SPARSE – KokkosKernels sparse interfaces

crsmatrix

template<class ScalarType, class OrdinalType, class Device, class MemoryTraits = void, class SizeType = KokkosKernels::default_size_type>
class CrsMatrix

Compressed sparse row implementation of a sparse matrix.

“Crs” stands for “compressed row sparse.” This is the phrase Trilinos traditionally uses to describe compressed sparse row storage for sparse matrices, as described, for example, in Saad (2nd ed.).

Template Parameters:

• ScalarType – The type of entries in the sparse matrix.

• OrdinalType – The type of column indices in the sparse matrix.

• Device – The Kokkos Device type.

• MemoryTraits – Traits describing how Kokkos manages and accesses data. The default parameter suffices for most users.

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.

staticcrsgraph_type graph

The graph (sparsity structure) of the sparse matrix.

values_type values

The 1-D array of values of the sparse matrix.

DeviceConfig dev_config

Launch configuration that can be used by overloads/specializations of MV_multiply().

This is a hack and needs to be replaced by a general state mechanism.

inline CrsMatrix()

Default constructor; constructs an empty sparse matrix.

FIXME (mfh 09 Aug 2013) numCols and nnz should be properties of the graph, not the matrix. Then CrsMatrix needs methods to get these from the graph.

template<typename InScalar, typename InOrdinal, class InDevice, class InMemTraits, typename InSizeType>
inline CrsMatrix(const CrsMatrix<InScalar, InOrdinal, InDevice, InMemTraits, InSizeType> &B)

Copy constructor (shallow copy).

template<typename InScalar, typename InOrdinal, typename InDevice, typename InMemTraits, typename InSizeType>
inline CrsMatrix(const std::string&, const CrsMatrix<InScalar, InOrdinal, InDevice, InMemTraits, InSizeType> &mat_)

Deep copy constructor (can cross spaces)

template<typename InOrdinal, typename InLayout, typename InDevice, typename InMemTraits, typename InSizeType>
inline CrsMatrix(const std::string &label, const StaticCrsGraph<InOrdinal, InLayout, InDevice, InMemTraits, InSizeType> &graph_)

Construct with a graph that will be shared.

Allocate the values array for subsequent fill.

template<typename InOrdinal, typename InLayout, typename InDevice, typename InMemTraits, typename InSizeType>
inline CrsMatrix(const std::string &label, const StaticCrsGraph<InOrdinal, InLayout, InDevice, InMemTraits, InSizeType> &graph_, const OrdinalType &ncols)

Constructor that accepts a a static graph, and numCols.

The matrix will store and use the row map, indices (by view, not by deep copy) and allocate the values view.

Parameters:

• label – [in] The sparse matrix’s label.

• graph_ – [in] The graph for storing the rowmap and col ids.

• ncols – [in] The number of columns.

template<typename InOrdinal, typename InLayout, typename InDevice, typename InMemTraits, typename InSizeType>
inline CrsMatrix(const std::string&, const OrdinalType &ncols, const values_type &vals, const StaticCrsGraph<InOrdinal, InLayout, InDevice, InMemTraits, InSizeType> &graph_)

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).

Parameters:

• ncols – [in] The number of columns.

• vals – [in/out] The entries.

• graph_ – The graph for storing the rowmap and col ids.

inline CrsMatrix(const std::string&, OrdinalType nrows, OrdinalType ncols, size_type annz, ScalarType *val, OrdinalType *rowmap, OrdinalType *cols)

Constructor that copies raw arrays of host data in 3-array CRS (compressed row storage) format.

On input, the entries must be sorted by row. rowmap determines where each row begins and ends. For each entry k (0 <= k < annz), cols[k] gives the adjacent column, and val[k] gives the corresponding matrix value.

This constructor is mainly useful for benchmarking or for reading the sparse matrix’s data from a file.

Parameters:

• nrows – [in] The number of rows.

• ncols – [in] The number of columns.

• annz – [in] The number of entries.

• val – [in] The values.

• rowmap – [in] The row offsets. The values/columns in row k begin at index rowmap[k] and end at rowmap[k+1]-1 (inclusive). This means the array must have length nrows+1.

• cols – [in] The column indices. cols[k] is the column index of entry k, with a corresponding value of val[k] .

inline CrsMatrix(const std::string&, const OrdinalType nrows, const OrdinalType ncols, const size_type annz, const values_type &vals, const row_map_type &rowmap, const index_type &cols)

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).

Parameters:

• nrows – [in] The number of rows.

• ncols – [in] The number of columns.

• annz – [in] The number of entries.

• vals – [in] The entries.

• rowmap – [in] The row map (containing the offsets to the data in each row).

• cols – [in] The column indices.

template<typename aScalarType, typename aOrdinalType, class aDevice, class aMemoryTraits, typename aSizeType>
inline CrsMatrix &operator=(const CrsMatrix<aScalarType, aOrdinalType, aDevice, aMemoryTraits, aSizeType> &mtx)

Attempt to assign the input matrix to *this.

inline ordinal_type numRows() const

The number of rows in the sparse matrix.

inline ordinal_type numCols() const

The number of columns in the sparse matrix.

inline ordinal_type numPointRows() const

The number of “point” (non-block) rows in the matrix. Since Crs is not blocked, this is just the number of regular rows.

inline ordinal_type numPointCols() const

The number of “point” (non-block) columns in the matrix. Since Crs is not blocked, this is just the number of regular columns.

inline size_type nnz() const

The number of stored entries in the sparse matrix.

inline SparseRowView<CrsMatrix> row(const ordinal_type i) const

Return a view of row i of the matrix.

If row i does not belong to the matrix, return an empty view.

The returned object view implements the following interface:

• view.length is the number of entries in the row

• view.value(k) returns a nonconst reference to the value of the k-th entry in the row

• view.colidx(k) returns a nonconst reference to the column index of the k-th entry in the row

k is not a column index; it just counts from 0 to view.length - 1.

Users should not rely on the return type of this method. They should instead assign to ‘auto’. That allows compile-time polymorphism for different kinds of sparse matrix formats (e.g., ELLPACK or Jagged Diagonal) that we may wish to support in the future.

Both row() and rowConst() used to take a “SizeType” template parameter, which was the type to use for row offsets. This is unnecessary, because the CrsMatrix specialization already has the row offset type available, via the size_type typedef. Our sparse matrix-vector multiply implementation for CrsMatrix safely uses ordinal_type rather than size_type to iterate over all the entries in a row of the sparse matrix. Since ordinal_type may be smaller than size_type, compilers may generate more efficient code. The row() and rowConst() methods first compute the difference of consecutive row offsets as size_type, and then cast to ordinal_type. If you want to do this yourself, here is an example:

for (ordinal_type lclRow = 0; lclRow < A.numRows (); ++lclRow) {
  const ordinal_type numEnt =
    static_cast<ordinal_type> (A.graph.row_map(i+1) - A.graph.row_map(i));
  for (ordinal_type k = 0; k < numEnt; ++k) {
    // etc.
  }
}

inline SparseRowViewConst<CrsMatrix> rowConst(const ordinal_type i) const

Return a const view of row i of the matrix.

