1 //===- llvm/ADT/SmallPtrSet.h - 'Normally small' pointer set ----*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file defines the SmallPtrSet class. See the doxygen comment for
11 // SmallPtrSetImplBase for more details on the algorithm used.
13 //===----------------------------------------------------------------------===//
15 #ifndef LLVM_ADT_SMALLPTRSET_H
16 #define LLVM_ADT_SMALLPTRSET_H
18 #include "llvm/Support/Compiler.h"
19 #include "llvm/Support/PointerLikeTypeTraits.h"
20 #include "llvm/Support/ReverseIteration.h"
21 #include "llvm/Support/type_traits.h"
26 #include <initializer_list>
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.
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.
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.
50 class SmallPtrSetImplBase {
51 friend class SmallPtrSetIteratorImpl;
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;
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
66 /// Number of tombstones in CurArray.
67 unsigned NumTombstones;
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);
75 explicit SmallPtrSetImplBase(const void **SmallStorage, unsigned SmallSize)
76 : SmallArray(SmallStorage), CurArray(SmallStorage),
77 CurArraySize(SmallSize), NumNonEmpty(0), NumTombstones(0) {
78 assert(SmallSize && (SmallSize & (SmallSize-1)) == 0 &&
79 "Initial size must be a power of two!");
82 ~SmallPtrSetImplBase() {
88 using size_type = unsigned;
90 SmallPtrSetImplBase &operator=(const SmallPtrSetImplBase &) = delete;
92 LLVM_NODISCARD bool empty() const { return size() == 0; }
93 size_type size() const { return NumNonEmpty - NumTombstones; }
96 // If the capacity of the array is huge, and the # elements used is small,
99 if (size() * 4 < CurArraySize && CurArraySize > 32)
100 return shrink_and_clear();
101 // Fill the array with empty markers.
102 memset(CurArray, -1, CurArraySize * sizeof(void *));
110 static void *getTombstoneMarker() { return reinterpret_cast<void*>(-2); }
112 static void *getEmptyMarker() {
113 // Note that -1 is chosen to make clear() efficiently implementable with
114 // memset and because it's not a valid pointer value.
115 return reinterpret_cast<void*>(-1);
118 const void **EndPointer() const {
119 return isSmall() ? CurArray + NumNonEmpty : CurArray + CurArraySize;
122 /// insert_imp - This returns true if the pointer was new to the set, false if
123 /// it was already in the set. This is hidden from the client so that the
124 /// derived class can check that the right type of pointer is passed in.
125 std::pair<const void *const *, bool> insert_imp(const void *Ptr) {
127 // Check to see if it is already in the set.
128 const void **LastTombstone = nullptr;
129 for (const void **APtr = SmallArray, **E = SmallArray + NumNonEmpty;
131 const void *Value = *APtr;
133 return std::make_pair(APtr, false);
134 if (Value == getTombstoneMarker())
135 LastTombstone = APtr;
138 // Did we find any tombstone marker?
139 if (LastTombstone != nullptr) {
140 *LastTombstone = Ptr;
142 return std::make_pair(LastTombstone, true);
145 // Nope, there isn't. If we stay small, just 'pushback' now.
146 if (NumNonEmpty < CurArraySize) {
147 SmallArray[NumNonEmpty++] = Ptr;
148 return std::make_pair(SmallArray + (NumNonEmpty - 1), true);
150 // Otherwise, hit the big set case, which will call grow.
152 return insert_imp_big(Ptr);
155 /// erase_imp - If the set contains the specified pointer, remove it and
156 /// return true, otherwise return false. This is hidden from the client so
157 /// that the derived class can check that the right type of pointer is passed
159 bool erase_imp(const void * Ptr) {
160 const void *const *P = find_imp(Ptr);
161 if (P == EndPointer())
164 const void **Loc = const_cast<const void **>(P);
165 assert(*Loc == Ptr && "broken find!");
166 *Loc = getTombstoneMarker();
171 /// Returns the raw pointer needed to construct an iterator. If element not
172 /// found, this will be EndPointer. Otherwise, it will be a pointer to the
173 /// slot which stores Ptr;
174 const void *const * find_imp(const void * Ptr) const {
176 // Linear search for the item.
