Hash.c 9.94 KB
Newer Older
1 2 3 4 5 6 7 8 9 10
/*-----------------------------------------------------------------------------
 *
 * (c) The AQUA Project, Glasgow University, 1995-1998
 * (c) The GHC Team, 1999
 *
 * Dynamically expanding linear hash tables, as described in
 * Per-\AAke Larson, ``Dynamic Hash Tables,'' CACM 31(4), April 1988,
 * pp. 446 -- 457.
 * -------------------------------------------------------------------------- */

11
#include "PosixSource.h"
12 13 14 15
#include "Rts.h"
#include "Hash.h"
#include "RtsUtils.h"

16 17 18
#include <stdlib.h>
#include <string.h>

19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37
#define HSEGSIZE    1024    /* Size of a single hash table segment */
			    /* Also the minimum size of a hash table */
#define HDIRSIZE    1024    /* Size of the segment directory */
			    /* Maximum hash table size is HSEGSIZE * HDIRSIZE */
#define HLOAD	    5	    /* Maximum average load of a single hash bucket */

#define HCHUNK	    (1024 * sizeof(W_) / sizeof(HashList))
			    /* Number of HashList cells to allocate in one go */


/* Linked list of (key, data) pairs for separate chaining */
struct hashlist {
    StgWord key;
    void *data;
    struct hashlist *next;  /* Next cell in bucket chain (same hash value) */
};

typedef struct hashlist HashList;

38 39 40
typedef int HashFunction(HashTable *table, StgWord key);
typedef int CompareFunction(StgWord key1, StgWord key2);

41 42 43 44 45 46 47 48
struct hashtable {
    int split;		    /* Next bucket to split when expanding */
    int max;		    /* Max bucket of smaller table */
    int mask1;		    /* Mask for doing the mod of h_1 (smaller table) */
    int mask2;		    /* Mask for doing the mod of h_2 (larger table) */
    int kcount;		    /* Number of keys */
    int bcount;		    /* Number of buckets */
    HashList **dir[HDIRSIZE];	/* Directory of segments */
49 50
    HashFunction *hash;		/* hash function */
    CompareFunction *compare;   /* key comparison function */
51 52 53 54 55 56 57 58
};

/* -----------------------------------------------------------------------------
 * Hash first using the smaller table.  If the bucket is less than the
 * next bucket to be split, re-hash using the larger table.
 * -------------------------------------------------------------------------- */

static int
59
hashWord(HashTable *table, StgWord key)
60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75
{
    int bucket;

    /* Strip the boring zero bits */
    key /= sizeof(StgWord);

    /* Mod the size of the hash table (a power of 2) */
    bucket = key & table->mask1;

    if (bucket < table->split) {
	/* Mod the size of the expanded hash table (also a power of 2) */
	bucket = key & table->mask2;
    }
    return bucket;
}

76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112
static int
hashStr(HashTable *table, char *key)
{
    int h, bucket;
    char *s;

    s = key;
    for (h=0; *s; s++) {
	h *= 128;
	h += *s;
	h = h % 1048583;	/* some random large prime */
    }

    /* Mod the size of the hash table (a power of 2) */
    bucket = h & table->mask1;

    if (bucket < table->split) {
	/* Mod the size of the expanded hash table (also a power of 2) */
	bucket = h & table->mask2;
    }

    return bucket;
}

static int
compareWord(StgWord key1, StgWord key2)
{
    return (key1 == key2);
}

static int
compareStr(StgWord key1, StgWord key2)
{
    return (strcmp((char *)key1, (char *)key2) == 0);
}


113 114 115 116 117 118 119 120 121 122 123
/* -----------------------------------------------------------------------------
 * Allocate a new segment of the dynamically growing hash table.
 * -------------------------------------------------------------------------- */

static void
allocSegment(HashTable *table, int segment)
{
    table->dir[segment] = stgMallocBytes(HSEGSIZE * sizeof(HashList *), 
					 "allocSegment");
}

