1 /*
2 * Copyright 2006 The Android Open Source Project
3 *
4 * Hash table. The dominant calls are add and lookup, with removals
5 * happening very infrequently. We use probing, and don't worry much
6 * about tombstone removal.
7 */
8 #include <stdlib.h>
9 #include <assert.h>
10
11 #define LOG_TAG "minzip"
12 #include "Log.h"
13 #include "Hash.h"
14
15 /* table load factor, i.e. how full can it get before we resize */
16 //#define LOAD_NUMER 3 // 75%
17 //#define LOAD_DENOM 4
18 #define LOAD_NUMER 5 // 62.5%
19 #define LOAD_DENOM 8
20 //#define LOAD_NUMER 1 // 50%
21 //#define LOAD_DENOM 2
22
23 /*
24 * Compute the capacity needed for a table to hold "size" elements.
25 */
mzHashSize(size_t size)26 size_t mzHashSize(size_t size) {
27 return (size * LOAD_DENOM) / LOAD_NUMER +1;
28 }
29
30 /*
31 * Round up to the next highest power of 2.
32 *
33 * Found on http://graphics.stanford.edu/~seander/bithacks.html.
34 */
roundUpPower2(unsigned int val)35 unsigned int roundUpPower2(unsigned int val)
36 {
37 val--;
38 val |= val >> 1;
39 val |= val >> 2;
40 val |= val >> 4;
41 val |= val >> 8;
42 val |= val >> 16;
43 val++;
44
45 return val;
46 }
47
48 /*
49 * Create and initialize a hash table.
50 */
mzHashTableCreate(size_t initialSize,HashFreeFunc freeFunc)51 HashTable* mzHashTableCreate(size_t initialSize, HashFreeFunc freeFunc)
52 {
53 HashTable* pHashTable;
54
55 assert(initialSize > 0);
56
57 pHashTable = (HashTable*) malloc(sizeof(*pHashTable));
58 if (pHashTable == NULL)
59 return NULL;
60
61 pHashTable->tableSize = roundUpPower2(initialSize);
62 pHashTable->numEntries = pHashTable->numDeadEntries = 0;
63 pHashTable->freeFunc = freeFunc;
64 pHashTable->pEntries =
65 (HashEntry*) calloc((size_t)pHashTable->tableSize, sizeof(HashTable));
66 if (pHashTable->pEntries == NULL) {
67 free(pHashTable);
68 return NULL;
69 }
70
71 return pHashTable;
72 }
73
74 /*
75 * Clear out all entries.
76 */
mzHashTableClear(HashTable * pHashTable)77 void mzHashTableClear(HashTable* pHashTable)
78 {
79 HashEntry* pEnt;
80 int i;
81
82 pEnt = pHashTable->pEntries;
83 for (i = 0; i < pHashTable->tableSize; i++, pEnt++) {
84 if (pEnt->data == HASH_TOMBSTONE) {
85 // nuke entry
86 pEnt->data = NULL;
87 } else if (pEnt->data != NULL) {
88 // call free func then nuke entry
89 if (pHashTable->freeFunc != NULL)
90 (*pHashTable->freeFunc)(pEnt->data);
91 pEnt->data = NULL;
92 }
93 }
94
95 pHashTable->numEntries = 0;
96 pHashTable->numDeadEntries = 0;
97 }
98
99 /*
100 * Free the table.
101 */
mzHashTableFree(HashTable * pHashTable)102 void mzHashTableFree(HashTable* pHashTable)
103 {
104 if (pHashTable == NULL)
105 return;
106 mzHashTableClear(pHashTable);
107 free(pHashTable->pEntries);
108 free(pHashTable);
109 }
110
111 #ifndef NDEBUG
112 /*
113 * Count up the number of tombstone entries in the hash table.
114 */
countTombStones(HashTable * pHashTable)115 static int countTombStones(HashTable* pHashTable)
116 {
117 int i, count;
118
119 for (count = i = 0; i < pHashTable->tableSize; i++) {
120 if (pHashTable->pEntries[i].data == HASH_TOMBSTONE)
121 count++;
122 }
123 return count;
124 }
125 #endif
126
127 /*
128 * Resize a hash table. We do this when adding an entry increased the
129 * size of the table beyond its comfy limit.
130 *
131 * This essentially requires re-inserting all elements into the new storage.
132 *
133 * If multiple threads can access the hash table, the table's lock should
134 * have been grabbed before issuing the "lookup+add" call that led to the
135 * resize, so we don't have a synchronization problem here.