177 for (const void *const *APtr = SmallArray,
178 *const *E = SmallArray + NumNonEmpty; APtr != E; ++APtr)
185 auto *Bucket = FindBucketFor(Ptr);
192 bool isSmall() const { return CurArray == SmallArray; }
194 std::pair<const void *const *, bool> insert_imp_big(const void *Ptr);
196 const void * const *FindBucketFor(const void *Ptr) const;
197 void shrink_and_clear();
199 /// Grow - Allocate a larger backing store for the buckets and move it over.
200 void Grow(unsigned NewSize);
203 /// swap - Swaps the elements of two sets.
204 /// Note: This method assumes that both sets have the same small size.
205 void swap(SmallPtrSetImplBase &RHS);
207 void CopyFrom(const SmallPtrSetImplBase &RHS);
208 void MoveFrom(unsigned SmallSize, SmallPtrSetImplBase &&RHS);
211 /// Code shared by MoveFrom() and move constructor.
212 void MoveHelper(unsigned SmallSize, SmallPtrSetImplBase &&RHS);
213 /// Code shared by CopyFrom() and copy constructor.
214 void CopyHelper(const SmallPtrSetImplBase &RHS);
217 /// SmallPtrSetIteratorImpl - This is the common base class shared between all
218 /// instances of SmallPtrSetIterator.
219 class SmallPtrSetIteratorImpl {
221 const void *const *Bucket;
222 const void *const *End;
225 explicit SmallPtrSetIteratorImpl(const void *const *BP, const void*const *E)
226 : Bucket(BP), End(E) {
227 #if LLVM_ENABLE_ABI_BREAKING_CHECKS
228 if (ReverseIterate<bool>::value) {
236 bool operator==(const SmallPtrSetIteratorImpl &RHS) const {
237 return Bucket == RHS.Bucket;
239 bool operator!=(const SmallPtrSetIteratorImpl &RHS) const {
240 return Bucket != RHS.Bucket;
244 /// AdvanceIfNotValid - If the current bucket isn't valid, advance to a bucket
245 /// that is. This is guaranteed to stop because the end() bucket is marked
247 void AdvanceIfNotValid() {
248 assert(Bucket <= End);
249 while (Bucket != End &&
250 (*Bucket == SmallPtrSetImplBase::getEmptyMarker() ||
251 *Bucket == SmallPtrSetImplBase::getTombstoneMarker()))
254 #if LLVM_ENABLE_ABI_BREAKING_CHECKS
255 void RetreatIfNotValid() {
256 assert(Bucket >= End);
257 while (Bucket != End &&
258 (Bucket[-1] == SmallPtrSetImplBase::getEmptyMarker() ||
259 Bucket[-1] == SmallPtrSetImplBase::getTombstoneMarker())) {
266 /// SmallPtrSetIterator - This implements a const_iterator for SmallPtrSet.
267 template<typename PtrTy>
268 class SmallPtrSetIterator : public SmallPtrSetIteratorImpl {
269 using PtrTraits = PointerLikeTypeTraits<PtrTy>;
272 using value_type = PtrTy;
273 using reference = PtrTy;
274 using pointer = PtrTy;
275 using difference_type = std::ptrdiff_t;
276 using iterator_category = std::forward_iterator_tag;
278 explicit SmallPtrSetIterator(const void *const *BP, const void *const *E)
279 : SmallPtrSetIteratorImpl(BP, E) {}
281 // Most methods provided by baseclass.
