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 // SmallPtrSetImplBase 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 /// SmallPtrSetImplBase - 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 SmallPtrSetImplBase {
     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   // Helpers to copy and move construct a SmallPtrSet.
     64   SmallPtrSetImplBase(const void **SmallStorage, const SmallPtrSetImplBase &that);
     65   SmallPtrSetImplBase(const void **SmallStorage, unsigned SmallSize,
     66                   SmallPtrSetImplBase &&that);
     67   explicit SmallPtrSetImplBase(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     clear();
     72   }
     73   ~SmallPtrSetImplBase();
     74 
     75 public:
     76   typedef unsigned size_type;
     77   bool LLVM_ATTRIBUTE_UNUSED_RESULT empty() const { return size() == 0; }
     78   size_type size() const { return NumElements; }
     79 
     80   void clear() {
     81     // If the capacity of the array is huge, and the # elements used is small,
     82     // shrink the array.
     83     if (!isSmall() && NumElements*4 < CurArraySize && CurArraySize > 32)
     84       return shrink_and_clear();
     85 
     86     // Fill the array with empty markers.
     87     memset(CurArray, -1, CurArraySize*sizeof(void*));
     88     NumElements = 0;
     89     NumTombstones = 0;
     90   }
     91 
     92 protected:
     93   static void *getTombstoneMarker() { return reinterpret_cast<void*>(-2); }
     94   static void *getEmptyMarker() {
     95     // Note that -1 is chosen to make clear() efficiently implementable with
     96     // memset and because it's not a valid pointer value.
     97     return reinterpret_cast<void*>(-1);
     98   }
     99 
    100   /// insert_imp - This returns true if the pointer was new to the set, false if
    101   /// it was already in the set.  This is hidden from the client so that the
    102   /// derived class can check that the right type of pointer is passed in.
    103   bool insert_imp(const void * Ptr);
    104 
    105   /// erase_imp - If the set contains the specified pointer, remove it and
    106   /// return true, otherwise return false.  This is hidden from the client so
    107   /// that the derived class can check that the right type of pointer is passed
    108   /// in.
    109   bool erase_imp(const void * Ptr);
    110 
    111   bool count_imp(const void * Ptr) const {
    112     if (isSmall()) {
    113       // Linear search for the item.
    114       for (const void *const *APtr = SmallArray,
    115                       *const *E = SmallArray+NumElements; APtr != E; ++APtr)
    116         if (*APtr == Ptr)
    117           return true;
    118       return false;
    119     }
    120 
    121     // Big set case.
    122     return *FindBucketFor(Ptr) == Ptr;
    123   }
    124 
    125 private:
    126   bool isSmall() const { return CurArray == SmallArray; }
    127 
    128   const void * const *FindBucketFor(const void *Ptr) const;
    129   void shrink_and_clear();
    130 
    131   /// Grow - Allocate a larger backing store for the buckets and move it over.
    132   void Grow(unsigned NewSize);
    133 
    134   void operator=(const SmallPtrSetImplBase &RHS) LLVM_DELETED_FUNCTION;
    135 protected:
    136   /// swap - Swaps the elements of two sets.
    137   /// Note: This method assumes that both sets have the same small size.
    138   void swap(SmallPtrSetImplBase &RHS);
    139 
    140   void CopyFrom(const SmallPtrSetImplBase &RHS);
    141   void MoveFrom(unsigned SmallSize, SmallPtrSetImplBase &&RHS);
    142 };
    143 
    144 /// SmallPtrSetIteratorImpl - This is the common base class shared between all
    145 /// instances of SmallPtrSetIterator.
    146 class SmallPtrSetIteratorImpl {
    147 protected:
    148   const void *const *Bucket;
    149   const void *const *End;
    150 public:
    151   explicit SmallPtrSetIteratorImpl(const void *const *BP, const void*const *E)
    152     : Bucket(BP), End(E) {
    153       AdvanceIfNotValid();
    154   }
    155 
    156   bool operator==(const SmallPtrSetIteratorImpl &RHS) const {
    157     return Bucket == RHS.Bucket;
    158   }
    159   bool operator!=(const SmallPtrSetIteratorImpl &RHS) const {
    160     return Bucket != RHS.Bucket;
    161   }
    162 
    163 protected:
    164   /// AdvanceIfNotValid - If the current bucket isn't valid, advance to a bucket
    165   /// that is.   This is guaranteed to stop because the end() bucket is marked
    166   /// valid.
