1//===- llvm/ADT/SmallPtrSet.h - 'Normally small' pointer set ----*- C++ -*-===//
2//
3//                     The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// This file defines the SmallPtrSet class.  See the doxygen comment for
11// SmallPtrSetImpl for more details on the algorithm used.
12//
13//===----------------------------------------------------------------------===//
14
15#ifndef LLVM_ADT_SMALLPTRSET_H
16#define LLVM_ADT_SMALLPTRSET_H
17
18#include <cassert>
19#include <cstddef>
20#include <cstring>
21#include <iterator>
22#include "llvm/Support/DataTypes.h"
23#include "llvm/Support/PointerLikeTypeTraits.h"
24
25namespace llvm {
26
27class SmallPtrSetIteratorImpl;
28
29/// SmallPtrSetImpl - This is the common code shared among all the
30/// SmallPtrSet<>'s, which is almost everything.  SmallPtrSet has two modes, one
31/// for small and one for large sets.
32///
33/// Small sets use an array of pointers allocated in the SmallPtrSet object,
34/// which is treated as a simple array of pointers.  When a pointer is added to
35/// the set, the array is scanned to see if the element already exists, if not
36/// the element is 'pushed back' onto the array.  If we run out of space in the
37/// array, we grow into the 'large set' case.  SmallSet should be used when the
38/// sets are often small.  In this case, no memory allocation is used, and only
39/// light-weight and cache-efficient scanning is used.
40///
41/// Large sets use a classic exponentially-probed hash table.  Empty buckets are
42/// represented with an illegal pointer value (-1) to allow null pointers to be
43/// inserted.  Tombstones are represented with another illegal pointer value
44/// (-2), to allow deletion.  The hash table is resized when the table is 3/4 or
45/// more.  When this happens, the table is doubled in size.
46///
47class SmallPtrSetImpl {
48  friend class SmallPtrSetIteratorImpl;
49protected:
50  /// SmallArray - Points to a fixed size set of buckets, used in 'small mode'.
51  const void **SmallArray;
52  /// CurArray - This is the current set of buckets.  If equal to SmallArray,
53  /// then the set is in 'small mode'.
54  const void **CurArray;
55  /// CurArraySize - The allocated size of CurArray, always a power of two.
56  /// Note that CurArray points to an array that has CurArraySize+1 elements in
57  /// it, so that the end iterator actually points to valid memory.
58  unsigned CurArraySize;
59
60  // If small, this is # elts allocated consecutively
61  unsigned NumElements;
62  unsigned NumTombstones;
63
64  // Helper to copy construct a SmallPtrSet.
65  SmallPtrSetImpl(const void **SmallStorage, const SmallPtrSetImpl& that);
66  explicit SmallPtrSetImpl(const void **SmallStorage, unsigned SmallSize) :
67    SmallArray(SmallStorage), CurArray(SmallStorage), CurArraySize(SmallSize) {
68    assert(SmallSize && (SmallSize & (SmallSize-1)) == 0 &&
69           "Initial size must be a power of two!");
70    // The end pointer, always valid, is set to a valid element to help the
71    // iterator.
72    CurArray[SmallSize] = 0;
73    clear();
74  }
75  ~SmallPtrSetImpl();
76
77public:
78  bool empty() const { return size() == 0; }
79  unsigned size() const { return NumElements; }
80
81  void clear() {
82    // If the capacity of the array is huge, and the # elements used is small,
83    // shrink the array.
84    if (!isSmall() && NumElements*4 < CurArraySize && CurArraySize > 32)
85      return shrink_and_clear();
86
87    // Fill the array with empty markers.
88    memset(CurArray, -1, CurArraySize*sizeof(void*));
89    NumElements = 0;
90    NumTombstones = 0;
91  }
92
93protected:
94  static void *getTombstoneMarker() { return reinterpret_cast<void*>(-2); }
95  static void *getEmptyMarker() {
96    // Note that -1 is chosen to make clear() efficiently implementable with
97    // memset and because it's not a valid pointer value.
