]> git.saurik.com Git - redis.git/blob - deps/jemalloc/src/ckh.c
Merge pull request #63 from djanowski/tcl
[redis.git] / deps / jemalloc / src / ckh.c
1 /*
2 *******************************************************************************
3 * Implementation of (2^1+,2) cuckoo hashing, where 2^1+ indicates that each
4 * hash bucket contains 2^n cells, for n >= 1, and 2 indicates that two hash
5 * functions are employed. The original cuckoo hashing algorithm was described
6 * in:
7 *
8 * Pagh, R., F.F. Rodler (2004) Cuckoo Hashing. Journal of Algorithms
9 * 51(2):122-144.
10 *
11 * Generalization of cuckoo hashing was discussed in:
12 *
13 * Erlingsson, U., M. Manasse, F. McSherry (2006) A cool and practical
14 * alternative to traditional hash tables. In Proceedings of the 7th
15 * Workshop on Distributed Data and Structures (WDAS'06), Santa Clara, CA,
16 * January 2006.
17 *
18 * This implementation uses precisely two hash functions because that is the
19 * fewest that can work, and supporting multiple hashes is an implementation
20 * burden. Here is a reproduction of Figure 1 from Erlingsson et al. (2006)
21 * that shows approximate expected maximum load factors for various
22 * configurations:
23 *
24 * | #cells/bucket |
25 * #hashes | 1 | 2 | 4 | 8 |
26 * --------+-------+-------+-------+-------+
27 * 1 | 0.006 | 0.006 | 0.03 | 0.12 |
28 * 2 | 0.49 | 0.86 |>0.93< |>0.96< |
29 * 3 | 0.91 | 0.97 | 0.98 | 0.999 |
30 * 4 | 0.97 | 0.99 | 0.999 | |
31 *
32 * The number of cells per bucket is chosen such that a bucket fits in one cache
33 * line. So, on 32- and 64-bit systems, we use (8,2) and (4,2) cuckoo hashing,
34 * respectively.
35 *
36 ******************************************************************************/
37 #define JEMALLOC_CKH_C_
38 #include "jemalloc/internal/jemalloc_internal.h"
39
40 /******************************************************************************/
41 /* Function prototypes for non-inline static functions. */
42
43 static bool ckh_grow(ckh_t *ckh);
44 static void ckh_shrink(ckh_t *ckh);
45
46 /******************************************************************************/
47
48 /*
49 * Search bucket for key and return the cell number if found; SIZE_T_MAX
50 * otherwise.
51 */
52 JEMALLOC_INLINE size_t
53 ckh_bucket_search(ckh_t *ckh, size_t bucket, const void *key)
54 {
55 ckhc_t *cell;
56 unsigned i;
57
58 for (i = 0; i < (ZU(1) << LG_CKH_BUCKET_CELLS); i++) {
59 cell = &ckh->tab[(bucket << LG_CKH_BUCKET_CELLS) + i];
60 if (cell->key != NULL && ckh->keycomp(key, cell->key))
61 return ((bucket << LG_CKH_BUCKET_CELLS) + i);
62 }
63
64 return (SIZE_T_MAX);
65 }
66
67 /*
68 * Search table for key and return cell number if found; SIZE_T_MAX otherwise.
69 */
70 JEMALLOC_INLINE size_t
71 ckh_isearch(ckh_t *ckh, const void *key)
72 {
73 size_t hash1, hash2, bucket, cell;
74
75 assert(ckh != NULL);
76 dassert(ckh->magic == CKH_MAGIC);
77
78 ckh->hash(key, ckh->lg_curbuckets, &hash1, &hash2);
79
80 /* Search primary bucket. */
81 bucket = hash1 & ((ZU(1) << ckh->lg_curbuckets) - 1);
82 cell = ckh_bucket_search(ckh, bucket, key);
83 if (cell != SIZE_T_MAX)
84 return (cell);
85
86 /* Search secondary bucket. */
87 bucket = hash2 & ((ZU(1) << ckh->lg_curbuckets) - 1);
88 cell = ckh_bucket_search(ckh, bucket, key);
89 return (cell);
90 }
91
92 JEMALLOC_INLINE bool
93 ckh_try_bucket_insert(ckh_t *ckh, size_t bucket, const void *key,
94 const void *data)
95 {
96 ckhc_t *cell;
97 unsigned offset, i;
98
99 /*
100 * Cycle through the cells in the bucket, starting at a random position.
