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   unsigned CurArraySize;
     58 
     59   // If small, this is # elts allocated consecutively
     60   unsigned NumElements;
     61   unsigned NumTombstones;
     62 
     63   // Helper to copy construct a SmallPtrSet.
     64   SmallPtrSetImpl(const void **SmallStorage, const SmallPtrSetImpl& that);
     65   explicit SmallPtrSetImpl(const void **SmallStorage, unsigned SmallSize) :
     66     SmallArray(SmallStorage), CurArray(SmallStorage), CurArraySize(SmallSize) {
     67     assert(SmallSize && (SmallSize & (SmallSize-1)) == 0 &&
     68            "Initial size must be a power of two!");
     69     clear();
     70   }
     71   ~SmallPtrSetImpl();
     72 
     73 public:
     74   bool empty() const { return size() == 0; }
     75   unsigned size() const { return NumElements; }
     76 
     77   void clear() {
     78     // If the capacity of the array is huge, and the # elements used is small,
     79     // shrink the array.
     80     if (!isSmall() && NumElements*4 < CurArraySize && CurArraySize > 32)
     81       return shrink_and_clear();
     82 
     83     // Fill the array with empty markers.
     84     memset(CurArray, -1, CurArraySize*sizeof(void*));
     85     NumElements = 0;
     86     NumTombstones = 0;
     87   }
     88 
     89 protected:
     90   static void *getTombstoneMarker() { return reinterpret_cast<void*>(-2); }
     91   static void *getEmptyMarker() {
     92     // Note that -1 is chosen to make clear() efficiently implementable with
     93     // memset and because it's not a valid pointer value.
     94     return reinterpret_cast<void*>(-1);
     95   }
     96 
     97   /// insert_imp - This returns true if the pointer was new to the set, false if
     98   /// it was already in the set.  This is hidden from the client so that the
     99   /// derived class can check that the right type of pointer is passed in.
    100   bool insert_imp(const void * Ptr);
    101 
    102   /// erase_imp - If the set contains the specified pointer, remove it and
    103   /// return true, otherwise return false.  This is hidden from the client so
    104   /// that the derived class can check that the right type of pointer is passed
    105   /// in.
    106   bool erase_imp(const void * Ptr);
    107 
    108   bool count_imp(const void * Ptr) const {
    109     if (isSmall()) {
    110       // Linear search for the item.
    111       for (const void *const *APtr = SmallArray,
    112                       *const *E = SmallArray+NumElements; APtr != E; ++APtr)
    113         if (*APtr == Ptr)
    114           return true;
    115       return false;
    116     }
    117 
    118     // Big set case.
    119     return *FindBucketFor(Ptr) == Ptr;
    120   }
    121 
    122 private:
    123   bool isSmall() const { return CurArray == SmallArray; }
    124 
    125   const void * const *FindBucketFor(const void *Ptr) const;
    126   void shrink_and_clear();
    127 
    128   /// Grow - Allocate a larger backing store for the buckets and move it over.
    129   void Grow(unsigned NewSize);
    130 
    131   void operator=(const SmallPtrSetImpl &RHS) LLVM_DELETED_FUNCTION;
    132 protected:
    133   /// swap - Swaps the elements of two sets.
    134   /// Note: This method assumes that both sets have the same small size.
    135   void swap(SmallPtrSetImpl &RHS);
    136 
    137   void CopyFrom(const SmallPtrSetImpl &RHS);
    138 };
    139 
    140 /// SmallPtrSetIteratorImpl - This is the common base class shared between all
    141 /// instances of SmallPtrSetIterator.
    142 class SmallPtrSetIteratorImpl {
    143 protected:
    144   const void *const *Bucket;
    145   const void *const *End;
    146 public:
    147   explicit SmallPtrSetIteratorImpl(const void *const *BP, const void*const *E)
    148     : Bucket(BP), End(E) {
    149       AdvanceIfNotValid();
    150   }
    151 
    152   bool operator==(const SmallPtrSetIteratorImpl &RHS) const {
    153     return Bucket == RHS.Bucket;
    154   }
    155   bool operator!=(const SmallPtrSetIteratorImpl &RHS) const {
    156     return Bucket != RHS.Bucket;
    157   }
    158 
    159 protected:
    160   /// AdvanceIfNotValid - If the current bucket isn't valid, advance to a bucket
    161   /// that is.   This is guaranteed to stop because the end() bucket is marked
    162   /// valid.
