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 /// SmallPtrSetImpl - This is the common code shared among all the
25 /// SmallPtrSet<>'s, which is almost everything. SmallPtrSet has two modes, one
26 /// for small and one for large sets.
28 /// Small sets use an array of pointers allocated in the SmallPtrSet object,
29 /// which is treated as a simple array of pointers. When a pointer is added to
30 /// the set, the array is scanned to see if the element already exists, if not
31 /// the element is 'pushed back' onto the array. If we run out of space in the
32 /// array, we grow into the 'large set' case. SmallSet should be used when the
33 /// sets are often small. In this case, no memory allocation is used, and only
34 /// light-weight and cache-efficient scanning is used.
36 /// Large sets use a classic exponentially-probed hash table. Empty buckets are
37 /// represented with an illegal pointer value (-1) to allow null pointers to be
38 /// inserted. Tombstones are represented with another illegal pointer value
39 /// (-2), to allow deletion. The hash table is resized when the table is 3/4 or
40 /// more. When this happens, the table is doubled in size.
42 class SmallPtrSetImpl {
44 /// CurArray - This is the current set of buckets. If it points to
45 /// SmallArray, then the set is in 'small mode'.
46 const void **CurArray;
47 /// CurArraySize - The allocated size of CurArray, always a power of two.
48 /// Note that CurArray points to an array that has CurArraySize+1 elements in
49 /// it, so that the end iterator actually points to valid memory.
50 unsigned CurArraySize;
52 // If small, this is # elts allocated consequtively
54 unsigned NumTombstones;
55 const void *SmallArray[1]; // Must be last ivar.
57 // Helper to copy construct a SmallPtrSet.
58 SmallPtrSetImpl(const SmallPtrSetImpl& that);
60 explicit SmallPtrSetImpl(unsigned SmallSize) {
61 assert(SmallSize && (SmallSize & (SmallSize-1)) == 0 &&
62 "Initial size must be a power of two!");
63 CurArray = &SmallArray[0];
64 CurArraySize = SmallSize;
65 // The end pointer, always valid, is set to a valid element to help the
67 CurArray[SmallSize] = 0;
72 bool empty() const { return size() == 0; }
73 unsigned size() const { return NumElements; }
75 static void *getTombstoneMarker() { return reinterpret_cast<void*>(-2); }
76 static void *getEmptyMarker() {
77 // Note that -1 is chosen to make clear() efficiently implementable with
78 // memset and because it's not a valid pointer value.
79 return reinterpret_cast<void*>(-1);
83 // If the capacity of the array is huge, and the # elements used is small,
85 if (!isSmall() && NumElements*4 < CurArraySize && CurArraySize > 32)
86 return shrink_and_clear();
88 // Fill the array with empty markers.
89 memset(CurArray, -1, CurArraySize*sizeof(void*));
95 /// insert_imp - This returns true if the pointer was new to the set, false if
96 /// it was already in the set. This is hidden from the client so that the
97 /// derived class can check that the right type of pointer is passed in.
98 bool insert_imp(const void * Ptr);
100 /// erase_imp - If the set contains the specified pointer, remove it and
101 /// return true, otherwise return false. This is hidden from the client so
102 /// that the derived class can check that the right type of pointer is passed
104 bool erase_imp(const void * Ptr);
106 bool count_imp(const void * Ptr) const {
108 // Linear search for the item.
109 for (const void *const *APtr = SmallArray,
110 *const *E = SmallArray+NumElements; APtr != E; ++APtr)
117 return *FindBucketFor(Ptr) == Ptr;
121 bool isSmall() const { return CurArray == &SmallArray[0]; }
123 unsigned Hash(const void *Ptr) const {
124 return static_cast<unsigned>(((uintptr_t)Ptr >> 4) & (CurArraySize-1));
126 const void * const *FindBucketFor(const void *Ptr) const;
127 void shrink_and_clear();
129 /// Grow - Allocate a larger backing store for the buckets and move it over.
132 void operator=(const SmallPtrSetImpl &RHS); // DO NOT IMPLEMENT.
134 void CopyFrom(const SmallPtrSetImpl &RHS);
137 /// SmallPtrSetIteratorImpl - This is the common base class shared between all
138 /// instances of SmallPtrSetIterator.