136 */
resizeHash(HashTable * pHashTable,int newSize)137 static bool resizeHash(HashTable* pHashTable, int newSize)
138 {
139 HashEntry* pNewEntries;
140 int i;
141
142 assert(countTombStones(pHashTable) == pHashTable->numDeadEntries);
143
144 pNewEntries = (HashEntry*) calloc(newSize, sizeof(HashTable));
145 if (pNewEntries == NULL)
146 return false;
147
148 for (i = 0; i < pHashTable->tableSize; i++) {
149 void* data = pHashTable->pEntries[i].data;
150 if (data != NULL && data != HASH_TOMBSTONE) {
151 int hashValue = pHashTable->pEntries[i].hashValue;
152 int newIdx;
153
154 /* probe for new spot, wrapping around */
155 newIdx = hashValue & (newSize-1);
156 while (pNewEntries[newIdx].data != NULL)
157 newIdx = (newIdx + 1) & (newSize-1);
158
159 pNewEntries[newIdx].hashValue = hashValue;
160 pNewEntries[newIdx].data = data;
161 }
162 }
163
164 free(pHashTable->pEntries);
165 pHashTable->pEntries = pNewEntries;
166 pHashTable->tableSize = newSize;
167 pHashTable->numDeadEntries = 0;
168
169 assert(countTombStones(pHashTable) == 0);
170 return true;
171 }
172
173 /*
174 * Look up an entry.
175 *
176 * We probe on collisions, wrapping around the table.
177 */
mzHashTableLookup(HashTable * pHashTable,unsigned int itemHash,void * item,HashCompareFunc cmpFunc,bool doAdd)178 void* mzHashTableLookup(HashTable* pHashTable, unsigned int itemHash, void* item,
179 HashCompareFunc cmpFunc, bool doAdd)
180 {
181 HashEntry* pEntry;
182 HashEntry* pEnd;
183 void* result = NULL;
184
185 assert(pHashTable->tableSize > 0);
186 assert(item != HASH_TOMBSTONE);
187 assert(item != NULL);
188
189 /* jump to the first entry and probe for a match */
190 pEntry = &pHashTable->pEntries[itemHash & (pHashTable->tableSize-1)];
191 pEnd = &pHashTable->pEntries[pHashTable->tableSize];
192 while (pEntry->data != NULL) {
193 if (pEntry->data != HASH_TOMBSTONE &&
194 pEntry->hashValue == itemHash &&
195 (*cmpFunc)(pEntry->data, item) == 0)
196 {
197 /* match */
198 break;
199 }
200
201 pEntry++;
202 if (pEntry == pEnd) { /* wrap around to start */
203 if (pHashTable->tableSize == 1)
204 break; /* edge case - single-entry table */
205 pEntry = pHashTable->pEntries;
206 }
207 }
208
209 if (pEntry->data == NULL) {
210 if (doAdd) {
211 pEntry->hashValue = itemHash;
212 pEntry->data = item;
213 pHashTable->numEntries++;
214
215 /*
216 * We've added an entry. See if this brings us too close to full.
217 */
218 if ((pHashTable->numEntries+pHashTable->numDeadEntries) * LOAD_DENOM
219 > pHashTable->tableSize * LOAD_NUMER)
220 {
221 if (!resizeHash(pHashTable, pHashTable->tableSize * 2)) {
222 /* don't really have a way to indicate failure */
223 LOGE("Dalvik hash resize failure\n");
224 abort();
225 }
226 /* note "pEntry" is now invalid */
227 }
228
229 /* full table is bad -- search for nonexistent never halts */
230 assert(pHashTable->numEntries < pHashTable->tableSize);
231 result = item;
232 } else {
233 assert(result == NULL);
234 }
235 } else {
236 result = pEntry->data;
237 }
238
239 return result;
240 }
241
242 /*
243 * Remove an entry from the table.
244 *
245 * Does NOT invoke the "free" function on the item.
