1 //===- llvm/ADT/SparseMultiSet.h - Sparse multiset --------------*- C++ -*-===//
2 //
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
6 //
7 //===----------------------------------------------------------------------===//
8 ///
9 /// \file
10 /// This file defines the SparseMultiSet class, which adds multiset behavior to
11 /// the SparseSet.
12 ///
13 /// A sparse multiset holds a small number of objects identified by integer keys
14 /// from a moderately sized universe. The sparse multiset uses more memory than
15 /// other containers in order to provide faster operations. Any key can map to
16 /// multiple values. A SparseMultiSetNode class is provided, which serves as a
17 /// convenient base class for the contents of a SparseMultiSet.
18 ///
19 //===----------------------------------------------------------------------===//
20 
21 #ifndef LLVM_ADT_SPARSEMULTISET_H
22 #define LLVM_ADT_SPARSEMULTISET_H
23 
24 #include "llvm/ADT/identity.h"
25 #include "llvm/ADT/SmallVector.h"
26 #include "llvm/ADT/SparseSet.h"
27 #include <cassert>
28 #include <cstdint>
29 #include <cstdlib>
30 #include <iterator>
31 #include <limits>
32 #include <utility>
33 
34 namespace llvm {
35 
36 /// Fast multiset implementation for objects that can be identified by small
37 /// unsigned keys.
38 ///
39 /// SparseMultiSet allocates memory proportional to the size of the key
40 /// universe, so it is not recommended for building composite data structures.
41 /// It is useful for algorithms that require a single set with fast operations.
42 ///
43 /// Compared to DenseSet and DenseMap, SparseMultiSet provides constant-time
44 /// fast clear() as fast as a vector.  The find(), insert(), and erase()
45 /// operations are all constant time, and typically faster than a hash table.
46 /// The iteration order doesn't depend on numerical key values, it only depends
47 /// on the order of insert() and erase() operations.  Iteration order is the
48 /// insertion order. Iteration is only provided over elements of equivalent
49 /// keys, but iterators are bidirectional.
50 ///
51 /// Compared to BitVector, SparseMultiSet<unsigned> uses 8x-40x more memory, but
52 /// offers constant-time clear() and size() operations as well as fast iteration
53 /// independent on the size of the universe.
54 ///
55 /// SparseMultiSet contains a dense vector holding all the objects and a sparse
56 /// array holding indexes into the dense vector.  Most of the memory is used by
57 /// the sparse array which is the size of the key universe. The SparseT template
58 /// parameter provides a space/speed tradeoff for sets holding many elements.
59 ///
60 /// When SparseT is uint32_t, find() only touches up to 3 cache lines, but the
61 /// sparse array uses 4 x Universe bytes.
62 ///
63 /// When SparseT is uint8_t (the default), find() touches up to 3+[N/256] cache
64 /// lines, but the sparse array is 4x smaller.  N is the number of elements in
65 /// the set.
66 ///
67 /// For sets that may grow to thousands of elements, SparseT should be set to
68 /// uint16_t or uint32_t.
69 ///
70 /// Multiset behavior is provided by providing doubly linked lists for values
71 /// that are inlined in the dense vector. SparseMultiSet is a good choice when
72 /// one desires a growable number of entries per key, as it will retain the
73 /// SparseSet algorithmic properties despite being growable. Thus, it is often a
74 /// better choice than a SparseSet of growable containers or a vector of
75 /// vectors. SparseMultiSet also keeps iterators valid after erasure (provided
76 /// the iterators don't point to the element erased), allowing for more
77 /// intuitive and fast removal.
78 ///
79 /// @tparam ValueT      The type of objects in the set.
80 /// @tparam KeyFunctorT A functor that computes an unsigned index from KeyT.
81 /// @tparam SparseT     An unsigned integer type. See above.
82 ///
83 template<typename ValueT,
84          typename KeyFunctorT = identity<unsigned>,
85          typename SparseT = uint8_t>
86 class SparseMultiSet {
87   static_assert(std::is_unsigned_v<SparseT>,
88                 "SparseT must be an unsigned integer type");
89 
90   /// The actual data that's stored, as a doubly-linked list implemented via
91   /// indices into the DenseVector.  The doubly linked list is implemented
92   /// circular in Prev indices, and INVALID-terminated in Next indices. This
93   /// provides efficient access to list tails. These nodes can also be
94   /// tombstones, in which case they are actually nodes in a single-linked
95   /// freelist of recyclable slots.
