2 * Copyright (C) 1995-2008 University of Karlsruhe. All right reserved.
4 * This file is part of libFirm.
6 * This file may be distributed and/or modified under the terms of the
7 * GNU General Public License version 2 as published by the Free Software
8 * Foundation and appearing in the file LICENSE.GPL included in the
9 * packaging of this file.
11 * Licensees holding valid libFirm Professional Edition licenses may use
12 * this file in accordance with the libFirm Commercial License.
13 * Agreement provided with the Software.
15 * This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
16 * WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * @brief Generic hashset implementation
23 * @author Matthias Braun, inspiration from densehash from google sparsehash
29 * You have to specialize this file by defining:
32 * <li><b>HashSet</b> The name of the hashset type</li>
33 * <li><b>HashSetIterator</b> The name of the hashset iterator type</li>
34 * <li><b>ValueType</b> The type of the stored data values</li>
35 * <li><b>NullValue</b> A special value representing no values</li>
36 * <li><b>DeletedValue</b> A special value representing deleted entries</li>
37 * <li><b>Hash(hashset,key)</b> calculates the hash value for a given key</li>
40 * Note that by default it is assumed that the data values themselfes are used
41 * as keys. However you can change that with additional defines:
44 * <li><b>KeyType</b> The type of the keys identifying data values.
45 * Defining this implies, that a data value contains
46 * more than just the key.</li>
47 * <li><b>GetKey(value)</b> Extracts the key from a data value</li>
48 * <li><b>KeysEqual(hashset,key1,key2)</b> Tests wether 2 keys are equal</li>
49 * <li><b>DO_REHASH</b> Instead of storing the hash-values, recalculate
50 * them on demand from the datavalues. (usefull if
51 * calculating the hash-values takes less time than
52 * a memory access)</li>
55 * You can further fine tune your hashset by defining the following:
58 * <li><b>JUMP(num_probes)</b> The probing method</li>
59 * <li><b>Alloc(count)</b> Allocates count hashset entries (NOT bytes)</li>
60 * <li><b>Free(ptr)</b> Frees a block of memory allocated by Alloc</li>
61 * <li><b>SetRangeEmpty(ptr,count)</b> Efficiently sets a range of elements to
63 * <li><b>ADDITIONAL_DATA<b> Additional fields appended to the hashset struct</li>
72 #include "bitfiddle.h"
75 /* quadratic probing */
77 #define JUMP(num_probes) (num_probes)
82 #define Hash(self,key) ((unsigned)(((char *)key) - (char *)0))
86 #define HashSetEntry ValueType
87 #define EntrySetHash(entry,new_hash)
88 #define EntryGetHash(self,entry) Hash(self, GetKey(entry))
89 #define EntryGetValue(entry) (entry)
90 #else /* ! DO_REHASH */
91 #define EntryGetHash(self,entry) (entry).hash
92 #define EntrySetHash(entry,new_hash) (entry).hash = (new_hash)
93 #define EntryGetValue(entry) (entry).data
94 #endif /* DO_REHASH */
98 #define Alloc(size) (HashSetEntry*) xmalloc((size) * sizeof(HashSetEntry))
99 #define Free(ptr) free(ptr)
103 #define InsertReturnValue int
104 #define GetInsertReturnValue(entry,found) (found)
105 #define NullReturnValue 0
106 #else /* ! ID_HASH */
108 #define InsertReturnValue ValueType
109 #define GetInsertReturnValue(entry,found) EntryGetValue(entry)
110 #define NullReturnValue NullValue
112 #define InsertReturnValue ValueType*
113 #define GetInsertReturnValue(entry,found) & EntryGetValue(entry)
114 #define NullReturnValue & NullValue
119 #define KeyType ValueType
120 #define GetKey(value) (value)
121 #define InitData(self,value,key) (value) = (key)
125 #define ConstKeyType const KeyType
126 #endif /* ConstKeyType */
128 #ifndef EntrySetEmpty
129 #define EntrySetEmpty(entry) EntryGetValue(entry) = NullValue