246 */
mzHashTableRemove(HashTable * pHashTable,unsigned int itemHash,void * item)247 bool mzHashTableRemove(HashTable* pHashTable, unsigned int itemHash, void* item)
248 {
249 HashEntry* pEntry;
250 HashEntry* pEnd;
251
252 assert(pHashTable->tableSize > 0);
253
254 /* jump to the first entry and probe for a match */
255 pEntry = &pHashTable->pEntries[itemHash & (pHashTable->tableSize-1)];
256 pEnd = &pHashTable->pEntries[pHashTable->tableSize];
257 while (pEntry->data != NULL) {
258 if (pEntry->data == item) {
259 pEntry->data = HASH_TOMBSTONE;
260 pHashTable->numEntries--;
261 pHashTable->numDeadEntries++;
262 return true;
263 }
264
265 pEntry++;
266 if (pEntry == pEnd) { /* wrap around to start */
267 if (pHashTable->tableSize == 1)
268 break; /* edge case - single-entry table */
269 pEntry = pHashTable->pEntries;
270 }
271 }
272
273 return false;
274 }
275
276 /*
277 * Execute a function on every entry in the hash table.
278 *
279 * If "func" returns a nonzero value, terminate early and return the value.
280 */
mzHashForeach(HashTable * pHashTable,HashForeachFunc func,void * arg)281 int mzHashForeach(HashTable* pHashTable, HashForeachFunc func, void* arg)
282 {
283 int i, val;
284
285 for (i = 0; i < pHashTable->tableSize; i++) {
286 HashEntry* pEnt = &pHashTable->pEntries[i];
287
288 if (pEnt->data != NULL && pEnt->data != HASH_TOMBSTONE) {
289 val = (*func)(pEnt->data, arg);
290 if (val != 0)
291 return val;
292 }
293 }
294
295 return 0;
296 }
297
298
299 /*
300 * Look up an entry, counting the number of times we have to probe.
301 *
302 * Returns -1 if the entry wasn't found.
303 */
countProbes(HashTable * pHashTable,unsigned int itemHash,const void * item,HashCompareFunc cmpFunc)304 int countProbes(HashTable* pHashTable, unsigned int itemHash, const void* item,
305 HashCompareFunc cmpFunc)
306 {
307 HashEntry* pEntry;
308 HashEntry* pEnd;
309 int count = 0;
310
311 assert(pHashTable->tableSize > 0);
312 assert(item != HASH_TOMBSTONE);
313 assert(item != NULL);
314
315 /* jump to the first entry and probe for a match */
316 pEntry = &pHashTable->pEntries[itemHash & (pHashTable->tableSize-1)];
317 pEnd = &pHashTable->pEntries[pHashTable->tableSize];
318 while (pEntry->data != NULL) {
319 if (pEntry->data != HASH_TOMBSTONE &&
320 pEntry->hashValue == itemHash &&
321 (*cmpFunc)(pEntry->data, item) == 0)
322 {
323 /* match */
324 break;
325 }
326
327 pEntry++;
328 if (pEntry == pEnd) { /* wrap around to start */
329 if (pHashTable->tableSize == 1)
330 break; /* edge case - single-entry table */
331 pEntry = pHashTable->pEntries;
332 }
333
334 count++;
335 }
336 if (pEntry->data == NULL)
337 return -1;
338
339 return count;
340 }
341
342 /*
343 * Evaluate the amount of probing required for the specified hash table.
344 *
345 * We do this by running through all entries in the hash table, computing
346 * the hash value and then doing a lookup.
347 *
348 * The caller should lock the table before calling here.
349 */
mzHashTableProbeCount(HashTable * pHashTable,HashCalcFunc calcFunc,HashCompareFunc cmpFunc)350 void mzHashTableProbeCount(HashTable* pHashTable, HashCalcFunc calcFunc,
351 HashCompareFunc cmpFunc)
352 {
353 int numEntries, minProbe, maxProbe, totalProbe;
354 HashIter iter;
355
356 numEntries = maxProbe = totalProbe = 0;
357 minProbe = 65536*32767;
358
359 for (mzHashIterBegin(pHashTable, &iter); !mzHashIterDone(&iter);
360 mzHashIterNext(&iter))
361 {
362 const void* data = (const void*)mzHashIterData(&iter);
363 int count;
364
365 count = countProbes(pHashTable, (*calcFunc)(data), data, cmpFunc);
366
367 numEntries++;
368
369 if (count < minProbe)
370 minProbe = count;
371 if (count > maxProbe)
372 maxProbe = count;
373 totalProbe += count;
374 }
375
376 LOGV("Probe: min=%d max=%d, total=%d in %d (%d), avg=%.3f\n",
377 minProbe, maxProbe, totalProbe, numEntries, pHashTable->tableSize,
378 (float) totalProbe / (float) numEntries);
379 }
380