Rename C++ header files to .hpp
[lttng-tools.git] / src / common / hashtable / utils.cpp
CommitLineData
819dc7d4 1/*
ab5be9fa 2 * Copyright (C) 2006 Bob Jenkins
21cf9b6b 3 * Copyright (C) 2011 EfficiOS Inc.
ab5be9fa 4 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
819dc7d4 5 *
c922647d 6 * SPDX-License-Identifier: LGPL-2.1-only
66c60361 7 *
66c60361
DG
8 */
9
10/*
819dc7d4
DG
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 */
890d8fe4 41
6c1c0768 42#define _LGPL_SOURCE
bec39940
DG
43#include <stdint.h> /* defines uint32_t etc */
44#include <stdio.h> /* defines printf for tests */
819dc7d4 45#include <string.h>
bec39940
DG
46#include <sys/param.h> /* attempt to define endianness */
47#include <time.h> /* defines time_t for timings in the test */
0df502fd 48#include <urcu/compiler.h>
819dc7d4 49
c9e313bc
SM
50#include "utils.hpp"
51#include <common/compat/endian.hpp> /* attempt to define endianness */
52#include <common/common.hpp>
53#include <common/hashtable/hashtable.hpp>
bec39940 54
819dc7d4
DG
55/*
56 * My best guess at if you are big-endian or little-endian. This may
57 * need adjustment.
58 */
6b5a354b
MJ
59#if (defined(BYTE_ORDER) && defined(LITTLE_ENDIAN) && \
60 BYTE_ORDER == LITTLE_ENDIAN) || \
819dc7d4
DG
61 (defined(i386) || defined(__i386__) || defined(__i486__) || \
62 defined(__i586__) || defined(__i686__) || defined(vax) || defined(MIPSEL))
63# define HASH_LITTLE_ENDIAN 1
64# define HASH_BIG_ENDIAN 0
6b5a354b
MJ
65#elif (defined(BYTE_ORDER) && defined(BIG_ENDIAN) && \
66 BYTE_ORDER == BIG_ENDIAN) || \
819dc7d4
DG
67 (defined(sparc) || defined(POWERPC) || defined(mc68000) || defined(sel))
68# define HASH_LITTLE_ENDIAN 0
69# define HASH_BIG_ENDIAN 1
70#else
71# define HASH_LITTLE_ENDIAN 0
72# define HASH_BIG_ENDIAN 0
73#endif
74
75#define hashsize(n) ((uint32_t)1<<(n))
76#define hashmask(n) (hashsize(n)-1)
77#define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
78
79/*
80 * mix -- mix 3 32-bit values reversibly.
81 *
82 * This is reversible, so any information in (a,b,c) before mix() is
83 * still in (a,b,c) after mix().
84 *
85 * If four pairs of (a,b,c) inputs are run through mix(), or through
86 * mix() in reverse, there are at least 32 bits of the output that
87 * are sometimes the same for one pair and different for another pair.
88 * This was tested for:
89 * * pairs that differed by one bit, by two bits, in any combination
90 * of top bits of (a,b,c), or in any combination of bottom bits of
91 * (a,b,c).
92 * * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
93 * the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
94 * is commonly produced by subtraction) look like a single 1-bit
95 * difference.
96 * * the base values were pseudorandom, all zero but one bit set, or
97 * all zero plus a counter that starts at zero.
98 *
99 * Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
100 * satisfy this are
101 * 4 6 8 16 19 4
102 * 9 15 3 18 27 15
103 * 14 9 3 7 17 3
104 * Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
105 * for "differ" defined as + with a one-bit base and a two-bit delta. I
106 * used http://burtleburtle.net/bob/hash/avalanche.html to choose
107 * the operations, constants, and arrangements of the variables.
108 *
109 * This does not achieve avalanche. There are input bits of (a,b,c)
110 * that fail to affect some output bits of (a,b,c), especially of a. The
111 * most thoroughly mixed value is c, but it doesn't really even achieve
112 * avalanche in c.
