SmallPtrSet.h revision e992a56ae93140171f5044079e8d317f784c8cc1
1//===- llvm/ADT/SmallPtrSet.h - 'Normally small' pointer set ----*- 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 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 <cstring> 20#include "llvm/Support/DataTypes.h" 21 22namespace llvm { 23 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. 27/// 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. 35/// 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. 41/// 42class SmallPtrSetImpl { 43protected: 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; 51 52 // If small, this is # elts allocated consequtively 53 unsigned NumElements; 54 unsigned NumTombstones; 55 const void *SmallArray[1]; // Must be last ivar. 56 57 // Helper to copy construct a SmallPtrSet. 58 SmallPtrSetImpl(const SmallPtrSetImpl& that); 59public: 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 66 // iterator. 67 CurArray[SmallSize] = 0; 68 clear(); 69 } 70 ~SmallPtrSetImpl(); 71 72 bool empty() const { return size() == 0; } 73 unsigned size() const { return NumElements; } 74 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); 80 } 81 82 void clear() { 83 // Fill the array with empty markers. 84 memset(CurArray, -1, CurArraySize*sizeof(void*)); 85 NumElements = 0; 86 NumTombstones = 0; 87 } 88 89 /// insert - This returns true if the pointer was new to the set, false if it 90 /// was already in the set. 91 bool insert(const void * Ptr); 92 93 template <typename IterT> 94 void insert(IterT I, IterT E) { 95 for (; I != E; ++I) 96 insert((void*)*I); 97 } 98 99 /// erase - If the set contains the specified pointer, remove it and return 100 /// true, otherwise return false. 101 bool erase(void * const Ptr); 102 103 bool count(void * const Ptr) const { 104 if (isSmall()) { 105 // Linear search for the item. 106 for (const void *const *APtr = SmallArray, *const *E = SmallArray+NumElements; 107 APtr != E; ++APtr) 108 if (*APtr == Ptr) 109 return true; 110 return false; 111 } 112 113 // Big set case. 114 return *FindBucketFor(Ptr) == Ptr; 115 } 116 117private: 118 bool isSmall() const { return CurArray == &SmallArray[0]; } 119 120 unsigned Hash(const void *Ptr) const { 121 return ((uintptr_t)Ptr >> 4) & (CurArraySize-1); 122 } 123 const void * const *FindBucketFor(const void *Ptr) const; 124 125 /// Grow - Allocate a larger backing store for the buckets and move it over. 126 void Grow(); 127 128 void operator=(const SmallPtrSetImpl &RHS); // DO NOT IMPLEMENT. 129protected: 130 void CopyFrom(const SmallPtrSetImpl &RHS); 131}; 132 133/// SmallPtrSetIteratorImpl - This is the common base class shared between all 134/// instances of SmallPtrSetIterator. 135class SmallPtrSetIteratorImpl { 136protected: 137 const void *const *Bucket; 138public: 139 SmallPtrSetIteratorImpl(const void *const *BP) : Bucket(BP) { 140 AdvanceIfNotValid(); 141 } 142 143 bool operator==(const SmallPtrSetIteratorImpl &RHS) const { 144 return Bucket == RHS.Bucket; 145 } 146 bool operator!=(const SmallPtrSetIteratorImpl &RHS) const { 147 return Bucket != RHS.Bucket; 148 } 149 150protected: 151 /// AdvanceIfNotValid - If the current bucket isn't valid, advance to a bucket 152 /// that is. This is guaranteed to stop because the end() bucket is marked 153 /// valid. 154 void AdvanceIfNotValid() { 155 while (*Bucket == SmallPtrSetImpl::getEmptyMarker() || 156 *Bucket == SmallPtrSetImpl::getTombstoneMarker()) 157 ++Bucket; 158 } 159}; 160 161/// SmallPtrSetIterator - This implements a const_iterator for SmallPtrSet. 162template<typename PtrTy> 163class SmallPtrSetIterator : public SmallPtrSetIteratorImpl { 164public: 165 SmallPtrSetIterator(const void *const *BP) : SmallPtrSetIteratorImpl(BP) {} 166 167 // Most methods provided by baseclass. 168 169 const PtrTy operator*() const { 170 return static_cast<const PtrTy>(const_cast<void*>(*Bucket)); 171 } 172 173 inline SmallPtrSetIterator& operator++() { // Preincrement 174 ++Bucket; 175 AdvanceIfNotValid(); 176 return *this; 177 } 178 179 SmallPtrSetIterator operator++(int) { // Postincrement 180 SmallPtrSetIterator tmp = *this; ++*this; return tmp; 181 } 182}; 183 184/// NextPowerOfTwo - This is a helper template that rounds N up to the next 185/// power of two. 186template<unsigned N> 187struct NextPowerOfTwo; 188 189/// NextPowerOfTwoH - If N is not a power of two, increase it. This is a helper 190/// template used to implement NextPowerOfTwo. 191template<unsigned N, bool isPowerTwo> 192struct NextPowerOfTwoH { 193 enum { Val = N }; 194}; 195template<unsigned N> 196struct NextPowerOfTwoH<N, false> { 197 enum { 198 // We could just use NextVal = N+1, but this converges faster. N|(N-1) sets 199 // the right-most zero bits to one all at once, e.g. 0b0011000 -> 0b0011111. 200 Val = NextPowerOfTwo<(N|(N-1)) + 1>::Val 201 }; 202}; 203 204template<unsigned N> 205struct NextPowerOfTwo { 206 enum { Val = NextPowerOfTwoH<N, (N&(N-1)) == 0>::Val }; 207}; 208 209 210/// SmallPtrSet - This class implements a set which is optimizer for holding 211/// SmallSize or less elements. This internally rounds up SmallSize to the next 212/// power of two if it is not already a power of two. See the comments above 213/// SmallPtrSetImpl for details of the algorithm. 214template<class PtrType, unsigned SmallSize> 215class SmallPtrSet : public SmallPtrSetImpl { 216 // Make sure that SmallSize is a power of two, round up if not. 217 enum { SmallSizePowTwo = NextPowerOfTwo<SmallSize>::Val }; 218 void *SmallArray[SmallSizePowTwo]; 219public: 220 SmallPtrSet() : SmallPtrSetImpl(NextPowerOfTwo<SmallSizePowTwo>::Val) {} 221 SmallPtrSet(const SmallPtrSet &that) : SmallPtrSetImpl(that) {} 222 223 template<typename It> 224 SmallPtrSet(It I, It E) 225 : SmallPtrSetImpl(NextPowerOfTwo<SmallSizePowTwo>::Val) { 226 insert(I, E); 227 } 228 229 typedef SmallPtrSetIterator<PtrType> iterator; 230 typedef SmallPtrSetIterator<PtrType> const_iterator; 231 inline iterator begin() const { 232 return iterator(CurArray); 233 } 234 inline iterator end() const { 235 return iterator(CurArray+CurArraySize); 236 } 237 238 // Allow assignment from any smallptrset with the same element type even if it 239 // doesn't have the same smallsize. 240 const SmallPtrSet<PtrType, SmallSize>& 241 operator=(const SmallPtrSet<PtrType, SmallSize> &RHS) { 242 CopyFrom(RHS); 243 return *this; 244 } 245 246}; 247 248} 249 250#endif 251