DenseMap.h revision 7b54452c84e478ab4d49ac08759ca4ec1badf2b2 
1//===- llvm/ADT/DenseMap.h - Dense probed hash table ------------*- C++ -*-===//
2//
3//                     The LLVM Compiler Infrastructure
4//
5// This file was developed by Chris Lattner and is distributed under
6// the University of Illinois Open Source License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// This file defines the DenseMap class.
11//
12//===----------------------------------------------------------------------===//
13
14#ifndef LLVM_ADT_DENSEMAP_H
15#define LLVM_ADT_DENSEMAP_H
16
17#include "llvm/Support/DataTypes.h"
18#include "llvm/Support/MathExtras.h"
19#include <cassert>
20#include <utility>
21
22namespace llvm {
23
24template<typename T>
25struct DenseMapKeyInfo {
26  //static inline T getEmptyKey();
27  //static inline T getTombstoneKey();
28  //static unsigned getHashValue(const T &Val);
29  //static bool isPod()
30};
31
32// Provide DenseMapKeyInfo for all pointers.
33template<typename T>
34struct DenseMapKeyInfo<T*> {
35  static inline T* getEmptyKey() { return (T*)-1; }
36  static inline T* getTombstoneKey() { return (T*)-2; }
37  static unsigned getHashValue(const T *PtrVal) {
38    return (unsigned)((uintptr_t)PtrVal >> 4) ^
39           (unsigned)((uintptr_t)PtrVal >> 9);
40  }
41  static bool isPod() { return true; }
42};
43
44template<typename KeyT, typename ValueT,
45         typename KeyInfoT = DenseMapKeyInfo<KeyT> >
46class DenseMapIterator;
47template<typename KeyT, typename ValueT,
48         typename KeyInfoT = DenseMapKeyInfo<KeyT> >
49class DenseMapConstIterator;
50
51template<typename KeyT, typename ValueT,
52         typename KeyInfoT = DenseMapKeyInfo<KeyT> >
53class DenseMap {
54  typedef std::pair<KeyT, ValueT> BucketT;
55  unsigned NumBuckets;
56  BucketT *Buckets;
57
58  unsigned NumEntries;
59  unsigned NumTombstones;
60  DenseMap(const DenseMap &); // not implemented.
61public:
62  explicit DenseMap(unsigned NumInitBuckets = 64) {
63    init(NumInitBuckets);
64  }
65  ~DenseMap() {
66    const KeyT EmptyKey = getEmptyKey(), TombstoneKey = getTombstoneKey();
67    for (BucketT *P = Buckets, *E = Buckets+NumBuckets; P != E; ++P) {
68      if (P->first != EmptyKey && P->first != TombstoneKey)
69        P->second.~ValueT();
70      P->first.~KeyT();
71    }
72    delete[] (char*)Buckets;
73  }
74
75  typedef DenseMapIterator<KeyT, ValueT, KeyInfoT> iterator;
76  typedef DenseMapConstIterator<KeyT, ValueT, KeyInfoT> const_iterator;
77  inline iterator begin() {
78     return iterator(Buckets, Buckets+NumBuckets);
79  }
80  inline iterator end() {
81    return iterator(Buckets+NumBuckets, Buckets+NumBuckets);
82  }
83  inline const_iterator begin() const {
84    return const_iterator(Buckets, Buckets+NumBuckets);
85  }
86  inline const_iterator end() const {
87    return const_iterator(Buckets+NumBuckets, Buckets+NumBuckets);
88  }
89
90  bool empty() const { return NumEntries == 0; }
91  unsigned size() const { return NumEntries; }
92
93  void clear() {
94    // If the capacity of the array is huge, and the # elements used is small,
95    // shrink the array.
