1 //===- llvm/ADT/SmallPtrSet.h - 'Normally small' pointer set ----*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file defines the SmallPtrSet class. See the doxygen comment for
11 // SmallPtrSetImpl for more details on the algorithm used.
13 //===----------------------------------------------------------------------===//
15 #ifndef LLVM_ADT_SMALLPTRSET_H
16 #define LLVM_ADT_SMALLPTRSET_H
20 #include "llvm/Support/DataTypes.h"
24 class SmallPtrSetIteratorImpl;
26 /// SmallPtrSetImpl - This is the common code shared among all the
27 /// SmallPtrSet<>'s, which is almost everything. SmallPtrSet has two modes, one
28 /// for small and one for large sets.
30 /// Small sets use an array of pointers allocated in the SmallPtrSet object,
31 /// which is treated as a simple array of pointers. When a pointer is added to
32 /// the set, the array is scanned to see if the element already exists, if not
33 /// the element is 'pushed back' onto the array. If we run out of space in the
34 /// array, we grow into the 'large set' case. SmallSet should be used when the
35 /// sets are often small. In this case, no memory allocation is used, and only
36 /// light-weight and cache-efficient scanning is used.
38 /// Large sets use a classic exponentially-probed hash table. Empty buckets are
39 /// represented with an illegal pointer value (-1) to allow null pointers to be
40 /// inserted. Tombstones are represented with another illegal pointer value
41 /// (-2), to allow deletion. The hash table is resized when the table is 3/4 or
42 /// more. When this happens, the table is doubled in size.
44 class SmallPtrSetImpl {
45 friend class SmallPtrSetIteratorImpl;
47 /// CurArray - This is the current set of buckets. If it points to
48 /// SmallArray, then the set is in 'small mode'.
49 const void **CurArray;
50 /// CurArraySize - The allocated size of CurArray, always a power of two.
51 /// Note that CurArray points to an array that has CurArraySize+1 elements in
52 /// it, so that the end iterator actually points to valid memory.
53 unsigned CurArraySize;
55 // If small, this is # elts allocated consequtively
57 unsigned NumTombstones;
58 const void *SmallArray[1]; // Must be last ivar.
60 // Helper to copy construct a SmallPtrSet.
61 SmallPtrSetImpl(const SmallPtrSetImpl& that);
62 explicit SmallPtrSetImpl(unsigned SmallSize) {
63 assert(SmallSize && (SmallSize & (SmallSize-1)) == 0 &&
64 "Initial size must be a power of two!");
65 CurArray = &SmallArray[0];
66 CurArraySize = SmallSize;
67 // The end pointer, always valid, is set to a valid element to help the
69 CurArray[SmallSize] = 0;
75 bool empty() const { return size() == 0; }
76 unsigned size() const { return NumElements; }
79 // If the capacity of the array is huge, and the # elements used is small,
81 if (!isSmall() && NumElements*4 < CurArraySize && CurArraySize > 32)
82 return shrink_and_clear();
84 // Fill the array with empty markers.
85 memset(CurArray, -1, CurArraySize*sizeof(void*));
91 static void *getTombstoneMarker() { return reinterpret_cast<void*>(-2); }
92 static void *getEmptyMarker() {
93 // Note that -1 is chosen to make clear() efficiently implementable with
94 // memset and because it's not a valid pointer value.
95 return reinterpret_cast<void*>(-1);
98 /// insert_imp - This returns true if the pointer was new to the set, false if
99 /// it was already in the set. This is hidden from the client so that the
100 /// derived class can check that the right type of pointer is passed in.
101 bool insert_imp(const void * Ptr);
103 /// erase_imp - If the set contains the specified pointer, remove it and
104 /// return true, otherwise return false. This is hidden from the client so
105 /// that the derived class can check that the right type of pointer is passed
107 bool erase_imp(const void * Ptr);
109 bool count_imp(const void * Ptr) const {
111 // Linear search for the item.
112 for (const void *const *APtr = SmallArray,
113 *const *E = SmallArray+NumElements; APtr != E; ++APtr)
120 return *FindBucketFor(Ptr) == Ptr;
124 bool isSmall() const { return CurArray == &SmallArray[0]; }
126 unsigned Hash(const void *Ptr) const {
127 return static_cast<unsigned>(((uintptr_t)Ptr >> 4) & (CurArraySize-1));
129 const void * const *FindBucketFor(const void *Ptr) const;
130 void shrink_and_clear();
132 /// Grow - Allocate a larger backing store for the buckets and move it over.
135 void operator=(const SmallPtrSetImpl &RHS); // DO NOT IMPLEMENT.