If row i does not belong to the matrix, return an empty view.

The returned object view implements the following interface:

• view.length is the number of entries in the row

• view.value(k) returns a const reference to the value of the k-th entry in the row

• view.colidx(k) returns a const reference to the column index of the k-th entry in the row

k is not a column index; it just counts from 0 to view.length - 1.

Users should not rely on the return type of this method. They should instead assign to ‘auto’. That allows compile-time polymorphism for different kinds of sparse matrix formats (e.g., ELLPACK or Jagged Diagonal) that we may wish to support in the future.

Both row() and rowConst() used to take a “SizeType” template parameter, which was the type to use for row offsets. This is unnecessary, because the CrsMatrix specialization already has the row offset type available, via the size_type typedef. Our sparse matrix-vector multiply implementation for CrsMatrix safely uses ordinal_type rather than size_type to iterate over all the entries in a row of the sparse matrix. Since ordinal_type may be smaller than size_type, compilers may generate more efficient code. The row() and rowConst() methods first compute the difference of consecutive row offsets as size_type, and then cast to ordinal_type. If you want to do this yourself, here is an example:

for (ordinal_type lclRow = 0; lclRow < A.numRows (); ++lclRow) {
  const ordinal_type numEnt =
    static_cast<ordinal_type> (A.graph.row_map(i+1) - A.graph.row_map(i));
  for (ordinal_type k = 0; k < numEnt; ++k) {
    // etc.
  }
}

Public Types

typedef Device::execution_space execution_space

Type of the matrix’s execution space.

typedef Device::memory_space memory_space

Type of the matrix’s memory space.

typedef Kokkos::Device<execution_space, memory_space> device_type

Canonical device type.

typedef ScalarType value_type

Type of each value in the matrix.

typedef OrdinalType ordinal_type

Type of each (column) index in the matrix.

typedef SizeType size_type

Type of each entry of the “row map.”.

The “row map” corresponds to the ptr array of row offsets in compressed sparse row (CSR) storage.

typedef CrsMatrix<ScalarType, OrdinalType, host_mirror_space, MemoryTraits, SizeType> HostMirror

Type of a host-memory mirror of the sparse matrix.

typedef StaticCrsGraph<ordinal_type, KokkosKernels::default_layout, device_type, memory_traits, size_type> StaticCrsGraphType

Type of the graph structure of the sparse matrix.

typedef StaticCrsGraph<ordinal_type, KokkosKernels::default_layout, device_type, memory_traits, size_type> staticcrsgraph_type

Type of the graph structure of the sparse matrix - consistent with Kokkos.

typedef staticcrsgraph_type::entries_type index_type

Type of column indices in the sparse matrix.

typedef index_type::const_value_type const_ordinal_type

Const version of the type of column indices in the sparse matrix.

typedef index_type::non_const_value_type non_const_ordinal_type

Nonconst version of the type of column indices in the sparse matrix.

typedef staticcrsgraph_type::row_map_type row_map_type

Type of the “row map” (which contains the offset for each row’s data).

typedef row_map_type::const_value_type const_size_type

Const version of the type of row offsets in the sparse matrix.

typedef row_map_type::non_const_value_type non_const_size_type

Nonconst version of the type of row offsets in the sparse matrix.

typedef Kokkos::View<value_type*, Kokkos::LayoutRight, device_type, MemoryTraits> values_type

Kokkos Array type of the entries (values) in the sparse matrix.

typedef values_type::const_value_type const_value_type

Const version of the type of the entries in the sparse matrix.

typedef values_type::non_const_value_type non_const_value_type

Nonconst version of the type of the entries in the sparse matrix.

ccsmatrix

template<class ScalarType, class OrdinalType, class Device, class MemoryTraits = void, class SizeType = typename Kokkos::ViewTraits<OrdinalType*, Device, void, void>::size_type>
class CcsMatrix

Compressed sparse column implementation of a sparse matrix.

“Ccs” stands for “compressed column sparse.”

Template Parameters:

• ScalarType – The type of entries in the sparse matrix.

• OrdinalType – The type of column indices in the sparse matrix.

• Device – The Kokkos Device type.

• MemoryTraits – Traits describing how Kokkos manages and accesses data. The default parameter suffices for most users.

Storage of the actual sparsity structure and values.

CcsMatrix uses the compressed sparse column (CCS) storage format to store the sparse matrix.

staticccsgraph_type graph

The graph (sparsity structure) of the sparse matrix.

values_type values

The 1-D array of values of the sparse matrix.

inline CcsMatrix()

Default constructor; constructs an empty sparse matrix.

inline CcsMatrix(const std::string&, const OrdinalType nrows, const OrdinalType ncols, const size_type annz, const values_type &vals, const col_map_type &colmap, const index_type &rows)

Constructor that accepts a column map, row indices, and values.

The matrix will store and use the column map, indices, and values directly (by view, not by deep copy).

Parameters:

• nrows – [in] The number of rows.

• ncols – [in] The number of columns.

• annz – [in] The number of entries.

• vals – [in] The entries.

• colmap – [in] The column map (containing the offsets to the data in each column).

• rows – [in] The row indices.

inline ordinal_type numCols() const

The number of rows in the sparse matrix.

inline ordinal_type numRows() const

The number of columns in the sparse matrix.

inline ordinal_type numPointRows() const

The number of “point” (non-block) rows in the matrix. Since Ccs is not blocked, this is just the number of regular rows.

inline ordinal_type numPointCols() const

The number of “point” (non-block) columns in the matrix. Since Ccs is not blocked, this is just the number of regular columns.

inline size_type nnz() const

The number of stored entries in the sparse matrix.

Public Types

typedef Device::execution_space execution_space

Type of the matrix’s execution space.

typedef Device::memory_space memory_space

Type of the matrix’s memory space.

typedef Kokkos::Device<execution_space, memory_space> device_type

Canonical device type.

typedef SizeType size_type

Type of each entry of the “column map.”.

The “column map” corresponds to the ptr array of column offsets in compressed sparse column (CCS) storage.

typedef ScalarType value_type

Type of each value in the matrix.

typedef OrdinalType ordinal_type

Type of each (column) index in the matrix.

typedef StaticCcsGraph<ordinal_type, KokkosKernels::default_layout, device_type, memory_traits, size_type> staticccsgraph_type

Type of the graph structure of the sparse matrix - consistent with Kokkos.

typedef staticccsgraph_type::col_map_type col_map_type

Type of the “column map” (which contains the offset for each column’s data).

typedef staticccsgraph_type::entries_type index_type

Type of column indices in the sparse matrix.

coomatrix

template<class ScalarType, class OrdinalType, class Device, class MemoryTraits = void, class SizeType = typename Kokkos::ViewTraits<OrdinalType*, Device, void, void>::size_type>
class CooMatrix

Coordinate format implementation of a sparse matrix.

Template Parameters:

• ScalarType – The type of scalar entries in the sparse matrix.

• OrdinalType – The type of index entries in the sparse matrix.

• Device – The Kokkos Device type.

• MemoryTraits – Traits describing how Kokkos manages and accesses data. The default parameter suffices for most users. “Coo” stands for “coordinate format”.