96    if (NumEntries * 4 < NumBuckets && NumBuckets > 64) {
97      shrink_and_clear();
98      return;
99    }
100
101    const KeyT EmptyKey = getEmptyKey(), TombstoneKey = getTombstoneKey();
102    for (BucketT *P = Buckets, *E = Buckets+NumBuckets; P != E; ++P) {
103      if (P->first != EmptyKey) {
104        if (P->first != TombstoneKey) {
105          P->second.~ValueT();
106          --NumEntries;
107        }
108        P->first = EmptyKey;
109      }
110    }
111    assert(NumEntries == 0 && "Node count imbalance!");
112    NumTombstones = 0;
113  }
114
115  /// count - Return true if the specified key is in the map.
116  bool count(const KeyT &Val) const {
117    BucketT *TheBucket;
118    return LookupBucketFor(Val, TheBucket);
119  }
120
121  iterator find(const KeyT &Val) {
122    BucketT *TheBucket;
123    if (LookupBucketFor(Val, TheBucket))
124      return iterator(TheBucket, Buckets+NumBuckets);
125    return end();
126  }
127  const_iterator find(const KeyT &Val) const {
128    BucketT *TheBucket;
129    if (LookupBucketFor(Val, TheBucket))
130      return const_iterator(TheBucket, Buckets+NumBuckets);
131    return end();
132  }
133
134  bool insert(const std::pair<KeyT, ValueT> &KV) {
135    BucketT *TheBucket;
136    if (LookupBucketFor(KV.first, TheBucket))
137      return false; // Already in map.
138
139    // Otherwise, insert the new element.
140    InsertIntoBucket(KV.first, KV.second, TheBucket);
141    return true;
142  }
143
144  bool erase(const KeyT &Val) {
145    BucketT *TheBucket;
146    if (!LookupBucketFor(Val, TheBucket))
147      return false; // not in map.
148
149    TheBucket->second.~ValueT();
150    TheBucket->first = getTombstoneKey();
151    --NumEntries;
152    ++NumTombstones;
153    return true;
154  }
155  bool erase(iterator I) {
156    BucketT *TheBucket = &*I;
157    TheBucket->second.~ValueT();
158    TheBucket->first = getTombstoneKey();
159    --NumEntries;
160    ++NumTombstones;
161    return true;
162  }
163
164  ValueT &operator[](const KeyT &Key) {
165    BucketT *TheBucket;
166    if (LookupBucketFor(Key, TheBucket))
167      return TheBucket->second;
168
169    return InsertIntoBucket(Key, ValueT(), TheBucket)->second;
170  }
171
172private:
173  BucketT *InsertIntoBucket(const KeyT &Key, const ValueT &Value,
174                            BucketT *TheBucket) {
175    // If the load of the hash table is more than 3/4, or if fewer than 1/8 of
176    // the buckets are empty (meaning that many are filled with tombstones),
177    // grow the table.
178    //
179    // The later case is tricky.  For example, if we had one empty bucket with
180    // tons of tombstones, failing lookups (e.g. for insertion) would have to
181    // probe almost the entire table until it found the empty bucket.  If the
182    // table completely filled with tombstones, no lookup would ever succeed,
183    // causing infinite loops in lookup.
184    if (NumEntries*4 >= NumBuckets*3 ||
185        NumBuckets-(NumEntries+NumTombstones) < NumBuckets/8) {
186      this->grow();
187      LookupBucketFor(Key, TheBucket);
188    }
189    ++NumEntries;
190
191    // If we are writing over a tombstone, remember this.
192    if (TheBucket->first != getEmptyKey())
193      --NumTombstones;
194
195    TheBucket->first = Key;
196    new (&TheBucket->second) ValueT(Value);
197    return TheBucket;
198  }
199
200  static unsigned getHashValue(const KeyT &Val) {
201    return KeyInfoT::getHashValue(Val);
202  }
203  static const KeyT getEmptyKey() {
204    return KeyInfoT::getEmptyKey();
205  }
206  static const KeyT getTombstoneKey() {
207    return KeyInfoT::getTombstoneKey();
208  }
209
210  /// LookupBucketFor - Lookup the appropriate bucket for Val, returning it in
211  /// FoundBucket.  If the bucket contains the key and a value, this returns
212  /// true, otherwise it returns a bucket with an empty marker or tombstone and
213  /// returns false.