137 void CopyFrom(const SmallPtrSetImpl &RHS);
140 /// SmallPtrSetIteratorImpl - This is the common base class shared between all
141 /// instances of SmallPtrSetIterator.
142 class SmallPtrSetIteratorImpl {
144 const void *const *Bucket;
146 explicit SmallPtrSetIteratorImpl(const void *const *BP) : Bucket(BP) {
150 bool operator==(const SmallPtrSetIteratorImpl &RHS) const {
151 return Bucket == RHS.Bucket;
153 bool operator!=(const SmallPtrSetIteratorImpl &RHS) const {
154 return Bucket != RHS.Bucket;
158 /// AdvanceIfNotValid - If the current bucket isn't valid, advance to a bucket
159 /// that is. This is guaranteed to stop because the end() bucket is marked
161 void AdvanceIfNotValid() {
162 while (*Bucket == SmallPtrSetImpl::getEmptyMarker() ||
163 *Bucket == SmallPtrSetImpl::getTombstoneMarker())
168 /// SmallPtrSetIterator - This implements a const_iterator for SmallPtrSet.
169 template<typename PtrTy>
170 class SmallPtrSetIterator : public SmallPtrSetIteratorImpl {
172 explicit SmallPtrSetIterator(const void *const *BP)
173 : SmallPtrSetIteratorImpl(BP) {}
175 // Most methods provided by baseclass.
177 const PtrTy operator*() const {
178 return static_cast<const PtrTy>(const_cast<void*>(*Bucket));
181 inline SmallPtrSetIterator& operator++() { // Preincrement
187 SmallPtrSetIterator operator++(int) { // Postincrement
188 SmallPtrSetIterator tmp = *this; ++*this; return tmp;
192 /// NextPowerOfTwo - This is a helper template that rounds N up to the next
195 struct NextPowerOfTwo;
197 /// NextPowerOfTwoH - If N is not a power of two, increase it. This is a helper
198 /// template used to implement NextPowerOfTwo.
199 template<unsigned N, bool isPowerTwo>
200 struct NextPowerOfTwoH {
204 struct NextPowerOfTwoH<N, false> {
206 // We could just use NextVal = N+1, but this converges faster. N|(N-1) sets
207 // the right-most zero bits to one all at once, e.g. 0b0011000 -> 0b0011111.
208 Val = NextPowerOfTwo<(N|(N-1)) + 1>::Val
213 struct NextPowerOfTwo {
214 enum { Val = NextPowerOfTwoH<N, (N&(N-1)) == 0>::Val };
218 /// SmallPtrSet - This class implements a set which is optimizer for holding
219 /// SmallSize or less elements. This internally rounds up SmallSize to the next
220 /// power of two if it is not already a power of two. See the comments above
221 /// SmallPtrSetImpl for details of the algorithm.
222 template<class PtrType, unsigned SmallSize>
223 class SmallPtrSet : public SmallPtrSetImpl {
224 // Make sure that SmallSize is a power of two, round up if not.
225 enum { SmallSizePowTwo = NextPowerOfTwo<SmallSize>::Val };
226 void *SmallArray[SmallSizePowTwo];
228 SmallPtrSet() : SmallPtrSetImpl(NextPowerOfTwo<SmallSizePowTwo>::Val) {}
229 SmallPtrSet(const SmallPtrSet &that) : SmallPtrSetImpl(that) {}
231 template<typename It>
232 SmallPtrSet(It I, It E)
233 : SmallPtrSetImpl(NextPowerOfTwo<SmallSizePowTwo>::Val) {
237 /// insert - This returns true if the pointer was new to the set, false if it
238 /// was already in the set.
239 bool insert(PtrType Ptr) { return insert_imp(Ptr); }
241 /// erase - If the set contains the specified pointer, remove it and return
242 /// true, otherwise return false.
243 bool erase(PtrType Ptr) { return erase_imp(Ptr); }
245 /// count - Return true if the specified pointer is in the set.
246 bool count(PtrType Ptr) const { return count_imp(Ptr); }
248 template <typename IterT>
249 void insert(IterT I, IterT E) {
254 typedef SmallPtrSetIterator<PtrType> iterator;
255 typedef SmallPtrSetIterator<PtrType> const_iterator;
256 inline iterator begin() const {
257 return iterator(CurArray);
259 inline iterator end() const {
260 return iterator(CurArray+CurArraySize);
263 // Allow assignment from any smallptrset with the same element type even if it
264 // doesn't have the same smallsize.
265 const SmallPtrSet<PtrType, SmallSize>&
266 operator=(const SmallPtrSet<PtrType, SmallSize> &RHS) {