101 * The randomness avoids worst-case search overhead as buckets fill up.
102 */
103 prn32(offset, LG_CKH_BUCKET_CELLS, ckh->prn_state, CKH_A, CKH_C);
104 for (i = 0; i < (ZU(1) << LG_CKH_BUCKET_CELLS); i++) {
105 cell = &ckh->tab[(bucket << LG_CKH_BUCKET_CELLS) +
106 ((i + offset) & ((ZU(1) << LG_CKH_BUCKET_CELLS) - 1))];
107 if (cell->key == NULL) {
108 cell->key = key;
109 cell->data = data;
110 ckh->count++;
111 return (false);
112 }
113 }
114
115 return (true);
116 }
117
118 /*
119 * No space is available in bucket. Randomly evict an item, then try to find an
120 * alternate location for that item. Iteratively repeat this
121 * eviction/relocation procedure until either success or detection of an
122 * eviction/relocation bucket cycle.
123 */
124 JEMALLOC_INLINE bool
125 ckh_evict_reloc_insert(ckh_t *ckh, size_t argbucket, void const **argkey,
126 void const **argdata)
127 {
128 const void *key, *data, *tkey, *tdata;
129 ckhc_t *cell;
130 size_t hash1, hash2, bucket, tbucket;
131 unsigned i;
132
133 bucket = argbucket;
134 key = *argkey;
135 data = *argdata;
136 while (true) {
137 /*
138 * Choose a random item within the bucket to evict. This is
139 * critical to correct function, because without (eventually)
140 * evicting all items within a bucket during iteration, it
141 * would be possible to get stuck in an infinite loop if there
142 * were an item for which both hashes indicated the same
143 * bucket.
144 */
145 prn32(i, LG_CKH_BUCKET_CELLS, ckh->prn_state, CKH_A, CKH_C);
146 cell = &ckh->tab[(bucket << LG_CKH_BUCKET_CELLS) + i];
147 assert(cell->key != NULL);
148
149 /* Swap cell->{key,data} and {key,data} (evict). */
150 tkey = cell->key; tdata = cell->data;
151 cell->key = key; cell->data = data;
152 key = tkey; data = tdata;
153
154 #ifdef CKH_COUNT
155 ckh->nrelocs++;
156 #endif
157
158 /* Find the alternate bucket for the evicted item. */
159 ckh->hash(key, ckh->lg_curbuckets, &hash1, &hash2);
160 tbucket = hash2 & ((ZU(1) << ckh->lg_curbuckets) - 1);
161 if (tbucket == bucket) {
162 tbucket = hash1 & ((ZU(1) << ckh->lg_curbuckets) - 1);
163 /*
164 * It may be that (tbucket == bucket) still, if the
165 * item's hashes both indicate this bucket. However,
166 * we are guaranteed to eventually escape this bucket
167 * during iteration, assuming pseudo-random item
168 * selection (true randomness would make infinite
169 * looping a remote possibility). The reason we can
170 * never get trapped forever is that there are two
171 * cases:
172 *
173 * 1) This bucket == argbucket, so we will quickly
174 * detect an eviction cycle and terminate.
175 * 2) An item was evicted to this bucket from another,
176 * which means that at least one item in this bucket
177 * has hashes that indicate distinct buckets.
178 */
179 }
180 /* Check for a cycle. */
181 if (tbucket == argbucket) {
182 *argkey = key;
183 *argdata = data;
184 return (true);
185 }
186
187 bucket = tbucket;
188 if (ckh_try_bucket_insert(ckh, bucket, key, data) == false)
189 return (false);
190 }
191 }
192
193 JEMALLOC_INLINE bool
194 ckh_try_insert(ckh_t *ckh, void const**argkey, void const**argdata)
195 {
196 size_t hash1, hash2, bucket;
197 const void *key = *argkey;
198 const void *data = *argdata;
199
200 ckh->hash(key, ckh->lg_curbuckets, &hash1, &hash2);
201
202 /* Try to insert in primary bucket. */
203 bucket = hash1 & ((ZU(1) << ckh->lg_curbuckets) - 1);
204 if (ckh_try_bucket_insert(ckh, bucket, key, data) == false)
205 return (false);
206
207 /* Try to insert in secondary bucket. */
208 bucket = hash2 & ((ZU(1) << ckh->lg_curbuckets) - 1);
209 if (ckh_try_bucket_insert(ckh, bucket, key, data) == false)
210 return (false);
211
212 /*
213 * Try to find a place for this item via iterative eviction/relocation.