214  bool LookupBucketFor(const KeyT &Val, BucketT *&FoundBucket) const {
215    unsigned BucketNo = getHashValue(Val);
216    unsigned ProbeAmt = 1;
217    BucketT *BucketsPtr = Buckets;
218
219    // FoundTombstone - Keep track of whether we find a tombstone while probing.
220    BucketT *FoundTombstone = 0;
221    const KeyT EmptyKey = getEmptyKey();
222    const KeyT TombstoneKey = getTombstoneKey();
223    assert(Val != EmptyKey && Val != TombstoneKey &&
224           "Empty/Tombstone value shouldn't be inserted into map!");
225
226    while (1) {
227      BucketT *ThisBucket = BucketsPtr + (BucketNo & (NumBuckets-1));
228      // Found Val's bucket?  If so, return it.
229      if (ThisBucket->first == Val) {
230        FoundBucket = ThisBucket;
231        return true;
232      }
233
234      // If we found an empty bucket, the key doesn't exist in the set.
235      // Insert it and return the default value.
236      if (ThisBucket->first == EmptyKey) {
237        // If we've already seen a tombstone while probing, fill it in instead
238        // of the empty bucket we eventually probed to.
239        if (FoundTombstone) ThisBucket = FoundTombstone;
240        FoundBucket = FoundTombstone ? FoundTombstone : ThisBucket;
241        return false;
242      }
243
244      // If this is a tombstone, remember it.  If Val ends up not in the map, we
245      // prefer to return it than something that would require more probing.
246      if (ThisBucket->first == TombstoneKey && !FoundTombstone)
247        FoundTombstone = ThisBucket;  // Remember the first tombstone found.
248
249      // Otherwise, it's a hash collision or a tombstone, continue quadratic
250      // probing.
251      BucketNo += ProbeAmt++;
252    }
253  }
254
255  void init(unsigned InitBuckets) {
256    NumEntries = 0;
257    NumTombstones = 0;
258    NumBuckets = InitBuckets;
259    assert(InitBuckets && (InitBuckets & InitBuckets-1) == 0 &&
260           "# initial buckets must be a power of two!");
261    Buckets = (BucketT*)new char[sizeof(BucketT)*InitBuckets];
262    // Initialize all the keys to EmptyKey.
263    const KeyT EmptyKey = getEmptyKey();
264    for (unsigned i = 0; i != InitBuckets; ++i)
265      new (&Buckets[i].first) KeyT(EmptyKey);
266  }
267
268  void grow() {
269    unsigned OldNumBuckets = NumBuckets;
270    BucketT *OldBuckets = Buckets;
271
272    // Double the number of buckets.
273    NumBuckets <<= 1;
274    NumTombstones = 0;
275    Buckets = (BucketT*)new char[sizeof(BucketT)*NumBuckets];
276
277    // Initialize all the keys to EmptyKey.
278    const KeyT EmptyKey = getEmptyKey();
279    for (unsigned i = 0, e = NumBuckets; i != e; ++i)
280      new (&Buckets[i].first) KeyT(EmptyKey);
281
282    // Insert all the old elements.
283    const KeyT TombstoneKey = getTombstoneKey();
284    for (BucketT *B = OldBuckets, *E = OldBuckets+OldNumBuckets; B != E; ++B) {
285      if (B->first != EmptyKey && B->first != TombstoneKey) {
286        // Insert the key/value into the new table.
287        BucketT *DestBucket;
288        bool FoundVal = LookupBucketFor(B->first, DestBucket);
289        FoundVal = FoundVal; // silence warning.
290        assert(!FoundVal && "Key already in new map?");
291        DestBucket->first = B->first;
292        new (&DestBucket->second) ValueT(B->second);
293
294        // Free the value.
