• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 <cstdlib>
25 #include <iterator>
26 #include <utility>
27 
28 namespace llvm {
29 
30 class SmallPtrSetIteratorImpl;
31 
32 /// SmallPtrSetImplBase - This is the common code shared among all the
33 /// SmallPtrSet<>'s, which is almost everything.  SmallPtrSet has two modes, one
34 /// for small and one for large sets.
35 ///
36 /// Small sets use an array of pointers allocated in the SmallPtrSet object,
37 /// which is treated as a simple array of pointers.  When a pointer is added to
38 /// the set, the array is scanned to see if the element already exists, if not
39 /// the element is 'pushed back' onto the array.  If we run out of space in the
40 /// array, we grow into the 'large set' case.  SmallSet should be used when the
41 /// sets are often small.  In this case, no memory allocation is used, and only
42 /// light-weight and cache-efficient scanning is used.
43 ///
44 /// Large sets use a classic exponentially-probed hash table.  Empty buckets are
45 /// represented with an illegal pointer value (-1) to allow null pointers to be
46 /// inserted.  Tombstones are represented with another illegal pointer value
47 /// (-2), to allow deletion.  The hash table is resized when the table is 3/4 or
48 /// more.  When this happens, the table is doubled in size.
49 ///
50 class SmallPtrSetImplBase {
51   friend class SmallPtrSetIteratorImpl;
52 
53 protected:
54   /// SmallArray - Points to a fixed size set of buckets, used in 'small mode'.
55   const void **SmallArray;
56   /// CurArray - This is the current set of buckets.  If equal to SmallArray,
57   /// then the set is in 'small mode'.
58   const void **CurArray;
59   /// CurArraySize - The allocated size of CurArray, always a power of two.
60   unsigned CurArraySize;
61 
62   /// Number of elements in CurArray that contain a value or are a tombstone.
63   /// If small, all these elements are at the beginning of CurArray and the rest
64   /// is uninitialized.
65   unsigned NumNonEmpty;
66   /// Number of tombstones in CurArray.
67   unsigned NumTombstones;
68 
69   // Helpers to copy and move construct a SmallPtrSet.
70   SmallPtrSetImplBase(const void **SmallStorage,
71                       const SmallPtrSetImplBase &that);
72   SmallPtrSetImplBase(const void **SmallStorage, unsigned SmallSize,
73                       SmallPtrSetImplBase &&that);
SmallPtrSetImplBase(const void ** SmallStorage,unsigned SmallSize)74   explicit SmallPtrSetImplBase(const void **SmallStorage, unsigned SmallSize)
75       : SmallArray(SmallStorage), CurArray(SmallStorage),
76         CurArraySize(SmallSize), NumNonEmpty(0), NumTombstones(0) {
77     assert(SmallSize && (SmallSize & (SmallSize-1)) == 0 &&
78            "Initial size must be a power of two!");
79   }
~SmallPtrSetImplBase()80   ~SmallPtrSetImplBase() {
81     if (!isSmall())
82       free(CurArray);
83   }
84 
85 public:
86   typedef unsigned size_type;
empty()87   bool LLVM_ATTRIBUTE_UNUSED_RESULT empty() const { return size() == 0; }
size()88   size_type size() const { return NumNonEmpty - NumTombstones; }
89 
clear()90   void clear() {
91     // If the capacity of the array is huge, and the # elements used is small,
92     // shrink the array.
93     if (!isSmall()) {
94       if (size() * 4 < CurArraySize && CurArraySize > 32)
95         return shrink_and_clear();
96       // Fill the array with empty markers.
97       memset(CurArray, -1, CurArraySize * sizeof(void *));
98     }
99 
100     NumNonEmpty = 0;
101     NumTombstones = 0;
102   }
103 
104 protected:
getTombstoneMarker()105   static void *getTombstoneMarker() { return reinterpret_cast<void*>(-2); }
getEmptyMarker()106   static void *getEmptyMarker() {
107     // Note that -1 is chosen to make clear() efficiently implementable with
108     // memset and because it's not a valid pointer value.
