LCOV - code coverage report
Current view: top level - include/llvm/ADT - SparseSet.h (source / functions) Hit Total Coverage
Test: llvm-toolchain.info Lines: 43 43 100.0 %
Date: 2017-09-14 15:23:50 Functions: 50 50 100.0 %
Legend: Lines: hit not hit

          Line data    Source code
       1             : //===- llvm/ADT/SparseSet.h - Sparse 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 SparseSet class derived from the version described in
      11             : // Briggs, Torczon, "An efficient representation for sparse sets", ACM Letters
      12             : // on Programming Languages and Systems, Volume 2 Issue 1-4, March-Dec.  1993.
      13             : //
      14             : // A sparse set holds a small number of objects identified by integer keys from
      15             : // a moderately sized universe. The sparse set uses more memory than other
      16             : // containers in order to provide faster operations.
      17             : //
      18             : //===----------------------------------------------------------------------===//
      19             : 
      20             : #ifndef LLVM_ADT_SPARSESET_H
      21             : #define LLVM_ADT_SPARSESET_H
      22             : 
      23             : #include "llvm/ADT/STLExtras.h"
      24             : #include "llvm/ADT/SmallVector.h"
      25             : #include <cassert>
      26             : #include <cstdint>
      27             : #include <cstdlib>
      28             : #include <limits>
      29             : #include <utility>
      30             : 
      31             : namespace llvm {
      32             : 
      33             : /// SparseSetValTraits - Objects in a SparseSet are identified by keys that can
      34             : /// be uniquely converted to a small integer less than the set's universe. This
      35             : /// class allows the set to hold values that differ from the set's key type as
      36             : /// long as an index can still be derived from the value. SparseSet never
      37             : /// directly compares ValueT, only their indices, so it can map keys to
      38             : /// arbitrary values. SparseSetValTraits computes the index from the value
      39             : /// object. To compute the index from a key, SparseSet uses a separate
      40             : /// KeyFunctorT template argument.
      41             : ///
      42             : /// A simple type declaration, SparseSet<Type>, handles these cases:
      43             : /// - unsigned key, identity index, identity value
      44             : /// - unsigned key, identity index, fat value providing getSparseSetIndex()
      45             : ///
      46             : /// The type declaration SparseSet<Type, UnaryFunction> handles:
      47             : /// - unsigned key, remapped index, identity value (virtual registers)
      48             : /// - pointer key, pointer-derived index, identity value (node+ID)
      49             : /// - pointer key, pointer-derived index, fat value with getSparseSetIndex()
      50             : ///
      51             : /// Only other, unexpected cases require specializing SparseSetValTraits.
      52             : ///
      53             : /// For best results, ValueT should not require a destructor.
      54             : ///
      55             : template<typename ValueT>
      56             : struct SparseSetValTraits {
      57             :   static unsigned getValIndex(const ValueT &Val) {
      58    71897886 :     return Val.getSparseSetIndex();
      59             :   }
      60             : };
      61             : 
      62             : /// SparseSetValFunctor - Helper class for selecting SparseSetValTraits. The
      63             : /// generic implementation handles ValueT classes which either provide
      64             : /// getSparseSetIndex() or specialize SparseSetValTraits<>.
      65             : ///
      66             : template<typename KeyT, typename ValueT, typename KeyFunctorT>
      67             : struct SparseSetValFunctor {
      68             :   unsigned operator()(const ValueT &Val) const {
      69    71916725 :     return SparseSetValTraits<ValueT>::getValIndex(Val);
      70             :   }
      71             : };
      72             : 
      73             : /// SparseSetValFunctor<KeyT, KeyT> - Helper class for the common case of
      74             : /// identity key/value sets.
      75             : template<typename KeyT, typename KeyFunctorT>
      76             : struct SparseSetValFunctor<KeyT, KeyT, KeyFunctorT> {
      77             :   unsigned operator()(const KeyT &Key) const {
      78    79289720 :     return KeyFunctorT()(Key);
      79             :   }
      80             : };
      81             : 
      82             : /// SparseSet - Fast set implmentation for objects that can be identified by
      83             : /// small unsigned keys.
      84             : ///
      85             : /// SparseSet allocates memory proportional to the size of the key universe, so
      86             : /// it is not recommended for building composite data structures.  It is useful
      87             : /// for algorithms that require a single set with fast operations.
      88             : ///
      89             : /// Compared to DenseSet and DenseMap, SparseSet provides constant-time fast
      90             : /// clear() and iteration as fast as a vector.  The find(), insert(), and
      91             : /// erase() operations are all constant time, and typically faster than a hash
      92             : /// table.  The iteration order doesn't depend on numerical key values, it only
      93             : /// depends on the order of insert() and erase() operations.  When no elements
      94             : /// have been erased, the iteration order is the insertion order.
      95             : ///
      96             : /// Compared to BitVector, SparseSet<unsigned> uses 8x-40x more memory, but
      97             : /// offers constant-time clear() and size() operations as well as fast
      98             : /// iteration independent on the size of the universe.
