1881092851f5cdac1854d71a52614b259b9e7dff
[ghc.git] / rts / Hash.c
1 /*-----------------------------------------------------------------------------
2 *
3 * (c) The AQUA Project, Glasgow University, 1995-1998
4 * (c) The GHC Team, 1999
5 *
6 * Dynamically expanding linear hash tables, as described in
7 * Per-\AAke Larson, ``Dynamic Hash Tables,'' CACM 31(4), April 1988,
8 * pp. 446 -- 457.
9 * -------------------------------------------------------------------------- */
10
11 #include "PosixSource.h"
12 #include "Rts.h"
13
14 #include "Hash.h"
15 #include "RtsUtils.h"
16
17 #include <string.h>
18
19 #define HSEGSIZE 1024 /* Size of a single hash table segment */
20 /* Also the minimum size of a hash table */
21 #define HDIRSIZE 1024 /* Size of the segment directory */
22 /* Maximum hash table size is HSEGSIZE * HDIRSIZE */
23 #define HLOAD 5 /* Maximum average load of a single hash bucket */
24
25 #define HCHUNK (1024 * sizeof(W_) / sizeof(HashList))
26 /* Number of HashList cells to allocate in one go */
27
28
29 /* Linked list of (key, data) pairs for separate chaining */
30 typedef struct hashlist {
31 StgWord key;
32 void *data;
33 struct hashlist *next; /* Next cell in bucket chain (same hash value) */
34 } HashList;
35
36 typedef struct chunklist {
37 HashList *chunk;
38 struct chunklist *next;
39 } HashListChunk;
40
41 struct hashtable {
42 int split; /* Next bucket to split when expanding */
43 int max; /* Max bucket of smaller table */
44 int mask1; /* Mask for doing the mod of h_1 (smaller table) */
45 int mask2; /* Mask for doing the mod of h_2 (larger table) */
46 int kcount; /* Number of keys */
47 int bcount; /* Number of buckets */
48 HashList **dir[HDIRSIZE]; /* Directory of segments */
49 HashList *freeList; /* free list of HashLists */
50 HashListChunk *chunks;
51 HashFunction *hash; /* hash function */
52 CompareFunction *compare; /* key comparison function */
53 };
54
55 /* -----------------------------------------------------------------------------
56 * Hash first using the smaller table. If the bucket is less than the
57 * next bucket to be split, re-hash using the larger table.
58 * -------------------------------------------------------------------------- */
59
60 int
61 hashWord(HashTable *table, StgWord key)
62 {
63 int bucket;
64
65 /* Strip the boring zero bits */
66 key /= sizeof(StgWord);
67
68 /* Mod the size of the hash table (a power of 2) */
69 bucket = key & table->mask1;
70
71 if (bucket < table->split) {
72 /* Mod the size of the expanded hash table (also a power of 2) */
73 bucket = key & table->mask2;
74 }
75 return bucket;
76 }
77
78 int
79 hashStr(HashTable *table, char *key)
80 {
81 int h, bucket;
82 char *s;
83
84 s = key;
85 for (h=0; *s; s++) {
86 h *= 128;
87 h += *s;
88 h = h % 1048583; /* some random large prime */
89 }
90
91 /* Mod the size of the hash table (a power of 2) */
92 bucket = h & table->mask1;
93
94 if (bucket < table->split) {
95 /* Mod the size of the expanded hash table (also a power of 2) */
96 bucket = h & table->mask2;
97 }
98
99 return bucket;
100 }
101
102 static int
103 compareWord(StgWord key1, StgWord key2)
104 {
105 return (key1 == key2);
106 }
107
108 static int
109 compareStr(StgWord key1, StgWord key2)
110 {
111 return (strcmp((char *)key1, (char *)key2) == 0);
112 }
113
114
115 /* -----------------------------------------------------------------------------
116 * Allocate a new segment of the dynamically growing hash table.
117 * -------------------------------------------------------------------------- */
118
119 static void
120 allocSegment(HashTable *table, int segment)
121 {
122 table->dir[segment] = stgMallocBytes(HSEGSIZE * sizeof(HashList *),
123 "allocSegment");
124 }
125
126
127 /* -----------------------------------------------------------------------------
128 * Expand the larger hash table by one bucket, and split one bucket
129 * from the smaller table into two parts. Only the bucket referenced
130 * by @table->split@ is affected by the expansion.
