|
1 /* -*- Mode: C; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- */ |
|
2 /* This Source Code Form is subject to the terms of the Mozilla Public |
|
3 * License, v. 2.0. If a copy of the MPL was not distributed with this |
|
4 * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ |
|
5 |
|
6 /* |
|
7 * Double hashing implementation. |
|
8 */ |
|
9 #include <stdio.h> |
|
10 #include <stdlib.h> |
|
11 #include <string.h> |
|
12 #include "pldhash.h" |
|
13 #include "mozilla/HashFunctions.h" |
|
14 #include "mozilla/MathAlgorithms.h" |
|
15 #include "nsDebug.h" /* for PR_ASSERT */ |
|
16 #include "nsAlgorithm.h" |
|
17 #include "mozilla/Likely.h" |
|
18 #include "mozilla/MemoryReporting.h" |
|
19 #include "mozilla/ChaosMode.h" |
|
20 |
|
21 #ifdef PL_DHASHMETER |
|
22 # define METER(x) x |
|
23 #else |
|
24 # define METER(x) /* nothing */ |
|
25 #endif |
|
26 |
|
27 /* |
|
28 * The following DEBUG-only code is used to assert that calls to one of |
|
29 * table->ops or to an enumerator do not cause re-entry into a call that |
|
30 * can mutate the table. |
|
31 */ |
|
32 #ifdef DEBUG |
|
33 |
|
34 /* |
|
35 * Most callers that assert about the recursion level don't care about |
|
36 * this magical value because they are asserting that mutation is |
|
37 * allowed (and therefore the level is 0 or 1, depending on whether they |
|
38 * incremented it). |
|
39 * |
|
40 * Only PL_DHashTableFinish needs to allow this special value. |
|
41 */ |
|
42 #define IMMUTABLE_RECURSION_LEVEL ((uint16_t)-1) |
|
43 |
|
44 #define RECURSION_LEVEL_SAFE_TO_FINISH(table_) \ |
|
45 (table_->recursionLevel == 0 || \ |
|
46 table_->recursionLevel == IMMUTABLE_RECURSION_LEVEL) |
|
47 |
|
48 #define INCREMENT_RECURSION_LEVEL(table_) \ |
|
49 do { \ |
|
50 if (table_->recursionLevel != IMMUTABLE_RECURSION_LEVEL) \ |
|
51 ++table_->recursionLevel; \ |
|
52 } while(0) |
|
53 #define DECREMENT_RECURSION_LEVEL(table_) \ |
|
54 do { \ |
|
55 if (table->recursionLevel != IMMUTABLE_RECURSION_LEVEL) { \ |
|
56 MOZ_ASSERT(table->recursionLevel > 0); \ |
|
57 --table->recursionLevel; \ |
|
58 } \ |
|
59 } while(0) |
|
60 |
|
61 #else |
|
62 |
|
63 #define INCREMENT_RECURSION_LEVEL(table_) do { } while(0) |
|
64 #define DECREMENT_RECURSION_LEVEL(table_) do { } while(0) |
|
65 |
|
66 #endif /* defined(DEBUG) */ |
|
67 |
|
68 using namespace mozilla; |
|
69 |
|
70 void * |
|
71 PL_DHashAllocTable(PLDHashTable *table, uint32_t nbytes) |
|
72 { |
|
73 return malloc(nbytes); |
|
74 } |
|
75 |
|
76 void |
|
77 PL_DHashFreeTable(PLDHashTable *table, void *ptr) |
|
78 { |
|
79 free(ptr); |
|
80 } |
|
81 |
|
82 PLDHashNumber |
|
83 PL_DHashStringKey(PLDHashTable *table, const void *key) |
|
84 { |
|
85 return HashString(static_cast<const char*>(key)); |
|
86 } |
|
87 |
|
88 PLDHashNumber |
|
89 PL_DHashVoidPtrKeyStub(PLDHashTable *table, const void *key) |
|
90 { |
|
91 return (PLDHashNumber)(ptrdiff_t)key >> 2; |
|
92 } |
|
93 |
|
94 bool |
|
95 PL_DHashMatchEntryStub(PLDHashTable *table, |
|
96 const PLDHashEntryHdr *entry, |
|
97 const void *key) |
|
98 { |
|
99 const PLDHashEntryStub *stub = (const PLDHashEntryStub *)entry; |
|
100 |
|
101 return stub->key == key; |
|
102 } |
|
103 |
|
104 bool |
|
105 PL_DHashMatchStringKey(PLDHashTable *table, |
|
106 const PLDHashEntryHdr *entry, |
|
107 const void *key) |
|
108 { |
|
109 const PLDHashEntryStub *stub = (const PLDHashEntryStub *)entry; |
|
110 |
|
111 /* XXX tolerate null keys on account of sloppy Mozilla callers. */ |
|
112 return stub->key == key || |
|
113 (stub->key && key && |
|
114 strcmp((const char *) stub->key, (const char *) key) == 0); |
|
115 } |
|
116 |
|
117 void |
|
118 PL_DHashMoveEntryStub(PLDHashTable *table, |
|
119 const PLDHashEntryHdr *from, |
|
120 PLDHashEntryHdr *to) |
|
121 { |
|
122 memcpy(to, from, table->entrySize); |