96   struct SMSNode {
97     static constexpr unsigned INVALID = ~0U;
98 
99     ValueT Data;
100     unsigned Prev;
101     unsigned Next;
102 
SMSNodeSMSNode103     SMSNode(ValueT D, unsigned P, unsigned N) : Data(D), Prev(P), Next(N) {}
104 
105     /// List tails have invalid Nexts.
isTailSMSNode106     bool isTail() const {
107       return Next == INVALID;
108     }
109 
110     /// Whether this node is a tombstone node, and thus is in our freelist.
isTombstoneSMSNode111     bool isTombstone() const {
112       return Prev == INVALID;
113     }
114 
115     /// Since the list is circular in Prev, all non-tombstone nodes have a valid
116     /// Prev.
isValidSMSNode117     bool isValid() const { return Prev != INVALID; }
118   };
119 
120   using KeyT = typename KeyFunctorT::argument_type;
121   using DenseT = SmallVector<SMSNode, 8>;
122   DenseT Dense;
123   SparseT *Sparse = nullptr;
124   unsigned Universe = 0;
125   KeyFunctorT KeyIndexOf;
126   SparseSetValFunctor<KeyT, ValueT, KeyFunctorT> ValIndexOf;
127 
128   /// We have a built-in recycler for reusing tombstone slots. This recycler
129   /// puts a singly-linked free list into tombstone slots, allowing us quick
130   /// erasure, iterator preservation, and dense size.
131   unsigned FreelistIdx = SMSNode::INVALID;
132   unsigned NumFree = 0;
133 
sparseIndex(const ValueT & Val)134   unsigned sparseIndex(const ValueT &Val) const {
135     assert(ValIndexOf(Val) < Universe &&
136            "Invalid key in set. Did object mutate?");
137     return ValIndexOf(Val);
138   }
sparseIndex(const SMSNode & N)139   unsigned sparseIndex(const SMSNode &N) const { return sparseIndex(N.Data); }
140 
141   /// Whether the given entry is the head of the list. List heads's previous
142   /// pointers are to the tail of the list, allowing for efficient access to the
143   /// list tail. D must be a valid entry node.
isHead(const SMSNode & D)144   bool isHead(const SMSNode &D) const {
145     assert(D.isValid() && "Invalid node for head");
146     return Dense[D.Prev].isTail();
147   }
148 
149   /// Whether the given entry is a singleton entry, i.e. the only entry with
150   /// that key.
isSingleton(const SMSNode & N)151   bool isSingleton(const SMSNode &N) const {
152     assert(N.isValid() && "Invalid node for singleton");
153     // Is N its own predecessor?
154     return &Dense[N.Prev] == &N;
155   }
156 
157   /// Add in the given SMSNode. Uses a free entry in our freelist if
158   /// available. Returns the index of the added node.
addValue(const ValueT & V,unsigned Prev,unsigned Next)159   unsigned addValue(const ValueT& V, unsigned Prev, unsigned Next) {
160     if (NumFree == 0) {
161       Dense.push_back(SMSNode(V, Prev, Next));
162       return Dense.size() - 1;
163     }
164 
165     // Peel off a free slot
166     unsigned Idx = FreelistIdx;
167     unsigned NextFree = Dense[Idx].Next;
168     assert(Dense[Idx].isTombstone() && "Non-tombstone free?");
169 
170     Dense[Idx] = SMSNode(V, Prev, Next);
171     FreelistIdx = NextFree;
172     --NumFree;
173     return Idx;
174   }
175 
176   /// Make the current index a new tombstone. Pushes it onto the freelist.
makeTombstone(unsigned Idx)177   void makeTombstone(unsigned Idx) {
178     Dense[Idx].Prev = SMSNode::INVALID;
179     Dense[Idx].Next = FreelistIdx;
180     FreelistIdx = Idx;
181     ++NumFree;
182   }
183 
184 public:
185   using value_type = ValueT;
186   using reference = ValueT &;
187   using const_reference = const ValueT &;
188   using pointer = ValueT *;
189   using const_pointer = const ValueT *;
190   using size_type = unsigned;
191 
192   SparseMultiSet() = default;
193   SparseMultiSet(const SparseMultiSet &) = delete;
194   SparseMultiSet &operator=(const SparseMultiSet &) = delete;
~SparseMultiSet()195   ~SparseMultiSet() { free(Sparse); }
196 
197   /// Set the universe size which determines the largest key the set can hold.