98    return reinterpret_cast<void*>(-1);
99  }
100
101  /// insert_imp - This returns true if the pointer was new to the set, false if
102  /// it was already in the set.  This is hidden from the client so that the
103  /// derived class can check that the right type of pointer is passed in.
104  bool insert_imp(const void * Ptr);
105
106  /// erase_imp - If the set contains the specified pointer, remove it and
107  /// return true, otherwise return false.  This is hidden from the client so
108  /// that the derived class can check that the right type of pointer is passed
109  /// in.
110  bool erase_imp(const void * Ptr);
111
112  bool count_imp(const void * Ptr) const {
113    if (isSmall()) {
114      // Linear search for the item.
115      for (const void *const *APtr = SmallArray,
116                      *const *E = SmallArray+NumElements; APtr != E; ++APtr)
117        if (*APtr == Ptr)
118          return true;
119      return false;
120    }
121
122    // Big set case.
123    return *FindBucketFor(Ptr) == Ptr;
124  }
125
126private:
127  bool isSmall() const { return CurArray == SmallArray; }
128
129  const void * const *FindBucketFor(const void *Ptr) const;
130  void shrink_and_clear();
131
132  /// Grow - Allocate a larger backing store for the buckets and move it over.
133  void Grow(unsigned NewSize);
134
135  void operator=(const SmallPtrSetImpl &RHS);  // DO NOT IMPLEMENT.
136protected:
137  /// swap - Swaps the elements of two sets.
138  /// Note: This method assumes that both sets have the same small size.
139  void swap(SmallPtrSetImpl &RHS);
140
141  void CopyFrom(const SmallPtrSetImpl &RHS);
142};
143
144/// SmallPtrSetIteratorImpl - This is the common base class shared between all
145/// instances of SmallPtrSetIterator.
146class SmallPtrSetIteratorImpl {
147protected:
148  const void *const *Bucket;
149public:
150  explicit SmallPtrSetIteratorImpl(const void *const *BP) : Bucket(BP) {
151    AdvanceIfNotValid();
152  }
153
154  bool operator==(const SmallPtrSetIteratorImpl &RHS) const {
155    return Bucket == RHS.Bucket;
156  }
157  bool operator!=(const SmallPtrSetIteratorImpl &RHS) const {
158    return Bucket != RHS.Bucket;
159  }
160
161protected:
162  /// AdvanceIfNotValid - If the current bucket isn't valid, advance to a bucket
163  /// that is.   This is guaranteed to stop because the end() bucket is marked
164  /// valid.
165  void AdvanceIfNotValid() {
166    while (*Bucket == SmallPtrSetImpl::getEmptyMarker() ||
167           *Bucket == SmallPtrSetImpl::getTombstoneMarker())
168      ++Bucket;
169  }
170};
171
172/// SmallPtrSetIterator - This implements a const_iterator for SmallPtrSet.
173template<typename PtrTy>
174class SmallPtrSetIterator : public SmallPtrSetIteratorImpl {
175  typedef PointerLikeTypeTraits<PtrTy> PtrTraits;
176
177public:
178  typedef PtrTy                     value_type;
179  typedef PtrTy                     reference;
180  typedef PtrTy                     pointer;
181  typedef std::ptrdiff_t            difference_type;
182  typedef std::forward_iterator_tag iterator_category;
183
184  explicit SmallPtrSetIterator(const void *const *BP)
185    : SmallPtrSetIteratorImpl(BP) {}
186
187  // Most methods provided by baseclass.
188
189  const PtrTy operator*() const {
190    return PtrTraits::getFromVoidPointer(const_cast<void*>(*Bucket));
191  }
192
193  inline SmallPtrSetIterator& operator++() {          // Preincrement
194    ++Bucket;
195    AdvanceIfNotValid();
196    return *this;
197  }
198
199  SmallPtrSetIterator operator++(int) {        // Postincrement
200    SmallPtrSetIterator tmp = *this; ++*this; return tmp;
201  }
202};
203
204/// RoundUpToPowerOfTwo - This is a helper template that rounds N up to the next
205/// power of two (which means N itself if N is already a power of two).
206template<unsigned N>
207struct RoundUpToPowerOfTwo;
208
209/// RoundUpToPowerOfTwoH - If N is not a power of two, increase it.  This is a
210/// helper template used to implement RoundUpToPowerOfTwo.