|
123 } |
|
124 |
|
125 void |
|
126 PL_DHashClearEntryStub(PLDHashTable *table, PLDHashEntryHdr *entry) |
|
127 { |
|
128 memset(entry, 0, table->entrySize); |
|
129 } |
|
130 |
|
131 void |
|
132 PL_DHashFreeStringKey(PLDHashTable *table, PLDHashEntryHdr *entry) |
|
133 { |
|
134 const PLDHashEntryStub *stub = (const PLDHashEntryStub *)entry; |
|
135 |
|
136 free((void *) stub->key); |
|
137 memset(entry, 0, table->entrySize); |
|
138 } |
|
139 |
|
140 void |
|
141 PL_DHashFinalizeStub(PLDHashTable *table) |
|
142 { |
|
143 } |
|
144 |
|
145 static const PLDHashTableOps stub_ops = { |
|
146 PL_DHashAllocTable, |
|
147 PL_DHashFreeTable, |
|
148 PL_DHashVoidPtrKeyStub, |
|
149 PL_DHashMatchEntryStub, |
|
150 PL_DHashMoveEntryStub, |
|
151 PL_DHashClearEntryStub, |
|
152 PL_DHashFinalizeStub, |
|
153 nullptr |
|
154 }; |
|
155 |
|
156 const PLDHashTableOps * |
|
157 PL_DHashGetStubOps(void) |
|
158 { |
|
159 return &stub_ops; |
|
160 } |
|
161 |
|
162 static bool |
|
163 SizeOfEntryStore(uint32_t capacity, uint32_t entrySize, uint32_t *nbytes) |
|
164 { |
|
165 uint64_t nbytes64 = uint64_t(capacity) * uint64_t(entrySize); |
|
166 *nbytes = capacity * entrySize; |
|
167 return uint64_t(*nbytes) == nbytes64; // returns false on overflow |
|
168 } |
|
169 |
|
170 PLDHashTable * |
|
171 PL_NewDHashTable(const PLDHashTableOps *ops, void *data, uint32_t entrySize, |
|
172 uint32_t capacity) |
|
173 { |
|
174 PLDHashTable *table = (PLDHashTable *) malloc(sizeof *table); |
|
175 if (!table) |
|
176 return nullptr; |
|
177 if (!PL_DHashTableInit(table, ops, data, entrySize, capacity, fallible_t())) { |
|
178 free(table); |
|
179 return nullptr; |
|
180 } |
|
181 return table; |
|
182 } |
|
183 |
|
184 void |
|
185 PL_DHashTableDestroy(PLDHashTable *table) |
|
186 { |
|
187 PL_DHashTableFinish(table); |
|
188 free(table); |
|
189 } |
|
190 |
|
191 bool |
|
192 PL_DHashTableInit(PLDHashTable *table, const PLDHashTableOps *ops, |
|
193 void *data, uint32_t entrySize, uint32_t capacity, |
|
194 const fallible_t& ) |
|
195 { |
|
196 #ifdef DEBUG |
|
197 if (entrySize > 16 * sizeof(void *)) { |
|
198 printf_stderr( |
|
199 "pldhash: for the table at address %p, the given entrySize" |
|
200 " of %lu definitely favors chaining over double hashing.\n", |
|
201 (void *) table, |
|
202 (unsigned long) entrySize); |
|
203 } |
|
204 #endif |
|
205 |
|
206 table->ops = ops; |
|
207 table->data = data; |
|
208 if (capacity < PL_DHASH_MIN_SIZE) |
|
209 capacity = PL_DHASH_MIN_SIZE; |
|
210 |
|
211 int log2 = CeilingLog2(capacity); |
|
212 |
|
213 capacity = 1u << log2; |
|
214 if (capacity > PL_DHASH_MAX_SIZE) |
|
215 return false; |
|
216 table->hashShift = PL_DHASH_BITS - log2; |
|
217 table->entrySize = entrySize; |
|
218 table->entryCount = table->removedCount = 0; |
|
219 table->generation = 0; |
|
220 uint32_t nbytes; |
|
221 if (!SizeOfEntryStore(capacity, entrySize, &nbytes)) |
|
222 return false; // overflowed |
|
223 |
|
224 table->entryStore = (char *) ops->allocTable(table, nbytes); |
|
225 if (!table->entryStore) |
|
226 return false; |
|
227 memset(table->entryStore, 0, nbytes); |
|
228 METER(memset(&table->stats, 0, sizeof table->stats)); |
|
229 |
|
230 #ifdef DEBUG |
|
231 table->recursionLevel = 0; |
|
232 #endif |
|
233 |
|
234 return true; |
|
235 } |
|
236 |
|
237 void |
|
238 PL_DHashTableInit(PLDHashTable *table, const PLDHashTableOps *ops, void *data, |
|
239 uint32_t entrySize, uint32_t capacity) |
|
240 { |
|
241 if (!PL_DHashTableInit(table, ops, data, entrySize, capacity, fallible_t())) { |
|
242 if (capacity > PL_DHASH_MAX_SIZE) { |
|
243 MOZ_CRASH(); |
|
244 } |
|
245 uint32_t nbytes; |
|
246 if (!SizeOfEntryStore(capacity, entrySize, &nbytes)) { |
|
247 MOZ_CRASH(); |
|
248 } |
|
249 NS_ABORT_OOM(nbytes); |
|
250 } |