214 */
215 return (ckh_evict_reloc_insert(ckh, bucket, argkey, argdata));
216 }
217
218 /*
219 * Try to rebuild the hash table from scratch by inserting all items from the
220 * old table into the new.
221 */
222 JEMALLOC_INLINE bool
223 ckh_rebuild(ckh_t *ckh, ckhc_t *aTab)
224 {
225 size_t count, i, nins;
226 const void *key, *data;
227
228 count = ckh->count;
229 ckh->count = 0;
230 for (i = nins = 0; nins < count; i++) {
231 if (aTab[i].key != NULL) {
232 key = aTab[i].key;
233 data = aTab[i].data;
234 if (ckh_try_insert(ckh, &key, &data)) {
235 ckh->count = count;
236 return (true);
237 }
238 nins++;
239 }
240 }
241
242 return (false);
243 }
244
245 static bool
246 ckh_grow(ckh_t *ckh)
247 {
248 bool ret;
249 ckhc_t *tab, *ttab;
250 size_t lg_curcells;
251 unsigned lg_prevbuckets;
252
253 #ifdef CKH_COUNT
254 ckh->ngrows++;
255 #endif
256
257 /*
258 * It is possible (though unlikely, given well behaved hashes) that the
259 * table will have to be doubled more than once in order to create a
260 * usable table.
261 */
262 lg_prevbuckets = ckh->lg_curbuckets;
263 lg_curcells = ckh->lg_curbuckets + LG_CKH_BUCKET_CELLS;
264 while (true) {
265 size_t usize;
266
267 lg_curcells++;
268 usize = sa2u(sizeof(ckhc_t) << lg_curcells, CACHELINE, NULL);
269 if (usize == 0) {
270 ret = true;
271 goto RETURN;
272 }
273 tab = (ckhc_t *)ipalloc(usize, CACHELINE, true);
274 if (tab == NULL) {
275 ret = true;
276 goto RETURN;
277 }
278 /* Swap in new table. */
279 ttab = ckh->tab;
280 ckh->tab = tab;
281 tab = ttab;
282 ckh->lg_curbuckets = lg_curcells - LG_CKH_BUCKET_CELLS;
283
284 if (ckh_rebuild(ckh, tab) == false) {
285 idalloc(tab);
286 break;
287 }
288
289 /* Rebuilding failed, so back out partially rebuilt table. */
290 idalloc(ckh->tab);
291 ckh->tab = tab;
292 ckh->lg_curbuckets = lg_prevbuckets;
293 }
294
295 ret = false;
296 RETURN:
297 return (ret);
298 }
299
300 static void
301 ckh_shrink(ckh_t *ckh)
302 {
303 ckhc_t *tab, *ttab;
304 size_t lg_curcells, usize;
305 unsigned lg_prevbuckets;
306
307 /*
308 * It is possible (though unlikely, given well behaved hashes) that the
309 * table rebuild will fail.
310 */
311 lg_prevbuckets = ckh->lg_curbuckets;
312 lg_curcells = ckh->lg_curbuckets + LG_CKH_BUCKET_CELLS - 1;
313 usize = sa2u(sizeof(ckhc_t) << lg_curcells, CACHELINE, NULL);
314 if (usize == 0)
315 return;
316 tab = (ckhc_t *)ipalloc(usize, CACHELINE, true);
317 if (tab == NULL) {
318 /*
319 * An OOM error isn't worth propagating, since it doesn't
320 * prevent this or future operations from proceeding.
321 */
322 return;
323 }
324 /* Swap in new table. */
325 ttab = ckh->tab;
326 ckh->tab = tab;
327 tab = ttab;
328 ckh->lg_curbuckets = lg_curcells - LG_CKH_BUCKET_CELLS;
329
330 if (ckh_rebuild(ckh, tab) == false) {
331 idalloc(tab);
332 #ifdef CKH_COUNT
333 ckh->nshrinks++;
334 #endif
335 return;
336 }
337
338 /* Rebuilding failed, so back out partially rebuilt table. */
339 idalloc(ckh->tab);
340 ckh->tab = tab;
341 ckh->lg_curbuckets = lg_prevbuckets;
342 #ifdef CKH_COUNT
343 ckh->nshrinkfails++;
344 #endif
345 }
346
347 bool
348 ckh_new(ckh_t *ckh, size_t minitems, ckh_hash_t *hash, ckh_keycomp_t *keycomp)
349 {
350 bool ret;
351 size_t mincells, usize;
352 unsigned lg_mincells;
353
354 assert(minitems > 0);
355 assert(hash != NULL);
356 assert(keycomp != NULL);
357
358 #ifdef CKH_COUNT
359 ckh->ngrows = 0;
360 ckh->nshrinks = 0;
361 ckh->nshrinkfails = 0;
362 ckh->ninserts = 0;
363 ckh->nrelocs = 0;
364 #endif
365 ckh->prn_state = 42; /* Value doesn't really matter. */
366 ckh->count = 0;
367
368 /*
369 * Find the minimum power of 2 that is large enough to fit aBaseCount
370 * entries. We are using (2+,2) cuckoo hashing, which has an expected
371 * maximum load factor of at least ~0.86, so 0.75 is a conservative load
372 * factor that will typically allow 2^aLgMinItems to fit without ever
373 * growing the table.