Public Types

using scalar_type = ScalarType

Type of each value in the matrix.

using const_scalar_type = const std::remove_const_t<scalar_type>

Type of each value in the const matrix.

using non_const_scalar_type = std::remove_const_t<scalar_type>

Non constant scalar type.

using ordinal_type = OrdinalType

Type of each index in the matrix.

using const_ordinal_type = const std::remove_const_t<ordinal_type>

Type of each value in the const matrix.

using non_const_ordinal_type = std::remove_const_t<ordinal_type>

Non constant ordinal type.

using row_type = ordinal_type

Type of each row index in the matrix.

using column_type = ordinal_type

Type of each column index in the matrix.

using device_type = Device

Type of the Kokkos::Device.

using execution_space = typename device_type::execution_space

Type of the Kokkos::Device::execution_space.

using memory_space = typename device_type::memory_space

Type of the Kokkos::Device::memory_space.

using memory_traits = MemoryTraits

Type of the Kokkos::MemoryTraits.

using size_type = SizeType

Type of all integral class members.

using row_view = Kokkos::View<row_type*, Kokkos::LayoutRight, device_type, memory_traits>

The type of the row index view in the matrix.

using column_view = Kokkos::View<column_type*, Kokkos::LayoutRight, device_type, memory_traits>

The type of the column index view in the matrix.

using scalar_view = Kokkos::View<scalar_type*, Kokkos::LayoutRight, device_type, memory_traits>

The type of the scalar values view in the matrix.

using const_type = CooMatrix<const_scalar_type, const_ordinal_type, device_type, memory_traits, size_type>

The type of a constant CooMatrix.

Public Functions

inline CooMatrix()

Default constructor; constructs an empty sparse matrix.

inline CooMatrix(size_type nrows, size_type ncols, row_view row_in, column_view col_in, scalar_view data_in)

Constructor that accepts a column indicies view, row indices view, and values view.

The matrix will store and use the column indices, rows indices, and values directly (by view, not by deep copy).

Parameters:

• nrows – [in] The number of rows.

• ncols – [in] The number of columns.

• row_in – [in] The row indexes.

• col_in – [in] The column indexes.

• data_in – [in] The values.

inline size_type numCols() const

The number of columns in the sparse matrix.

inline size_type numRows() const

The number of rows in the sparse matrix.

inline size_type nnz() const

The number of stored entries in the sparse matrix, including zeros.

inline row_view row() const

The row indexes of the matrix.

inline column_view col() const

The column indexes of the matrix.

inline scalar_view data() const

The scalar values of the matrix.

crs2ccs

template<class OrdinalType, class SizeType, class ValViewType, class RowMapViewType, class ColIdViewType>
auto KokkosSparse::crs2ccs(OrdinalType nrows, OrdinalType ncols, SizeType nnz, ValViewType vals, RowMapViewType row_map, ColIdViewType col_ids)

Blocking function that converts a CrsMatrix to a CcsMatrix. Crs values are copied from row-contiguous layout into column-contiguous layout.

Note

In KokkosKernels sparse code, adj stands for adjacency list and here we’re passing in a crs matrix with xadj=row_map and adj=col_ids.

Template Parameters:

• OrdinalType – The view value type associated with the RowIdViewType

• SizeType – The type of nnz

• ValViewType – The values view type

• RowMapViewType – The column map view type

• ColIdViewType – The row ids view type

Parameters:

• nrows – The number of rows in the crs matrix

• ncols – The number of columns in the crs matrix

• nnz – The number of non-zeros in the crs matrix

• vals – The values view of the crs matrix

• row_map – The row map view of the crs matrix

• col_ids – The col ids view of the crs matrix

Returns:

A KokkosSparse::CcsMatrix.

template<typename ScalarType, typename OrdinalType, class DeviceType, class MemoryTraitsType, typename SizeType>
auto KokkosSparse::crs2ccs(KokkosSparse::CrsMatrix<ScalarType, OrdinalType, DeviceType, MemoryTraitsType, SizeType> &crsMatrix)

Blocking function that converts a crs matrix to a CcsMatrix. Crs values are copied from row-contiguous layout into column-contiguous layout.

Template Parameters:

• ScalarType – The crsMatrix::scalar_type

• OrdinalType – The crsMatrix::ordinal_type

• DeviceType – The crsMatrix::device_type

• MemoryTraits – The crsMatrix::memory_traits

• SizeType – The crsMatrix::size_type

Parameters:

crsMatrix – The KokkosSparse::CrsMatrix.

Returns:

A KokkosSparse::CcsMatrix.

ccs2crs

template<class OrdinalType, class SizeType, class ValViewType, class ColMapViewType, class RowIdViewType>
auto KokkosSparse::ccs2crs(OrdinalType nrows, OrdinalType ncols, SizeType nnz, ValViewType vals, ColMapViewType col_map, RowIdViewType row_ids)

Blocking function that converts a ccs matrix to a CrsMatrix. Ccs values are copied from column-contiguous layout into row-contiguous layout.

Note

In KokkosKernels sparse code, adj stands for adjacency list and here we’re passing in a ccs matrix with xadj=col_map and adj=row_ids.

Template Parameters:

• OrdinalType – The view value type associated with the RowIdViewType

• SizeType – The type of nnz

• ValViewType – The values view type

• ColMapViewType – The column map view type

• RowIdViewType – The row ids view type

Parameters:

• nrows – The number of rows in the ccs matrix

• ncols – The number of columns in the ccs matrix

• nnz – The number of non-zeros in the ccs matrix

• vals – The values view of the ccs matrix

• col_map – The column map view of the ccs matrix

• row_ids – The row ids view of the ccs matrix

Returns:

A KokkosSparse::CrsMatrix.

template<typename ScalarType, typename OrdinalType, class DeviceType, class MemoryTraitsType, typename SizeType>
auto KokkosSparse::ccs2crs(KokkosSparse::CcsMatrix<ScalarType, OrdinalType, DeviceType, MemoryTraitsType, SizeType> &ccsMatrix)

Blocking function that converts a crs matrix to a CcsMatrix. Ccs values are copied from column-contiguous layout into row-contiguous layout.

Template Parameters:

• ScalarType – The ccsMatrix::scalar_type

• OrdinalType – The ccsMatrix::ordinal_type

• DeviceType – The ccsMatrix::device_type

• MemoryTraits – The ccsMatrix::memory_traits

• SizeType – The ccsMatrix::size_type

Parameters:

ccsMatrix – The KokkosSparse::CcsMatrix.

Returns:

A KokkosSparse::CrsMatrix.

coo2crs

template<class DimType, class RowViewType, class ColViewType, class DataViewType>
auto KokkosSparse::coo2crs(DimType m, DimType n, RowViewType row, ColViewType col, DataViewType data)

Blocking function that converts a CooMatrix into a CrsMatrix. Values are summed.

Template Parameters:

• DimType – the dimension type

• RowViewType – The row array view type

• ColViewType – The column array view type

• DataViewType – The data array view type

Parameters:

• m – the number of rows

• n – the number of columns

• row – the array of row ids

• col – the array of col ids

• data – the array of data

Returns:

A KokkosSparse::CrsMatrix.

template<typename ScalarType, typename OrdinalType, class DeviceType, class MemoryTraitsType, typename SizeType>
auto KokkosSparse::coo2crs(KokkosSparse::CooMatrix<ScalarType, OrdinalType, DeviceType, MemoryTraitsType, SizeType> &cooMatrix)

Blocking function that converts a CooMatrix into a CrsMatrix. Values are summed.