      99             : ///
     100             : /// SparseSet contains a dense vector holding all the objects and a sparse
     101             : /// array holding indexes into the dense vector.  Most of the memory is used by
     102             : /// the sparse array which is the size of the key universe.  The SparseT
     103             : /// template parameter provides a space/speed tradeoff for sets holding many
     104             : /// elements.
     105             : ///
     106             : /// When SparseT is uint32_t, find() only touches 2 cache lines, but the sparse
     107             : /// array uses 4 x Universe bytes.
     108             : ///
     109             : /// When SparseT is uint8_t (the default), find() touches up to 2+[N/256] cache
     110             : /// lines, but the sparse array is 4x smaller.  N is the number of elements in
     111             : /// the set.
     112             : ///
     113             : /// For sets that may grow to thousands of elements, SparseT should be set to
     114             : /// uint16_t or uint32_t.
     115             : ///
     116             : /// @tparam ValueT      The type of objects in the set.
     117             : /// @tparam KeyFunctorT A functor that computes an unsigned index from KeyT.
     118             : /// @tparam SparseT     An unsigned integer type. See above.
     119             : ///
     120             : template<typename ValueT,
     121             :          typename KeyFunctorT = identity<unsigned>,
     122             :          typename SparseT = uint8_t>
     123             : class SparseSet {
     124             :   static_assert(std::numeric_limits<SparseT>::is_integer &&
     125             :                 !std::numeric_limits<SparseT>::is_signed,
     126             :                 "SparseT must be an unsigned integer type");
     127             : 
     128             :   using KeyT = typename KeyFunctorT::argument_type;
     129             :   using DenseT = SmallVector<ValueT, 8>;
     130             :   using size_type = unsigned;
     131             :   DenseT Dense;
     132             :   SparseT *Sparse = nullptr;
     133             :   unsigned Universe = 0;
     134             :   KeyFunctorT KeyIndexOf;
     135             :   SparseSetValFunctor<KeyT, ValueT, KeyFunctorT> ValIndexOf;
     136             : 
     137             : public:
     138             :   using value_type = ValueT;
     139             :   using reference = ValueT &;
     140             :   using const_reference = const ValueT &;
     141             :   using pointer = ValueT *;
     142             :   using const_pointer = const ValueT *;
     143             : 
     144     2482842 :   SparseSet() = default;
     145             :   SparseSet(const SparseSet &) = delete;
     146             :   SparseSet &operator=(const SparseSet &) = delete;
     147     2481944 :   ~SparseSet() { free(Sparse); }
     148             : 
     149             :   /// setUniverse - Set the universe size which determines the largest key the
     150             :   /// set can hold.  The universe must be sized before any elements can be
     151             :   /// added.
     152             :   ///
     153             :   /// @param U Universe size. All object keys must be less than U.
     154             :   ///
     155      975801 :   void setUniverse(unsigned U) {
     156             :     // It's not hard to resize the universe on a non-empty set, but it doesn't
     157             :     // seem like a likely use case, so we can add that code when we need it.
     158             :     assert(empty() && "Can only resize universe on an empty map");
     159             :     // Hysteresis prevents needless reallocations.
     160      975801 :     if (U >= Universe/4 && U <= Universe)
     161             :       return;
     162      473265 :     free(Sparse);
     163             :     // The Sparse array doesn't actually need to be initialized, so malloc
     164             :     // would be enough here, but that will cause tools like valgrind to
     165             :     // complain about branching on uninitialized data.
     166      473265 :     Sparse = reinterpret_cast<SparseT*>(calloc(U, sizeof(SparseT)));
     167      473265 :     Universe = U;
     168             :   }
     169             : 
     170             :   // Import trivial vector stuff from DenseT.
     171             :   using iterator = typename DenseT::iterator;
     172             :   using const_iterator = typename DenseT::const_iterator;
     173             : 
     174      774926 :   const_iterator begin() const { return Dense.begin(); }
     175    14055796 :   const_iterator end() const { return Dense.end(); }
     176     7679619 :   iterator begin() { return Dense.begin(); }
     177   120025985 :   iterator end() { return Dense.end(); }
     178             : 
     179             :   /// empty - Returns true if the set is empty.
     180             :   ///
     181             :   /// This is not the same as BitVector::empty().
     182             :   ///
     183     2615080 :   bool empty() const { return Dense.empty(); }
     184             : 
     185             :   /// size - Returns the number of elements in the set.
     186             :   ///
     187             :   /// This is not the same as BitVector::size() which returns the size of the
     188             :   /// universe.
     189             :   ///
     190   181936786 :   size_type size() const { return Dense.size(); }
     191             : 
     192             :   /// clear - Clears the set.  This is a very fast constant time operation.
     193             :   ///
     194             :   void clear() {
     195             :     // Sparse does not need to be cleared, see find().
     196     4940736 :     Dense.clear();
     197             :   }
     198             : 
     199             :   /// findIndex - Find an element by its index.
     200             :   ///
     201             :   /// @param   Idx A valid index to find.
     202             :   /// @returns An iterator to the element identified by key, or end().
