3 * Hashtable implementation
6 * Miguel de Icaza (miguel@novell.com)
8 * (C) 2006 Novell, Inc.
10 * Permission is hereby granted, free of charge, to any person obtaining
11 * a copy of this software and associated documentation files (the
12 * "Software"), to deal in the Software without restriction, including
13 * without limitation the rights to use, copy, modify, merge, publish,
14 * distribute, sublicense, and/or sell copies of the Software, and to
15 * permit persons to whom the Software is furnished to do so, subject to
16 * the following conditions:
18 * The above copyright notice and this permission notice shall be
19 * included in all copies or substantial portions of the Software.
21 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
22 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
23 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
24 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
25 * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
26 * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
27 * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
33 #include "mono-hash.h"
34 #include "metadata/gc-internals.h"
35 #include <mono/utils/checked-build.h>
36 #include <mono/utils/mono-threads-coop.h>
38 int mono_g_hash_table_max_chain_length;
41 #define mg_new0(type,n) ((type *) GC_MALLOC(sizeof(type) * (n)))
42 #define mg_new(type,n) ((type *) GC_MALLOC(sizeof(type) * (n)))
43 #define mg_free(x) do { } while (0)
45 #define mg_new0(x,n) g_new0(x,n)
46 #define mg_new(type,n) g_new(type,n)
47 #define mg_free(x) g_free(x)
50 struct _MonoGHashTable {
52 GEqualFunc key_equal_func;
58 GDestroyNotify value_destroy_func, key_destroy_func;
59 MonoGHashGCType gc_type;
60 MonoGCRootSource source;
70 for (n = 3; n< (int)sqrt (x); n += 2) {
76 // There is only one even prime - 2.
85 for (i = (x & (~1))-1; i< G_MAXINT32; i += 2) {
93 #define HASH_TABLE_MAX_LOAD_FACTOR 0.7f
94 /* We didn't really do compaction before, keep it lenient for now */
95 #define HASH_TABLE_MIN_LOAD_FACTOR 0.05f
96 /* We triple the table size at rehash time, similar with previous implementation */
97 #define HASH_TABLE_RESIZE_RATIO 3
99 static inline void mono_g_hash_table_key_store (MonoGHashTable *hash, int slot, MonoObject* key)
101 MonoObject **key_addr = &hash->keys [slot];
102 if (hash->gc_type & MONO_HASH_KEY_GC)
103 mono_gc_wbarrier_generic_store (key_addr, key);
108 static inline void mono_g_hash_table_value_store (MonoGHashTable *hash, int slot, MonoObject* value)
110 MonoObject **value_addr = &hash->values [slot];
111 if (hash->gc_type & MONO_HASH_VALUE_GC)
112 mono_gc_wbarrier_generic_store (value_addr, value);
117 /* Returns position of key or of an empty slot for it */
118 static inline int mono_g_hash_table_find_slot (MonoGHashTable *hash, const MonoObject *key)
120 guint start = ((*hash->hash_func) (key)) % hash->table_size;
123 if (hash->key_equal_func) {
124 GEqualFunc equal = hash->key_equal_func;
126 while (hash->keys [i] && !(*equal) (hash->keys [i], key)) {
128 if (i == hash->table_size)
132 while (hash->keys [i] && hash->keys [i] != key) {
134 if (i == hash->table_size)
139 if (i > start && (i - start) > mono_g_hash_table_max_chain_length)
140 mono_g_hash_table_max_chain_length = i - start;
141 else if (i < start && (hash->table_size - (start - i)) > mono_g_hash_table_max_chain_length)
142 mono_g_hash_table_max_chain_length = hash->table_size - (start - i);
148 mono_g_hash_table_new_type (GHashFunc hash_func, GEqualFunc key_equal_func, MonoGHashGCType type, MonoGCRootSource source, const char *msg)
150 MonoGHashTable *hash;
153 hash_func = g_direct_hash;
156 hash = mg_new0 (MonoGHashTable, 1);
158 hash = mono_gc_alloc_fixed (sizeof (MonoGHashTable), MONO_GC_ROOT_DESCR_FOR_FIXED (sizeof (MonoGHashTable)), source, msg);
161 hash->hash_func = hash_func;
162 hash->key_equal_func = key_equal_func;
164 hash->table_size = g_spaced_primes_closest (1);
165 hash->keys = mg_new0 (MonoObject*, hash->table_size);
166 hash->values = mg_new0 (MonoObject*, hash->table_size);
168 hash->gc_type = type;
169 hash->source = source;
172 if (type > MONO_HASH_KEY_VALUE_GC)
173 g_error ("wrong type for gc hashtable");
176 if (hash->gc_type & MONO_HASH_KEY_GC)
177 mono_gc_register_root_wbarrier ((char*)hash->keys, sizeof (MonoObject*) * hash->table_size, mono_gc_make_vector_descr (), hash->source, hash->msg);