113 *
114 * This allows some parallelism. Read-after-writes are good at doubling
115 * the number of bits affected, so the goal of mixing pulls in the opposite
116 * direction as the goal of parallelism. I did what I could. Rotates
117 * seem to cost as much as shifts on every machine I could lay my hands
118 * on, and rotates are much kinder to the top and bottom bits, so I used
119 * rotates.
120 */
121#define mix(a,b,c) \
122{ \
123 a -= c; a ^= rot(c, 4); c += b; \
124 b -= a; b ^= rot(a, 6); a += c; \
125 c -= b; c ^= rot(b, 8); b += a; \
126 a -= c; a ^= rot(c,16); c += b; \
127 b -= a; b ^= rot(a,19); a += c; \
128 c -= b; c ^= rot(b, 4); b += a; \
129}
130
131/*
132 * final -- final mixing of 3 32-bit values (a,b,c) into c
133 *
134 * Pairs of (a,b,c) values differing in only a few bits will usually
135 * produce values of c that look totally different. This was tested for
136 * * pairs that differed by one bit, by two bits, in any combination
137 * of top bits of (a,b,c), or in any combination of bottom bits of
138 * (a,b,c).
139 * * "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
140 * the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
141 * is commonly produced by subtraction) look like a single 1-bit
142 * difference.
143 * * the base values were pseudorandom, all zero but one bit set, or
144 * all zero plus a counter that starts at zero.
145 *
146 * These constants passed:
147 * 14 11 25 16 4 14 24
148 * 12 14 25 16 4 14 24
149 * and these came close:
150 * 4 8 15 26 3 22 24
151 * 10 8 15 26 3 22 24
152 * 11 8 15 26 3 22 24
153 */
154#define final(a,b,c) \
155{ \
156 c ^= b; c -= rot(b,14); \
157 a ^= c; a -= rot(c,11); \
158 b ^= a; b -= rot(a,25); \
159 c ^= b; c -= rot(b,16); \
160 a ^= c; a -= rot(c,4); \
161 b ^= a; b -= rot(a,14); \
162 c ^= b; c -= rot(b,24); \
163}
164
bec39940
DG
165/*
166 * k - the key, an array of uint32_t values
167 * length - the length of the key, in uint32_ts
168 * initval - the previous hash, or an arbitrary value
169 */
170static uint32_t __attribute__((unused)) hashword(const uint32_t *k,
171 size_t length, uint32_t initval)
0df502fd
MD
172{
173 uint32_t a, b, c;
174
175 /* Set up the internal state */
176 a = b = c = 0xdeadbeef + (((uint32_t) length) << 2) + initval;
177
178 /*----------------------------------------- handle most of the key */
179 while (length > 3) {
180 a += k[0];
181 b += k[1];
182 c += k[2];
183 mix(a, b, c);
184 length -= 3;
185 k += 3;
186 }
187
188 /*----------------------------------- handle the last 3 uint32_t's */
189 switch (length) { /* all the case statements fall through */
a95e960e
MJ
190 case 3: c += k[2]; /* fall through */
191 case 2: b += k[1]; /* fall through */
0df502fd
MD
192 case 1: a += k[0];
193 final(a, b, c);
194 case 0: /* case 0: nothing left to add */
195 break;
196 }
197 /*---------------------------------------------- report the result */
198 return c;
199}
200
201
819dc7d4
DG
202/*
203 * hashword2() -- same as hashword(), but take two seeds and return two 32-bit
204 * values. pc and pb must both be nonnull, and *pc and *pb must both be
205 * initialized with seeds. If you pass in (*pb)==0, the output (*pc) will be
206 * the same as the return value from hashword().