     203             :   ///
     204             :   iterator findIndex(unsigned Idx) {
     205             :     assert(Idx < Universe && "Key out of range");
     206    67433877 :     const unsigned Stride = std::numeric_limits<SparseT>::max() + 1u;
     207   161434497 :     for (unsigned i = Sparse[Idx], e = size(); i < e; i += Stride) {
     208   157810476 :       const unsigned FoundIdx = ValIndexOf(Dense[i]);
     209             :       assert(FoundIdx < Universe && "Invalid key in set. Did object mutate?");
     210    52603492 :       if (Idx == FoundIdx)
     211             :         return begin() + i;
     212             :       // Stride is 0 when SparseT >= unsigned.  We don't need to loop.
     213             :       if (!Stride)
     214             :         break;
     215             :     }
     216             :     return end();
     217             :   }
     218             : 
     219             :   /// find - Find an element by its key.
     220             :   ///
     221             :   /// @param   Key A valid key to find.
     222             :   /// @returns An iterator to the element identified by key, or end().
     223             :   ///
     224             :   iterator find(const KeyT &Key) {
     225    40837730 :     return findIndex(KeyIndexOf(Key));
     226             :   }
     227             : 
     228             :   const_iterator find(const KeyT &Key) const {
     229    26816749 :     return const_cast<SparseSet*>(this)->findIndex(KeyIndexOf(Key));
     230             :   }
     231             : 
     232             :   /// count - Returns 1 if this set contains an element identified by Key,
     233             :   /// 0 otherwise.
     234             :   ///
     235             :   size_type count(const KeyT &Key) const {
     236    21292734 :     return find(Key) == end() ? 0 : 1;
     237             :   }
     238             : 
     239             :   /// insert - Attempts to insert a new element.
     240             :   ///
     241             :   /// If Val is successfully inserted, return (I, true), where I is an iterator
     242             :   /// pointing to the newly inserted element.
     243             :   ///
     244             :   /// If the set already contains an element with the same key as Val, return
     245             :   /// (I, false), where I is an iterator pointing to the existing element.
     246             :   ///
     247             :   /// Insertion invalidates all iterators.
     248             :   ///
     249    33679127 :   std::pair<iterator, bool> insert(const ValueT &Val) {
     250    67358254 :     unsigned Idx = ValIndexOf(Val);
     251    33679127 :     iterator I = findIndex(Idx);
     252    33679127 :     if (I != end())
     253    10513371 :       return std::make_pair(I, false);
     254    23165756 :     Sparse[Idx] = size();
     255    23165756 :     Dense.push_back(Val);
     256    46331512 :     return std::make_pair(end() - 1, true);
     257             :   }
     258             : 
     259             :   /// array subscript - If an element already exists with this key, return it.
     260             :   /// Otherwise, automatically construct a new value from Key, insert it,
     261             :   /// and return the newly inserted element.
     262             :   ValueT &operator[](const KeyT &Key) {
     263       11564 :     return *insert(ValueT(Key)).first;
     264             :   }
     265             : 
     266             :   ValueT pop_back_val() {
     267             :     // Sparse does not need to be cleared, see find().
     268     4264052 :     return Dense.pop_back_val();
     269             :   }
     270             : 
     271             :   /// erase - Erases an existing element identified by a valid iterator.
     272             :   ///
     273             :   /// This invalidates all iterators, but erase() returns an iterator pointing
     274             :   /// to the next element.  This makes it possible to erase selected elements
     275             :   /// while iterating over the set:
     276             :   ///
     277             :   ///   for (SparseSet::iterator I = Set.begin(); I != Set.end();)
     278             :   ///     if (test(*I))
     279             :   ///       I = Set.erase(I);
     280             :   ///     else
     281             :   ///       ++I;
     282             :   ///
     283             :   /// Note that end() changes when elements are erased, unlike std::list.
     284             :   ///
     285             :   iterator erase(iterator I) {
     286             :     assert(unsigned(I - begin()) < size() && "Invalid iterator");
     287     9402687 :     if (I != end() - 1) {
     288    12402372 :       *I = Dense.back();
     289    18603558 :       unsigned BackIdx = ValIndexOf(Dense.back());
     290             :       assert(BackIdx < Universe && "Invalid key in set. Did object mutate?");
     291    12402372 :       Sparse[BackIdx] = I - begin();
     292             :     }
     293             :     // This depends on SmallVector::pop_back() not invalidating iterators.
     294             :     // std::vector::pop_back() doesn't give that guarantee.
     295    18805374 :     Dense.pop_back();
     296             :     return I;
     297             :   }
     298             : 
     299             :   /// erase - Erases an element identified by Key, if it exists.
     300             :   ///
     301             :   /// @param   Key The key identifying the element to erase.
     302             :   /// @returns True when an element was erased, false if no element was found.
     303             :   ///
     304    18518771 :   bool erase(const KeyT &Key) {
     305    37037542 :     iterator I = find(Key);
     306    18518771 :     if (I == end())
     307             :       return false;
     308     8270977 :     erase(I);
     309             :     return true;
     310             :   }
     311             : };
     312             : 
     313             : } // end namespace llvm
     314             : 
     315             : #endif // LLVM_ADT_SPARSESET_H

Generated by: LCOV version 1.13