178 if (hash->gc_type & MONO_HASH_VALUE_GC)
179 mono_gc_register_root_wbarrier ((char*)hash->values, sizeof (MonoObject*) * hash->table_size, mono_gc_make_vector_descr (), hash->source, hash->msg);
186 MonoGHashTable *hash;
193 do_rehash (void *_data)
195 RehashData *data = (RehashData *)_data;
196 MonoGHashTable *hash = data->hash;
198 MonoObject **old_keys;
199 MonoObject **old_values;
201 current_size = hash->table_size;
202 hash->table_size = data->new_size;
203 old_keys = hash->keys;
204 old_values = hash->values;
205 hash->keys = data->keys;
206 hash->values = data->values;
208 for (i = 0; i < current_size; i++) {
210 int slot = mono_g_hash_table_find_slot (hash, old_keys [i]);
211 mono_g_hash_table_key_store (hash, slot, old_keys [i]);
212 mono_g_hash_table_value_store (hash, slot, old_values [i]);
219 rehash (MonoGHashTable *hash)
221 MONO_REQ_GC_UNSAFE_MODE; //we must run in unsafe mode to make rehash safe
224 void *old_keys G_GNUC_UNUSED = hash->keys; /* unused on Boehm */
225 void *old_values G_GNUC_UNUSED = hash->values; /* unused on Boehm */
229 * Rehash to a size that can fit the current elements. Rehash relative to in_use
230 * to allow also for compaction.
232 data.new_size = g_spaced_primes_closest (hash->in_use / HASH_TABLE_MAX_LOAD_FACTOR * HASH_TABLE_RESIZE_RATIO);
233 data.keys = mg_new0 (MonoObject*, data.new_size);
234 data.values = mg_new0 (MonoObject*, data.new_size);
237 if (hash->gc_type & MONO_HASH_KEY_GC)
238 mono_gc_register_root_wbarrier ((char*)data.keys, sizeof (MonoObject*) * data.new_size, mono_gc_make_vector_descr (), hash->source, hash->msg);
239 if (hash->gc_type & MONO_HASH_VALUE_GC)
240 mono_gc_register_root_wbarrier ((char*)data.values, sizeof (MonoObject*) * data.new_size, mono_gc_make_vector_descr (), hash->source, hash->msg);
243 if (!mono_threads_is_coop_enabled ()) {
244 mono_gc_invoke_with_gc_lock (do_rehash, &data);
246 /* We cannot be preempted */
251 if (hash->gc_type & MONO_HASH_KEY_GC)
252 mono_gc_deregister_root ((char*)old_keys);
253 if (hash->gc_type & MONO_HASH_VALUE_GC)
254 mono_gc_deregister_root ((char*)old_values);
257 mg_free (old_values);
261 mono_g_hash_table_size (MonoGHashTable *hash)
263 g_return_val_if_fail (hash != NULL, 0);
269 mono_g_hash_table_lookup (MonoGHashTable *hash, gconstpointer key)
271 gpointer orig_key, value;
273 if (mono_g_hash_table_lookup_extended (hash, key, &orig_key, &value))
280 mono_g_hash_table_lookup_extended (MonoGHashTable *hash, gconstpointer key, gpointer *orig_key, gpointer *value)
284 g_return_val_if_fail (hash != NULL, FALSE);
286 slot = mono_g_hash_table_find_slot (hash, key);
288 if (hash->keys [slot]) {
289 *orig_key = hash->keys [slot];
290 *value = hash->values [slot];
298 mono_g_hash_table_foreach (MonoGHashTable *hash, GHFunc func, gpointer user_data)
302 g_return_if_fail (hash != NULL);
303 g_return_if_fail (func != NULL);
305 for (i = 0; i < hash->table_size; i++) {
307 (*func)(hash->keys [i], hash->values [i], user_data);
312 mono_g_hash_table_find (MonoGHashTable *hash, GHRFunc predicate, gpointer user_data)
316 g_return_val_if_fail (hash != NULL, NULL);
317 g_return_val_if_fail (predicate != NULL, NULL);
319 for (i = 0; i < hash->table_size; i++) {
320 if (hash->keys [i] && (*predicate)(hash->keys [i], hash->values [i], user_data))
321 return hash->values [i];
327 mono_g_hash_table_remove (MonoGHashTable *hash, gconstpointer key)
329 int slot, last_clear_slot;
331 g_return_val_if_fail (hash != NULL, FALSE);
332 slot = mono_g_hash_table_find_slot (hash, key);
334 if (!hash->keys [slot])
337 if (hash->key_destroy_func)
338 (*hash->key_destroy_func)(hash->keys [slot]);
339 hash->keys [slot] = NULL;
340 if (hash->value_destroy_func)
341 (*hash->value_destroy_func)(hash->values [slot]);
342 hash->values [slot] = NULL;
346 * When we insert in the hashtable, if the required position is occupied we
347 * consecutively try out following positions. In order to be able to find
348 * if a key exists or not in the array (without traversing the entire hash)
349 * we maintain the constraint that there can be no free slots between two
350 * entries that are hashed to the same position. This means that, at search
351 * time, when we encounter a free slot we can stop looking for collissions.
