* fall through */
375 case 6:
376 b+=k[2];
377 a+=k[0]+(((uint32_t)k[1])<<16);
378 break;
379 case 5:
380 b+=k8[4]; /* fall through */
381 case 4:
382 a+=k[0]+(((uint32_t)k[1])<<16);
383 break;
384 case 3:
385 a+=((uint32_t)k8[2])<<16; /* fall through */
386 case 2:
387 a+=k[0];
388 break;
389 case 1:
390 a+=k8[0];
391 break;
392 case 0:
393 return c; /* zero length requires no mixing */
394 }
395
396 } else { /* need to read the key one byte at a time */
397 const uint8_t *k = (const uint8_t *)key;
398
399 while (length > 12) {
400 a += k[0];
401 a += ((uint32_t)k[1])<<8;
402 a += ((uint32_t)k[2])<<16;
403 a += ((uint32_t)k[3])<<24;
404 b += k[4];
405 b += ((uint32_t)k[5])<<8;
406 b += ((uint32_t)k[6])<<16;
407 b += ((uint32_t)k[7])<<24;
408 c += k[8];
409 c += ((uint32_t)k[9])<<8;
410 c += ((uint32_t)k[10])<<16;
411 c += ((uint32_t)k[11])<<24;
412 mix(a,b,c);
413 length -= 12;
414 k += 12;
415 }
416
417 switch(length) { /* all the case statements fall through */
418 case 12: c+=((uint32_t)k[11])<<24;
419 case 11: c+=((uint32_t)k[10])<<16;
420 case 10: c+=((uint32_t)k[9])<<8;
421 case 9: c+=k[8];
422 case 8: b+=((uint32_t)k[7])<<24;
423 case 7: b+=((uint32_t)k[6])<<16;
424 case 6: b+=((uint32_t)k[5])<<8;
425 case 5: b+=k[4];
426 case 4: a+=((uint32_t)k[3])<<24;
427 case 3: a+=((uint32_t)k[2])<<16;
428 case 2: a+=((uint32_t)k[1])<<8;
429 case 1:
430 a+=k[0];
431 break;
432 case 0:
433 return c;
434 }
435 }
436
437 final(a,b,c);
438 return c;
439 }
440
441 LTTNG_HIDDEN
442 unsigned long hash_key_u64(const void *_key, unsigned long seed)
443 {
444 union {
445 uint64_t v64;
446 uint32_t v32[2];
447 } v;
448 union {
449 uint64_t v64;
450 uint32_t v32[2];
451 } key;
452
453 v.v64 = (uint64_t) seed;
454 key.v64 = *(const uint64_t *) _key;
455 hashword2(key.v32, 2, &v.v32[0], &v.v32[1]);
456 return v.v64;
457 }
458
459 #if (CAA_BITS_PER_LONG == 64)
460 /*
461 * Hash function for number value.
462 * Pass the value itself as the key, not its address.
463 */
464 LTTNG_HIDDEN
465 unsigned long hash_key_ulong(const void *_key, unsigned long seed)
466 {
467 uint64_t __key = (uint64_t) _key;
468 return (unsigned long) hash_key_u64(&__key, seed);
469 }
470 #else
471 /*
472 * Hash function for number value.
473 * Pass the value itself as the key, not its address.
474 */
475 LTTNG_HIDDEN
476 unsigned long hash_key_ulong(const void *_key, unsigned long seed)
477 {
478 uint32_t key = (uint32_t) _key;
479
480 return hashword(&key, 1, seed);
481 }
482 #endif /* CAA_BITS_PER_LONG */
483
484 /*
485 * Hash function for string.
486 */
487 LTTNG_HIDDEN
488 unsigned long hash_key_str(const void *key, unsigned long seed)
489 {
490 return hashlittle(key, strlen((const char *) key), seed);
491 }
492
493 /*
494 * Hash function for two uint64_t.
495 */
496 LTTNG_HIDDEN
497 unsigned long hash_key_two_u64(const void *key, unsigned long seed)
498 {
499 const struct lttng_ht_two_u64 *k =
500 (const struct lttng_ht_two_u64 *) key;
501
502 return hash_key_u64(&k->key1, seed) ^ hash_key_u64(&k->key2, seed);
503 }
504
505 /*
506 * Hash function compare for number value.
507 */
508 LTTNG_HIDDEN
509 int hash_match_key_ulong(const void *key1, const void *key2)
510 {
511 if (key1 == key2) {
512 return 1;
513 }
514
515 return 0;
516 }
517
518 /*
519 * Hash function compare for number value.
520 */
521 LTTNG_HIDDEN
522 int hash_match_key_u64(const void *key1, const void *key2)
523 {
524 if (*(const uint64_t *) key1 == *(const uint64_t *) key2) {
525 return 1;
526 }
527
528 return 0;
529 }
530
531 /*
532 * Hash compare function for string.
533 */
534 LTTNG_HIDDEN
535 int hash_match_key_str(const void *key1, const void *key2)
536 {
537 if (strcmp(key1, key2) == 0) {
538 return 1;
539 }
540
541 return 0;
542 }
543
544 /*
545 * Hash function compare two uint64_t.
546 */
547 LTTNG_HIDDEN
548 int hash_match_key_two_u64(const void *key1, const void *key2)
549 {
550 const struct lttng_ht_two_u64 *k1 =
551 (const struct lttng_ht_two_u64 *) key1;
552 const struct lttng_ht_two_u64 *k2 =
553 (const struct lttng_ht_two_u64 *) key2;
554
555 if (hash_match_key_u64(&k1->key1, &k2->key1) &&
556 hash_match_key_u64(&k1->key2, &k2->key2)) {
557 return 1;
558 }
559
560 return 0;
561 }
This page took 0.062798 seconds and 3 git commands to generate.