CrsMatrix

CrsMatrix class

The CrsMatrix class provides a compressed sparse row implementation of a sparse matrix, as described, for example, in Saad (2nd ed.).

KokkosSparse::CrsMatrix API

template<class ScalarType,
         class OrdinalType,
         class Device,
         class MemoryTraits = void,
         class SizeType = typename Kokkos::ViewTraits<OrdinalType*, Device, void, void>::size_type>
class CrsMatrix {
public:
  //! Type of the matrix's execution space.
  typedef typename Device::execution_space execution_space;
  //! Type of the matrix's memory space.
  typedef typename Device::memory_space memory_space;
  //! Type of the matrix's device type.
  typedef Kokkos::Device<execution_space, memory_space> device_type;

  //! Type of each value in the matrix.
  typedef ScalarType value_type;
  //! Type of each (column) index in the matrix.
  typedef OrdinalType ordinal_type;
  typedef MemoryTraits memory_traits;
  /// \brief Type of each entry of the "row map."
  ///
  /// The "row map" corresponds to the \c ptr array of row offsets in
  /// compressed sparse row (CSR) storage.
  typedef SizeType size_type;

  //! Type of a host-memory mirror of the sparse matrix.
  typedef CrsMatrix<ScalarType, OrdinalType, host_mirror_space, MemoryTraits> HostMirror;
  //! Type of the graph structure of the sparse matrix.
  typedef Kokkos::StaticCrsGraph<ordinal_type, Kokkos::LayoutLeft, execution_space, memory_traits, size_type> StaticCrsGraphType;
  //! Type of the graph structure of the sparse matrix - consistent with Kokkos.
  typedef Kokkos::StaticCrsGraph<ordinal_type, Kokkos::LayoutLeft, execution_space, memory_traits, size_type> staticcrsgraph_type;
  //! Type of column indices in the sparse matrix.
  typedef typename staticcrsgraph_type::entries_type index_type;
  //! Const version of the type of column indices in the sparse matrix.
  typedef typename index_type::const_value_type const_ordinal_type;
  //! Nonconst version of the type of column indices in the sparse matrix.
  typedef typename index_type::non_const_value_type non_const_ordinal_type;
  //! Type of the "row map" (which contains the offset for each row's data).
  typedef typename staticcrsgraph_type::row_map_type row_map_type;
  //! Const version of the type of row offsets in the sparse matrix.
  typedef typename row_map_type::const_value_type const_size_type;
  //! Nonconst version of the type of row offsets in the sparse matrix.
  typedef typename row_map_type::non_const_value_type non_const_size_type;
  //! Kokkos Array type of the entries (values) in the sparse matrix.
  typedef Kokkos::View<value_type*, Kokkos::LayoutRight, device_type, MemoryTraits> values_type;
  //! Const version of the type of the entries in the sparse matrix.
  typedef typename values_type::const_value_type const_value_type;
  //! Nonconst version of the type of the entries in the sparse matrix.
  typedef typename values_type::non_const_value_type non_const_value_type;


  /// \name Storage of the actual sparsity structure and values.
  ///
  /// CrsMatrix uses the compressed sparse row (CSR) storage format to
  /// store the sparse matrix.  CSR is also called "compressed row
  /// storage"; hence the name, which it inherits from Tpetra and from
  /// Epetra before it.
  //@{
  //! The graph (sparsity structure) of the sparse matrix.
  staticcrsgraph_type graph;
  //! The 1-D array of values of the sparse matrix.
  values_type values;
  //@}

  /// \brief Default constructor; constructs an empty sparse matrix.
  KOKKOS_INLINE_FUNCTION
  CrsMatrix () :
    numCols_ (0)
  {}

  //! Copy constructor (shallow copy).
  template<typename SType,
           typename OType,
           class DType,
           class MTType,
           typename IType>
  KOKKOS_INLINE_FUNCTION
  CrsMatrix (const CrsMatrix<SType,OType,DType,MTType,IType> & B)
}

  /// \brief Construct with a graph that will be shared.
  ///
  /// Allocate the values array for subsquent fill.
  CrsMatrix (const std::string& arg_label,
             const staticcrsgraph_type& arg_graph)

  /// \brief Constructor that copies raw arrays of host data in
  ///   coordinate format.
  ///
  /// On input, each entry of the sparse matrix is stored in val[k],
  /// with row index rows[k] and column index cols[k].  We assume that
  /// the entries are sorted in increasing order by row index.
  ///
  /// This constructor is mainly useful for benchmarking or for
  /// reading the sparse matrix's data from a file.
  ///
  /// \param label [in] The sparse matrix's label.
  /// \param nrows [in] The number of rows.
  /// \param ncols [in] The number of columns.
  /// \param annz [in] The number of entries.
  /// \param val [in] The entries.
  /// \param rows [in] The row indices.  rows[k] is the row index of
  ///   val[k].
  /// \param cols [in] The column indices.  cols[k] is the column
  ///   index of val[k].
  /// \param pad [in] If true, pad the sparse matrix's storage with
  ///   zeros in order to improve cache alignment and / or
  ///   vectorization.
  CrsMatrix (const std::string &label,
             OrdinalType nrows,
             OrdinalType ncols,
             size_type annz,
             ScalarType* val,
             OrdinalType* rows,
             OrdinalType* cols)

  /// \brief Constructor that accepts a row map, column indices, and
  ///   values.
  ///
  /// The matrix will store and use the row map, indices, and values
  /// directly (by view, not by deep copy).
  ///
  /// \param label [in] The sparse matrix's label.
  /// \param nrows [in] The number of rows.
  /// \param ncols [in] The number of columns.
  /// \param annz [in] The number of entries.
  /// \param vals [in/out] The entries.
  /// \param rows [in/out] The row map (containing the offsets to the
  ///   data in each row).
  /// \param cols [in/out] The column indices.
  CrsMatrix (const std::string& /* label */,
             const OrdinalType nrows,
             const OrdinalType ncols,
             const size_type annz,
             const values_type& vals,
             const row_map_type& rows,
             const index_type& cols)

  /// \brief Constructor that accepts a a static graph, and values.
  ///
  /// The matrix will store and use the row map, indices, and values
  /// directly (by view, not by deep copy).
  ///
  /// \param label [in] The sparse matrix's label.
  /// \param nrows [in] The number of rows.
  /// \param ncols [in] The number of columns.
  /// \param annz [in] The number of entries.
  /// \param vals [in/out] The entries.
  /// \param rows [in/out] The row map (containing the offsets to the
  ///   data in each row).
  /// \param cols [in/out] The column indices.
  CrsMatrix (const std::string& /* label */,
             const OrdinalType& ncols,
             const values_type& vals,
             const staticcrsgraph_type& graph_)

  //! Attempt to assign the input matrix to \c *this.
  template<typename aScalarType, typename aOrdinalType, class aDevice, class aMemoryTraits,typename aSizeType>
  CrsMatrix&
  operator= (const CrsMatrix<aScalarType, aOrdinalType, aDevice, aMemoryTraits, aSizeType>& mtx)

  //! The number of rows in the sparse matrix.
  KOKKOS_INLINE_FUNCTION ordinal_type numRows () const

  //! The number of columns in the sparse matrix.
  KOKKOS_INLINE_FUNCTION ordinal_type numCols () const

  //! The number of stored entries in the sparse matrix.
  KOKKOS_INLINE_FUNCTION size_type nnz () const

Sparse linear algebra functions that support a CrsMatrix interface

1. SpMV

New interfaces under development

2. SpADD
3. SpGEMM