283 const PtrTy operator*() const {
284 #if LLVM_ENABLE_ABI_BREAKING_CHECKS
285 if (ReverseIterate<bool>::value) {
286 assert(Bucket > End);
287 return PtrTraits::getFromVoidPointer(const_cast<void *>(Bucket[-1]));
290 assert(Bucket < End);
291 return PtrTraits::getFromVoidPointer(const_cast<void*>(*Bucket));
294 inline SmallPtrSetIterator& operator++() { // Preincrement
295 #if LLVM_ENABLE_ABI_BREAKING_CHECKS
296 if (ReverseIterate<bool>::value) {
307 SmallPtrSetIterator operator++(int) { // Postincrement
308 SmallPtrSetIterator tmp = *this;
314 /// RoundUpToPowerOfTwo - This is a helper template that rounds N up to the next
315 /// power of two (which means N itself if N is already a power of two).
317 struct RoundUpToPowerOfTwo;
319 /// RoundUpToPowerOfTwoH - If N is not a power of two, increase it. This is a
320 /// helper template used to implement RoundUpToPowerOfTwo.
321 template<unsigned N, bool isPowerTwo>
322 struct RoundUpToPowerOfTwoH {
326 struct RoundUpToPowerOfTwoH<N, false> {
328 // We could just use NextVal = N+1, but this converges faster. N|(N-1) sets
329 // the right-most zero bits to one all at once, e.g. 0b0011000 -> 0b0011111.
330 Val = RoundUpToPowerOfTwo<(N|(N-1)) + 1>::Val
335 struct RoundUpToPowerOfTwo {
336 enum { Val = RoundUpToPowerOfTwoH<N, (N&(N-1)) == 0>::Val };
339 /// \brief A templated base class for \c SmallPtrSet which provides the
340 /// typesafe interface that is common across all small sizes.
342 /// This is particularly useful for passing around between interface boundaries
343 /// to avoid encoding a particular small size in the interface boundary.
344 template <typename PtrType>
345 class SmallPtrSetImpl : public SmallPtrSetImplBase {
346 using ConstPtrType = typename add_const_past_pointer<PtrType>::type;
347 using PtrTraits = PointerLikeTypeTraits<PtrType>;
348 using ConstPtrTraits = PointerLikeTypeTraits<ConstPtrType>;
351 // Constructors that forward to the base.
352 SmallPtrSetImpl(const void **SmallStorage, const SmallPtrSetImpl &that)
353 : SmallPtrSetImplBase(SmallStorage, that) {}
354 SmallPtrSetImpl(const void **SmallStorage, unsigned SmallSize,
355 SmallPtrSetImpl &&that)
356 : SmallPtrSetImplBase(SmallStorage, SmallSize, std::move(that)) {}
357 explicit SmallPtrSetImpl(const void **SmallStorage, unsigned SmallSize)
358 : SmallPtrSetImplBase(SmallStorage, SmallSize) {}
361 using iterator = SmallPtrSetIterator<PtrType>;
362 using const_iterator = SmallPtrSetIterator<PtrType>;
363 using key_type = ConstPtrType;
364 using value_type = PtrType;
366 SmallPtrSetImpl(const SmallPtrSetImpl &) = delete;
368 /// Inserts Ptr if and only if there is no element in the container equal to
369 /// Ptr. The bool component of the returned pair is true if and only if the
370 /// insertion takes place, and the iterator component of the pair points to
371 /// the element equal to Ptr.
372 std::pair<iterator, bool> insert(PtrType Ptr) {
373 auto p = insert_imp(PtrTraits::getAsVoidPointer(Ptr));
374 return std::make_pair(makeIterator(p.first), p.second);
377 /// erase - If the set contains the specified pointer, remove it and return
378 /// true, otherwise return false.
379 bool erase(PtrType Ptr) {
380 return erase_imp(PtrTraits::getAsVoidPointer(Ptr));
382 /// count - Return 1 if the specified pointer is in the set, 0 otherwise.