139 class SmallPtrSetIteratorImpl {
141 const void *const *Bucket;
143 explicit SmallPtrSetIteratorImpl(const void *const *BP) : Bucket(BP) {
147 bool operator==(const SmallPtrSetIteratorImpl &RHS) const {
148 return Bucket == RHS.Bucket;
150 bool operator!=(const SmallPtrSetIteratorImpl &RHS) const {
151 return Bucket != RHS.Bucket;
155 /// AdvanceIfNotValid - If the current bucket isn't valid, advance to a bucket
156 /// that is. This is guaranteed to stop because the end() bucket is marked
158 void AdvanceIfNotValid() {
159 while (*Bucket == SmallPtrSetImpl::getEmptyMarker() ||
160 *Bucket == SmallPtrSetImpl::getTombstoneMarker())
165 /// SmallPtrSetIterator - This implements a const_iterator for SmallPtrSet.
166 template<typename PtrTy>
167 class SmallPtrSetIterator : public SmallPtrSetIteratorImpl {
169 explicit SmallPtrSetIterator(const void *const *BP)
170 : SmallPtrSetIteratorImpl(BP) {}
172 // Most methods provided by baseclass.
174 const PtrTy operator*() const {
175 return static_cast<const PtrTy>(const_cast<void*>(*Bucket));
178 inline SmallPtrSetIterator& operator++() { // Preincrement
184 SmallPtrSetIterator operator++(int) { // Postincrement
185 SmallPtrSetIterator tmp = *this; ++*this; return tmp;
189 /// NextPowerOfTwo - This is a helper template that rounds N up to the next
192 struct NextPowerOfTwo;
194 /// NextPowerOfTwoH - If N is not a power of two, increase it. This is a helper
195 /// template used to implement NextPowerOfTwo.
196 template<unsigned N, bool isPowerTwo>
197 struct NextPowerOfTwoH {
201 struct NextPowerOfTwoH<N, false> {
203 // We could just use NextVal = N+1, but this converges faster. N|(N-1) sets
204 // the right-most zero bits to one all at once, e.g. 0b0011000 -> 0b0011111.
205 Val = NextPowerOfTwo<(N|(N-1)) + 1>::Val
210 struct NextPowerOfTwo {
211 enum { Val = NextPowerOfTwoH<N, (N&(N-1)) == 0>::Val };
215 /// SmallPtrSet - This class implements a set which is optimizer for holding
216 /// SmallSize or less elements. This internally rounds up SmallSize to the next
217 /// power of two if it is not already a power of two. See the comments above
218 /// SmallPtrSetImpl for details of the algorithm.
219 template<class PtrType, unsigned SmallSize>
220 class SmallPtrSet : public SmallPtrSetImpl {
221 // Make sure that SmallSize is a power of two, round up if not.
222 enum { SmallSizePowTwo = NextPowerOfTwo<SmallSize>::Val };
223 void *SmallArray[SmallSizePowTwo];
225 SmallPtrSet() : SmallPtrSetImpl(NextPowerOfTwo<SmallSizePowTwo>::Val) {}
226 SmallPtrSet(const SmallPtrSet &that) : SmallPtrSetImpl(that) {}
228 template<typename It>
229 SmallPtrSet(It I, It E)
230 : SmallPtrSetImpl(NextPowerOfTwo<SmallSizePowTwo>::Val) {
234 /// insert - This returns true if the pointer was new to the set, false if it
235 /// was already in the set.
236 bool insert(PtrType Ptr) { return insert_imp(Ptr); }
238 /// erase - If the set contains the specified pointer, remove it and return
239 /// true, otherwise return false.
240 bool erase(PtrType Ptr) { return erase_imp(Ptr); }
242 /// count - Return true if the specified pointer is in the set.
243 bool count(PtrType Ptr) const { return count_imp(Ptr); }
245 template <typename IterT>
246 void insert(IterT I, IterT E) {
251 typedef SmallPtrSetIterator<PtrType> iterator;
252 typedef SmallPtrSetIterator<PtrType> const_iterator;
253 inline iterator begin() const {
254 return iterator(CurArray);
256 inline iterator end() const {
257 return iterator(CurArray+CurArraySize);
260 // Allow assignment from any smallptrset with the same element type even if it
261 // doesn't have the same smallsize.
262 const SmallPtrSet<PtrType, SmallSize>&
263 operator=(const SmallPtrSet<PtrType, SmallSize> &RHS) {