|
251 } |
|
252 |
|
253 /* |
|
254 * Compute max and min load numbers (entry counts). We have a secondary max |
|
255 * that allows us to overload a table reasonably if it cannot be grown further |
|
256 * (i.e. if ChangeTable() fails). The table slows down drastically if the |
|
257 * secondary max is too close to 1, but 0.96875 gives only a slight slowdown |
|
258 * while allowing 1.3x more elements. |
|
259 */ |
|
260 static inline uint32_t MaxLoad(uint32_t size) { |
|
261 return size - (size >> 2); // == size * 0.75 |
|
262 } |
|
263 static inline uint32_t MaxLoadOnGrowthFailure(uint32_t size) { |
|
264 return size - (size >> 5); // == size * 0.96875 |
|
265 } |
|
266 static inline uint32_t MinLoad(uint32_t size) { |
|
267 return size >> 2; // == size * 0.25 |
|
268 } |
|
269 |
|
270 /* |
|
271 * Double hashing needs the second hash code to be relatively prime to table |
|
272 * size, so we simply make hash2 odd. |
|
273 */ |
|
274 #define HASH1(hash0, shift) ((hash0) >> (shift)) |
|
275 #define HASH2(hash0,log2,shift) ((((hash0) << (log2)) >> (shift)) | 1) |
|
276 |
|
277 /* |
|
278 * Reserve keyHash 0 for free entries and 1 for removed-entry sentinels. Note |
|
279 * that a removed-entry sentinel need be stored only if the removed entry had |
|
280 * a colliding entry added after it. Therefore we can use 1 as the collision |
|
281 * flag in addition to the removed-entry sentinel value. Multiplicative hash |
|
282 * uses the high order bits of keyHash, so this least-significant reservation |
|
283 * should not hurt the hash function's effectiveness much. |
|
284 * |
|
285 * If you change any of these magic numbers, also update PL_DHASH_ENTRY_IS_LIVE |
|
286 * in pldhash.h. It used to be private to pldhash.c, but then became public to |
|
287 * assist iterator writers who inspect table->entryStore directly. |
|
288 */ |
|
289 #define COLLISION_FLAG ((PLDHashNumber) 1) |
|
290 #define MARK_ENTRY_FREE(entry) ((entry)->keyHash = 0) |
|
291 #define MARK_ENTRY_REMOVED(entry) ((entry)->keyHash = 1) |
|
292 #define ENTRY_IS_REMOVED(entry) ((entry)->keyHash == 1) |
|
293 #define ENTRY_IS_LIVE(entry) PL_DHASH_ENTRY_IS_LIVE(entry) |
|
294 #define ENSURE_LIVE_KEYHASH(hash0) if (hash0 < 2) hash0 -= 2; else (void)0 |
|
295 |
|
296 /* Match an entry's keyHash against an unstored one computed from a key. */ |
|
297 #define MATCH_ENTRY_KEYHASH(entry,hash0) \ |
|
298 (((entry)->keyHash & ~COLLISION_FLAG) == (hash0)) |
|
299 |
|
300 /* Compute the address of the indexed entry in table. */ |
|
301 #define ADDRESS_ENTRY(table, index) \ |
|
302 ((PLDHashEntryHdr *)((table)->entryStore + (index) * (table)->entrySize)) |
|
303 |
|
304 void |
|
305 PL_DHashTableFinish(PLDHashTable *table) |
|
306 { |
|
307 INCREMENT_RECURSION_LEVEL(table); |
|
308 |
|
309 /* Call finalize before clearing entries, so it can enumerate them. */ |
|
310 table->ops->finalize(table); |
|
311 |
|
312 /* Clear any remaining live entries. */ |
|
313 char *entryAddr = table->entryStore; |
|
314 uint32_t entrySize = table->entrySize; |
|
315 char *entryLimit = entryAddr + PL_DHASH_TABLE_SIZE(table) * entrySize; |
|
316 while (entryAddr < entryLimit) { |
|
317 PLDHashEntryHdr *entry = (PLDHashEntryHdr *)entryAddr; |
|
318 if (ENTRY_IS_LIVE(entry)) { |
|
319 METER(table->stats.removeEnums++); |
|
320 table->ops->clearEntry(table, entry); |
|
321 } |
|
322 entryAddr += entrySize; |
|
323 } |
|
324 |
|
325 DECREMENT_RECURSION_LEVEL(table); |
|
326 MOZ_ASSERT(RECURSION_LEVEL_SAFE_TO_FINISH(table)); |
|
327 |
|
328 /* Free entry storage last. */ |
|
329 table->ops->freeTable(table, table->entryStore); |
|
330 } |
|
331 |
|
332 static PLDHashEntryHdr * PL_DHASH_FASTCALL |
|
333 SearchTable(PLDHashTable *table, const void *key, PLDHashNumber keyHash, |
|
334 PLDHashOperator op) |
|
335 { |
|