130 #endif /* EntrySetEmpty */
131 #ifndef EntrySetDeleted
132 #define EntrySetDeleted(entry) EntryGetValue(entry) = DeletedValue
133 #endif /* EntrySetDeleted */
135 #define EntryIsEmpty(entry) (EntryGetValue(entry) == NullValue)
136 #endif /* EntryIsEmpty */
137 #ifndef EntryIsDeleted
138 #define EntryIsDeleted(entry) (EntryGetValue(entry) == DeletedValue)
139 #endif /* EntryIsDeleted */
140 #ifndef SetRangeEmpty
141 #define SetRangeEmpty(ptr,size) \
144 size_t _size = (size); \
145 HashSetEntry *entries = (ptr); \
146 for(_i = 0; _i < _size; ++_i) { \
147 HashSetEntry *entry = & entries[_i]; \
148 EntrySetEmpty(*entry); \
151 #endif /* SetRangeEmpty */
153 #ifndef HT_OCCUPANCY_FLT
154 /** how full before we double size */
155 #define HT_OCCUPANCY_FLT(x) ((x)/2)
156 #endif /* HT_OCCUPANCY_FLT */
157 #ifndef HT_1_DIV_OCCUPANCY_FLT
158 #define HT_1_DIV_OCCUPANCY_FLT 2
162 /** how empty before we half size */
163 #define HT_EMPTY_FLT(x) ((x)/5)
164 #endif /* HT_EMPTY_FLT */
166 #ifndef HT_MIN_BUCKETS
167 /** default smallest bucket size */
168 #define HT_MIN_BUCKETS 32
169 #endif /* HT_MIN_BUCKETS */
171 #define ILLEGAL_POS ((size_t)-1)
173 /* check that all needed functions are defined */
175 #error You have to redefine hashset_init
177 #ifndef hashset_init_size
178 #error You have to redefine hashset_init_size
180 #ifndef hashset_destroy
181 #error You have to redefine hashset_destroy
183 #ifndef hashset_insert
184 #error You have to redefine hashset_insert
186 #ifndef hashset_remove
187 #error You have to redefine hashset_remove
190 #error You have to redefine hashset_find
193 #error You have to redefine hashset_size
197 #ifndef hashset_iterator_init
198 #error You have to redefine hashset_iterator_init
200 #ifndef hashset_iterator_next
201 #error You have to redefine hashset_iterator_next
203 #ifndef hashset_remove_iterator
204 #error You have to redefine hashset_remove_iterator
209 * Returns the number of elements in the hashset
211 size_t hashset_size(const HashSet *self)
213 return self->num_elements - self->num_deleted;
217 * Inserts an element into a hashset without growing the set (you have to make
218 * sure there's enough room for that.
219 * @note also see comments for hashset_insert()
223 InsertReturnValue insert_nogrow(HashSet *self, KeyType key)
225 size_t num_probes = 0;
226 size_t num_buckets = self->num_buckets;
227 size_t hashmask = num_buckets - 1;
228 unsigned hash = Hash(self, key);
229 size_t bucknum = hash & hashmask;
230 size_t insert_pos = ILLEGAL_POS;
232 assert((num_buckets & (num_buckets - 1)) == 0);
235 HashSetEntry *entry = & self->entries[bucknum];
237 if(EntryIsEmpty(*entry)) {
239 HashSetEntry *nentry;
241 if(insert_pos != ILLEGAL_POS) {
247 nentry = &self->entries[p];
248 InitData(self, EntryGetValue(*nentry), key);
249 EntrySetHash(*nentry, hash);
250 self->num_elements++;
251 return GetInsertReturnValue(*nentry, 0);
253 if(EntryIsDeleted(*entry)) {
254 if(insert_pos == ILLEGAL_POS)
255 insert_pos = bucknum;
256 } else if(EntryGetHash(self, *entry) == hash) {
257 if(KeysEqual(self, GetKey(EntryGetValue(*entry)), key)) {
258 // Value already in the set, return it
259 return GetInsertReturnValue(*entry, 1);
264 bucknum = (bucknum + JUMP(num_probes)) & hashmask;
265 assert(num_probes < num_buckets);
270 * calculate shrink and enlarge limits
274 void reset_thresholds(HashSet *self)
276 self->enlarge_threshold = (size_t) HT_OCCUPANCY_FLT(self->num_buckets);
277 self->shrink_threshold = (size_t) HT_EMPTY_FLT(self->num_buckets);
278 self->consider_shrink = 0;
281 #ifndef HAVE_OWN_RESIZE
283 * Inserts an element into a hashset under the assumption that the hashset
284 * contains no deleted entries and the element doesn't exist in the hashset yet.