352 * Similarly, at remove time, we need to shift all following slots to their
353 * normal slot, until we reach an empty slot.
355 last_clear_slot = slot;
356 slot = (slot + 1) % hash->table_size;
357 while (hash->keys [slot]) {
358 guint hashcode = ((*hash->hash_func)(hash->keys [slot])) % hash->table_size;
360 * We try to move the current element to last_clear_slot, but only if
361 * it brings it closer to its normal position (hashcode)
363 if ((last_clear_slot < slot && (hashcode > slot || hashcode <= last_clear_slot)) ||
364 (last_clear_slot > slot && (hashcode > slot && hashcode <= last_clear_slot))) {
365 mono_g_hash_table_key_store (hash, last_clear_slot, hash->keys [slot]);
366 mono_g_hash_table_value_store (hash, last_clear_slot, hash->values [slot]);
367 hash->keys [slot] = NULL;
368 hash->values [slot] = NULL;
369 last_clear_slot = slot;
372 if (slot == hash->table_size)
379 mono_g_hash_table_foreach_remove (MonoGHashTable *hash, GHRFunc func, gpointer user_data)
384 g_return_val_if_fail (hash != NULL, 0);
385 g_return_val_if_fail (func != NULL, 0);
387 for (i = 0; i < hash->table_size; i++) {
388 if (hash->keys [i] && (*func)(hash->keys [i], hash->values [i], user_data)) {
389 mono_g_hash_table_remove (hash, hash->keys [i]);
391 /* Retry current slot in case the removal shifted elements */
395 if (hash->in_use < hash->table_size * HASH_TABLE_MIN_LOAD_FACTOR)
401 mono_g_hash_table_destroy (MonoGHashTable *hash)
405 g_return_if_fail (hash != NULL);
408 if (hash->gc_type & MONO_HASH_KEY_GC)
409 mono_gc_deregister_root ((char*)hash->keys);
410 if (hash->gc_type & MONO_HASH_VALUE_GC)
411 mono_gc_deregister_root ((char*)hash->values);
414 for (i = 0; i < hash->table_size; i++) {
415 if (hash->keys [i]) {
416 if (hash->key_destroy_func)
417 (*hash->key_destroy_func)(hash->keys [i]);
418 if (hash->value_destroy_func)
419 (*hash->value_destroy_func)(hash->values [i]);
422 mg_free (hash->keys);
423 mg_free (hash->values);
427 mono_gc_free_fixed (hash);
432 mono_g_hash_table_insert_replace (MonoGHashTable *hash, gpointer key, gpointer value, gboolean replace)
435 g_return_if_fail (hash != NULL);
437 if (hash->in_use > (hash->table_size * HASH_TABLE_MAX_LOAD_FACTOR))
440 slot = mono_g_hash_table_find_slot (hash, key);
442 if (hash->keys [slot]) {
444 if (hash->key_destroy_func)
445 (*hash->key_destroy_func)(hash->keys [slot]);
446 mono_g_hash_table_key_store (hash, slot, (MonoObject*)key);
448 if (hash->value_destroy_func)
449 (*hash->value_destroy_func) (hash->values [slot]);
450 mono_g_hash_table_value_store (hash, slot, (MonoObject*)value);
452 mono_g_hash_table_key_store (hash, slot, (MonoObject*)key);
453 mono_g_hash_table_value_store (hash, slot, (MonoObject*)value);
459 mono_g_hash_table_insert (MonoGHashTable *h, gpointer k, gpointer v)
461 mono_g_hash_table_insert_replace (h, k, v, FALSE);
465 mono_g_hash_table_replace(MonoGHashTable *h, gpointer k, gpointer v)
467 mono_g_hash_table_insert_replace (h, k, v, TRUE);
471 mono_g_hash_table_print_stats (MonoGHashTable *hash)
473 int i = 0, chain_size = 0, max_chain_size = 0;
474 gboolean wrapped_around = FALSE;
477 if (hash->keys [i]) {
480 max_chain_size = MAX(max_chain_size, chain_size);
486 if (i == (hash->table_size - 1)) {
487 wrapped_around = TRUE;
493 /* Rehash to a size that can fit the current elements */
494 printf ("Size: %d Table Size: %d Max Chain Length: %d\n", hash->in_use, hash->table_size, max_chain_size);