109     return reinterpret_cast<void*>(-1);
110   }
111 
EndPointer()112   const void **EndPointer() const {
113     return isSmall() ? CurArray + NumNonEmpty : CurArray + CurArraySize;
114   }
115 
116   /// insert_imp - This returns true if the pointer was new to the set, false if
117   /// it was already in the set.  This is hidden from the client so that the
118   /// derived class can check that the right type of pointer is passed in.
insert_imp(const void * Ptr)119   std::pair<const void *const *, bool> insert_imp(const void *Ptr) {
120     if (isSmall()) {
121       // Check to see if it is already in the set.
122       const void **LastTombstone = nullptr;
123       for (const void **APtr = SmallArray, **E = SmallArray + NumNonEmpty;
124            APtr != E; ++APtr) {
125         const void *Value = *APtr;
126         if (Value == Ptr)
127           return std::make_pair(APtr, false);
128         if (Value == getTombstoneMarker())
129           LastTombstone = APtr;
130       }
131 
132       // Did we find any tombstone marker?
133       if (LastTombstone != nullptr) {
134         *LastTombstone = Ptr;
135         --NumTombstones;
136         return std::make_pair(LastTombstone, true);
137       }
138 
139       // Nope, there isn't.  If we stay small, just 'pushback' now.
140       if (NumNonEmpty < CurArraySize) {
141         SmallArray[NumNonEmpty++] = Ptr;
142         return std::make_pair(SmallArray + (NumNonEmpty - 1), true);
143       }
144       // Otherwise, hit the big set case, which will call grow.
145     }
146     return insert_imp_big(Ptr);
147   }
148 
149   /// erase_imp - If the set contains the specified pointer, remove it and
150   /// return true, otherwise return false.  This is hidden from the client so
151   /// that the derived class can check that the right type of pointer is passed
152   /// in.
153   bool erase_imp(const void * Ptr);
154 
count_imp(const void * Ptr)155   bool count_imp(const void * Ptr) const {
156     if (isSmall()) {
157       // Linear search for the item.
158       for (const void *const *APtr = SmallArray,
159                       *const *E = SmallArray + NumNonEmpty; APtr != E; ++APtr)
160         if (*APtr == Ptr)
161           return true;
162       return false;
163     }
164 
165     // Big set case.
166     return *FindBucketFor(Ptr) == Ptr;
167   }
168 
169 private:
isSmall()170   bool isSmall() const { return CurArray == SmallArray; }
171 
172   std::pair<const void *const *, bool> insert_imp_big(const void *Ptr);
173 
174   const void * const *FindBucketFor(const void *Ptr) const;
175   void shrink_and_clear();
176 
177   /// Grow - Allocate a larger backing store for the buckets and move it over.
178   void Grow(unsigned NewSize);
179 
180   void operator=(const SmallPtrSetImplBase &RHS) = delete;
181 
182 protected:
183   /// swap - Swaps the elements of two sets.
184   /// Note: This method assumes that both sets have the same small size.
185   void swap(SmallPtrSetImplBase &RHS);
186 
187   void CopyFrom(const SmallPtrSetImplBase &RHS);
188   void MoveFrom(unsigned SmallSize, SmallPtrSetImplBase &&RHS);
189 
190 private:
191   /// Code shared by MoveFrom() and move constructor.
192   void MoveHelper(unsigned SmallSize, SmallPtrSetImplBase &&RHS);
193   /// Code shared by CopyFrom() and copy constructor.
194   void CopyHelper(const SmallPtrSetImplBase &RHS);
195 };
196 
197 /// SmallPtrSetIteratorImpl - This is the common base class shared between all
198 /// instances of SmallPtrSetIterator.
199 class SmallPtrSetIteratorImpl {
200 protected:
201   const void *const *Bucket;
202   const void *const *End;
203 
204 public:
SmallPtrSetIteratorImpl(const void * const * BP,const void * const * E)205   explicit SmallPtrSetIteratorImpl(const void *const *BP, const void*const *E)
206     : Bucket(BP), End(E) {
207     AdvanceIfNotValid();
208   }
209 
210   bool operator==(const SmallPtrSetIteratorImpl &RHS) const {
211     return Bucket == RHS.Bucket;
212   }
213   bool operator!=(const SmallPtrSetIteratorImpl &RHS) const {
214     return Bucket != RHS.Bucket;
215   }
216 
217 protected:
218   /// AdvanceIfNotValid - If the current bucket isn't valid, advance to a bucket
219   /// that is.   This is guaranteed to stop because the end() bucket is marked
220   /// valid.