124

125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172
/* -----------------------------------------------------------------------------
 * Expand the larger hash table by one bucket, and split one bucket
 * from the smaller table into two parts.  Only the bucket referenced
 * by @table->split@ is affected by the expansion.
 * -------------------------------------------------------------------------- */

static void
expand(HashTable *table)
{
    int oldsegment;
    int oldindex;
    int newbucket;
    int newsegment;
    int newindex;
    HashList *hl;
    HashList *next;
    HashList *old, *new;

    if (table->split + table->max >= HDIRSIZE * HSEGSIZE)
	/* Wow!  That's big.  Too big, so don't expand. */
	return;

    /* Calculate indices of bucket to split */
    oldsegment = table->split / HSEGSIZE;
    oldindex = table->split % HSEGSIZE;

    newbucket = table->max + table->split;

    /* And the indices of the new bucket */
    newsegment = newbucket / HSEGSIZE;
    newindex = newbucket % HSEGSIZE;

    if (newindex == 0)
	allocSegment(table, newsegment);

    if (++table->split == table->max) {
	table->split = 0;
	table->max *= 2;
	table->mask1 = table->mask2;
	table->mask2 = table->mask2 << 1 | 1;
    }
    table->bcount++;

    /* Split the bucket, paying no attention to the original order */

    old = new = NULL;
    for (hl = table->dir[oldsegment][oldindex]; hl != NULL; hl = next) {
	next = hl->next;
173
	if (table->hash(table, hl->key) == newbucket) {
174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194
	    hl->next = new;
	    new = hl;
	} else {
	    hl->next = old;
	    old = hl;
	}
    }
    table->dir[oldsegment][oldindex] = old;
    table->dir[newsegment][newindex] = new;

    return;
}

void *
lookupHashTable(HashTable *table, StgWord key)
{
    int bucket;
    int segment;
    int index;
    HashList *hl;

195
    bucket = table->hash(table, key);
196 197 198 199
    segment = bucket / HSEGSIZE;
    index = bucket % HSEGSIZE;

    for (hl = table->dir[segment][index]; hl != NULL; hl = hl->next)
200
	if (table->compare(hl->key, key))
201 202 203 204 205 206
	    return hl->data;

    /* It's not there */
    return NULL;
}

207 208 209 210 211 212 213 214
/* -----------------------------------------------------------------------------
 * We allocate the hashlist cells in large chunks to cut down on malloc
 * overhead.  Although we keep a free list of hashlist cells, we make
 * no effort to actually return the space to the malloc arena.
 * -------------------------------------------------------------------------- */

static HashList *freeList = NULL;

215 216 217 218 219
static struct chunkList {
  void *chunk;
  struct chunkList *next;
} *chunks;

220 221 222 223
static HashList *
allocHashList(void)
{
    HashList *hl, *p;
224
    struct chunkList *cl;
225 226 227 228 229

    if ((hl = freeList) != NULL) {
	freeList = hl->next;
    } else {
        hl = stgMallocBytes(HCHUNK * sizeof(HashList), "allocHashList");
230 231 232 233
	cl = stgMallocBytes(sizeof (*cl), "allocHashList: chunkList");
	cl->chunk = hl;
	cl->next = chunks;
	chunks = cl;
234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249

	freeList = hl + 1;
	for (p = freeList; p < hl + HCHUNK - 1; p++)
	    p->next = p + 1;
	p->next = NULL;
    }
    return hl;
}

static void
freeHashList(HashList *hl)
{
    hl->next = freeList;
    freeList = hl;
}

250 251 252 253 254 255 256 257
void
insertHashTable(HashTable *table, StgWord key, void *data)
{
    int bucket;
    int segment;
    int index;
    HashList *hl;

258 259 260
    // Disable this assert; sometimes it's useful to be able to
    // overwrite entries in the hash table.
    // ASSERT(lookupHashTable(table, key) == NULL);
261

262 263 264 265
    /* When the average load gets too high, we expand the table */
    if (++table->kcount >= HLOAD * table->bcount)
	expand(table);

