1 //===- llvm/ADT/SmallPtrSet.h - 'Normally small' pointer set ----*- C++ -*-===//
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
7 //===----------------------------------------------------------------------===//
9 // This file defines the SmallPtrSet class. See the doxygen comment for
10 // SmallPtrSetImplBase for more details on the algorithm used.
12 //===----------------------------------------------------------------------===//
14 #ifndef LLVM_ADT_SMALLPTRSET_H
15 #define LLVM_ADT_SMALLPTRSET_H
17 #include "llvm/ADT/EpochTracker.h"
18 #include "llvm/Support/Compiler.h"
19 #include "llvm/Support/ReverseIteration.h"
20 #include "llvm/Support/type_traits.h"
25 #include <initializer_list>
31 /// SmallPtrSetImplBase - This is the common code shared among all the
32 /// SmallPtrSet<>'s, which is almost everything. SmallPtrSet has two modes, one
33 /// for small and one for large sets.
35 /// Small sets use an array of pointers allocated in the SmallPtrSet object,
36 /// which is treated as a simple array of pointers. When a pointer is added to
37 /// the set, the array is scanned to see if the element already exists, if not
38 /// the element is 'pushed back' onto the array. If we run out of space in the
39 /// array, we grow into the 'large set' case. SmallSet should be used when the
40 /// sets are often small. In this case, no memory allocation is used, and only
41 /// light-weight and cache-efficient scanning is used.
43 /// Large sets use a classic exponentially-probed hash table. Empty buckets are
44 /// represented with an illegal pointer value (-1) to allow null pointers to be
45 /// inserted. Tombstones are represented with another illegal pointer value
46 /// (-2), to allow deletion. The hash table is resized when the table is 3/4 or
47 /// more. When this happens, the table is doubled in size.
49 class SmallPtrSetImplBase : public DebugEpochBase {
50 friend class SmallPtrSetIteratorImpl;
53 /// SmallArray - Points to a fixed size set of buckets, used in 'small mode'.
54 const void **SmallArray;
55 /// CurArray - This is the current set of buckets. If equal to SmallArray,
56 /// then the set is in 'small mode'.
57 const void **CurArray;
58 /// CurArraySize - The allocated size of CurArray, always a power of two.
59 unsigned CurArraySize;
61 /// Number of elements in CurArray that contain a value or are a tombstone.
62 /// If small, all these elements are at the beginning of CurArray and the rest
65 /// Number of tombstones in CurArray.
66 unsigned NumTombstones;
68 // Helpers to copy and move construct a SmallPtrSet.
69 SmallPtrSetImplBase(const void **SmallStorage,
70 const SmallPtrSetImplBase &that);
71 SmallPtrSetImplBase(const void **SmallStorage, unsigned SmallSize,
72 SmallPtrSetImplBase &&that);
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!");
81 ~SmallPtrSetImplBase() {
87 using size_type = unsigned;
89 SmallPtrSetImplBase &operator=(const SmallPtrSetImplBase &) = delete;
91 LLVM_NODISCARD bool empty() const { return size() == 0; }
92 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;
143 return std::make_pair(LastTombstone, true);
146 // Nope, there isn't. If we stay small, just 'pushback' now.
147 if (NumNonEmpty < CurArraySize) {
148 SmallArray[NumNonEmpty++] = Ptr;
150 return std::make_pair(SmallArray + (NumNonEmpty - 1), true);
152 // Otherwise, hit the big set case, which will call grow.
154 return insert_imp_big(Ptr);
157 /// erase_imp - If the set contains the specified pointer, remove it and
158 /// return true, otherwise return false. This is hidden from the client so
159 /// that the derived class can check that the right type of pointer is passed
161 bool erase_imp(const void * Ptr) {
162 const void *const *P = find_imp(Ptr);
163 if (P == EndPointer())
166 const void **Loc = const_cast<const void **>(P);
167 assert(*Loc == Ptr && "broken find!");
168 *Loc = getTombstoneMarker();
173 /// Returns the raw pointer needed to construct an iterator. If element not
174 /// found, this will be EndPointer. Otherwise, it will be a pointer to the
175 /// slot which stores Ptr;
176 const void *const * find_imp(const void * Ptr) const {
178 // Linear search for the item.
179 for (const void *const *APtr = SmallArray,
180 *const *E = SmallArray + NumNonEmpty; APtr != E; ++APtr)
187 auto *Bucket = FindBucketFor(Ptr);
194 bool isSmall() const { return CurArray == SmallArray; }
196 std::pair<const void *const *, bool> insert_imp_big(const void *Ptr);
198 const void * const *FindBucketFor(const void *Ptr) const;
199 void shrink_and_clear();
201 /// Grow - Allocate a larger backing store for the buckets and move it over.
202 void Grow(unsigned NewSize);
205 /// swap - Swaps the elements of two sets.
206 /// Note: This method assumes that both sets have the same small size.