198   /// The universe must be sized before any elements can be added.
199   ///
200   /// @param U Universe size. All object keys must be less than U.
201   ///
setUniverse(unsigned U)202   void setUniverse(unsigned U) {
203     // It's not hard to resize the universe on a non-empty set, but it doesn't
204     // seem like a likely use case, so we can add that code when we need it.
205     assert(empty() && "Can only resize universe on an empty map");
206     // Hysteresis prevents needless reallocations.
207     if (U >= Universe/4 && U <= Universe)
208       return;
209     free(Sparse);
210     // The Sparse array doesn't actually need to be initialized, so malloc
211     // would be enough here, but that will cause tools like valgrind to
212     // complain about branching on uninitialized data.
213     Sparse = static_cast<SparseT*>(safe_calloc(U, sizeof(SparseT)));
214     Universe = U;
215   }
216 
217   /// Our iterators are iterators over the collection of objects that share a
218   /// key.
219   template <typename SMSPtrTy> class iterator_base {
220     friend class SparseMultiSet;
221 
222   public:
223     using iterator_category = std::bidirectional_iterator_tag;
224     using value_type = ValueT;
225     using difference_type = std::ptrdiff_t;
226     using pointer = value_type *;
227     using reference = value_type &;
228 
229   private:
230     SMSPtrTy SMS;
231     unsigned Idx;
232     unsigned SparseIdx;
233 
iterator_base(SMSPtrTy P,unsigned I,unsigned SI)234     iterator_base(SMSPtrTy P, unsigned I, unsigned SI)
235       : SMS(P), Idx(I), SparseIdx(SI) {}
236 
237     /// Whether our iterator has fallen outside our dense vector.
isEnd()238     bool isEnd() const {
239       if (Idx == SMSNode::INVALID)
240         return true;
241 
242       assert(Idx < SMS->Dense.size() && "Out of range, non-INVALID Idx?");
243       return false;
244     }
245 
246     /// Whether our iterator is properly keyed, i.e. the SparseIdx is valid
isKeyed()247     bool isKeyed() const { return SparseIdx < SMS->Universe; }
248 
Prev()249     unsigned Prev() const { return SMS->Dense[Idx].Prev; }
Next()250     unsigned Next() const { return SMS->Dense[Idx].Next; }
251 
setPrev(unsigned P)252     void setPrev(unsigned P) { SMS->Dense[Idx].Prev = P; }
setNext(unsigned N)253     void setNext(unsigned N) { SMS->Dense[Idx].Next = N; }
254 
255   public:
256     reference operator*() const {
257       assert(isKeyed() && SMS->sparseIndex(SMS->Dense[Idx].Data) == SparseIdx &&
258              "Dereferencing iterator of invalid key or index");
259 
260       return SMS->Dense[Idx].Data;
261     }
262     pointer operator->() const { return &operator*(); }
263 
264     /// Comparison operators
265     bool operator==(const iterator_base &RHS) const {
266       // end compares equal
267       if (SMS == RHS.SMS && Idx == RHS.Idx) {
268         assert((isEnd() || SparseIdx == RHS.SparseIdx) &&
269                "Same dense entry, but different keys?");
270         return true;
271       }
272 
273       return false;
274     }
275 
276     bool operator!=(const iterator_base &RHS) const {
277       return !operator==(RHS);
278     }
279 
280     /// Increment and decrement operators
281     iterator_base &operator--() { // predecrement - Back up
282       assert(isKeyed() && "Decrementing an invalid iterator");
283       assert((isEnd() || !SMS->isHead(SMS->Dense[Idx])) &&
284              "Decrementing head of list");
285 
286       // If we're at the end, then issue a new find()
287       if (isEnd())
288         Idx = SMS->findIndex(SparseIdx).Prev();
289       else
290         Idx = Prev();
291 
292       return *this;
293     }
294     iterator_base &operator++() { // preincrement - Advance
295       assert(!isEnd() && isKeyed() && "Incrementing an invalid/end iterator");
296       Idx = Next();
297       return *this;
298     }
299     iterator_base operator--(int) { // postdecrement
300       iterator_base I(*this);
301       --*this;
302       return I;
303     }
304     iterator_base operator++(int) { // postincrement
305       iterator_base I(*this);
306       ++*this;
307       return I;
308     }
309   };
310 
311   using iterator = iterator_base<SparseMultiSet *>;
312   using const_iterator = iterator_base<const SparseMultiSet *>;
313 
314   // Convenience types
315   using RangePair = std::pair<iterator, iterator>;
316 
317   /// Returns an iterator past this container. Note that such an iterator cannot
318   /// be decremented, but will compare equal to other end iterators.