207 */
bec39940 208static void __attribute__((unused)) hashword2(const uint32_t *k, size_t length,
819dc7d4
DG
209 uint32_t *pc, uint32_t *pb)
210{
211 uint32_t a, b, c;
212
213 /* Set up the internal state */
214 a = b = c = 0xdeadbeef + ((uint32_t) (length << 2)) + *pc;
215 c += *pb;
216
217 while (length > 3) {
218 a += k[0];
219 b += k[1];
220 c += k[2];
221 mix(a, b, c);
222 length -= 3;
223 k += 3;
224 }
225
226 switch (length) {
227 case 3 :
228 c += k[2];
30eb3927 229 /* fall through */
819dc7d4
DG
230 case 2 :
231 b += k[1];
30eb3927 232 /* fall through */
819dc7d4
DG
233 case 1 :
234 a += k[0];
235 final(a, b, c);
30eb3927 236 /* fall through */
819dc7d4
DG
237 case 0: /* case 0: nothing left to add */
238 break;
239 }
240
241 *pc = c;
242 *pb = b;
243}
244
245/*
246 * hashlittle() -- hash a variable-length key into a 32-bit value
247 * k : the key (the unaligned variable-length array of bytes)
248 * length : the length of the key, counting by bytes
249 * initval : can be any 4-byte value
250 * Returns a 32-bit value. Every bit of the key affects every bit of
251 * the return value. Two keys differing by one or two bits will have
252 * totally different hash values.
253 *
254 * The best hash table sizes are powers of 2. There is no need to do
255 * mod a prime (mod is sooo slow!). If you need less than 32 bits,
256 * use a bitmask. For example, if you need only 10 bits, do
257 * h = (h & hashmask(10));
258 * In which case, the hash table should have hashsize(10) elements.
259 *
260 * If you are hashing n strings (uint8_t **)k, do it like this:
261 * for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h);
262 *
263 * By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this
264 * code any way you wish, private, educational, or commercial. It's free.
265 *
266 * Use for hash table lookup, or anything where one collision in 2^^32 is
267 * acceptable. Do NOT use for cryptographic purposes.
268 */
1405051a
FD
269LTTNG_NO_SANITIZE_ADDRESS
270__attribute__((unused))
271static uint32_t hashlittle(const void *key,
bec39940 272 size_t length, uint32_t initval)
819dc7d4
DG
273{
274 uint32_t a,b,c;
275 union {
276 const void *ptr;
277 size_t i;
278 } u; /* needed for Mac Powerbook G4 */
279
280 /* Set up the internal state */
281 a = b = c = 0xdeadbeef + ((uint32_t)length) + initval;
282
283 u.ptr = key;
284 if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
285 const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */
286
287 /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
288 while (length > 12) {
289 a += k[0];
290 b += k[1];
291 c += k[2];
292 mix(a,b,c);
293 length -= 12;
294 k += 3;
295 }
296
297 /*
298 * "k[2]&0xffffff" actually reads beyond the end of the string, but
299 * then masks off the part it's not allowed to read. Because the
300 * string is aligned, the masked-off tail is in the same word as the
301 * rest of the string. Every machine with memory protection I've seen
302 * does it on word boundaries, so is OK with this. But VALGRIND will
303 * still catch it and complain. The masking trick does make the hash
304 * noticably faster for short strings (like English words).
305 */
306#ifndef VALGRIND
307
308 switch (length) {
309 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
310 case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
311 case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break;
312 case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break;
313 case 8 : b+=k[1]; a+=k[0]; break;
314 case 7 : b+=k[1]&0xffffff; a+=k[0]; break;
315 case 6 : b+=k[1]&0xffff; a+=k[0]; break;
316 case 5 : b+=k[1]&0xff; a+=k[0]; break;
317 case 4 : a+=k[0]; break;
318 case 3 : a+=k[0]&0xffffff; break;
319 case 2 : a+=k[0]&0xffff; break;
320 case 1 : a+=k[0]&0xff; break;
321 case 0 : return c; /* zero length strings require no mixing */
322 }
323#else /* make valgrind happy */