    163   void AdvanceIfNotValid() {
    164     assert(Bucket <= End);
    165     while (Bucket != End &&
    166            (*Bucket == SmallPtrSetImpl::getEmptyMarker() ||
    167             *Bucket == SmallPtrSetImpl::getTombstoneMarker()))
    168       ++Bucket;
    169   }
    170 };
    171 
    172 /// SmallPtrSetIterator - This implements a const_iterator for SmallPtrSet.
    173 template<typename PtrTy>
    174 class SmallPtrSetIterator : public SmallPtrSetIteratorImpl {
    175   typedef PointerLikeTypeTraits<PtrTy> PtrTraits;
    176 
    177 public:
    178   typedef PtrTy                     value_type;
    179   typedef PtrTy                     reference;
    180   typedef PtrTy                     pointer;
    181   typedef std::ptrdiff_t            difference_type;
    182   typedef std::forward_iterator_tag iterator_category;
    183 
    184   explicit SmallPtrSetIterator(const void *const *BP, const void *const *E)
    185     : SmallPtrSetIteratorImpl(BP, E) {}
    186 
    187   // Most methods provided by baseclass.
    188 
    189   const PtrTy operator*() const {
    190     assert(Bucket < End);
    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'.
    240   const void *SmallStorage[SmallSizePowTwo];
    241   typedef PointerLikeTypeTraits<PtrType> PtrTraits;
    242 public:
    243   SmallPtrSet() : SmallPtrSetImpl(SmallStorage, SmallSizePowTwo) {}
    244   SmallPtrSet(const SmallPtrSet &that) : SmallPtrSetImpl(SmallStorage, that) {}
    245 
    246   template<typename It>
    247   SmallPtrSet(It I, It E) : SmallPtrSetImpl(SmallStorage, SmallSizePowTwo) {
    248     insert(I, E);
    249   }
    250 
    251   /// insert - This returns true if the pointer was new to the set, false if it
    252   /// was already in the set.
    253   bool insert(PtrType Ptr) {
    254     return insert_imp(PtrTraits::getAsVoidPointer(Ptr));
    255   }
    256 
    257   /// erase - If the set contains the specified pointer, remove it and return
    258   /// true, otherwise return false.
    259   bool erase(PtrType Ptr) {
    260     return erase_imp(PtrTraits::getAsVoidPointer(Ptr));
    261   }
    262 
    263   /// count - Return true if the specified pointer is in the set.
    264   bool count(PtrType Ptr) const {
    265     return count_imp(PtrTraits::getAsVoidPointer(Ptr));
    266   }
    267 
    268   template <typename IterT>
    269   void insert(IterT I, IterT E) {
    270     for (; I != E; ++I)
    271       insert(*I);
    272   }
    273 
    274   typedef SmallPtrSetIterator<PtrType> iterator;
    275   typedef SmallPtrSetIterator<PtrType> const_iterator;
    276   inline iterator begin() const {
    277     return iterator(CurArray, CurArray+CurArraySize);
    278   }
    279   inline iterator end() const {
    280     return iterator(CurArray+CurArraySize, CurArray+CurArraySize);
    281   }
    282 
    283   // Allow assignment from any smallptrset with the same element type even if it
    284   // doesn't have the same smallsize.
    285   const SmallPtrSet<PtrType, SmallSize>&
    286   operator=(const SmallPtrSet<PtrType, SmallSize> &RHS) {
    287     CopyFrom(RHS);
    288     return *this;
    289   }
    290 
    291   /// swap - Swaps the elements of two sets.
    292   void swap(SmallPtrSet<PtrType, SmallSize> &RHS) {
    293     SmallPtrSetImpl::swap(RHS);
    294   }
    295 };
    296 
    297 }
    298 
    299 namespace std {
    300   /// Implement std::swap in terms of SmallPtrSet swap.
    301   template<class T, unsigned N>
    302   inline void swap(llvm::SmallPtrSet<T, N> &LHS, llvm::SmallPtrSet<T, N> &RHS) {
    303     LHS.swap(RHS);
    304   }
    305 }
    306 
    307 #endif
    308