374 */
375 assert(LG_CKH_BUCKET_CELLS > 0);
376 mincells = ((minitems + (3 - (minitems % 3))) / 3) << 2;
377 for (lg_mincells = LG_CKH_BUCKET_CELLS;
378 (ZU(1) << lg_mincells) < mincells;
379 lg_mincells++)
380 ; /* Do nothing. */
381 ckh->lg_minbuckets = lg_mincells - LG_CKH_BUCKET_CELLS;
382 ckh->lg_curbuckets = lg_mincells - LG_CKH_BUCKET_CELLS;
383 ckh->hash = hash;
384 ckh->keycomp = keycomp;
385
386 usize = sa2u(sizeof(ckhc_t) << lg_mincells, CACHELINE, NULL);
387 if (usize == 0) {
388 ret = true;
389 goto RETURN;
390 }
391 ckh->tab = (ckhc_t *)ipalloc(usize, CACHELINE, true);
392 if (ckh->tab == NULL) {
393 ret = true;
394 goto RETURN;
395 }
396
397 #ifdef JEMALLOC_DEBUG
398 ckh->magic = CKH_MAGIC;
399 #endif
400
401 ret = false;
402 RETURN:
403 return (ret);
404 }
405
406 void
407 ckh_delete(ckh_t *ckh)
408 {
409
410 assert(ckh != NULL);
411 dassert(ckh->magic == CKH_MAGIC);
412
413 #ifdef CKH_VERBOSE
414 malloc_printf(
415 "%s(%p): ngrows: %"PRIu64", nshrinks: %"PRIu64","
416 " nshrinkfails: %"PRIu64", ninserts: %"PRIu64","
417 " nrelocs: %"PRIu64"\n", __func__, ckh,
418 (unsigned long long)ckh->ngrows,
419 (unsigned long long)ckh->nshrinks,
420 (unsigned long long)ckh->nshrinkfails,
421 (unsigned long long)ckh->ninserts,
422 (unsigned long long)ckh->nrelocs);
423 #endif
424
425 idalloc(ckh->tab);
426 #ifdef JEMALLOC_DEBUG
427 memset(ckh, 0x5a, sizeof(ckh_t));
428 #endif
429 }
430
431 size_t
432 ckh_count(ckh_t *ckh)
433 {
434
435 assert(ckh != NULL);
436 dassert(ckh->magic == CKH_MAGIC);
437
438 return (ckh->count);
439 }
440
441 bool
442 ckh_iter(ckh_t *ckh, size_t *tabind, void **key, void **data)
443 {
444 size_t i, ncells;
445
446 for (i = *tabind, ncells = (ZU(1) << (ckh->lg_curbuckets +
447 LG_CKH_BUCKET_CELLS)); i < ncells; i++) {
448 if (ckh->tab[i].key != NULL) {
449 if (key != NULL)
450 *key = (void *)ckh->tab[i].key;
451 if (data != NULL)
452 *data = (void *)ckh->tab[i].data;
453 *tabind = i + 1;
454 return (false);
455 }
456 }
457
458 return (true);
459 }
460
461 bool
462 ckh_insert(ckh_t *ckh, const void *key, const void *data)
463 {
464 bool ret;
465
466 assert(ckh != NULL);
467 dassert(ckh->magic == CKH_MAGIC);
468 assert(ckh_search(ckh, key, NULL, NULL));
469
470 #ifdef CKH_COUNT
471 ckh->ninserts++;
472 #endif
473
474 while (ckh_try_insert(ckh, &key, &data)) {
475 if (ckh_grow(ckh)) {
476 ret = true;
477 goto RETURN;
478 }
479 }
480
481 ret = false;
482 RETURN:
483 return (ret);
484 }
485
486 bool
487 ckh_remove(ckh_t *ckh, const void *searchkey, void **key, void **data)
488 {
489 size_t cell;
490
491 assert(ckh != NULL);
492 dassert(ckh->magic == CKH_MAGIC);
493
494 cell = ckh_isearch(ckh, searchkey);
495 if (cell != SIZE_T_MAX) {
496 if (key != NULL)
497 *key = (void *)ckh->tab[cell].key;
