Home | History | Annotate | Download | only in ADT
      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 "llvm/Support/Compiler.h"
     19 #include "llvm/Support/DataTypes.h"
     20 #include "llvm/Support/PointerLikeTypeTraits.h"
     21 #include <cassert>
     22 #include <cstddef>
     23 #include <cstring>
     24 #include <iterator>
     25 
     26 namespace llvm {
     27 
     28 class SmallPtrSetIteratorImpl;
     29 
     30 /// SmallPtrSetImpl - This is the common code shared among all the
     31 /// SmallPtrSet<>'s, which is almost everything.  SmallPtrSet has two modes, one
     32 /// for small and one for large sets.
     33 ///
     34 /// Small sets use an array of pointers allocated in the SmallPtrSet object,
     35 /// which is treated as a simple array of pointers.  When a pointer is added to
     36 /// the set, the array is scanned to see if the element already exists, if not
     37 /// the element is 'pushed back' onto the array.  If we run out of space in the
     38 /// array, we grow into the 'large set' case.  SmallSet should be used when the
     39 /// sets are often small.  In this case, no memory allocation is used, and only
     40 /// light-weight and cache-efficient scanning is used.
     41 ///
     42 /// Large sets use a classic exponentially-probed hash table.  Empty buckets are
     43 /// represented with an illegal pointer value (-1) to allow null pointers to be
     44 /// inserted.  Tombstones are represented with another illegal pointer value
     45 /// (-2), to allow deletion.  The hash table is resized when the table is 3/4 or
     46 /// more.  When this happens, the table is doubled in size.
     47 ///
     48 class SmallPtrSetImpl {
     49   friend class SmallPtrSetIteratorImpl;
     50 protected:
     51   /// SmallArray - Points to a fixed size set of buckets, used in 'small mode'.
     52   const void **SmallArray;
     53   /// CurArray - This is the current set of buckets.  If equal to SmallArray,
     54   /// then the set is in 'small mode'.
     55   const void **CurArray;
     56   /// CurArraySize - The allocated size of CurArray, always a power of two.
     57   /// Note that CurArray points to an array that has CurArraySize+1 elements in
     58   /// it, so that the end iterator actually points to valid memory.
     59   unsigned CurArraySize;
     60 
     61   // If small, this is # elts allocated consecutively
     62   unsigned NumElements;
     63   unsigned NumTombstones;
     64 
     65   // Helper to copy construct a SmallPtrSet.
     66   SmallPtrSetImpl(const void **SmallStorage, const SmallPtrSetImpl& that);
     67   explicit SmallPtrSetImpl(const void **SmallStorage, unsigned SmallSize) :
     68     SmallArray(SmallStorage), CurArray(SmallStorage), CurArraySize(SmallSize) {
     69     assert(SmallSize && (SmallSize & (SmallSize-1)) == 0 &&
     70            "Initial size must be a power of two!");
     71     // The end pointer, always valid, is set to a valid element to help the
     72     // iterator.
     73     CurArray[SmallSize] = 0;
     74     clear();
     75   }
     76   ~SmallPtrSetImpl();
     77 
     78 public:
     79   bool empty() const { return size() == 0; }
     80   unsigned size() const { return NumElements; }
     81 
     82   void clear() {
     83     // If the capacity of the array is huge, and the # elements used is small,
     84     // shrink the array.
     85     if (!isSmall() && NumElements*4 < CurArraySize && CurArraySize > 32)
     86       return shrink_and_clear();
     87 
     88     // Fill the array with empty markers.
     89     memset(CurArray, -1, CurArraySize*sizeof(void*));
     90     NumElements = 0;
     91     NumTombstones = 0;
     92   }
     93 
     94 protected:
     95   static void *getTombstoneMarker() { return reinterpret_cast<void*>(-2); }
     96   static void *getEmptyMarker() {
     97     // Note that -1 is chosen to make clear() efficiently implementable with
     98     // memset and because it's not a valid pointer value.
     99     return reinterpret_cast<void*>(-1);
    100   }
    101 
    102   /// insert_imp - This returns true if the pointer was new to the set, false if
    103   /// it was already in the set.  This is hidden from the client so that the
    104   /// derived class can check that the right type of pointer is passed in.