end()319   iterator end() { return iterator(this, SMSNode::INVALID, SMSNode::INVALID); }
end()320   const_iterator end() const {
321     return const_iterator(this, SMSNode::INVALID, SMSNode::INVALID);
322   }
323 
324   /// Returns true if the set is empty.
325   ///
326   /// This is not the same as BitVector::empty().
327   ///
empty()328   bool empty() const { return size() == 0; }
329 
330   /// Returns the number of elements in the set.
331   ///
332   /// This is not the same as BitVector::size() which returns the size of the
333   /// universe.
334   ///
size()335   size_type size() const {
336     assert(NumFree <= Dense.size() && "Out-of-bounds free entries");
337     return Dense.size() - NumFree;
338   }
339 
340   /// Clears the set.  This is a very fast constant time operation.
341   ///
clear()342   void clear() {
343     // Sparse does not need to be cleared, see find().
344     Dense.clear();
345     NumFree = 0;
346     FreelistIdx = SMSNode::INVALID;
347   }
348 
349   /// Find an element by its index.
350   ///
351   /// @param   Idx A valid index to find.
352   /// @returns An iterator to the element identified by key, or end().
353   ///
findIndex(unsigned Idx)354   iterator findIndex(unsigned Idx) {
355     assert(Idx < Universe && "Key out of range");
356     const unsigned Stride = std::numeric_limits<SparseT>::max() + 1u;
357     for (unsigned i = Sparse[Idx], e = Dense.size(); i < e; i += Stride) {
358       const unsigned FoundIdx = sparseIndex(Dense[i]);
359       // Check that we're pointing at the correct entry and that it is the head
360       // of a valid list.
361       if (Idx == FoundIdx && Dense[i].isValid() && isHead(Dense[i]))
362         return iterator(this, i, Idx);
363       // Stride is 0 when SparseT >= unsigned.  We don't need to loop.
364       if (!Stride)
365         break;
366     }
367     return end();
368   }
369 
370   /// Find an element by its key.
371   ///
372   /// @param   Key A valid key to find.
373   /// @returns An iterator to the element identified by key, or end().
374   ///
find(const KeyT & Key)375   iterator find(const KeyT &Key) {
376     return findIndex(KeyIndexOf(Key));
377   }
378 
find(const KeyT & Key)379   const_iterator find(const KeyT &Key) const {
380     iterator I = const_cast<SparseMultiSet*>(this)->findIndex(KeyIndexOf(Key));
381     return const_iterator(I.SMS, I.Idx, KeyIndexOf(Key));
382   }
383 
384   /// Returns the number of elements identified by Key. This will be linear in
385   /// the number of elements of that key.
count(const KeyT & Key)386   size_type count(const KeyT &Key) const {
387     unsigned Ret = 0;
388     for (const_iterator It = find(Key); It != end(); ++It)
389       ++Ret;
390 
391     return Ret;
392   }
393 
394   /// Returns true if this set contains an element identified by Key.
contains(const KeyT & Key)395   bool contains(const KeyT &Key) const {
396     return find(Key) != end();
397   }
398 
399   /// Return the head and tail of the subset's list, otherwise returns end().
getHead(const KeyT & Key)400   iterator getHead(const KeyT &Key) { return find(Key); }
getTail(const KeyT & Key)401   iterator getTail(const KeyT &Key) {
402     iterator I = find(Key);
403     if (I != end())
404       I = iterator(this, I.Prev(), KeyIndexOf(Key));
405     return I;
406   }
407 
408   /// The bounds of the range of items sharing Key K. First member is the head
409   /// of the list, and the second member is a decrementable end iterator for
410   /// that key.