498 if (data != NULL)
499 *data = (void *)ckh->tab[cell].data;
500 ckh->tab[cell].key = NULL;
501 ckh->tab[cell].data = NULL; /* Not necessary. */
502
503 ckh->count--;
504 /* Try to halve the table if it is less than 1/4 full. */
505 if (ckh->count < (ZU(1) << (ckh->lg_curbuckets
506 + LG_CKH_BUCKET_CELLS - 2)) && ckh->lg_curbuckets
507 > ckh->lg_minbuckets) {
508 /* Ignore error due to OOM. */
509 ckh_shrink(ckh);
510 }
511
512 return (false);
513 }
514
515 return (true);
516 }
517
518 bool
519 ckh_search(ckh_t *ckh, const void *searchkey, void **key, void **data)
520 {
521 size_t cell;
522
523 assert(ckh != NULL);
524 dassert(ckh->magic == CKH_MAGIC);
525
526 cell = ckh_isearch(ckh, searchkey);
527 if (cell != SIZE_T_MAX) {
528 if (key != NULL)
529 *key = (void *)ckh->tab[cell].key;
530 if (data != NULL)
531 *data = (void *)ckh->tab[cell].data;
532 return (false);
533 }
534
535 return (true);
536 }
537
538 void
539 ckh_string_hash(const void *key, unsigned minbits, size_t *hash1, size_t *hash2)
540 {
541 size_t ret1, ret2;
542 uint64_t h;
543
544 assert(minbits <= 32 || (SIZEOF_PTR == 8 && minbits <= 64));
545 assert(hash1 != NULL);
546 assert(hash2 != NULL);
547
548 h = hash(key, strlen((const char *)key), 0x94122f335b332aeaLLU);
549 if (minbits <= 32) {
550 /*
551 * Avoid doing multiple hashes, since a single hash provides
552 * enough bits.
553 */
554 ret1 = h & ZU(0xffffffffU);
555 ret2 = h >> 32;
556 } else {
557 ret1 = h;
558 ret2 = hash(key, strlen((const char *)key),
559 0x8432a476666bbc13U);
560 }
561
562 *hash1 = ret1;
563 *hash2 = ret2;
564 }
565
566 bool
567 ckh_string_keycomp(const void *k1, const void *k2)
568 {
569
570 assert(k1 != NULL);
571 assert(k2 != NULL);
572
573 return (strcmp((char *)k1, (char *)k2) ? false : true);
574 }
575
576 void
577 ckh_pointer_hash(const void *key, unsigned minbits, size_t *hash1,
578 size_t *hash2)
579 {
580 size_t ret1, ret2;
581 uint64_t h;
582 union {
583 const void *v;
584 uint64_t i;
585 } u;
586
587 assert(minbits <= 32 || (SIZEOF_PTR == 8 && minbits <= 64));
588 assert(hash1 != NULL);
589 assert(hash2 != NULL);
590
591 assert(sizeof(u.v) == sizeof(u.i));
592 #if (LG_SIZEOF_PTR != LG_SIZEOF_INT)
593 u.i = 0;
594 #endif
595 u.v = key;
596 h = hash(&u.i, sizeof(u.i), 0xd983396e68886082LLU);
597 if (minbits <= 32) {
598 /*
599 * Avoid doing multiple hashes, since a single hash provides
600 * enough bits.
601 */
602 ret1 = h & ZU(0xffffffffU);
603 ret2 = h >> 32;
604 } else {
605 assert(SIZEOF_PTR == 8);
606 ret1 = h;
607 ret2 = hash(&u.i, sizeof(u.i), 0x5e2be9aff8709a5dLLU);
608 }
609
610 *hash1 = ret1;
611 *hash2 = ret2;
612 }
613
614 bool
615 ckh_pointer_keycomp(const void *k1, const void *k2)
616 {
617
618 return ((k1 == k2) ? true : false);
619 }