266
    bucket = table->hash(table, key);
267 268 269
    segment = bucket / HSEGSIZE;
    index = bucket % HSEGSIZE;

270
    hl = allocHashList();
271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287

    hl->key = key;
    hl->data = data;
    hl->next = table->dir[segment][index];
    table->dir[segment][index] = hl;

}

void *
removeHashTable(HashTable *table, StgWord key, void *data)
{
    int bucket;
    int segment;
    int index;
    HashList *hl;
    HashList *prev = NULL;

288
    bucket = table->hash(table, key);
289 290 291 292
    segment = bucket / HSEGSIZE;
    index = bucket % HSEGSIZE;

    for (hl = table->dir[segment][index]; hl != NULL; hl = hl->next) {
293
	if (table->compare(hl->key,key) && (data == NULL || hl->data == data)) {
294 295 296 297
	    if (prev == NULL)
		table->dir[segment][index] = hl->next;
	    else
		prev->next = hl->next;
298
	    freeHashList(hl);
299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316
	    table->kcount--;
	    return hl->data;
	}
	prev = hl;
    }

    /* It's not there */
    ASSERT(data == NULL);
    return NULL;
}

/* -----------------------------------------------------------------------------
 * When we free a hash table, we are also good enough to free the
 * data part of each (key, data) pair, as long as our caller can tell
 * us how to do it.
 * -------------------------------------------------------------------------- */

void
317
freeHashTable(HashTable *table, void (*freeDataFun)(void *) )
318 319 320 321 322 323 324 325 326
{
    long segment;
    long index;
    HashList *hl;
    HashList *next;

    /* The last bucket with something in it is table->max + table->split - 1 */
    segment = (table->max + table->split - 1) / HSEGSIZE;
    index = (table->max + table->split - 1) % HSEGSIZE;
327

328
    while (segment >= 0) {
329 330 331 332
	while (index >= 0) {
	    for (hl = table->dir[segment][index]; hl != NULL; hl = next) {
		next = hl->next;
		if (freeDataFun != NULL)
333
		    (*freeDataFun)(hl->data);
334
		freeHashList(hl);
335
	    }
336
	    index--;
337
	}
sof's avatar
sof committed
338
	stgFree(table->dir[segment]);
339 340 341
	segment--;
	index = HSEGSIZE - 1;
    }
sof's avatar
sof committed
342
    stgFree(table);
343 344 345 346 347 348 349
}

/* -----------------------------------------------------------------------------
 * When we initialize a hash table, we set up the first segment as well,
 * initializing all of the first segment's hash buckets to NULL.
 * -------------------------------------------------------------------------- */

350 351
static HashTable *
allocHashTable_(HashFunction *hash, CompareFunction *compare)
352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368
{
    HashTable *table;
    HashList **hb;

    table = stgMallocBytes(sizeof(HashTable),"allocHashTable");

    allocSegment(table, 0);

    for (hb = table->dir[0]; hb < table->dir[0] + HSEGSIZE; hb++)
	*hb = NULL;

    table->split = 0;
    table->max = HSEGSIZE;
    table->mask1 = HSEGSIZE - 1;
    table->mask2 = 2 * HSEGSIZE - 1;
    table->kcount = 0;
    table->bcount = HSEGSIZE;
369 370
    table->hash = hash;
    table->compare = compare;
371 372 373

    return table;
}
374 375 376 377 378 379 380 381 382 383 384 385 386

HashTable *
allocHashTable(void)
{
    return allocHashTable_(hashWord, compareWord);
}

HashTable *
allocStrHashTable(void)
{
    return allocHashTable_((HashFunction *)hashStr, 
			   (CompareFunction *)compareStr);
}
387 388 389 390 391 392 393 394 395 396 397 398

void
exitHashTable(void)
{
  struct chunkList *cl;

  while ((cl = chunks) != NULL) {
    chunks = cl->next;
    stgFree(cl->chunk);
    stgFree(cl);
  }
}