Template Parameters:

• ScalarType – The KokkosSparse::CooMatrix::scalar_type

• OrdinalType – The KokkosSparse::CooMatrix::ordinal_type

• DeviceType – The KokkosSparse::CooMatrix::device_type

• MemoryTraits – The KokkosSparse::CooMatrix::memory_traits

• SizeType – The KokkosSparse::CooMatrix::size_type

Parameters:

cooMatrix – The sparse matrix stored in coordinate (“Coo”) format.

Returns:

A KokkosSparse::CrsMatrix.

crs2coo

template<class OrdinalType, class SizeType, class ValViewType, class RowMapViewType, class ColIdViewType>
auto KokkosSparse::crs2coo(OrdinalType nrows, OrdinalType ncols, SizeType nnz, ValViewType vals, RowMapViewType row_map, ColIdViewType col_ids)

Blocking function that converts a CrsMatrix to a CooMatrix. Crs values are copied into the CooMatrix in the order they appear within the CrsMatrix, starting from row 0 to row nrows - 1.

Template Parameters:

• OrdinalType – The view value type associated with the RowIdViewType

• SizeType – The type of nnz

• ValViewType – The values view type

• RowMapViewType – The column map view type

• ColIdViewType – The row ids view type

Parameters:

• nrows – The number of rows in the crs matrix

• ncols – The number of columns in the crs matrix

• nnz – The number of non-zeros in the crs matrix

• vals – The values view of the crs matrix

• row_map – The row map view of the crs matrix

• col_ids – The col ids view of the crs matrix

Returns:

A KokkosSparse::CooMatrix.

template<typename ScalarType, typename OrdinalType, class DeviceType, class MemoryTraitsType, typename SizeType>
auto KokkosSparse::crs2coo(KokkosSparse::CrsMatrix<ScalarType, OrdinalType, DeviceType, MemoryTraitsType, SizeType> &crsMatrix)

Blocking function that converts a CrsMatrix to a CooMatrix. Crs values are copied into the CooMatrix in the order they appear within the CrsMatrix, starting from row 0 to row nrows - 1.

Template Parameters:

• ScalarType – The crsMatrix::scalar_type

• OrdinalType – The crsMatrix::ordinal_type

• DeviceType – The crsMatrix::device_type

• MemoryTraits – The crsMatrix::memory_traits

• SizeType – The crsMatrix::size_type

Parameters:

crsMatrix – The KokkosSparse::CrsMatrix.

Returns:

A KokkosSparse::CooMatrix.

spmv

template<class ExecutionSpace, class AlphaType, class AMatrix, class XVector, class BetaType, class YVector>
void KokkosSparse::spmv(const ExecutionSpace &space, KokkosKernels::Experimental::Controls controls, const char mode[], const AlphaType &alpha, const AMatrix &A, const XVector &x, const BetaType &beta, const YVector &y)

template<class AlphaType, class AMatrix, class XVector, class BetaType, class YVector>
void KokkosSparse::spmv(KokkosKernels::Experimental::Controls controls, const char mode[], const AlphaType &alpha, const AMatrix &A, const XVector &x, const BetaType &beta, const YVector &y)

template<class ExecutionSpace, class AlphaType, class AMatrix, class XVector, class BetaType, class YVector, typename = std::enable_if_t<Kokkos::is_execution_space<ExecutionSpace>::value>>
void KokkosSparse::spmv(const ExecutionSpace &space, const char mode[], const AlphaType &alpha, const AMatrix &A, const XVector &x, const BetaType &beta, const YVector &y)

Kokkos sparse matrix-vector multiply. Computes y := alpha*Op(A)*x + beta*y, where Op(A) is controlled by mode (see below).

Template Parameters:

• ExecutionSpace – A Kokkos execution space. Must be able to access the memory spaces of A, x, and y.

• AlphaType – Type of coefficient alpha. Must be convertible to YVector::value_type.

• AMatrix – A KokkosSparse::CrsMatrix, or KokkosSparse::Experimental::BsrMatrix

• XVector – Type of x, must be a rank-1 or rank-2 Kokkos::View

• BetaType – Type of coefficient beta. Must be convertible to YVector::value_type.

• YVector – Type of y, must be a Kokkos::View and its rank must match that of XVector

Parameters:

• space – [in] The execution space instance on which to run the kernel.

• mode – [in] Select A’s operator mode: “N” for normal, “T” for transpose, “C” for conjugate or “H” for conjugate transpose.

• alpha – [in] Scalar multiplier for the matrix A.

• A – [in] The sparse matrix A.

• x – [in] A vector to multiply on the left by A.

• beta – [in] Scalar multiplier for the vector y.

• y – [in/out] Result vector.

template<class AlphaType, class AMatrix, class XVector, class BetaType, class YVector>
void KokkosSparse::spmv(const char mode[], const AlphaType &alpha, const AMatrix &A, const XVector &x, const BetaType &beta, const YVector &y)

Kokkos sparse matrix-vector multiply. Computes y := alpha*Op(A)*x + beta*y, where Op(A) is controlled by mode (see below).

Template Parameters:

• AlphaType – Type of coefficient alpha. Must be convertible to YVector::value_type.

• AMatrix – A KokkosSparse::CrsMatrix, or KokkosSparse::Experimental::BsrMatrix

• XVector – Type of x, must be a rank-1 or rank-2 Kokkos::View

• BetaType – Type of coefficient beta. Must be convertible to YVector::value_type.

• YVector – Type of y, must be a Kokkos::View and its rank must match that of XVector

Parameters:

• mode – [in] Select A’s operator mode: “N” for normal, “T” for transpose, “C” for conjugate or “H” for conjugate transpose.

• alpha – [in] Scalar multiplier for the matrix A.

• A – [in] The sparse matrix A.

• x – [in] A vector to multiply on the left by A.

• beta – [in] Scalar multiplier for the vector y.

• y – [in/out] Result vector.

trsv

template<class AMatrix, class BMV, class XMV>
void KokkosSparse::trsv(const char uplo[], const char trans[], const char diag[], const AMatrix &A, const BMV &b, const XMV &x)

Solve the triangular sparse linear system Op(A) x = b.

Template Parameters:

• AMatrix – KokkosSparse::CrsMatrix specialization.

• BMV – The type of the input (right-hand side) (multi)vector.

• XMV – The type of the output (left-hand side) (multi)vector.

Parameters:

• uplo – [in] “U” (for upper triangular) or “L” (for lower triangular).

• trans – [in] “C” (for conjugate transpose), “T” (for transpose), or “N” (for no transpose).

• diag – [in] “U” (for implicit unit diagonal) or “N” (for not).

• A – [in] The input matrix A; must be upper triangular or lower triangular.

• b – [in] The input (right-hand side) (multi)vector.