    167   void AdvanceIfNotValid() {
    168     assert(Bucket <= End);
    169     while (Bucket != End &&
    170            (*Bucket == SmallPtrSetImplBase::getEmptyMarker() ||
    171             *Bucket == SmallPtrSetImplBase::getTombstoneMarker()))
    172       ++Bucket;
    173   }
    174 };
    175 
    176 /// SmallPtrSetIterator - This implements a const_iterator for SmallPtrSet.
    177 template<typename PtrTy>
    178 class SmallPtrSetIterator : public SmallPtrSetIteratorImpl {
    179   typedef PointerLikeTypeTraits<PtrTy> PtrTraits;
    180 
    181 public:
    182   typedef PtrTy                     value_type;
    183   typedef PtrTy                     reference;
    184   typedef PtrTy                     pointer;
    185   typedef std::ptrdiff_t            difference_type;
    186   typedef std::forward_iterator_tag iterator_category;
    187 
    188   explicit SmallPtrSetIterator(const void *const *BP, const void *const *E)
    189     : SmallPtrSetIteratorImpl(BP, E) {}
    190 
    191   // Most methods provided by baseclass.
    192 
    193   const PtrTy operator*() const {
    194     assert(Bucket < End);
    195     return PtrTraits::getFromVoidPointer(const_cast<void*>(*Bucket));
    196   }
    197 
    198   inline SmallPtrSetIterator& operator++() {          // Preincrement
    199     ++Bucket;
    200     AdvanceIfNotValid();
    201     return *this;
    202   }
    203 
    204   SmallPtrSetIterator operator++(int) {        // Postincrement
    205     SmallPtrSetIterator tmp = *this; ++*this; return tmp;
    206   }
    207 };
    208 
    209 /// RoundUpToPowerOfTwo - This is a helper template that rounds N up to the next
    210 /// power of two (which means N itself if N is already a power of two).
    211 template<unsigned N>
    212 struct RoundUpToPowerOfTwo;
    213 
    214 /// RoundUpToPowerOfTwoH - If N is not a power of two, increase it.  This is a
    215 /// helper template used to implement RoundUpToPowerOfTwo.
    216 template<unsigned N, bool isPowerTwo>
    217 struct RoundUpToPowerOfTwoH {
    218   enum { Val = N };
    219 };
    220 template<unsigned N>
    221 struct RoundUpToPowerOfTwoH<N, false> {
    222   enum {
    223     // We could just use NextVal = N+1, but this converges faster.  N|(N-1) sets
    224     // the right-most zero bits to one all at once, e.g. 0b0011000 -> 0b0011111.
    225     Val = RoundUpToPowerOfTwo<(N|(N-1)) + 1>::Val
    226   };
    227 };
    228 
    229 template<unsigned N>
    230 struct RoundUpToPowerOfTwo {
    231   enum { Val = RoundUpToPowerOfTwoH<N, (N&(N-1)) == 0>::Val };
    232 };
    233 
    234 
    235 /// \brief A templated base class for \c SmallPtrSet which provides the
    236 /// typesafe interface that is common across all small sizes.
    237 ///
    238 /// This is particularly useful for passing around between interface boundaries
    239 /// to avoid encoding a particular small size in the interface boundary.
    240 template <typename PtrType>
    241 class SmallPtrSetImpl : public SmallPtrSetImplBase {
    242   typedef PointerLikeTypeTraits<PtrType> PtrTraits;
    243 protected:
    244   // Constructors that forward to the base.
    245   SmallPtrSetImpl(const void **SmallStorage, const SmallPtrSetImpl &that)
    246       : SmallPtrSetImplBase(SmallStorage, that) {}
    247   SmallPtrSetImpl(const void **SmallStorage, unsigned SmallSize,
    248                   SmallPtrSetImpl &&that)
    249       : SmallPtrSetImplBase(SmallStorage, SmallSize, std::move(that)) {}
    250   explicit SmallPtrSetImpl(const void **SmallStorage, unsigned SmallSize)
    251       : SmallPtrSetImplBase(SmallStorage, SmallSize) {}
    252 
    253 public:
    254   /// insert - This returns true if the pointer was new to the set, false if it
    255   /// was already in the set.