336 METER(table->stats.searches++); |
|
337 NS_ASSERTION(!(keyHash & COLLISION_FLAG), |
|
338 "!(keyHash & COLLISION_FLAG)"); |
|
339 |
|
340 /* Compute the primary hash address. */ |
|
341 int hashShift = table->hashShift; |
|
342 PLDHashNumber hash1 = HASH1(keyHash, hashShift); |
|
343 PLDHashEntryHdr *entry = ADDRESS_ENTRY(table, hash1); |
|
344 |
|
345 /* Miss: return space for a new entry. */ |
|
346 if (PL_DHASH_ENTRY_IS_FREE(entry)) { |
|
347 METER(table->stats.misses++); |
|
348 return entry; |
|
349 } |
|
350 |
|
351 /* Hit: return entry. */ |
|
352 PLDHashMatchEntry matchEntry = table->ops->matchEntry; |
|
353 if (MATCH_ENTRY_KEYHASH(entry, keyHash) && matchEntry(table, entry, key)) { |
|
354 METER(table->stats.hits++); |
|
355 return entry; |
|
356 } |
|
357 |
|
358 /* Collision: double hash. */ |
|
359 int sizeLog2 = PL_DHASH_BITS - table->hashShift; |
|
360 PLDHashNumber hash2 = HASH2(keyHash, sizeLog2, hashShift); |
|
361 uint32_t sizeMask = (1u << sizeLog2) - 1; |
|
362 |
|
363 /* Save the first removed entry pointer so PL_DHASH_ADD can recycle it. */ |
|
364 PLDHashEntryHdr *firstRemoved = nullptr; |
|
365 |
|
366 for (;;) { |
|
367 if (MOZ_UNLIKELY(ENTRY_IS_REMOVED(entry))) { |
|
368 if (!firstRemoved) |
|
369 firstRemoved = entry; |
|
370 } else { |
|
371 if (op == PL_DHASH_ADD) |
|
372 entry->keyHash |= COLLISION_FLAG; |
|
373 } |
|
374 |
|
375 METER(table->stats.steps++); |
|
376 hash1 -= hash2; |
|
377 hash1 &= sizeMask; |
|
378 |
|
379 entry = ADDRESS_ENTRY(table, hash1); |
|
380 if (PL_DHASH_ENTRY_IS_FREE(entry)) { |
|
381 METER(table->stats.misses++); |
|
382 return (firstRemoved && op == PL_DHASH_ADD) ? firstRemoved : entry; |
|
383 } |
|
384 |
|
385 if (MATCH_ENTRY_KEYHASH(entry, keyHash) && |
|
386 matchEntry(table, entry, key)) { |
|
387 METER(table->stats.hits++); |
|
388 return entry; |
|
389 } |
|
390 } |
|
391 |
|
392 /* NOTREACHED */ |
|
393 return nullptr; |
|
394 } |
|
395 |
|
396 /* |
|
397 * This is a copy of SearchTable, used by ChangeTable, hardcoded to |
|
398 * 1. assume |op == PL_DHASH_ADD|, |
|
399 * 2. assume that |key| will never match an existing entry, and |
|
400 * 3. assume that no entries have been removed from the current table |
|
401 * structure. |
|
402 * Avoiding the need for |key| means we can avoid needing a way to map |
|
403 * entries to keys, which means callers can use complex key types more |
|
404 * easily. |
|
405 */ |
|
406 static PLDHashEntryHdr * PL_DHASH_FASTCALL |
|
407 FindFreeEntry(PLDHashTable *table, PLDHashNumber keyHash) |
|
408 { |
|
409 METER(table->stats.searches++); |
|
410 NS_ASSERTION(!(keyHash & COLLISION_FLAG), |
|
411 "!(keyHash & COLLISION_FLAG)"); |
|
412 |
|
413 /* Compute the primary hash address. */ |
|
414 int hashShift = table->hashShift; |
|
415 PLDHashNumber hash1 = HASH1(keyHash, hashShift); |
|
416 PLDHashEntryHdr *entry = ADDRESS_ENTRY(table, hash1); |
|
417 |
|
418 /* Miss: return space for a new entry. */ |
|
419 if (PL_DHASH_ENTRY_IS_FREE(entry)) { |
|
420 METER(table->stats.misses++); |
|
421 return entry; |
|
422 } |
|
423 |
|
424 /* Collision: double hash. */ |
|
425 int sizeLog2 = PL_DHASH_BITS - table->hashShift; |
|
426 PLDHashNumber hash2 = HASH2(keyHash, sizeLog2, hashShift); |
|
427 uint32_t sizeMask = (1u << sizeLog2) - 1; |
|
428 |
|
429 for (;;) { |
|
430 NS_ASSERTION(!ENTRY_IS_REMOVED(entry), |
|
431 "!ENTRY_IS_REMOVED(entry)"); |
|
432 entry->keyHash |= COLLISION_FLAG; |
|
433 |
|
434 METER(table->stats.steps++); |
|
435 hash1 -= hash2; |
|
436 hash1 &= sizeMask; |
|
437 |
|
438 entry = ADDRESS_ENTRY(table, hash1); |
|
439 if (PL_DHASH_ENTRY_IS_FREE(entry)) { |
|
440 METER(table->stats.misses++); |
|
441 return entry; |
|
442 } |
|
443 } |
|
444 |
|
445 /* NOTREACHED */ |
|
446 return nullptr; |
|
447 } |
|
448 |
|
449 static bool |
|