288 void insert_new(HashSet *self, unsigned hash, ValueType value)
290 size_t num_probes = 0;
291 size_t num_buckets = self->num_buckets;
292 size_t hashmask = num_buckets - 1;
293 size_t bucknum = hash & hashmask;
294 size_t insert_pos = ILLEGAL_POS;
296 //assert(value != NullValue);
299 HashSetEntry *entry = & self->entries[bucknum];
301 if(EntryIsEmpty(*entry)) {
303 HashSetEntry *nentry;
305 if(insert_pos != ILLEGAL_POS) {
310 nentry = &self->entries[p];
312 EntryGetValue(*nentry) = value;
313 EntrySetHash(*nentry, hash);
314 self->num_elements++;
317 assert(!EntryIsDeleted(*entry));
320 bucknum = (bucknum + JUMP(num_probes)) & hashmask;
321 assert(num_probes < num_buckets);
330 void resize(HashSet *self, size_t new_size)
332 size_t num_buckets = self->num_buckets;
334 HashSetEntry *old_entries = self->entries;
335 HashSetEntry *new_entries;
337 /* allocate a new array with double size */
338 new_entries = Alloc(new_size);
339 SetRangeEmpty(new_entries, new_size);
341 /* use the new array */
342 self->entries = new_entries;
343 self->num_buckets = new_size;
344 self->num_elements = 0;
345 self->num_deleted = 0;
347 self->entries_version++;
349 reset_thresholds(self);
351 /* reinsert all elements */
352 for(i = 0; i < num_buckets; ++i) {
353 HashSetEntry *entry = & old_entries[i];
354 if(EntryIsEmpty(*entry) || EntryIsDeleted(*entry))
357 insert_new(self, EntryGetHash(self, *entry), EntryGetValue(*entry));
360 /* now we can free the old array */
365 /* resize must be defined outside */
366 static INLINE void resize(HashSet *self, size_t new_size);
371 * grow the hashset if adding 1 more elements would make it too crowded
375 void maybe_grow(HashSet *self)
379 if(LIKELY(self->num_elements + 1 <= self->enlarge_threshold))
382 /* double table size */
383 resize_to = self->num_buckets * 2;
384 resize(self, resize_to);
388 * shrink the hashset if it is only sparsely filled
392 void maybe_shrink(HashSet *self)
397 if(!self->consider_shrink)
400 self->consider_shrink = 0;
401 size = hashset_size(self);
402 if(size <= HT_MIN_BUCKETS)
405 if(LIKELY(size > self->shrink_threshold))
408 resize_to = ceil_po2(size);
413 resize(self, resize_to);
417 * Insert an element into the hashset. If no element with the given key exists yet,
418 * then a new one is created and initialized with the InitData function.
419 * Otherwise the existing element is returned (for hashs where key is equal to
420 * value, nothing is returned.)
422 * @param self the hashset
423 * @param key the key that identifies the data
424 * @returns the existing or newly created data element (or nothing in case of hashs where keys are the while value)
426 InsertReturnValue hashset_insert(HashSet *self, KeyType key)
429 self->entries_version++;
434 return insert_nogrow(self, key);
438 * Searches for an element with key @p key.