295        B->second.~ValueT();
296      }
297      B->first.~KeyT();
298    }
299
300    // Free the old table.
301    delete[] (char*)OldBuckets;
302  }
303
304  void shrink_and_clear() {
305    unsigned OldNumBuckets = NumBuckets;
306    BucketT *OldBuckets = Buckets;
307
308    // Reduce the number of buckets.
309    NumBuckets = NumEntries > 32 ? 1 << (Log2_32_Ceil(NumEntries) + 1)
310                                 : 64;
311    NumTombstones = 0;
312    Buckets = (BucketT*)new char[sizeof(BucketT)*NumBuckets];
313
314    // Initialize all the keys to EmptyKey.
315    const KeyT EmptyKey = getEmptyKey();
316    for (unsigned i = 0, e = NumBuckets; i != e; ++i)
317      new (&Buckets[i].first) KeyT(EmptyKey);
318
319    // Free the old buckets.
320    const KeyT TombstoneKey = getTombstoneKey();
321    for (BucketT *B = OldBuckets, *E = OldBuckets+OldNumBuckets; B != E; ++B) {
322      if (B->first != EmptyKey && B->first != TombstoneKey) {
323        // Free the value.
324        B->second.~ValueT();
325      }
326      B->first.~KeyT();
327    }
328
329    // Free the old table.
330    delete[] (char*)OldBuckets;
331
332    NumEntries = 0;
333  }
334};
335
336template<typename KeyT, typename ValueT, typename KeyInfoT>
337class DenseMapIterator {
338  typedef std::pair<KeyT, ValueT> BucketT;
339protected:
340  const BucketT *Ptr, *End;
341public:
342  DenseMapIterator(const BucketT *Pos, const BucketT *E) : Ptr(Pos), End(E) {
343    AdvancePastEmptyBuckets();
344  }
345
346  std::pair<KeyT, ValueT> &operator*() const {
347    return *const_cast<BucketT*>(Ptr);
348  }
349  std::pair<KeyT, ValueT> *operator->() const {
350    return const_cast<BucketT*>(Ptr);
351  }
352
353  bool operator==(const DenseMapIterator &RHS) const {
354    return Ptr == RHS.Ptr;
355  }
356  bool operator!=(const DenseMapIterator &RHS) const {
357    return Ptr != RHS.Ptr;
358  }
359
360  inline DenseMapIterator& operator++() {          // Preincrement
361    ++Ptr;
362    AdvancePastEmptyBuckets();
363    return *this;
364  }
365  DenseMapIterator operator++(int) {        // Postincrement
366    DenseMapIterator tmp = *this; ++*this; return tmp;
367  }
368
369private:
370  void AdvancePastEmptyBuckets() {
371    const KeyT Empty = KeyInfoT::getEmptyKey();
372    const KeyT Tombstone = KeyInfoT::getTombstoneKey();
373
374    while (Ptr != End && (Ptr->first == Empty || Ptr->first == Tombstone))
375      ++Ptr;
376  }
377};
378
379template<typename KeyT, typename ValueT, typename KeyInfoT>
380class DenseMapConstIterator : public DenseMapIterator<KeyT, ValueT, KeyInfoT> {
381public:
382  DenseMapConstIterator(const std::pair<KeyT, ValueT> *Pos,
383                        const std::pair<KeyT, ValueT> *E)
384    : DenseMapIterator<KeyT, ValueT, KeyInfoT>(Pos, E) {
385  }
386  const std::pair<KeyT, ValueT> &operator*() const {
387    return *this->Ptr;
388  }
389  const std::pair<KeyT, ValueT> *operator->() const {
390    return this->Ptr;
391  }
392};
393
394} // end namespace llvm
395
396#endif
397