211template<unsigned N, bool isPowerTwo>
212struct RoundUpToPowerOfTwoH {
213  enum { Val = N };
214};
215template<unsigned N>
216struct RoundUpToPowerOfTwoH<N, false> {
217  enum {
218    // We could just use NextVal = N+1, but this converges faster.  N|(N-1) sets
219    // the right-most zero bits to one all at once, e.g. 0b0011000 -> 0b0011111.
220    Val = RoundUpToPowerOfTwo<(N|(N-1)) + 1>::Val
221  };
222};
223
224template<unsigned N>
225struct RoundUpToPowerOfTwo {
226  enum { Val = RoundUpToPowerOfTwoH<N, (N&(N-1)) == 0>::Val };
227};
228
229
230/// SmallPtrSet - This class implements a set which is optimized for holding
231/// SmallSize or less elements.  This internally rounds up SmallSize to the next
232/// power of two if it is not already a power of two.  See the comments above
233/// SmallPtrSetImpl for details of the algorithm.
234template<class PtrType, unsigned SmallSize>
235class SmallPtrSet : public SmallPtrSetImpl {
236  // Make sure that SmallSize is a power of two, round up if not.
237  enum { SmallSizePowTwo = RoundUpToPowerOfTwo<SmallSize>::Val };
238  /// SmallStorage - Fixed size storage used in 'small mode'.  The extra element
239  /// ensures that the end iterator actually points to valid memory.
240  const void *SmallStorage[SmallSizePowTwo+1];
241  typedef PointerLikeTypeTraits<PtrType> PtrTraits;
242public:
243  SmallPtrSet() : SmallPtrSetImpl(SmallStorage, SmallSizePowTwo) {}
244  SmallPtrSet(const SmallPtrSet &that) : SmallPtrSetImpl(SmallStorage, that) {}
245
246  template<typename It>
247  SmallPtrSet(It I, It E) : SmallPtrSetImpl(SmallStorage, SmallSizePowTwo) {
248    insert(I, E);
249  }
250
251  /// insert - This returns true if the pointer was new to the set, false if it
252  /// was already in the set.
253  bool insert(PtrType Ptr) {
254    return insert_imp(PtrTraits::getAsVoidPointer(Ptr));
255  }
256
257  /// erase - If the set contains the specified pointer, remove it and return
258  /// true, otherwise return false.
259  bool erase(PtrType Ptr) {
260    return erase_imp(PtrTraits::getAsVoidPointer(Ptr));
261  }
262
263  /// count - Return true if the specified pointer is in the set.
264  bool count(PtrType Ptr) const {
265    return count_imp(PtrTraits::getAsVoidPointer(Ptr));
266  }
267
268  template <typename IterT>
269  void insert(IterT I, IterT E) {
270    for (; I != E; ++I)
271      insert(*I);
272  }
273
274  typedef SmallPtrSetIterator<PtrType> iterator;
275  typedef SmallPtrSetIterator<PtrType> const_iterator;
276  inline iterator begin() const {
277    return iterator(CurArray);
278  }
279  inline iterator end() const {
280    return iterator(CurArray+CurArraySize);
281  }
282
283  // Allow assignment from any smallptrset with the same element type even if it
284  // doesn't have the same smallsize.
285  const SmallPtrSet<PtrType, SmallSize>&
286  operator=(const SmallPtrSet<PtrType, SmallSize> &RHS) {
287    CopyFrom(RHS);
288    return *this;
289  }
290
291  /// swap - Swaps the elements of two sets.
292  void swap(SmallPtrSet<PtrType, SmallSize> &RHS) {
293    SmallPtrSetImpl::swap(RHS);
294  }
295};
296
297}
298
299namespace std {
300  /// Implement std::swap in terms of SmallPtrSet swap.
301  template<class T, unsigned N>
302  inline void swap(llvm::SmallPtrSet<T, N> &LHS, llvm::SmallPtrSet<T, N> &RHS) {
303    LHS.swap(RHS);
304  }
305}
306
307#endif
308