• x – [in] The output (left-hand side) (multi)vector.

spgemm

template<class KernelHandle, class AMatrix, class BMatrix, class CMatrix>
void KokkosSparse::spgemm_symbolic(KernelHandle &kh, const AMatrix &A, const bool Amode, const BMatrix &B, const bool Bmode, CMatrix &C)

Template Parameters:

• KernelHandle –

• AMatrix –

• BMatrix –

• CMatrix –

Parameters:

• kh –

• A –

• Amode –

• B –

• Bmode –

• C –

template<class KernelHandle, class AMatrix, class BMatrix, class CMatrix>
void KokkosSparse::spgemm_numeric(KernelHandle &kh, const AMatrix &A, const bool Amode, const BMatrix &B, const bool Bmode, CMatrix &C)

Template Parameters:

• KernelHandle –

• AMatrix –

• BMatrix –

• CMatrix –

Parameters:

• kh –

• A –

• Amode –

• B –

• Bmode –

• C –

template<class CMatrix, class AMatrix, class BMatrix>
CMatrix KokkosSparse::spgemm(const AMatrix &A, const bool Amode, const BMatrix &B, const bool Bmode)

Template Parameters:

• CMatrix –

• AMatrix –

• BMatrix –

Parameters:

• A –

• Amode –

• B –

• Bmode –

Returns:

CMatrix

block_spgemm

template<class KernelHandle, class AMatrixType, class BMatrixType, class CMatrixType>
void KokkosSparse::block_spgemm_symbolic(KernelHandle &kh, const AMatrixType &A, const bool transposeA, const BMatrixType &B, const bool transposeB, CMatrixType &C)

Symbolic phase for block SpGEMM (BSR matrices)

Template Parameters:

• KernelHandle –

• AMatrixType –

• BMatrixType –

• CMatrixType –

Parameters:

• kh –

• A –

• transposeA –

• B –

• transposeB –

• C –

template<class KernelHandle, class AMatrix, class BMatrix, class CMatrix>
void KokkosSparse::block_spgemm_numeric(KernelHandle &kh, const AMatrix &A, const bool Amode, const BMatrix &B, const bool Bmode, CMatrix &C)

Template Parameters:

• KernelHandle –

• AMatrix –

• BMatrix –

• CMatrix –

Parameters:

• kh –

• A –

• Amode –

• B –

• Bmode –

• C –

gauss_seidel

inline void KokkosKernels::Experimental::KokkosKernelsHandle::create_gs_handle(KokkosSparse::GSAlgorithm gs_algorithm = KokkosSparse::GS_DEFAULT, KokkosGraph::ColoringAlgorithm coloring_algorithm = KokkosGraph::COLORING_DEFAULT)

Create a gauss seidel handle object.

Parameters:

• gs_algorithm – Specifies which algorithm to use:

                  KokkosSpace::GS_DEFAULT  PointGaussSeidel or BlockGaussSeidel, depending on matrix type.
                  KokkosSpace::GS_PERMUTED Reorders rows/cols into colors to improve locality. Uses RangePolicy over rows.
                  KokkosSpace::GS_TEAM     Uses TeamPolicy over batches of rows with ThreadVector within rows.
                  KokkosSpace::GS_CLUSTER  Uses independent clusters of nodes in the graph. Within a cluster, x is updated sequentially.
                                           For more information, see: https://arxiv.org/pdf/2204.02934.pdf.
                  KokkosSpace::GS_TWOSTAGE Uses spmv to parallelize inner sweeps of x.
                                           For more information, see: https://arxiv.org/pdf/2104.01196.pdf.
  

• coloring_algorithm – Specifies which coloring algorithm to color the graph with:

                        KokkosGraph::COLORING_DEFAULT  Depends on execution space:
                                                         COLORING_SERIAL on Kokkos::Serial;
                                                         COLORING_EB on GPUs;
                                                         COLORING_VBBIT on Kokkos::Sycl or elsewhere.
                        KokkosGraph::COLORING_SERIAL   Serial Greedy Coloring
                        KokkosGraph::COLORING_VB       Vertex Based Coloring
                        KokkosGraph::COLORING_VBBIT    Vertex Based Coloring with bit array
                        KokkosGraph::COLORING_VBCS     Vertex Based Color Set
                        KokkosGraph::COLORING_VBD      Vertex Based Deterministic Coloring
                        KokkosGraph::COLORING_VBDBIT   Vertex Based Deterministic Coloring with bit array
                        KokkosGraph::COLORING_EB       Edge Based Coloring
                        KokkosGraph::COLORING_SERIAL2  Serial Distance-2 Graph Coloring (kept here for
                                                        backwards compatibility for SPGEMM and other use cases)
  

inline void KokkosKernels::Experimental::KokkosKernelsHandle::create_gs_handle(const HandleExecSpace &handle_exec_space, int num_streams, KokkosSparse::GSAlgorithm gs_algorithm = KokkosSparse::GS_DEFAULT, KokkosGraph::ColoringAlgorithm coloring_algorithm = KokkosGraph::COLORING_DEFAULT)

Create a gauss seidel handle object.

Parameters:

• handle_exec_space – The execution space instance to execute kernels on.

• num_streams – The number of streams to allocate memory for.

• gs_algorithm – Specifies which algorithm to use:

                  KokkosSpace::GS_DEFAULT  PointGaussSeidel
                  KokkosSpace::GS_PERMUTED ??
                  KokkosSpace::GS_TEAM     ??
                  KokkosSpace::GS_CLUSTER  ??
                  KokkosSpace::GS_TWOSTAGE ??
  

• coloring_algorithm – Specifies which coloring algorithm to color the graph with:

                        KokkosGraph::COLORING_DEFAULT  ??
                        KokkosGraph::COLORING_SERIAL   Serial Greedy Coloring
                        KokkosGraph::COLORING_VB       Vertex Based Coloring
                        KokkosGraph::COLORING_VBBIT    Vertex Based Coloring with bit array
                        KokkosGraph::COLORING_VBCS     Vertex Based Color Set
                        KokkosGraph::COLORING_VBD      Vertex Based Deterministic Coloring
                        KokkosGraph::COLORING_VBDBIT   Vertex Based Deterministic Coloring with bit array
                        KokkosGraph::COLORING_EB       Edge Based Coloring
                        KokkosGraph::COLORING_SERIAL2  Serial Distance-2 Graph Coloring (kept here for
                                                        backwards compatibility for SPGEMM and other use cases)
  

inline void KokkosKernels::Experimental::KokkosKernelsHandle::create_gs_handle(KokkosSparse::ClusteringAlgorithm clusterAlgo, nnz_lno_t hint_verts_per_cluster, KokkosGraph::ColoringAlgorithm coloring_algorithm = KokkosGraph::COLORING_DEFAULT)

Create a gs handle object.

Parameters:

• clusterAlgo – Specifies which clustering algorithm to use:

                 KokkosSparse::CLUSTER_DEFAULT           ??
                 KokkosSparse::CLUSTER_MIS2              ??
                 KokkosSparse::CLUSTER_BALLOON           ??
                 KokkosSparse::NUM_CLUSTERING_ALGORITHMS ??
  