324 const uint8_t *k8;
325
326 k8 = (const uint8_t *)k;
327 switch (length) {
328 case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
329 case 11: c+=((uint32_t)k8[10])<<16; /* fall through */
330 case 10: c+=((uint32_t)k8[9])<<8; /* fall through */
331 case 9 : c+=k8[8]; /* fall through */
332 case 8 : b+=k[1]; a+=k[0]; break;
333 case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */
334 case 6 : b+=((uint32_t)k8[5])<<8; /* fall through */
335 case 5 : b+=k8[4]; /* fall through */
336 case 4 : a+=k[0]; break;
337 case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */
338 case 2 : a+=((uint32_t)k8[1])<<8; /* fall through */
339 case 1 : a+=k8[0]; break;
340 case 0 : return c;
341 }
342#endif /* !valgrind */
343 } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
344 const uint16_t *k = (const uint16_t *)key; /* read 16-bit chunks */
345 const uint8_t *k8;
346
347 /*--------------- all but last block: aligned reads and different mixing */
348 while (length > 12) {
349 a += k[0] + (((uint32_t)k[1])<<16);
350 b += k[2] + (((uint32_t)k[3])<<16);
351 c += k[4] + (((uint32_t)k[5])<<16);
352 mix(a,b,c);
353 length -= 12;
354 k += 6;
355 }
356
357 k8 = (const uint8_t *)k;
358 switch (length) {
359 case 12:
360 c+=k[4]+(((uint32_t)k[5])<<16);
361 b+=k[2]+(((uint32_t)k[3])<<16);
362 a+=k[0]+(((uint32_t)k[1])<<16);
363 break;
364 case 11:
365 c+=((uint32_t)k8[10])<<16; /* fall through */
366 case 10:
367 c+=k[4];
368 b+=k[2]+(((uint32_t)k[3])<<16);
369 a+=k[0]+(((uint32_t)k[1])<<16);
370 break;
371 case 9:
372 c+=k8[8]; /* fall through */
373 case 8:
374 b+=k[2]+(((uint32_t)k[3])<<16);
375 a+=k[0]+(((uint32_t)k[1])<<16);
376 break;
377 case 7:
378 b+=((uint32_t)k8[6])<<16; /* fall through */
379 case 6:
380 b+=k[2];
381 a+=k[0]+(((uint32_t)k[1])<<16);
382 break;
383 case 5:
384 b+=k8[4]; /* fall through */
385 case 4:
386 a+=k[0]+(((uint32_t)k[1])<<16);
387 break;
388 case 3:
389 a+=((uint32_t)k8[2])<<16; /* fall through */
390 case 2:
391 a+=k[0];
392 break;
393 case 1:
394 a+=k8[0];
395 break;
396 case 0:
397 return c; /* zero length requires no mixing */
398 }
399
400 } else { /* need to read the key one byte at a time */
401 const uint8_t *k = (const uint8_t *)key;
402
403 while (length > 12) {
404 a += k[0];
405 a += ((uint32_t)k[1])<<8;
406 a += ((uint32_t)k[2])<<16;
407 a += ((uint32_t)k[3])<<24;
408 b += k[4];
409 b += ((uint32_t)k[5])<<8;
410 b += ((uint32_t)k[6])<<16;
411 b += ((uint32_t)k[7])<<24;
412 c += k[8];
413 c += ((uint32_t)k[9])<<8;
414 c += ((uint32_t)k[10])<<16;
415 c += ((uint32_t)k[11])<<24;
416 mix(a,b,c);
417 length -= 12;
418 k += 12;
419 }
420
421 switch(length) { /* all the case statements fall through */
30eb3927
MJ
422 case 12: c+=((uint32_t)k[11])<<24; /* fall through */
423 case 11: c+=((uint32_t)k[10])<<16; /* fall through */
424 case 10: c+=((uint32_t)k[9])<<8; /* fall through */
425 case 9: c+=k[8]; /* fall through */
426 case 8: b+=((uint32_t)k[7])<<24; /* fall through */
427 case 7: b+=((uint32_t)k[6])<<16; /* fall through */
428 case 6: b+=((uint32_t)k[5])<<8; /* fall through */
429 case 5: b+=k[4]; /* fall through */
430 case 4: a+=((uint32_t)k[3])<<24; /* fall through */
431 case 3: a+=((uint32_t)k[2])<<16; /* fall through */
432 case 2: a+=((uint32_t)k[1])<<8; /* fall through */
819dc7d4
DG
433 case 1:
434 a+=k[0];
435 break;
436 case 0:
437 return c;
438 }
439 }
440
441 final(a,b,c);
442 return c;
443}
444
bcd52dd9 445unsigned long hash_key_u64(const void *_key, unsigned long seed)
0df502fd
MD
446{
447 union {
448 uint64_t v64;
449 uint32_t v32[2];
450 } v;
451 union {
452 uint64_t v64;
453 uint32_t v32[2];
454 } key;
455
0df502fd 456 v.v64 = (uint64_t) seed;
bcd52dd9 457 key.v64 = *(const uint64_t *) _key;
0df502fd
MD
458 hashword2(key.v32, 2, &v.v32[0], &v.v32[1]);
459 return v.v64;
460}
d88aee68
DG
461
462#if (CAA_BITS_PER_LONG == 64)
463/*
464 * Hash function for number value.