    256   bool insert(PtrType Ptr) {
    257     return insert_imp(PtrTraits::getAsVoidPointer(Ptr));
    258   }
    259 
    260   /// erase - If the set contains the specified pointer, remove it and return
    261   /// true, otherwise return false.
    262   bool erase(PtrType Ptr) {
    263     return erase_imp(PtrTraits::getAsVoidPointer(Ptr));
    264   }
    265 
    266   /// count - Return 1 if the specified pointer is in the set, 0 otherwise.
    267   size_type count(PtrType Ptr) const {
    268     return count_imp(PtrTraits::getAsVoidPointer(Ptr)) ? 1 : 0;
    269   }
    270 
    271   template <typename IterT>
    272   void insert(IterT I, IterT E) {
    273     for (; I != E; ++I)
    274       insert(*I);
    275   }
    276 
    277   typedef SmallPtrSetIterator<PtrType> iterator;
    278   typedef SmallPtrSetIterator<PtrType> const_iterator;
    279   inline iterator begin() const {
    280     return iterator(CurArray, CurArray+CurArraySize);
    281   }
    282   inline iterator end() const {
    283     return iterator(CurArray+CurArraySize, CurArray+CurArraySize);
    284   }
    285 };
    286 
    287 /// SmallPtrSet - This class implements a set which is optimized for holding
    288 /// SmallSize or less elements.  This internally rounds up SmallSize to the next
    289 /// power of two if it is not already a power of two.  See the comments above
    290 /// SmallPtrSetImplBase for details of the algorithm.
    291 template<class PtrType, unsigned SmallSize>
    292 class SmallPtrSet : public SmallPtrSetImpl<PtrType> {
    293   typedef SmallPtrSetImpl<PtrType> BaseT;
    294 
    295   // Make sure that SmallSize is a power of two, round up if not.
    296   enum { SmallSizePowTwo = RoundUpToPowerOfTwo<SmallSize>::Val };
    297   /// SmallStorage - Fixed size storage used in 'small mode'.
    298   const void *SmallStorage[SmallSizePowTwo];
    299 public:
    300   SmallPtrSet() : BaseT(SmallStorage, SmallSizePowTwo) {}
    301   SmallPtrSet(const SmallPtrSet &that) : BaseT(SmallStorage, that) {}
    302   SmallPtrSet(SmallPtrSet &&that)
    303       : BaseT(SmallStorage, SmallSizePowTwo, std::move(that)) {}
    304 
    305   template<typename It>
    306   SmallPtrSet(It I, It E) : BaseT(SmallStorage, SmallSizePowTwo) {
    307     this->insert(I, E);
    308   }
    309 
    310   SmallPtrSet<PtrType, SmallSize> &
    311   operator=(const SmallPtrSet<PtrType, SmallSize> &RHS) {
    312     if (&RHS != this)
    313       this->CopyFrom(RHS);
    314     return *this;
    315   }
    316 
    317   SmallPtrSet<PtrType, SmallSize>&
    318   operator=(SmallPtrSet<PtrType, SmallSize> &&RHS) {
    319     if (&RHS != this)
    320       this->MoveFrom(SmallSizePowTwo, std::move(RHS));
    321     return *this;
    322   }
    323 
    324   /// swap - Swaps the elements of two sets.
    325   void swap(SmallPtrSet<PtrType, SmallSize> &RHS) {
    326     SmallPtrSetImplBase::swap(RHS);
    327   }
    328 };
    329 
    330 }
    331 
    332 namespace std {
    333   /// Implement std::swap in terms of SmallPtrSet swap.
    334   template<class T, unsigned N>
    335   inline void swap(llvm::SmallPtrSet<T, N> &LHS, llvm::SmallPtrSet<T, N> &RHS) {
    336     LHS.swap(RHS);
    337   }
    338 }
    339 
    340 #endif
    341