1 //===- llvm/ADT/SmallPtrSet.h - 'Normally small' pointer set ----*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
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.
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'.
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 void *SmallArray[1]; // Must be last ivar.
57 // Helper to copy construct a SmallPtrSet.
58 SmallPtrSetImpl(const SmallPtrSetImpl& that);
60 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;
75 bool empty() const { return size() == 0; }
76 unsigned size() const { return NumElements; }
78 static void *getTombstoneMarker() { return reinterpret_cast<void*>(-2); }
79 static void *getEmptyMarker() {
80 // Note that -1 is chosen to make clear() efficiently implementable with
81 // memset and because it's not a valid pointer value.
82 return reinterpret_cast<void*>(-1);
86 // Fill the array with empty markers.
87 memset(CurArray, -1, CurArraySize*sizeof(void*));
92 /// insert - This returns true if the pointer was new to the set, false if it
93 /// was already in the set.
94 bool insert(void *Ptr);
96 template <typename IterT>
97 void insert(IterT I, IterT E) {
102 /// erase - If the set contains the specified pointer, remove it and return
103 /// true, otherwise return false.
104 bool erase(void *Ptr);
106 bool count(void *Ptr) const {
108 // Linear search for the item.
109 for (void *const *APtr = SmallArray, *const *E = SmallArray+NumElements;
117 return *FindBucketFor(Ptr) == Ptr;
121 bool isSmall() const { return CurArray == &SmallArray[0]; }
123 unsigned Hash(void *Ptr) const {
124 return ((uintptr_t)Ptr >> 4) & (CurArraySize-1);
126 void * const *FindBucketFor(void *Ptr) const;
128 /// Grow - Allocate a larger backing store for the buckets and move it over.
131 void operator=(const SmallPtrSetImpl &RHS); // DO NOT IMPLEMENT.
133 void CopyFrom(const SmallPtrSetImpl &RHS);
136 /// SmallPtrSetIteratorImpl - This is the common base class shared between all
137 /// instances of SmallPtrSetIterator.
138 class SmallPtrSetIteratorImpl {
142 SmallPtrSetIteratorImpl(void *const *BP) : Bucket(BP) {
146 bool operator==(const SmallPtrSetIteratorImpl &RHS) const {
147 return Bucket == RHS.Bucket;
149 bool operator!=(const SmallPtrSetIteratorImpl &RHS) const {
150 return Bucket != RHS.Bucket;
154 /// AdvanceIfNotValid - If the current bucket isn't valid, advance to a bucket
155 /// that is. This is guaranteed to stop because the end() bucket is marked
157 void AdvanceIfNotValid() {
158 while (*Bucket == SmallPtrSetImpl::getEmptyMarker() ||
159 *Bucket == SmallPtrSetImpl::getTombstoneMarker())
164 /// SmallPtrSetIterator - This implements a const_iterator for SmallPtrSet.
165 template<typename PtrTy>
166 class SmallPtrSetIterator : public SmallPtrSetIteratorImpl {
168 SmallPtrSetIterator(void *const *BP) : SmallPtrSetIteratorImpl(BP) {}
170 // Most methods provided by baseclass.
172 PtrTy operator*() const {
173 return static_cast<PtrTy>(*Bucket);
176 inline SmallPtrSetIterator& operator++() { // Preincrement
182 SmallPtrSetIterator operator++(int) { // Postincrement
183 SmallPtrSetIterator tmp = *this; ++*this; return tmp;
187 /// NextPowerOfTwo - This is a helper template that rounds N up to the next
190 struct NextPowerOfTwo;
192 /// NextPowerOfTwoH - If N is not a power of two, increase it. This is a helper
193 /// template used to implement NextPowerOfTwo.
194 template<unsigned N, bool isPowerTwo>
195 struct NextPowerOfTwoH {
199 struct NextPowerOfTwoH<N, false> {
201 // We could just use NextVal = N+1, but this converges faster. N|(N-1) sets
202 // the right-most zero bits to one all at once, e.g. 0b0011000 -> 0b0011111.
203 Val = NextPowerOfTwo<(N|(N-1)) + 1>::Val
208 struct NextPowerOfTwo {
209 enum { Val = NextPowerOfTwoH<N, (N&(N-1)) == 0>::Val };
213 /// SmallPtrSet - This class implements a set which is optimizer for holding
214 /// SmallSize or less elements. This internally rounds up SmallSize to the next
215 /// power of two if it is not already a power of two. See the comments above
216 /// SmallPtrSetImpl for details of the algorithm.
217 template<class PtrType, unsigned SmallSize>
218 class SmallPtrSet : public SmallPtrSetImpl {
219 // Make sure that SmallSize is a power of two, round up if not.
220 enum { SmallSizePowTwo = NextPowerOfTwo<SmallSize>::Val };
221 void *SmallArray[SmallSizePowTwo];
223 SmallPtrSet() : SmallPtrSetImpl(NextPowerOfTwo<SmallSizePowTwo>::Val) {}
224 SmallPtrSet(const SmallPtrSet &that) : SmallPtrSetImpl(that) {}
226 template<typename It>
227 SmallPtrSet(It I, It E)
228 : SmallPtrSetImpl(NextPowerOfTwo<SmallSizePowTwo>::Val) {
232 typedef SmallPtrSetIterator<PtrType> iterator;
233 typedef SmallPtrSetIterator<PtrType> const_iterator;
234 inline iterator begin() const {
235 return iterator(CurArray);
237 inline iterator end() const {
238 return iterator(CurArray+CurArraySize);
241 // Allow assignment from any smallptrset with the same element type even if it
242 // doesn't have the same smallsize.
243 const SmallPtrSet<PtrType, SmallSize>&
244 operator=(const SmallPtrSet<PtrType, SmallSize> &RHS) {