450 ChangeTable(PLDHashTable *table, int deltaLog2) |
|
451 { |
|
452 /* Look, but don't touch, until we succeed in getting new entry store. */ |
|
453 int oldLog2 = PL_DHASH_BITS - table->hashShift; |
|
454 int newLog2 = oldLog2 + deltaLog2; |
|
455 uint32_t newCapacity = 1u << newLog2; |
|
456 if (newCapacity > PL_DHASH_MAX_SIZE) |
|
457 return false; |
|
458 |
|
459 uint32_t entrySize = table->entrySize; |
|
460 uint32_t nbytes; |
|
461 if (!SizeOfEntryStore(newCapacity, entrySize, &nbytes)) |
|
462 return false; // overflowed |
|
463 |
|
464 char *newEntryStore = (char *) table->ops->allocTable(table, nbytes); |
|
465 if (!newEntryStore) |
|
466 return false; |
|
467 |
|
468 /* We can't fail from here on, so update table parameters. */ |
|
469 #ifdef DEBUG |
|
470 uint32_t recursionLevel = table->recursionLevel; |
|
471 #endif |
|
472 table->hashShift = PL_DHASH_BITS - newLog2; |
|
473 table->removedCount = 0; |
|
474 table->generation++; |
|
475 |
|
476 /* Assign the new entry store to table. */ |
|
477 memset(newEntryStore, 0, nbytes); |
|
478 char *oldEntryStore, *oldEntryAddr; |
|
479 oldEntryAddr = oldEntryStore = table->entryStore; |
|
480 table->entryStore = newEntryStore; |
|
481 PLDHashMoveEntry moveEntry = table->ops->moveEntry; |
|
482 #ifdef DEBUG |
|
483 table->recursionLevel = recursionLevel; |
|
484 #endif |
|
485 |
|
486 /* Copy only live entries, leaving removed ones behind. */ |
|
487 uint32_t oldCapacity = 1u << oldLog2; |
|
488 for (uint32_t i = 0; i < oldCapacity; i++) { |
|
489 PLDHashEntryHdr *oldEntry = (PLDHashEntryHdr *)oldEntryAddr; |
|
490 if (ENTRY_IS_LIVE(oldEntry)) { |
|
491 oldEntry->keyHash &= ~COLLISION_FLAG; |
|
492 PLDHashEntryHdr *newEntry = FindFreeEntry(table, oldEntry->keyHash); |
|
493 NS_ASSERTION(PL_DHASH_ENTRY_IS_FREE(newEntry), |
|
494 "PL_DHASH_ENTRY_IS_FREE(newEntry)"); |
|
495 moveEntry(table, oldEntry, newEntry); |
|
496 newEntry->keyHash = oldEntry->keyHash; |
|
497 } |
|
498 oldEntryAddr += entrySize; |
|
499 } |
|
500 |
|
501 table->ops->freeTable(table, oldEntryStore); |
|
502 return true; |
|
503 } |
|
504 |
|
505 PLDHashEntryHdr * PL_DHASH_FASTCALL |
|
506 PL_DHashTableOperate(PLDHashTable *table, const void *key, PLDHashOperator op) |
|
507 { |
|
508 PLDHashEntryHdr *entry; |
|
509 |
|
510 MOZ_ASSERT(op == PL_DHASH_LOOKUP || table->recursionLevel == 0); |
|
511 INCREMENT_RECURSION_LEVEL(table); |
|
512 |
|
513 PLDHashNumber keyHash = table->ops->hashKey(table, key); |
|
514 keyHash *= PL_DHASH_GOLDEN_RATIO; |
|
515 |
|
516 /* Avoid 0 and 1 hash codes, they indicate free and removed entries. */ |
|
517 ENSURE_LIVE_KEYHASH(keyHash); |
|
518 keyHash &= ~COLLISION_FLAG; |
|
519 |
|
520 switch (op) { |
|
521 case PL_DHASH_LOOKUP: |
|
522 METER(table->stats.lookups++); |
|
523 entry = SearchTable(table, key, keyHash, op); |
|
524 break; |
|
525 |
|
526 case PL_DHASH_ADD: { |
|
527 /* |
|
528 * If alpha is >= .75, grow or compress the table. If key is already |
|
529 * in the table, we may grow once more than necessary, but only if we |
|
530 * are on the edge of being overloaded. |
|
531 */ |
|
532 uint32_t size = PL_DHASH_TABLE_SIZE(table); |
|
533 if (table->entryCount + table->removedCount >= MaxLoad(size)) { |
|
534 /* Compress if a quarter or more of all entries are removed. */ |
|
535 int deltaLog2; |
|
536 if (table->removedCount >= size >> 2) { |
|
537 METER(table->stats.compresses++); |
|
538 deltaLog2 = 0; |
|
539 } else { |
|
540 METER(table->stats.grows++); |
|
541 deltaLog2 = 1; |
|
542 } |
|
543 |
|
544 /* |
|
545 * Grow or compress table. If ChangeTable() fails, allow |
|
546 * overloading up to the secondary max. Once we hit the secondary |
|
547 * max, return null. |
|
548 */ |
|
549 if (!ChangeTable(table, deltaLog2) && |
|
550 table->entryCount + table->removedCount >= |