131 * -------------------------------------------------------------------------- */
132
133 static void
134 expand(HashTable *table)
135 {
136 int oldsegment;
137 int oldindex;
138 int newbucket;
139 int newsegment;
140 int newindex;
141 HashList *hl;
142 HashList *next;
143 HashList *old, *new;
144
145 if (table->split + table->max >= HDIRSIZE * HSEGSIZE)
146 /* Wow! That's big. Too big, so don't expand. */
147 return;
148
149 /* Calculate indices of bucket to split */
150 oldsegment = table->split / HSEGSIZE;
151 oldindex = table->split % HSEGSIZE;
152
153 newbucket = table->max + table->split;
154
155 /* And the indices of the new bucket */
156 newsegment = newbucket / HSEGSIZE;
157 newindex = newbucket % HSEGSIZE;
158
159 if (newindex == 0)
160 allocSegment(table, newsegment);
161
162 if (++table->split == table->max) {
163 table->split = 0;
164 table->max *= 2;
165 table->mask1 = table->mask2;
166 table->mask2 = table->mask2 << 1 | 1;
167 }
168 table->bcount++;
169
170 /* Split the bucket, paying no attention to the original order */
171
172 old = new = NULL;
173 for (hl = table->dir[oldsegment][oldindex]; hl != NULL; hl = next) {
174 next = hl->next;
175 if (table->hash(table, hl->key) == newbucket) {
176 hl->next = new;
177 new = hl;
178 } else {
179 hl->next = old;
180 old = hl;
181 }
182 }
183 table->dir[oldsegment][oldindex] = old;
184 table->dir[newsegment][newindex] = new;
185
186 return;
187 }
188
189 void *
190 lookupHashTable(HashTable *table, StgWord key)
191 {
192 int bucket;
193 int segment;
194 int index;
195 HashList *hl;
196
197 bucket = table->hash(table, key);
198 segment = bucket / HSEGSIZE;
199 index = bucket % HSEGSIZE;
200
201 for (hl = table->dir[segment][index]; hl != NULL; hl = hl->next)
202 if (table->compare(hl->key, key))
203 return hl->data;
204
205 /* It's not there */
206 return NULL;
207 }
208
209 // Puts up to keys_sz keys of the hash table into the given array. Returns the
210 // actual amount of keys that have been retrieved.
211 //
212 // If the table is modified concurrently, the function behavior is undefined.
213 //
214 int keysHashTable(HashTable *table, StgWord keys[], int szKeys) {
215 int segment;
216 int k = 0;
217 for(segment=0;segment<HDIRSIZE && table->dir[segment];segment+=1) {
218 int index;
219 for(index=0;index<HSEGSIZE;index+=1) {
220 HashList *hl;
221 for(hl=table->dir[segment][index];hl;hl=hl->next) {
222 if (k == szKeys)
223 return k;
224 keys[k] = hl->key;
225 k += 1;
226 }
227 }
228 }
229 return k;
230 }
231
232 /* -----------------------------------------------------------------------------
233 * We allocate the hashlist cells in large chunks to cut down on malloc
234 * overhead. Although we keep a free list of hashlist cells, we make
235 * no effort to actually return the space to the malloc arena.
236 * -------------------------------------------------------------------------- */
237
238 static HashList *
239 allocHashList (HashTable *table)
240 {
241 HashList *hl, *p;
242 HashListChunk *cl;
243
244 if ((hl = table->freeList) != NULL) {
245 table->freeList = hl->next;
246 } else {
247 hl = stgMallocBytes(HCHUNK * sizeof(HashList), "allocHashList");
248 cl = stgMallocBytes(sizeof (*cl), "allocHashList: chunkList");
249 cl->chunk = hl;
250 cl->next = table->chunks;
251 table->chunks = cl;
252
253 table->freeList = hl + 1;
254 for (p = table->freeList; p < hl + HCHUNK - 1; p++)
255 p->next = p + 1;
256 p->next = NULL;
257 }
258 return hl;
259 }
260
261 static void
262 freeHashList (HashTable *table, HashList *hl)
263 {
264 hl->next = table->freeList;
265 table->freeList = hl;
266 }
267
268 void
269 insertHashTable(HashTable *table, StgWord key, void *data)
270 {
271 int bucket;
272 int segment;
273 int index;
274 HashList *hl;
275
276 // Disable this assert; sometimes it's useful to be able to
277 // overwrite entries in the hash table.