AdvanceIfNotValid()221   void AdvanceIfNotValid() {
222     assert(Bucket <= End);
223     while (Bucket != End &&
224            (*Bucket == SmallPtrSetImplBase::getEmptyMarker() ||
225             *Bucket == SmallPtrSetImplBase::getTombstoneMarker()))
226       ++Bucket;
227   }
228 };
229 
230 /// SmallPtrSetIterator - This implements a const_iterator for SmallPtrSet.
231 template<typename PtrTy>
232 class SmallPtrSetIterator : public SmallPtrSetIteratorImpl {
233   typedef PointerLikeTypeTraits<PtrTy> PtrTraits;
234 
235 public:
236   typedef PtrTy                     value_type;
237   typedef PtrTy                     reference;
238   typedef PtrTy                     pointer;
239   typedef std::ptrdiff_t            difference_type;
240   typedef std::forward_iterator_tag iterator_category;
241 
SmallPtrSetIterator(const void * const * BP,const void * const * E)242   explicit SmallPtrSetIterator(const void *const *BP, const void *const *E)
243     : SmallPtrSetIteratorImpl(BP, E) {}
244 
245   // Most methods provided by baseclass.
246 
247   const PtrTy operator*() const {
248     assert(Bucket < End);
249     return PtrTraits::getFromVoidPointer(const_cast<void*>(*Bucket));
250   }
251 
252   inline SmallPtrSetIterator& operator++() {          // Preincrement
253     ++Bucket;
254     AdvanceIfNotValid();
255     return *this;
256   }
257 
258   SmallPtrSetIterator operator++(int) {        // Postincrement
259     SmallPtrSetIterator tmp = *this; ++*this; return tmp;
260   }
261 };
262 
263 /// RoundUpToPowerOfTwo - This is a helper template that rounds N up to the next
264 /// power of two (which means N itself if N is already a power of two).
265 template<unsigned N>
266 struct RoundUpToPowerOfTwo;
267 
268 /// RoundUpToPowerOfTwoH - If N is not a power of two, increase it.  This is a
269 /// helper template used to implement RoundUpToPowerOfTwo.
270 template<unsigned N, bool isPowerTwo>
271 struct RoundUpToPowerOfTwoH {
272   enum { Val = N };
273 };
274 template<unsigned N>
275 struct RoundUpToPowerOfTwoH<N, false> {
276   enum {
277     // We could just use NextVal = N+1, but this converges faster.  N|(N-1) sets
278     // the right-most zero bits to one all at once, e.g. 0b0011000 -> 0b0011111.
279     Val = RoundUpToPowerOfTwo<(N|(N-1)) + 1>::Val
280   };
281 };
282 
283 template<unsigned N>
284 struct RoundUpToPowerOfTwo {
285   enum { Val = RoundUpToPowerOfTwoH<N, (N&(N-1)) == 0>::Val };
286 };
287 
288 /// \brief A templated base class for \c SmallPtrSet which provides the
289 /// typesafe interface that is common across all small sizes.
290 ///
291 /// This is particularly useful for passing around between interface boundaries
292 /// to avoid encoding a particular small size in the interface boundary.
293 template <typename PtrType>
294 class SmallPtrSetImpl : public SmallPtrSetImplBase {
295   typedef PointerLikeTypeTraits<PtrType> PtrTraits;
296 
297   SmallPtrSetImpl(const SmallPtrSetImpl &) = delete;
298 
299 protected:
300   // Constructors that forward to the base.
301   SmallPtrSetImpl(const void **SmallStorage, const SmallPtrSetImpl &that)
302       : SmallPtrSetImplBase(SmallStorage, that) {}
303   SmallPtrSetImpl(const void **SmallStorage, unsigned SmallSize,
304                   SmallPtrSetImpl &&that)
305       : SmallPtrSetImplBase(SmallStorage, SmallSize, std::move(that)) {}
306   explicit SmallPtrSetImpl(const void **SmallStorage, unsigned SmallSize)
307       : SmallPtrSetImplBase(SmallStorage, SmallSize) {}
308 
309 public:
310   typedef SmallPtrSetIterator<PtrType> iterator;
311   typedef SmallPtrSetIterator<PtrType> const_iterator;
312 
313   /// Inserts Ptr if and only if there is no element in the container equal to
314   /// Ptr. The bool component of the returned pair is true if and only if the
315   /// insertion takes place, and the iterator component of the pair points to
316   /// the element equal to Ptr.