|
551 MaxLoadOnGrowthFailure(size)) |
|
552 { |
|
553 METER(table->stats.addFailures++); |
|
554 entry = nullptr; |
|
555 break; |
|
556 } |
|
557 } |
|
558 |
|
559 /* |
|
560 * Look for entry after possibly growing, so we don't have to add it, |
|
561 * then skip it while growing the table and re-add it after. |
|
562 */ |
|
563 entry = SearchTable(table, key, keyHash, op); |
|
564 if (!ENTRY_IS_LIVE(entry)) { |
|
565 /* Initialize the entry, indicating that it's no longer free. */ |
|
566 METER(table->stats.addMisses++); |
|
567 if (ENTRY_IS_REMOVED(entry)) { |
|
568 METER(table->stats.addOverRemoved++); |
|
569 table->removedCount--; |
|
570 keyHash |= COLLISION_FLAG; |
|
571 } |
|
572 if (table->ops->initEntry && |
|
573 !table->ops->initEntry(table, entry, key)) { |
|
574 /* We haven't claimed entry yet; fail with null return. */ |
|
575 memset(entry + 1, 0, table->entrySize - sizeof *entry); |
|
576 entry = nullptr; |
|
577 break; |
|
578 } |
|
579 entry->keyHash = keyHash; |
|
580 table->entryCount++; |
|
581 } |
|
582 METER(else table->stats.addHits++); |
|
583 break; |
|
584 } |
|
585 |
|
586 case PL_DHASH_REMOVE: |
|
587 entry = SearchTable(table, key, keyHash, op); |
|
588 if (ENTRY_IS_LIVE(entry)) { |
|
589 /* Clear this entry and mark it as "removed". */ |
|
590 METER(table->stats.removeHits++); |
|
591 PL_DHashTableRawRemove(table, entry); |
|
592 |
|
593 /* Shrink if alpha is <= .25 and table isn't too small already. */ |
|
594 uint32_t size = PL_DHASH_TABLE_SIZE(table); |
|
595 if (size > PL_DHASH_MIN_SIZE && |
|
596 table->entryCount <= MinLoad(size)) { |
|
597 METER(table->stats.shrinks++); |
|
598 (void) ChangeTable(table, -1); |
|
599 } |
|
600 } |
|
601 METER(else table->stats.removeMisses++); |
|
602 entry = nullptr; |
|
603 break; |
|
604 |
|
605 default: |
|
606 NS_NOTREACHED("0"); |
|
607 entry = nullptr; |
|
608 } |
|
609 |
|
610 DECREMENT_RECURSION_LEVEL(table); |
|
611 |
|
612 return entry; |
|
613 } |
|
614 |
|
615 void |
|
616 PL_DHashTableRawRemove(PLDHashTable *table, PLDHashEntryHdr *entry) |
|
617 { |
|
618 MOZ_ASSERT(table->recursionLevel != IMMUTABLE_RECURSION_LEVEL); |
|
619 |
|
620 NS_ASSERTION(PL_DHASH_ENTRY_IS_LIVE(entry), |
|
621 "PL_DHASH_ENTRY_IS_LIVE(entry)"); |
|
622 |
|
623 /* Load keyHash first in case clearEntry() goofs it. */ |
|
624 PLDHashNumber keyHash = entry->keyHash; |
|
625 table->ops->clearEntry(table, entry); |
|
626 if (keyHash & COLLISION_FLAG) { |
|
627 MARK_ENTRY_REMOVED(entry); |
|
628 table->removedCount++; |
|
629 } else { |
|
630 METER(table->stats.removeFrees++); |
|
631 MARK_ENTRY_FREE(entry); |
|
632 } |
|
633 table->entryCount--; |
|
634 } |
|
635 |
|
636 uint32_t |
|
637 PL_DHashTableEnumerate(PLDHashTable *table, PLDHashEnumerator etor, void *arg) |
|
638 { |
|
639 INCREMENT_RECURSION_LEVEL(table); |
|
640 |
|
641 char *entryAddr = table->entryStore; |
|
642 uint32_t entrySize = table->entrySize; |
|
643 uint32_t capacity = PL_DHASH_TABLE_SIZE(table); |
|
644 uint32_t tableSize = capacity * entrySize; |
|
645 char *entryLimit = entryAddr + tableSize; |
|
646 uint32_t i = 0; |
|
647 bool didRemove = false; |
|
648 |
|
649 if (ChaosMode::isActive()) { |
|
650 // Start iterating at a random point in the hashtable. It would be |
|
651 // even more chaotic to iterate in fully random order, but that's a lot |
|
652 // more work. |
|
653 entryAddr += ChaosMode::randomUint32LessThan(capacity) * entrySize; |
|
654 if (entryAddr >= entryLimit) { |
|
655 entryAddr -= tableSize; |
|
656 } |
|
657 } |
|
658 |
|
659 for (uint32_t e = 0; e < capacity; ++e) { |
|
660 PLDHashEntryHdr *entry = (PLDHashEntryHdr *)entryAddr; |
|
661 if (ENTRY_IS_LIVE(entry)) { |
|
662 PLDHashOperator op = etor(table, entry, i++, arg); |
|
663 if (op & PL_DHASH_REMOVE) { |
|
664 METER(table->stats.removeEnums++); |