440 * @param self the hashset
441 * @param key the key to search for
442 * @returns the found value or NullValue if nothing was found
444 InsertReturnValue hashset_find(const HashSet *self, ConstKeyType key)
446 size_t num_probes = 0;
447 size_t num_buckets = self->num_buckets;
448 size_t hashmask = num_buckets - 1;
449 unsigned hash = Hash(self, key);
450 size_t bucknum = hash & hashmask;
453 HashSetEntry *entry = & self->entries[bucknum];
455 if(EntryIsEmpty(*entry)) {
456 return NullReturnValue;
458 if(EntryIsDeleted(*entry)) {
460 } else if(EntryGetHash(self, *entry) == hash) {
461 if(KeysEqual(self, GetKey(EntryGetValue(*entry)), key)) {
463 return GetInsertReturnValue(*entry, 1);
468 bucknum = (bucknum + JUMP(num_probes)) & hashmask;
469 assert(num_probes < num_buckets);
474 * Removes an element from a hashset. Does nothing if the set doesn't contain
477 * @param self the hashset
478 * @param key key that identifies the data to remove
480 void hashset_remove(HashSet *self, ConstKeyType key)
482 size_t num_probes = 0;
483 size_t num_buckets = self->num_buckets;
484 size_t hashmask = num_buckets - 1;
485 unsigned hash = Hash(self, key);
486 size_t bucknum = hash & hashmask;
489 self->entries_version++;
493 HashSetEntry *entry = & self->entries[bucknum];
495 if(EntryIsEmpty(*entry)) {
498 if(EntryIsDeleted(*entry)) {
500 } else if(EntryGetHash(self, *entry) == hash) {
501 if(KeysEqual(self, GetKey(EntryGetValue(*entry)), key)) {
502 EntrySetDeleted(*entry);
504 self->consider_shrink = 1;
510 bucknum = (bucknum + JUMP(num_probes)) & hashmask;
511 assert(num_probes < num_buckets);
516 * Initializes hashset with a specific size
520 void init_size(HashSet *self, size_t initial_size)
525 self->entries = Alloc(initial_size);
526 SetRangeEmpty(self->entries, initial_size);
527 self->num_buckets = initial_size;
528 self->consider_shrink = 0;
529 self->num_elements = 0;
530 self->num_deleted = 0;
532 self->entries_version = 0;
534 #ifdef ADDITIONAL_INIT
538 reset_thresholds(self);
542 * Initializes a hashset with the default size. The memory for the set has to
545 void hashset_init(HashSet *self)
547 init_size(self, HT_MIN_BUCKETS);
551 * Destroys a hashset, freeing all used memory (except the memory for the
552 * HashSet struct itself).
554 void hashset_destroy(HashSet *self)
556 #ifdef ADDITIONAL_TERM
561 self->entries = NULL;
566 * Initializes a hashset expecting expected_element size.
568 void hashset_init_size(HashSet *self, size_t expected_elements)
573 if(expected_elements >= UINT_MAX/2) {
577 needed_size = expected_elements * HT_1_DIV_OCCUPANCY_FLT;
578 po2size = ceil_po2(needed_size);
579 init_size(self, po2size);
584 * Initializes a hashset iterator. The memory for the allocator has to be
586 * @note it is not allowed to remove or insert elements while iterating
588 void hashset_iterator_init(HashSetIterator *self, const HashSet *hashset)
590 self->current_bucket = hashset->entries - 1;
591 self->end = hashset->entries + hashset->num_buckets;
594 self->entries_version = hashset->entries_version;
599 * Returns the next value in the iterator or NULL if no value is left
601 * @note it is not allowed to remove or insert elements while iterating
603 ValueType hashset_iterator_next(HashSetIterator *self)
605 HashSetEntry *current_bucket = self->current_bucket;
606 HashSetEntry *end = self->end;
608 /* using hashset_insert or hashset_remove is not allowed while iterating */
609 assert(self->entries_version == self->set->entries_version);
613 if(current_bucket >= end)
615 } while(EntryIsEmpty(*current_bucket) || EntryIsDeleted(*current_bucket));
617 self->current_bucket = current_bucket;
618 return EntryGetValue(*current_bucket);
622 * Removes the element the iterator points to. Removing an element a second time
625 void hashset_remove_iterator(HashSet *self, const HashSetIterator *iter)
627 HashSetEntry *entry = iter->current_bucket;
629 /* iterator_next needs to have been called at least once */
630 assert(entry >= self->entries);
631 /* needs to be on a valid element */
632 assert(entry < self->entries + self->num_buckets);
634 if(EntryIsDeleted(*entry))
637 EntrySetDeleted(*entry);
639 self->consider_shrink = 1;
641 #endif /* NO_ITERATOR */