207 void swap(SmallPtrSetImplBase &RHS);
209 void CopyFrom(const SmallPtrSetImplBase &RHS);
210 void MoveFrom(unsigned SmallSize, SmallPtrSetImplBase &&RHS);
213 /// Code shared by MoveFrom() and move constructor.
214 void MoveHelper(unsigned SmallSize, SmallPtrSetImplBase &&RHS);
215 /// Code shared by CopyFrom() and copy constructor.
216 void CopyHelper(const SmallPtrSetImplBase &RHS);
219 /// SmallPtrSetIteratorImpl - This is the common base class shared between all
220 /// instances of SmallPtrSetIterator.
221 class SmallPtrSetIteratorImpl {
223 const void *const *Bucket;
224 const void *const *End;
227 explicit SmallPtrSetIteratorImpl(const void *const *BP, const void*const *E)
228 : Bucket(BP), End(E) {
229 if (shouldReverseIterate()) {
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 void RetreatIfNotValid() {
255 assert(Bucket >= End);
256 while (Bucket != End &&
257 (Bucket[-1] == SmallPtrSetImplBase::getEmptyMarker() ||
258 Bucket[-1] == SmallPtrSetImplBase::getTombstoneMarker())) {
264 /// SmallPtrSetIterator - This implements a const_iterator for SmallPtrSet.
265 template <typename PtrTy>
266 class SmallPtrSetIterator : public SmallPtrSetIteratorImpl,
267 DebugEpochBase::HandleBase {
268 using PtrTraits = PointerLikeTypeTraits<PtrTy>;
271 using value_type = PtrTy;
272 using reference = PtrTy;
273 using pointer = PtrTy;
274 using difference_type = std::ptrdiff_t;
275 using iterator_category = std::forward_iterator_tag;
277 explicit SmallPtrSetIterator(const void *const *BP, const void *const *E,
278 const DebugEpochBase &Epoch)
279 : SmallPtrSetIteratorImpl(BP, E), DebugEpochBase::HandleBase(&Epoch) {}
281 // Most methods are provided by the base class.
283 const PtrTy operator*() const {
284 assert(isHandleInSync() && "invalid iterator access!");
285 if (shouldReverseIterate()) {
286 assert(Bucket > End);
287 return PtrTraits::getFromVoidPointer(const_cast<void *>(Bucket[-1]));
289 assert(Bucket < End);
290 return PtrTraits::getFromVoidPointer(const_cast<void*>(*Bucket));
293 inline SmallPtrSetIterator& operator++() { // Preincrement
294 assert(isHandleInSync() && "invalid iterator access!");
295 if (shouldReverseIterate()) {
305 SmallPtrSetIterator operator++(int) { // Postincrement
306 SmallPtrSetIterator tmp = *this;
312 /// RoundUpToPowerOfTwo - This is a helper template that rounds N up to the next
313 /// power of two (which means N itself if N is already a power of two).
315 struct RoundUpToPowerOfTwo;
317 /// RoundUpToPowerOfTwoH - If N is not a power of two, increase it. This is a
318 /// helper template used to implement RoundUpToPowerOfTwo.
319 template<unsigned N, bool isPowerTwo>
320 struct RoundUpToPowerOfTwoH {
324 struct RoundUpToPowerOfTwoH<N, false> {
326 // We could just use NextVal = N+1, but this converges faster. N|(N-1) sets
327 // the right-most zero bits to one all at once, e.g. 0b0011000 -> 0b0011111.
328 Val = RoundUpToPowerOfTwo<(N|(N-1)) + 1>::Val
333 struct RoundUpToPowerOfTwo {
334 enum { Val = RoundUpToPowerOfTwoH<N, (N&(N-1)) == 0>::Val };
337 /// A templated base class for \c SmallPtrSet which provides the
338 /// typesafe interface that is common across all small sizes.
340 /// This is particularly useful for passing around between interface boundaries
341 /// to avoid encoding a particular small size in the interface boundary.
342 template <typename PtrType>
343 class SmallPtrSetImpl : public SmallPtrSetImplBase {
344 using ConstPtrType = typename add_const_past_pointer<PtrType>::type;
345 using PtrTraits = PointerLikeTypeTraits<PtrType>;
346 using ConstPtrTraits = PointerLikeTypeTraits<ConstPtrType>;
349 // Forward constructors to the base.
350 using SmallPtrSetImplBase::SmallPtrSetImplBase;
353 using iterator = SmallPtrSetIterator<PtrType>;
354 using const_iterator = SmallPtrSetIterator<PtrType>;
355 using key_type = ConstPtrType;
356 using value_type = PtrType;
358 SmallPtrSetImpl(const SmallPtrSetImpl &) = delete;
360 /// Inserts Ptr if and only if there is no element in the container equal to
361 /// Ptr. The bool component of the returned pair is true if and only if the
362 /// insertion takes place, and the iterator component of the pair points to
363 /// the element equal to Ptr.
364 std::pair<iterator, bool> insert(PtrType Ptr) {
365 auto p = insert_imp(PtrTraits::getAsVoidPointer(Ptr));
366 return std::make_pair(makeIterator(p.first), p.second);
369 /// erase - If the set contains the specified pointer, remove it and return
370 /// true, otherwise return false.