|
665 PL_DHashTableRawRemove(table, entry); |
|
666 didRemove = true; |
|
667 } |
|
668 if (op & PL_DHASH_STOP) |
|
669 break; |
|
670 } |
|
671 entryAddr += entrySize; |
|
672 if (entryAddr >= entryLimit) { |
|
673 entryAddr -= tableSize; |
|
674 } |
|
675 } |
|
676 |
|
677 MOZ_ASSERT(!didRemove || table->recursionLevel == 1); |
|
678 |
|
679 /* |
|
680 * Shrink or compress if a quarter or more of all entries are removed, or |
|
681 * if the table is underloaded according to the minimum alpha, and is not |
|
682 * minimal-size already. Do this only if we removed above, so non-removing |
|
683 * enumerations can count on stable table->entryStore until the next |
|
684 * non-lookup-Operate or removing-Enumerate. |
|
685 */ |
|
686 if (didRemove && |
|
687 (table->removedCount >= capacity >> 2 || |
|
688 (capacity > PL_DHASH_MIN_SIZE && |
|
689 table->entryCount <= MinLoad(capacity)))) { |
|
690 METER(table->stats.enumShrinks++); |
|
691 capacity = table->entryCount; |
|
692 capacity += capacity >> 1; |
|
693 if (capacity < PL_DHASH_MIN_SIZE) |
|
694 capacity = PL_DHASH_MIN_SIZE; |
|
695 |
|
696 uint32_t ceiling = CeilingLog2(capacity); |
|
697 ceiling -= PL_DHASH_BITS - table->hashShift; |
|
698 |
|
699 (void) ChangeTable(table, ceiling); |
|
700 } |
|
701 |
|
702 DECREMENT_RECURSION_LEVEL(table); |
|
703 |
|
704 return i; |
|
705 } |
|
706 |
|
707 struct SizeOfEntryExcludingThisArg |
|
708 { |
|
709 size_t total; |
|
710 PLDHashSizeOfEntryExcludingThisFun sizeOfEntryExcludingThis; |
|
711 MallocSizeOf mallocSizeOf; |
|
712 void *arg; // the arg passed by the user |
|
713 }; |
|
714 |
|
715 static PLDHashOperator |
|
716 SizeOfEntryExcludingThisEnumerator(PLDHashTable *table, PLDHashEntryHdr *hdr, |
|
717 uint32_t number, void *arg) |
|
718 { |
|
719 SizeOfEntryExcludingThisArg *e = (SizeOfEntryExcludingThisArg *)arg; |
|
720 e->total += e->sizeOfEntryExcludingThis(hdr, e->mallocSizeOf, e->arg); |
|
721 return PL_DHASH_NEXT; |
|
722 } |
|
723 |
|
724 size_t |
|
725 PL_DHashTableSizeOfExcludingThis(const PLDHashTable *table, |
|
726 PLDHashSizeOfEntryExcludingThisFun sizeOfEntryExcludingThis, |
|
727 MallocSizeOf mallocSizeOf, |
|
728 void *arg /* = nullptr */) |
|
729 { |
|
730 size_t n = 0; |
|
731 n += mallocSizeOf(table->entryStore); |
|
732 if (sizeOfEntryExcludingThis) { |
|
733 SizeOfEntryExcludingThisArg arg2 = { 0, sizeOfEntryExcludingThis, mallocSizeOf, arg }; |
|
734 PL_DHashTableEnumerate(const_cast<PLDHashTable *>(table), |
|
735 SizeOfEntryExcludingThisEnumerator, &arg2); |
|
736 n += arg2.total; |
|
737 } |
|
738 return n; |
|
739 } |
|
740 |
|
741 size_t |
|
742 PL_DHashTableSizeOfIncludingThis(const PLDHashTable *table, |
|
743 PLDHashSizeOfEntryExcludingThisFun sizeOfEntryExcludingThis, |
|
744 MallocSizeOf mallocSizeOf, |
|
745 void *arg /* = nullptr */) |
|
746 { |
|
747 return mallocSizeOf(table) + |
|
748 PL_DHashTableSizeOfExcludingThis(table, sizeOfEntryExcludingThis, |
|
749 mallocSizeOf, arg); |
|
750 } |
|
751 |
|
752 #ifdef DEBUG |
|
753 void |
|
754 PL_DHashMarkTableImmutable(PLDHashTable *table) |
|
755 { |
|
756 table->recursionLevel = IMMUTABLE_RECURSION_LEVEL; |
|
757 } |
|
758 #endif |
|
759 |
|
760 #ifdef PL_DHASHMETER |
|
761 #include <math.h> |
|
762 |
|
763 void |
|
764 PL_DHashTableDumpMeter(PLDHashTable *table, PLDHashEnumerator dump, FILE *fp) |
|
765 { |
|
766 PLDHashNumber hash1, hash2, maxChainHash1, maxChainHash2; |
|
767 double sqsum, mean, variance, sigma; |
|
768 PLDHashEntryHdr *entry; |
|
769 |
|
770 char *entryAddr = table->entryStore; |
|
771 uint32_t entrySize = table->entrySize; |
|
772 int hashShift = table->hashShift; |
|
773 int sizeLog2 = PL_DHASH_BITS - hashShift; |
|
774 uint32_t tableSize = PL_DHASH_TABLE_SIZE(table); |
|
775 uint32_t sizeMask = (1u << sizeLog2) - 1; |
|