• hint_verts_per_cluster – Hint how many verticies to use per cluster

• coloring_algorithm – Specifies which coloring algorithm to color the graph with:

                        KokkosGraph::COLORING_DEFAULT  ??
                        KokkosGraph::COLORING_SERIAL   Serial Greedy Coloring
                        KokkosGraph::COLORING_VB       Vertex Based Coloring
                        KokkosGraph::COLORING_VBBIT    Vertex Based Coloring with bit array
                        KokkosGraph::COLORING_VBCS     Vertex Based Color Set
                        KokkosGraph::COLORING_VBD      Vertex Based Deterministic Coloring
                        KokkosGraph::COLORING_VBDBIT   Vertex Based Deterministic Coloring with bit array
                        KokkosGraph::COLORING_EB       Edge Based Coloring
                        KokkosGraph::COLORING_SERIAL2  Serial Distance-2 Graph Coloring (kept here for
                                                        backwards compatibility for SPGEMM and other use cases)
  

template<typename ExecutionSpace, typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_>
void KokkosSparse::gauss_seidel_symbolic(const ExecutionSpace &space, KernelHandle *handle, typename KernelHandle::const_nnz_lno_t num_rows, typename KernelHandle::const_nnz_lno_t num_cols, lno_row_view_t_ row_map, lno_nnz_view_t_ entries, bool is_graph_symmetric = true)

Gauss-Seidel preconditioner setup (first phase, based on sparsity pattern only)

Template Parameters:

• ExecutionSpace – This kernels execution space type.

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The matrix’s rowmap type

• lno_nnz_view_t_ – The matrix’s entries type

Parameters:

• space – The execution space instance this kernel will be run on.

• handle – KernelHandle instance

• num_rows – Number of rows in the matrix

• num_cols – Number of columns in the matrix

• row_map – The matrix’s rowmap

• entries – The matrix’s entries

• is_graph_symmetric – Whether the upper-left num_rows x num_rows submatrix of A is structurally symmetric

Pre:

handle->create_gs_handle(...) has been called previously

template<typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_>
void KokkosSparse::gauss_seidel_symbolic(KernelHandle *handle, typename KernelHandle::const_nnz_lno_t num_rows, typename KernelHandle::const_nnz_lno_t num_cols, lno_row_view_t_ row_map, lno_nnz_view_t_ entries, bool is_graph_symmetric = true)

Gauss-Seidel preconditioner setup (first phase, based on sparsity pattern only)

Template Parameters:

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The matrix’s rowmap type

• lno_nnz_view_t_ – The matrix’s entries type

Parameters:

• handle – KernelHandle instance

• num_rows – Number of rows in the matrix

• num_cols – Number of columns in the matrix

• row_map – The matrix’s rowmap

• entries – The matrix’s entries

• is_graph_symmetric – Whether the upper-left num_rows x num_rows submatrix of A is structurally symmetric

Pre:

handle->create_gs_handle(...) has been called previously

template<class ExecutionSpace, KokkosSparse::SparseMatrixFormat format = KokkosSparse::SparseMatrixFormat::CRS, typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_, typename scalar_nnz_view_t_>
void KokkosSparse::gauss_seidel_numeric(const ExecutionSpace &space, KernelHandle *handle, typename KernelHandle::const_nnz_lno_t num_rows, typename KernelHandle::const_nnz_lno_t num_cols, lno_row_view_t_ row_map, lno_nnz_view_t_ entries, scalar_nnz_view_t_ values, bool is_graph_symmetric = true)

Gauss-Seidel preconditioner setup (second phase, based on matrix’s numeric values)

Template Parameters:

• ExecutionSpace – This kernels execution space type.

• format – The matrix storage format, CRS or BSR

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The matrix’s rowmap type

• lno_nnz_view_t_ – The matrix’s entries type

• scalar_nnz_view_t_ – The matrix’s values type

Parameters:

• space – The execution space instance this kernel will be run on.

• handle – KernelHandle instance

• num_rows – Number of rows in the matrix

• num_cols – Number of columns in the matrix

• row_map – The matrix’s rowmap

• entries – The matrix’s entries

• values – The matrix’s values

• is_graph_symmetric – Whether the upper-left num_rows x num_rows submatrix of A is structurally symmetric

template<KokkosSparse::SparseMatrixFormat format = KokkosSparse::SparseMatrixFormat::CRS, typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_, typename scalar_nnz_view_t_>
void KokkosSparse::gauss_seidel_numeric(KernelHandle *handle, typename KernelHandle::const_nnz_lno_t num_rows, typename KernelHandle::const_nnz_lno_t num_cols, lno_row_view_t_ row_map, lno_nnz_view_t_ entries, scalar_nnz_view_t_ values, bool is_graph_symmetric = true)

Gauss-Seidel preconditioner setup (second phase, based on matrix’s numeric values). This version accepts the matrix’s inverse diagonal from the user.

Remark

If the inverse diagonal is not already available, it’s best to call the version of gauss_seidel_numeric that doesn’t take it as an argument. The inverse diagonal will be computed internally.

Template Parameters:

• format – The matrix storage format, CRS or BSR

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The matrix’s rowmap type

• lno_nnz_view_t_ – The matrix’s entries type

• scalar_nnz_view_t_ – The matrix’s values type. The user-provided inverse diagonal must share this type.

Parameters:

• handle – KernelHandle instance

• num_rows – Number of rows in the matrix

• num_cols – Number of columns in the matrix

• row_map – The matrix’s rowmap

• entries – The matrix’s entries

• values – The matrix’s values

• is_graph_symmetric – Whether the upper-left num_rows x num_rows submatrix of A is structurally symmetric

template<class ExecutionSpace, KokkosSparse::SparseMatrixFormat format = KokkosSparse::SparseMatrixFormat::CRS, typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_, typename scalar_nnz_view_t_>
void KokkosSparse::gauss_seidel_numeric(const ExecutionSpace &space, KernelHandle *handle, typename KernelHandle::const_nnz_lno_t num_rows, typename KernelHandle::const_nnz_lno_t num_cols, lno_row_view_t_ row_map, lno_nnz_view_t_ entries, scalar_nnz_view_t_ values, scalar_nnz_view_t_ given_inverse_diagonal, bool is_graph_symmetric = true)

Gauss-Seidel preconditioner setup (second phase, based on matrix’s numeric values). This version accepts the matrix’s inverse diagonal from the user.

Remark

If the inverse diagonal is not already available, it’s best to call the version of gauss_seidel_numeric that doesn’t take it as an argument. The inverse diagonal will be computed internally.

Template Parameters:

• ExecutionSpace – This kernels execution space type.

• format – The matrix storage format, CRS or BSR

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The matrix’s rowmap type

• lno_nnz_view_t_ – The matrix’s entries type

• scalar_nnz_view_t_ – The matrix’s values type. The user-provided inverse diagonal must share this type.

Parameters:

• space – The execution space instance this kernel will be run on.