317   std::pair<iterator, bool> insert(PtrType Ptr) {
318     auto p = insert_imp(PtrTraits::getAsVoidPointer(Ptr));
319     return std::make_pair(iterator(p.first, EndPointer()), p.second);
320   }
321 
322   /// erase - If the set contains the specified pointer, remove it and return
323   /// true, otherwise return false.
324   bool erase(PtrType Ptr) {
325     return erase_imp(PtrTraits::getAsVoidPointer(Ptr));
326   }
327 
328   /// count - Return 1 if the specified pointer is in the set, 0 otherwise.
329   size_type count(PtrType Ptr) const {
330     return count_imp(PtrTraits::getAsVoidPointer(Ptr)) ? 1 : 0;
331   }
332 
333   template <typename IterT>
334   void insert(IterT I, IterT E) {
335     for (; I != E; ++I)
336       insert(*I);
337   }
338 
339   inline iterator begin() const {
340     return iterator(CurArray, EndPointer());
341   }
342   inline iterator end() const {
343     const void *const *End = EndPointer();
344     return iterator(End, End);
345   }
346 };
347 
348 /// SmallPtrSet - This class implements a set which is optimized for holding
349 /// SmallSize or less elements.  This internally rounds up SmallSize to the next
350 /// power of two if it is not already a power of two.  See the comments above
351 /// SmallPtrSetImplBase for details of the algorithm.
352 template<class PtrType, unsigned SmallSize>
353 class SmallPtrSet : public SmallPtrSetImpl<PtrType> {
354   // In small mode SmallPtrSet uses linear search for the elements, so it is
355   // not a good idea to choose this value too high. You may consider using a
356   // DenseSet<> instead if you expect many elements in the set.
357   static_assert(SmallSize <= 32, "SmallSize should be small");
358 
359   typedef SmallPtrSetImpl<PtrType> BaseT;
360 
361   // Make sure that SmallSize is a power of two, round up if not.
362   enum { SmallSizePowTwo = RoundUpToPowerOfTwo<SmallSize>::Val };
363   /// SmallStorage - Fixed size storage used in 'small mode'.
364   const void *SmallStorage[SmallSizePowTwo];
365 
366 public:
367   SmallPtrSet() : BaseT(SmallStorage, SmallSizePowTwo) {}
368   SmallPtrSet(const SmallPtrSet &that) : BaseT(SmallStorage, that) {}
369   SmallPtrSet(SmallPtrSet &&that)
370       : BaseT(SmallStorage, SmallSizePowTwo, std::move(that)) {}
371 
372   template<typename It>
373   SmallPtrSet(It I, It E) : BaseT(SmallStorage, SmallSizePowTwo) {
374     this->insert(I, E);
375   }
376 
377   SmallPtrSet<PtrType, SmallSize> &
378   operator=(const SmallPtrSet<PtrType, SmallSize> &RHS) {
379     if (&RHS != this)
380       this->CopyFrom(RHS);
381     return *this;
382   }
383 
384   SmallPtrSet<PtrType, SmallSize>&
385   operator=(SmallPtrSet<PtrType, SmallSize> &&RHS) {
386     if (&RHS != this)
387       this->MoveFrom(SmallSizePowTwo, std::move(RHS));
388     return *this;
389   }
390 
391   /// swap - Swaps the elements of two sets.
392   void swap(SmallPtrSet<PtrType, SmallSize> &RHS) {
393     SmallPtrSetImplBase::swap(RHS);
394   }
395 };
396 }
397 
398 namespace std {
399   /// Implement std::swap in terms of SmallPtrSet swap.
400   template<class T, unsigned N>
401   inline void swap(llvm::SmallPtrSet<T, N> &LHS, llvm::SmallPtrSet<T, N> &RHS) {
402     LHS.swap(RHS);
403   }
404 }
405 
406 #endif
407