278 // ASSERT(lookupHashTable(table, key) == NULL);
279
280 /* When the average load gets too high, we expand the table */
281 if (++table->kcount >= HLOAD * table->bcount)
282 expand(table);
283
284 bucket = table->hash(table, key);
285 segment = bucket / HSEGSIZE;
286 index = bucket % HSEGSIZE;
287
288 hl = allocHashList(table);
289
290 hl->key = key;
291 hl->data = data;
292 hl->next = table->dir[segment][index];
293 table->dir[segment][index] = hl;
294
295 }
296
297 void *
298 removeHashTable(HashTable *table, StgWord key, void *data)
299 {
300 int bucket;
301 int segment;
302 int index;
303 HashList *hl;
304 HashList *prev = NULL;
305
306 bucket = table->hash(table, key);
307 segment = bucket / HSEGSIZE;
308 index = bucket % HSEGSIZE;
309
310 for (hl = table->dir[segment][index]; hl != NULL; hl = hl->next) {
311 if (table->compare(hl->key,key) && (data == NULL || hl->data == data)) {
312 if (prev == NULL)
313 table->dir[segment][index] = hl->next;
314 else
315 prev->next = hl->next;
316 freeHashList(table,hl);
317 table->kcount--;
318 return hl->data;
319 }
320 prev = hl;
321 }
322
323 /* It's not there */
324 ASSERT(data == NULL);
325 return NULL;
326 }
327
328 /* -----------------------------------------------------------------------------
329 * When we free a hash table, we are also good enough to free the
330 * data part of each (key, data) pair, as long as our caller can tell
331 * us how to do it.
332 * -------------------------------------------------------------------------- */
333
334 void
335 freeHashTable(HashTable *table, void (*freeDataFun)(void *) )
336 {
337 long segment;
338 long index;
339 HashList *hl;
340 HashList *next;
341 HashListChunk *cl, *cl_next;
342
343 /* The last bucket with something in it is table->max + table->split - 1 */
344 segment = (table->max + table->split - 1) / HSEGSIZE;
345 index = (table->max + table->split - 1) % HSEGSIZE;
346
347 while (segment >= 0) {
348 while (index >= 0) {
349 for (hl = table->dir[segment][index]; hl != NULL; hl = next) {
350 next = hl->next;
351 if (freeDataFun != NULL)
352 (*freeDataFun)(hl->data);
353 }
354 index--;
355 }
356 stgFree(table->dir[segment]);
357 segment--;
358 index = HSEGSIZE - 1;
359 }
360 for (cl = table->chunks; cl != NULL; cl = cl_next) {
361 cl_next = cl->next;
362 stgFree(cl->chunk);
363 stgFree(cl);
364 }
365 stgFree(table);
366 }
367
368 /* -----------------------------------------------------------------------------
369 * When we initialize a hash table, we set up the first segment as well,
370 * initializing all of the first segment's hash buckets to NULL.
371 * -------------------------------------------------------------------------- */
372
373 HashTable *
374 allocHashTable_(HashFunction *hash, CompareFunction *compare)
375 {
376 HashTable *table;
377 HashList **hb;
378
379 table = stgMallocBytes(sizeof(HashTable),"allocHashTable");
380
381 allocSegment(table, 0);
382
383 for (hb = table->dir[0]; hb < table->dir[0] + HSEGSIZE; hb++)
384 *hb = NULL;
385
386 table->split = 0;
387 table->max = HSEGSIZE;
388 table->mask1 = HSEGSIZE - 1;
389 table->mask2 = 2 * HSEGSIZE - 1;
390 table->kcount = 0;
391 table->bcount = HSEGSIZE;
392 table->freeList = NULL;
393 table->chunks = NULL;
394 table->hash = hash;
395 table->compare = compare;
396
397 return table;
398 }
399
400 HashTable *
401 allocHashTable(void)
402 {
403 return allocHashTable_(hashWord, compareWord);
404 }
405
406 HashTable *
407 allocStrHashTable(void)
408 {
409 return allocHashTable_((HashFunction *)hashStr,
410 (CompareFunction *)compareStr);
411 }
412
413 void
414 exitHashTable(void)
415 {
416 /* nothing to do */
417 }
418
419 int keyCountHashTable (HashTable *table)
420 {
421 return table->kcount;
422 }