• handle – KernelHandle instance

• num_rows – Number of rows in the matrix

• num_cols – Number of columns in the matrix

• row_map – The matrix’s rowmap

• entries – The matrix’s entries

• values – The matrix’s values

• given_inverse_diagonal – The inverse (reciprocal) of diagonal

• is_graph_symmetric – Whether the upper-left num_rows x num_rows submatrix of A is structurally symmetric

template<KokkosSparse::SparseMatrixFormat format = KokkosSparse::SparseMatrixFormat::CRS, typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_, typename scalar_nnz_view_t_>
void KokkosSparse::gauss_seidel_numeric(KernelHandle *handle, typename KernelHandle::const_nnz_lno_t num_rows, typename KernelHandle::const_nnz_lno_t num_cols, lno_row_view_t_ row_map, lno_nnz_view_t_ entries, scalar_nnz_view_t_ values, scalar_nnz_view_t_ given_inverse_diagonal, bool is_graph_symmetric = true)

Gauss-Seidel preconditioner setup (second phase, based on matrix’s numeric values). This version accepts the matrix’s inverse diagonal from the user.

Remark

If the inverse diagonal is not already available, it’s best to call the version of gauss_seidel_numeric that doesn’t take it as an argument. The inverse diagonal will be computed internally.

Template Parameters:

• format – The matrix storage format, CRS or BSR

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The matrix’s rowmap type

• lno_nnz_view_t_ – The matrix’s entries type

• scalar_nnz_view_t_ – The matrix’s values type. The user-provided inverse diagonal must share this type.

Parameters:

• handle – KernelHandle instance

• num_rows – Number of rows in the matrix

• num_cols – Number of columns in the matrix

• row_map – The matrix’s rowmap

• entries – The matrix’s entries

• values – The matrix’s values

• given_inverse_diagonal – The inverse (reciprocal) of diagonal

• is_graph_symmetric – Whether the upper-left num_rows x num_rows submatrix of A is structurally symmetric

template<class ExecutionSpace, KokkosSparse::SparseMatrixFormat format = KokkosSparse::SparseMatrixFormat::CRS, typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_, typename scalar_nnz_view_t_, typename x_scalar_view_t, typename y_scalar_view_t>
void KokkosSparse::symmetric_gauss_seidel_apply(const ExecutionSpace &space, KernelHandle *handle, typename KernelHandle::const_nnz_lno_t num_rows, typename KernelHandle::const_nnz_lno_t num_cols, lno_row_view_t_ row_map, lno_nnz_view_t_ entries, scalar_nnz_view_t_ values, x_scalar_view_t x_lhs_output_vec, y_scalar_view_t y_rhs_input_vec, bool init_zero_x_vector, bool update_y_vector, typename KernelHandle::nnz_scalar_t omega, int numIter)

Apply symmetric (forward + backward) Gauss-Seidel preconditioner to system AX=Y.

Template Parameters:

• ExecutionSpace – This kernels execution space type.

• format – The matrix storage format, CRS or BSR

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The matrix’s rowmap type

• lno_nnz_view_t_ – The matrix’s entries type

• scalar_nnz_view_t_ – The matrix’s values type

• x_scalar_view_t – The type of the X (left-hand side, unknown) vector. May be rank-1 or rank-2 View.

• y_scalar_view_t – The type of the Y (right-hand side) vector. May be rank-1 or rank-2 View.

Parameters:

• space – The execution space instance this kernel will be run on. NOTE: Currently only used for GS_DEFAULT.

• handle – handle A KokkosKernelsHandle instance

• num_rows – Number of rows in the matrix

• num_cols – Number of columns in the matrix

• row_map – The matrix’s rowmap

• entries – The matrix’s entries

• values – The matrix’s values

• x_lhs_output_vec – The X (left-hand side, unknown) vector

• y_rhs_input_vec – The Y (right-hand side) vector

• init_zero_x_vector – Whether to zero out X before applying

• update_y_vector – Whether Y has changed since the last call to apply

• omega – The damping factor for successive over-relaxation

• numIter – How many iterations to run (forward and backward counts as 1)

Pre:

x_lhs_output_vec.extent(0) == num_cols

Pre:

y_rhs_input_vec.extent(0) == num_rows

Pre:

x_lhs_output_vec.extent(1) == y_rhs_input_vec.extent(1)

template<KokkosSparse::SparseMatrixFormat format = KokkosSparse::SparseMatrixFormat::CRS, typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_, typename scalar_nnz_view_t_, typename x_scalar_view_t, typename y_scalar_view_t>
void KokkosSparse::symmetric_gauss_seidel_apply(KernelHandle *handle, typename KernelHandle::const_nnz_lno_t num_rows, typename KernelHandle::const_nnz_lno_t num_cols, lno_row_view_t_ row_map, lno_nnz_view_t_ entries, scalar_nnz_view_t_ values, x_scalar_view_t x_lhs_output_vec, y_scalar_view_t y_rhs_input_vec, bool init_zero_x_vector, bool update_y_vector, typename KernelHandle::nnz_scalar_t omega, int numIter)

Apply symmetric (forward + backward) Gauss-Seidel preconditioner to system AX=Y.

Template Parameters:

• format – The matrix storage format, CRS or BSR

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The matrix’s rowmap type

• lno_nnz_view_t_ – The matrix’s entries type

• scalar_nnz_view_t_ – The matrix’s values type

• x_scalar_view_t – The type of the X (left-hand side, unknown) vector. May be rank-1 or rank-2 View.

• y_scalar_view_t – The type of the Y (right-hand side) vector. May be rank-1 or rank-2 View.

Parameters:

• handle – handle A KokkosKernelsHandle instance

• num_rows – Number of rows in the matrix

• num_cols – Number of columns in the matrix

• row_map – The matrix’s rowmap

• entries – The matrix’s entries

• values – The matrix’s values

• x_lhs_output_vec – The X (left-hand side, unknown) vector

• y_rhs_input_vec – The Y (right-hand side) vector

• init_zero_x_vector – Whether to zero out X before applying

• update_y_vector – Whether Y has changed since the last call to apply

• omega – The damping factor for successive over-relaxation

• numIter – How many iterations to run (forward and backward counts as 1)

Pre:

x_lhs_output_vec.extent(0) == num_cols

Pre:

y_rhs_input_vec.extent(0) == num_rows

Pre:

x_lhs_output_vec.extent(1) == y_rhs_input_vec.extent(1)

template<class ExecutionSpace, KokkosSparse::SparseMatrixFormat format = KokkosSparse::SparseMatrixFormat::CRS, class KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_, typename scalar_nnz_view_t_, typename x_scalar_view_t, typename y_scalar_view_t>
void KokkosSparse::forward_sweep_gauss_seidel_apply(const ExecutionSpace &space, KernelHandle *handle, typename KernelHandle::const_nnz_lno_t num_rows, typename KernelHandle::const_nnz_lno_t num_cols, lno_row_view_t_ row_map, lno_nnz_view_t_ entries, scalar_nnz_view_t_ values, x_scalar_view_t x_lhs_output_vec, y_scalar_view_t y_rhs_input_vec, bool init_zero_x_vector, bool update_y_vector, typename KernelHandle::nnz_scalar_t omega, int numIter)

Apply forward Gauss-Seidel preconditioner to system AX=Y.