    105   bool insert_imp(const void * Ptr);
    106 
    107   /// erase_imp - If the set contains the specified pointer, remove it and
    108   /// return true, otherwise return false.  This is hidden from the client so
    109   /// that the derived class can check that the right type of pointer is passed
    110   /// in.
    111   bool erase_imp(const void * Ptr);
    112 
    113   bool count_imp(const void * Ptr) const {
    114     if (isSmall()) {
    115       // Linear search for the item.
    116       for (const void *const *APtr = SmallArray,
    117                       *const *E = SmallArray+NumElements; APtr != E; ++APtr)
    118         if (*APtr == Ptr)
    119           return true;
    120       return false;
    121     }
    122 
    123     // Big set case.
    124     return *FindBucketFor(Ptr) == Ptr;
    125   }
    126 
    127 private:
    128   bool isSmall() const { return CurArray == SmallArray; }
    129 
    130   const void * const *FindBucketFor(const void *Ptr) const;
    131   void shrink_and_clear();
    132 
    133   /// Grow - Allocate a larger backing store for the buckets and move it over.
    134   void Grow(unsigned NewSize);
    135 
    136   void operator=(const SmallPtrSetImpl &RHS) LLVM_DELETED_FUNCTION;
    137 protected:
    138   /// swap - Swaps the elements of two sets.
    139   /// Note: This method assumes that both sets have the same small size.
    140   void swap(SmallPtrSetImpl &RHS);
    141 
    142   void CopyFrom(const SmallPtrSetImpl &RHS);
    143 };
    144 
    145 /// SmallPtrSetIteratorImpl - This is the common base class shared between all
    146 /// instances of SmallPtrSetIterator.
    147 class SmallPtrSetIteratorImpl {
    148 protected:
    149   const void *const *Bucket;
    150 public:
    151   explicit SmallPtrSetIteratorImpl(const void *const *BP) : Bucket(BP) {
    152     AdvanceIfNotValid();
    153   }
    154 
    155   bool operator==(const SmallPtrSetIteratorImpl &RHS) const {
    156     return Bucket == RHS.Bucket;
    157   }
    158   bool operator!=(const SmallPtrSetIteratorImpl &RHS) const {
    159     return Bucket != RHS.Bucket;
    160   }
    161 
    162 protected:
    163   /// AdvanceIfNotValid - If the current bucket isn't valid, advance to a bucket
    164   /// that is.   This is guaranteed to stop because the end() bucket is marked
    165   /// valid.
    166   void AdvanceIfNotValid() {
    167     while (*Bucket == SmallPtrSetImpl::getEmptyMarker() ||
    168            *Bucket == SmallPtrSetImpl::getTombstoneMarker())
    169       ++Bucket;
    170   }
    171 };
    172 
    173 /// SmallPtrSetIterator - This implements a const_iterator for SmallPtrSet.
    174 template<typename PtrTy>
    175 class SmallPtrSetIterator : public SmallPtrSetIteratorImpl {
    176   typedef PointerLikeTypeTraits<PtrTy> PtrTraits;
    177 
    178 public:
    179   typedef PtrTy                     value_type;
    180   typedef PtrTy                     reference;
    181   typedef PtrTy                     pointer;
    182   typedef std::ptrdiff_t            difference_type;
    183   typedef std::forward_iterator_tag iterator_category;
    184 
    185   explicit SmallPtrSetIterator(const void *const *BP)
    186     : SmallPtrSetIteratorImpl(BP) {}
    187 
    188   // Most methods provided by baseclass.
    189 
    190   const PtrTy operator*() const {
    191     return PtrTraits::getFromVoidPointer(const_cast<void*>(*Bucket));
    192   }
    193 
    194   inline SmallPtrSetIterator& operator++() {          // Preincrement
    195     ++Bucket;
    196     AdvanceIfNotValid();
    197     return *this;
    198   }
    199 
    200   SmallPtrSetIterator operator++(int) {        // Postincrement
    201     SmallPtrSetIterator tmp = *this; ++*this; return tmp;
    202   }
    203 };
    204 
    205 /// RoundUpToPowerOfTwo - This is a helper template that rounds N up to the next
    206 /// power of two (which means N itself if N is already a power of two).
    207 template<unsigned N>
    208 struct RoundUpToPowerOfTwo;
    209 
    210 /// RoundUpToPowerOfTwoH - If N is not a power of two, increase it.  This is a
    211 /// helper template used to implement RoundUpToPowerOfTwo.