383 size_type count(ConstPtrType Ptr) const { return find(Ptr) != end() ? 1 : 0; }
384 iterator find(ConstPtrType Ptr) const {
385 return makeIterator(find_imp(ConstPtrTraits::getAsVoidPointer(Ptr)));
388 template <typename IterT>
389 void insert(IterT I, IterT E) {
394 void insert(std::initializer_list<PtrType> IL) {
395 insert(IL.begin(), IL.end());
398 iterator begin() const {
399 #if LLVM_ENABLE_ABI_BREAKING_CHECKS
400 if (ReverseIterate<bool>::value)
401 return makeIterator(EndPointer() - 1);
403 return makeIterator(CurArray);
405 iterator end() const { return makeIterator(EndPointer()); }
408 /// Create an iterator that dereferences to same place as the given pointer.
409 iterator makeIterator(const void *const *P) const {
410 #if LLVM_ENABLE_ABI_BREAKING_CHECKS
411 if (ReverseIterate<bool>::value)
412 return iterator(P == EndPointer() ? CurArray : P + 1, CurArray);
414 return iterator(P, EndPointer());
418 /// SmallPtrSet - This class implements a set which is optimized for holding
419 /// SmallSize or less elements. This internally rounds up SmallSize to the next
420 /// power of two if it is not already a power of two. See the comments above
421 /// SmallPtrSetImplBase for details of the algorithm.
422 template<class PtrType, unsigned SmallSize>
423 class SmallPtrSet : public SmallPtrSetImpl<PtrType> {
424 // In small mode SmallPtrSet uses linear search for the elements, so it is
425 // not a good idea to choose this value too high. You may consider using a
426 // DenseSet<> instead if you expect many elements in the set.
427 static_assert(SmallSize <= 32, "SmallSize should be small");
429 using BaseT = SmallPtrSetImpl<PtrType>;
431 // Make sure that SmallSize is a power of two, round up if not.
432 enum { SmallSizePowTwo = RoundUpToPowerOfTwo<SmallSize>::Val };
433 /// SmallStorage - Fixed size storage used in 'small mode'.
434 const void *SmallStorage[SmallSizePowTwo];
437 SmallPtrSet() : BaseT(SmallStorage, SmallSizePowTwo) {}
438 SmallPtrSet(const SmallPtrSet &that) : BaseT(SmallStorage, that) {}
439 SmallPtrSet(SmallPtrSet &&that)
440 : BaseT(SmallStorage, SmallSizePowTwo, std::move(that)) {}
442 template<typename It>
443 SmallPtrSet(It I, It E) : BaseT(SmallStorage, SmallSizePowTwo) {
447 SmallPtrSet(std::initializer_list<PtrType> IL)
448 : BaseT(SmallStorage, SmallSizePowTwo) {
449 this->insert(IL.begin(), IL.end());
452 SmallPtrSet<PtrType, SmallSize> &
453 operator=(const SmallPtrSet<PtrType, SmallSize> &RHS) {
459 SmallPtrSet<PtrType, SmallSize> &
460 operator=(SmallPtrSet<PtrType, SmallSize> &&RHS) {
462 this->MoveFrom(SmallSizePowTwo, std::move(RHS));
466 SmallPtrSet<PtrType, SmallSize> &
467 operator=(std::initializer_list<PtrType> IL) {
469 this->insert(IL.begin(), IL.end());
473 /// swap - Swaps the elements of two sets.
474 void swap(SmallPtrSet<PtrType, SmallSize> &RHS) {
475 SmallPtrSetImplBase::swap(RHS);
479 } // end namespace llvm
483 /// Implement std::swap in terms of SmallPtrSet swap.
484 template<class T, unsigned N>
485 inline void swap(llvm::SmallPtrSet<T, N> &LHS, llvm::SmallPtrSet<T, N> &RHS) {
489 } // end namespace std
491 #endif // LLVM_ADT_SMALLPTRSET_H