equal_range(const KeyT & K)411   RangePair equal_range(const KeyT &K) {
412     iterator B = find(K);
413     iterator E = iterator(this, SMSNode::INVALID, B.SparseIdx);
414     return std::make_pair(B, E);
415   }
416 
417   /// Insert a new element at the tail of the subset list. Returns an iterator
418   /// to the newly added entry.
insert(const ValueT & Val)419   iterator insert(const ValueT &Val) {
420     unsigned Idx = sparseIndex(Val);
421     iterator I = findIndex(Idx);
422 
423     unsigned NodeIdx = addValue(Val, SMSNode::INVALID, SMSNode::INVALID);
424 
425     if (I == end()) {
426       // Make a singleton list
427       Sparse[Idx] = NodeIdx;
428       Dense[NodeIdx].Prev = NodeIdx;
429       return iterator(this, NodeIdx, Idx);
430     }
431 
432     // Stick it at the end.
433     unsigned HeadIdx = I.Idx;
434     unsigned TailIdx = I.Prev();
435     Dense[TailIdx].Next = NodeIdx;
436     Dense[HeadIdx].Prev = NodeIdx;
437     Dense[NodeIdx].Prev = TailIdx;
438 
439     return iterator(this, NodeIdx, Idx);
440   }
441 
442   /// Erases an existing element identified by a valid iterator.
443   ///
444   /// This invalidates iterators pointing at the same entry, but erase() returns
445   /// an iterator pointing to the next element in the subset's list. This makes
446   /// it possible to erase selected elements while iterating over the subset:
447   ///
448   ///   tie(I, E) = Set.equal_range(Key);
449   ///   while (I != E)
450   ///     if (test(*I))
451   ///       I = Set.erase(I);
452   ///     else
453   ///       ++I;
454   ///
455   /// Note that if the last element in the subset list is erased, this will
456   /// return an end iterator which can be decremented to get the new tail (if it
457   /// exists):
458   ///
459   ///  tie(B, I) = Set.equal_range(Key);
460   ///  for (bool isBegin = B == I; !isBegin; /* empty */) {
461   ///    isBegin = (--I) == B;
462   ///    if (test(I))
463   ///      break;
464   ///    I = erase(I);
465   ///  }
erase(iterator I)466   iterator erase(iterator I) {
467     assert(I.isKeyed() && !I.isEnd() && !Dense[I.Idx].isTombstone() &&
468            "erasing invalid/end/tombstone iterator");
469 
470     // First, unlink the node from its list. Then swap the node out with the
471     // dense vector's last entry
472     iterator NextI = unlink(Dense[I.Idx]);
473 
474     // Put in a tombstone.
475     makeTombstone(I.Idx);
476 
477     return NextI;
478   }
479 
480   /// Erase all elements with the given key. This invalidates all
481   /// iterators of that key.
eraseAll(const KeyT & K)482   void eraseAll(const KeyT &K) {
483     for (iterator I = find(K); I != end(); /* empty */)
484       I = erase(I);
485   }
486 
487 private:
488   /// Unlink the node from its list. Returns the next node in the list.
unlink(const SMSNode & N)489   iterator unlink(const SMSNode &N) {
490     if (isSingleton(N)) {
491       // Singleton is already unlinked
492       assert(N.Next == SMSNode::INVALID && "Singleton has next?");
493       return iterator(this, SMSNode::INVALID, ValIndexOf(N.Data));
494     }
495 
496     if (isHead(N)) {
497       // If we're the head, then update the sparse array and our next.
498       Sparse[sparseIndex(N)] = N.Next;
499       Dense[N.Next].Prev = N.Prev;
500       return iterator(this, N.Next, ValIndexOf(N.Data));
501     }
502 
503     if (N.isTail()) {
504       // If we're the tail, then update our head and our previous.
505       findIndex(sparseIndex(N)).setPrev(N.Prev);
506       Dense[N.Prev].Next = N.Next;
507 
508       // Give back an end iterator that can be decremented
509       iterator I(this, N.Prev, ValIndexOf(N.Data));
510       return ++I;
511     }
512 
513     // Otherwise, just drop us
514     Dense[N.Next].Prev = N.Prev;
515     Dense[N.Prev].Next = N.Next;
516     return iterator(this, N.Next, ValIndexOf(N.Data));
517   }
518 };
519 
520 } // end namespace llvm
521 
522 #endif // LLVM_ADT_SPARSEMULTISET_H
523