9a2746aa 465 * Pass the value itself as the key, not its address.
d88aee68 466 */
bcd52dd9 467unsigned long hash_key_ulong(const void *_key, unsigned long seed)
d88aee68
DG
468{
469 uint64_t __key = (uint64_t) _key;
470 return (unsigned long) hash_key_u64(&__key, seed);
471}
0df502fd 472#else
819dc7d4
DG
473/*
474 * Hash function for number value.
9a2746aa 475 * Pass the value itself as the key, not its address.
819dc7d4 476 */
bcd52dd9 477unsigned long hash_key_ulong(const void *_key, unsigned long seed)
819dc7d4 478{
8da9ba32 479 uint32_t key = (uint32_t) _key;
0df502fd 480
0df502fd 481 return hashword(&key, 1, seed);
819dc7d4 482}
bec39940 483#endif /* CAA_BITS_PER_LONG */
819dc7d4
DG
484
485/*
486 * Hash function for string.
487 */
bcd52dd9 488unsigned long hash_key_str(const void *key, unsigned long seed)
819dc7d4 489{
bcd52dd9 490 return hashlittle(key, strlen((const char *) key), seed);
819dc7d4
DG
491}
492
3c4599b9
JD
493/*
494 * Hash function for two uint64_t.
495 */
bcd52dd9 496unsigned long hash_key_two_u64(const void *key, unsigned long seed)
3c4599b9 497{
bcd52dd9
JG
498 const struct lttng_ht_two_u64 *k =
499 (const struct lttng_ht_two_u64 *) key;
3c4599b9
JD
500
501 return hash_key_u64(&k->key1, seed) ^ hash_key_u64(&k->key2, seed);
502}
503
819dc7d4
DG
504/*
505 * Hash function compare for number value.
506 */
bcd52dd9 507int hash_match_key_ulong(const void *key1, const void *key2)
819dc7d4 508{
819dc7d4 509 if (key1 == key2) {
bec39940 510 return 1;
819dc7d4
DG
511 }
512
bec39940 513 return 0;
819dc7d4
DG
514}
515
d88aee68
DG
516/*
517 * Hash function compare for number value.
518 */
bcd52dd9 519int hash_match_key_u64(const void *key1, const void *key2)
d88aee68 520{
bcd52dd9 521 if (*(const uint64_t *) key1 == *(const uint64_t *) key2) {
d88aee68
DG
522 return 1;
523 }
524
525 return 0;
526}
527
819dc7d4
DG
528/*
529 * Hash compare function for string.
530 */
bcd52dd9 531int hash_match_key_str(const void *key1, const void *key2)
819dc7d4 532{
6dca8ba7 533 if (strcmp((const char *) key1, (const char *) key2) == 0) {
bec39940 534 return 1;
819dc7d4
DG
535 }
536
bec39940 537 return 0;
819dc7d4 538}
3c4599b9
JD
539
540/*
541 * Hash function compare two uint64_t.
542 */
bcd52dd9 543int hash_match_key_two_u64(const void *key1, const void *key2)
3c4599b9 544{
bcd52dd9
JG
545 const struct lttng_ht_two_u64 *k1 =
546 (const struct lttng_ht_two_u64 *) key1;
547 const struct lttng_ht_two_u64 *k2 =
548 (const struct lttng_ht_two_u64 *) key2;
3c4599b9
JD
549
550 if (hash_match_key_u64(&k1->key1, &k2->key1) &&
551 hash_match_key_u64(&k1->key2, &k2->key2)) {
552 return 1;
553 }
554
555 return 0;
556}
This page took 0.084041 seconds and 4 git commands to generate.