776 uint32_t chainCount = 0, maxChainLen = 0; |
|
777 hash2 = 0; |
|
778 sqsum = 0; |
|
779 |
|
780 for (uint32_t i = 0; i < tableSize; i++) { |
|
781 entry = (PLDHashEntryHdr *)entryAddr; |
|
782 entryAddr += entrySize; |
|
783 if (!ENTRY_IS_LIVE(entry)) |
|
784 continue; |
|
785 hash1 = HASH1(entry->keyHash & ~COLLISION_FLAG, hashShift); |
|
786 PLDHashNumber saveHash1 = hash1; |
|
787 PLDHashEntryHdr *probe = ADDRESS_ENTRY(table, hash1); |
|
788 uint32_t chainLen = 1; |
|
789 if (probe == entry) { |
|
790 /* Start of a (possibly unit-length) chain. */ |
|
791 chainCount++; |
|
792 } else { |
|
793 hash2 = HASH2(entry->keyHash & ~COLLISION_FLAG, sizeLog2, |
|
794 hashShift); |
|
795 do { |
|
796 chainLen++; |
|
797 hash1 -= hash2; |
|
798 hash1 &= sizeMask; |
|
799 probe = ADDRESS_ENTRY(table, hash1); |
|
800 } while (probe != entry); |
|
801 } |
|
802 sqsum += chainLen * chainLen; |
|
803 if (chainLen > maxChainLen) { |
|
804 maxChainLen = chainLen; |
|
805 maxChainHash1 = saveHash1; |
|
806 maxChainHash2 = hash2; |
|
807 } |
|
808 } |
|
809 |
|
810 uint32_t entryCount = table->entryCount; |
|
811 if (entryCount && chainCount) { |
|
812 mean = (double)entryCount / chainCount; |
|
813 variance = chainCount * sqsum - entryCount * entryCount; |
|
814 if (variance < 0 || chainCount == 1) |
|
815 variance = 0; |
|
816 else |
|
817 variance /= chainCount * (chainCount - 1); |
|
818 sigma = sqrt(variance); |
|
819 } else { |
|
820 mean = sigma = 0; |
|
821 } |
|
822 |
|
823 fprintf(fp, "Double hashing statistics:\n"); |
|
824 fprintf(fp, " table size (in entries): %u\n", tableSize); |
|
825 fprintf(fp, " number of entries: %u\n", table->entryCount); |
|
826 fprintf(fp, " number of removed entries: %u\n", table->removedCount); |
|
827 fprintf(fp, " number of searches: %u\n", table->stats.searches); |
|
828 fprintf(fp, " number of hits: %u\n", table->stats.hits); |
|
829 fprintf(fp, " number of misses: %u\n", table->stats.misses); |
|
830 fprintf(fp, " mean steps per search: %g\n", table->stats.searches ? |
|
831 (double)table->stats.steps |
|
832 / table->stats.searches : |
|
833 0.); |
|
834 fprintf(fp, " mean hash chain length: %g\n", mean); |
|
835 fprintf(fp, " standard deviation: %g\n", sigma); |
|
836 fprintf(fp, " maximum hash chain length: %u\n", maxChainLen); |
|
837 fprintf(fp, " number of lookups: %u\n", table->stats.lookups); |
|
838 fprintf(fp, " adds that made a new entry: %u\n", table->stats.addMisses); |
|
839 fprintf(fp, "adds that recycled removeds: %u\n", table->stats.addOverRemoved); |
|
840 fprintf(fp, " adds that found an entry: %u\n", table->stats.addHits); |
|
841 fprintf(fp, " add failures: %u\n", table->stats.addFailures); |
|
842 fprintf(fp, " useful removes: %u\n", table->stats.removeHits); |
|
843 fprintf(fp, " useless removes: %u\n", table->stats.removeMisses); |
|
844 fprintf(fp, "removes that freed an entry: %u\n", table->stats.removeFrees); |
|
845 fprintf(fp, " removes while enumerating: %u\n", table->stats.removeEnums); |
|
846 fprintf(fp, " number of grows: %u\n", table->stats.grows); |
|
847 fprintf(fp, " number of shrinks: %u\n", table->stats.shrinks); |
|
848 fprintf(fp, " number of compresses: %u\n", table->stats.compresses); |
|
849 fprintf(fp, "number of enumerate shrinks: %u\n", table->stats.enumShrinks); |
|
850 |
|
851 if (dump && maxChainLen && hash2) { |
|
852 fputs("Maximum hash chain:\n", fp); |
|
853 hash1 = maxChainHash1; |
|
854 hash2 = maxChainHash2; |
|
855 entry = ADDRESS_ENTRY(table, hash1); |
|
856 uint32_t i = 0; |
|
857 do { |
|
858 if (dump(table, entry, i++, fp) != PL_DHASH_NEXT) |
|
859 break; |
|
860 hash1 -= hash2; |
|
861 hash1 &= sizeMask; |
|
862 entry = ADDRESS_ENTRY(table, hash1); |
|
863 } while (PL_DHASH_ENTRY_IS_BUSY(entry)); |
|
864 } |
|
865 } |
|
866 #endif /* PL_DHASHMETER */ |