371 bool erase(PtrType Ptr) {
372 return erase_imp(PtrTraits::getAsVoidPointer(Ptr));
374 /// count - Return 1 if the specified pointer is in the set, 0 otherwise.
375 size_type count(ConstPtrType Ptr) const {
376 return find_imp(ConstPtrTraits::getAsVoidPointer(Ptr)) != EndPointer();
378 iterator find(ConstPtrType Ptr) const {
379 return makeIterator(find_imp(ConstPtrTraits::getAsVoidPointer(Ptr)));
381 bool contains(ConstPtrType Ptr) const {
382 return find_imp(ConstPtrTraits::getAsVoidPointer(Ptr)) != EndPointer();
385 template <typename IterT>
386 void insert(IterT I, IterT E) {
391 void insert(std::initializer_list<PtrType> IL) {
392 insert(IL.begin(), IL.end());
395 iterator begin() const {
396 if (shouldReverseIterate())
397 return makeIterator(EndPointer() - 1);
398 return makeIterator(CurArray);
400 iterator end() const { return makeIterator(EndPointer()); }
403 /// Create an iterator that dereferences to same place as the given pointer.
404 iterator makeIterator(const void *const *P) const {
405 if (shouldReverseIterate())
406 return iterator(P == EndPointer() ? CurArray : P + 1, CurArray, *this);
407 return iterator(P, EndPointer(), *this);
411 /// Equality comparison for SmallPtrSet.
413 /// Iterates over elements of LHS confirming that each value from LHS is also in
414 /// RHS, and that no additional values are in RHS.
415 template <typename PtrType>
416 bool operator==(const SmallPtrSetImpl<PtrType> &LHS,
417 const SmallPtrSetImpl<PtrType> &RHS) {
418 if (LHS.size() != RHS.size())
421 for (const auto *KV : LHS)
428 /// Inequality comparison for SmallPtrSet.
430 /// Equivalent to !(LHS == RHS).
431 template <typename PtrType>
432 bool operator!=(const SmallPtrSetImpl<PtrType> &LHS,
433 const SmallPtrSetImpl<PtrType> &RHS) {
434 return !(LHS == RHS);
437 /// SmallPtrSet - This class implements a set which is optimized for holding
438 /// SmallSize or less elements. This internally rounds up SmallSize to the next
439 /// power of two if it is not already a power of two. See the comments above
440 /// SmallPtrSetImplBase for details of the algorithm.
441 template<class PtrType, unsigned SmallSize>
442 class SmallPtrSet : public SmallPtrSetImpl<PtrType> {
443 // In small mode SmallPtrSet uses linear search for the elements, so it is
444 // not a good idea to choose this value too high. You may consider using a
445 // DenseSet<> instead if you expect many elements in the set.
446 static_assert(SmallSize <= 32, "SmallSize should be small");
448 using BaseT = SmallPtrSetImpl<PtrType>;
450 // Make sure that SmallSize is a power of two, round up if not.
451 enum { SmallSizePowTwo = RoundUpToPowerOfTwo<SmallSize>::Val };
452 /// SmallStorage - Fixed size storage used in 'small mode'.
453 const void *SmallStorage[SmallSizePowTwo];
456 SmallPtrSet() : BaseT(SmallStorage, SmallSizePowTwo) {}
457 SmallPtrSet(const SmallPtrSet &that) : BaseT(SmallStorage, that) {}
458 SmallPtrSet(SmallPtrSet &&that)
459 : BaseT(SmallStorage, SmallSizePowTwo, std::move(that)) {}
461 template<typename It>
462 SmallPtrSet(It I, It E) : BaseT(SmallStorage, SmallSizePowTwo) {
466 SmallPtrSet(std::initializer_list<PtrType> IL)
467 : BaseT(SmallStorage, SmallSizePowTwo) {
468 this->insert(IL.begin(), IL.end());
471 SmallPtrSet<PtrType, SmallSize> &
472 operator=(const SmallPtrSet<PtrType, SmallSize> &RHS) {
478 SmallPtrSet<PtrType, SmallSize> &
479 operator=(SmallPtrSet<PtrType, SmallSize> &&RHS) {
481 this->MoveFrom(SmallSizePowTwo, std::move(RHS));
485 SmallPtrSet<PtrType, SmallSize> &
486 operator=(std::initializer_list<PtrType> IL) {
488 this->insert(IL.begin(), IL.end());
492 /// swap - Swaps the elements of two sets.
493 void swap(SmallPtrSet<PtrType, SmallSize> &RHS) {
494 SmallPtrSetImplBase::swap(RHS);
498 } // end namespace llvm
502 /// Implement std::swap in terms of SmallPtrSet swap.
503 template<class T, unsigned N>
504 inline void swap(llvm::SmallPtrSet<T, N> &LHS, llvm::SmallPtrSet<T, N> &RHS) {
508 } // end namespace std
510 #endif // LLVM_ADT_SMALLPTRSET_H