    212 template<unsigned N, bool isPowerTwo>
    213 struct RoundUpToPowerOfTwoH {
    214   enum { Val = N };
    215 };
    216 template<unsigned N>
    217 struct RoundUpToPowerOfTwoH<N, false> {
    218   enum {
    219     // We could just use NextVal = N+1, but this converges faster.  N|(N-1) sets
    220     // the right-most zero bits to one all at once, e.g. 0b0011000 -> 0b0011111.
    221     Val = RoundUpToPowerOfTwo<(N|(N-1)) + 1>::Val
    222   };
    223 };
    224 
    225 template<unsigned N>
    226 struct RoundUpToPowerOfTwo {
    227   enum { Val = RoundUpToPowerOfTwoH<N, (N&(N-1)) == 0>::Val };
    228 };
    229 
    230 
    231 /// SmallPtrSet - This class implements a set which is optimized for holding
    232 /// SmallSize or less elements.  This internally rounds up SmallSize to the next
    233 /// power of two if it is not already a power of two.  See the comments above
    234 /// SmallPtrSetImpl for details of the algorithm.
    235 template<class PtrType, unsigned SmallSize>
    236 class SmallPtrSet : public SmallPtrSetImpl {
    237   // Make sure that SmallSize is a power of two, round up if not.
    238   enum { SmallSizePowTwo = RoundUpToPowerOfTwo<SmallSize>::Val };
    239   /// SmallStorage - Fixed size storage used in 'small mode'.  The extra element
    240   /// ensures that the end iterator actually points to valid memory.
    241   const void *SmallStorage[SmallSizePowTwo+1];
    242   typedef PointerLikeTypeTraits<PtrType> PtrTraits;
    243 public:
    244   SmallPtrSet() : SmallPtrSetImpl(SmallStorage, SmallSizePowTwo) {}
    245   SmallPtrSet(const SmallPtrSet &that) : SmallPtrSetImpl(SmallStorage, that) {}
    246 
    247   template<typename It>
    248   SmallPtrSet(It I, It E) : SmallPtrSetImpl(SmallStorage, SmallSizePowTwo) {
    249     insert(I, E);
    250   }
    251 
    252   /// insert - This returns true if the pointer was new to the set, false if it
    253   /// was already in the set.
    254   bool insert(PtrType Ptr) {
    255     return insert_imp(PtrTraits::getAsVoidPointer(Ptr));
    256   }
    257 
    258   /// erase - If the set contains the specified pointer, remove it and return
    259   /// true, otherwise return false.
    260   bool erase(PtrType Ptr) {
    261     return erase_imp(PtrTraits::getAsVoidPointer(Ptr));
    262   }
    263 
    264   /// count - Return true if the specified pointer is in the set.
    265   bool count(PtrType Ptr) const {
    266     return count_imp(PtrTraits::getAsVoidPointer(Ptr));
    267   }
    268 
    269   template <typename IterT>
    270   void insert(IterT I, IterT E) {
    271     for (; I != E; ++I)
    272       insert(*I);
    273   }
    274 
    275   typedef SmallPtrSetIterator<PtrType> iterator;
    276   typedef SmallPtrSetIterator<PtrType> const_iterator;
    277   inline iterator begin() const {
    278     return iterator(CurArray);
    279   }
    280   inline iterator end() const {
    281     return iterator(CurArray+CurArraySize);
    282   }
    283 
    284   // Allow assignment from any smallptrset with the same element type even if it
    285   // doesn't have the same smallsize.
    286   const SmallPtrSet<PtrType, SmallSize>&
    287   operator=(const SmallPtrSet<PtrType, SmallSize> &RHS) {
    288     CopyFrom(RHS);
    289     return *this;
    290   }
    291 
    292   /// swap - Swaps the elements of two sets.
    293   void swap(SmallPtrSet<PtrType, SmallSize> &RHS) {
    294     SmallPtrSetImpl::swap(RHS);
    295   }
    296 };
    297 
    298 }
    299 
    300 namespace std {
    301   /// Implement std::swap in terms of SmallPtrSet swap.
    302   template<class T, unsigned N>
    303   inline void swap(llvm::SmallPtrSet<T, N> &LHS, llvm::SmallPtrSet<T, N> &RHS) {
    304     LHS.swap(RHS);
    305   }
    306 }
    307 
    308 #endif
    309