Template Parameters:

• format – The matrix storage format, CRS or BSR

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The matrix’s rowmap type

• lno_nnz_view_t_ – The matrix’s entries type

• scalar_nnz_view_t_ – The matrix’s values type

• x_scalar_view_t – The type of the X (left-hand side, unknown) vector. May be rank-1 or rank-2 View.

• y_scalar_view_t – The type of the Y (right-hand side) vector. May be rank-1 or rank-2 View.

Parameters:

• space – The execution space instance this kernel will be run on. NOTE: Currently only used for GS_DEFAULT.

• handle – KernelHandle instance

• num_rows – Number of rows in the matrix

• num_cols – Number of columns in the matrix

• row_map – The matrix’s rowmap

• entries – The matrix’s entries

• values – The matrix’s values

• x_lhs_output_vec – The X (left-hand side, unknown) vector

• y_rhs_input_vec – The Y (right-hand side) vector

• init_zero_x_vector – Whether to zero out X before applying

• update_y_vector – Whether Y has changed since the last call to apply

• omega – The damping factor for successive over-relaxation

• numIter – How many iterations to run

Pre:

x_lhs_output_vec.extent(0) == num_cols

Pre:

y_rhs_input_vec.extent(0) == num_rows

Pre:

x_lhs_output_vec.extent(1) == y_rhs_input_vec.extent(1)

template<KokkosSparse::SparseMatrixFormat format = KokkosSparse::SparseMatrixFormat::CRS, class KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_, typename scalar_nnz_view_t_, typename x_scalar_view_t, typename y_scalar_view_t>
void KokkosSparse::forward_sweep_gauss_seidel_apply(KernelHandle *handle, typename KernelHandle::const_nnz_lno_t num_rows, typename KernelHandle::const_nnz_lno_t num_cols, lno_row_view_t_ row_map, lno_nnz_view_t_ entries, scalar_nnz_view_t_ values, x_scalar_view_t x_lhs_output_vec, y_scalar_view_t y_rhs_input_vec, bool init_zero_x_vector, bool update_y_vector, typename KernelHandle::nnz_scalar_t omega, int numIter)

Apply forward Gauss-Seidel preconditioner to system AX=Y.

Template Parameters:

• format – The matrix storage format, CRS or BSR

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The matrix’s rowmap type

• lno_nnz_view_t_ – The matrix’s entries type

• scalar_nnz_view_t_ – The matrix’s values type

• x_scalar_view_t – The type of the X (left-hand side, unknown) vector. May be rank-1 or rank-2 View.

• y_scalar_view_t – The type of the Y (right-hand side) vector. May be rank-1 or rank-2 View.

Parameters:

• handle – KernelHandle instance

• num_rows – Number of rows in the matrix

• num_cols – Number of columns in the matrix

• row_map – The matrix’s rowmap

• entries – The matrix’s entries

• values – The matrix’s values

• x_lhs_output_vec – The X (left-hand side, unknown) vector

• y_rhs_input_vec – The Y (right-hand side) vector

• init_zero_x_vector – Whether to zero out X before applying

• update_y_vector – Whether Y has changed since the last call to apply

• omega – The damping factor for successive over-relaxation

• numIter – How many iterations to run

Pre:

x_lhs_output_vec.extent(0) == num_cols

Pre:

y_rhs_input_vec.extent(0) == num_rows

Pre:

x_lhs_output_vec.extent(1) == y_rhs_input_vec.extent(1)

template<class ExecutionSpace, KokkosSparse::SparseMatrixFormat format = KokkosSparse::SparseMatrixFormat::CRS, class KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_, typename scalar_nnz_view_t_, typename x_scalar_view_t, typename y_scalar_view_t>
void KokkosSparse::backward_sweep_gauss_seidel_apply(const ExecutionSpace &space, KernelHandle *handle, typename KernelHandle::const_nnz_lno_t num_rows, typename KernelHandle::const_nnz_lno_t num_cols, lno_row_view_t_ row_map, lno_nnz_view_t_ entries, scalar_nnz_view_t_ values, x_scalar_view_t x_lhs_output_vec, y_scalar_view_t y_rhs_input_vec, bool init_zero_x_vector, bool update_y_vector, typename KernelHandle::nnz_scalar_t omega, int numIter)

Apply backward Gauss-Seidel preconditioner to system AX=Y.

Template Parameters:

• ExecutionSpace – This kernels execution space type.

• format – The matrix storage format, CRS or BSR

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The matrix’s rowmap type

• lno_nnz_view_t_ – The matrix’s entries type

• scalar_nnz_view_t_ – The matrix’s values type

• x_scalar_view_t – The type of the X (left-hand side, unknown) vector. May be rank-1 or rank-2 View.

• y_scalar_view_t – The type of the Y (right-hand side) vector. May be rank-1 or rank-2 View.

Parameters:

• space – The execution space instance this kernel will be run on. NOTE: Currently only used for GS_DEFAULT.

• handle – KernelHandle instance

• num_rows – Number of rows in the matrix

• num_cols – Number of columns in the matrix

• row_map – The matrix’s rowmap

• entries – The matrix’s entries

• values – The matrix’s values

• x_lhs_output_vec – The X (left-hand side, unknown) vector

• y_rhs_input_vec – The Y (right-hand side) vector

• init_zero_x_vector – Whether to zero out X before applying

• update_y_vector – Whether Y has changed since the last call to apply

• omega – The damping factor for successive over-relaxation

• numIter – How many iterations to run

Pre:

x_lhs_output_vec.extent(0) == num_cols

Pre:

y_rhs_input_vec.extent(0) == num_rows

Pre:

x_lhs_output_vec.extent(1) == y_rhs_input_vec.extent(1)

template<KokkosSparse::SparseMatrixFormat format = KokkosSparse::SparseMatrixFormat::CRS, class KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_, typename scalar_nnz_view_t_, typename x_scalar_view_t, typename y_scalar_view_t>
void KokkosSparse::backward_sweep_gauss_seidel_apply(KernelHandle *handle, typename KernelHandle::const_nnz_lno_t num_rows, typename KernelHandle::const_nnz_lno_t num_cols, lno_row_view_t_ row_map, lno_nnz_view_t_ entries, scalar_nnz_view_t_ values, x_scalar_view_t x_lhs_output_vec, y_scalar_view_t y_rhs_input_vec, bool init_zero_x_vector, bool update_y_vector, typename KernelHandle::nnz_scalar_t omega, int numIter)

Apply backward Gauss-Seidel preconditioner to system AX=Y.

Template Parameters:

• format – The matrix storage format, CRS or BSR

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The matrix’s rowmap type

• lno_nnz_view_t_ – The matrix’s entries type

• scalar_nnz_view_t_ – The matrix’s values type

• x_scalar_view_t – The type of the X (left-hand side, unknown) vector. May be rank-1 or rank-2 View.

• y_scalar_view_t – The type of the Y (right-hand side) vector. May be rank-1 or rank-2 View.

Parameters:

• handle – KernelHandle instance

• num_rows – Number of rows in the matrix

• num_cols – Number of columns in the matrix

• row_map – The matrix’s rowmap

• entries – The matrix’s entries

• values – The matrix’s values

• x_lhs_output_vec – The X (left-hand side, unknown) vector

• y_rhs_input_vec – The Y (right-hand side) vector

• init_zero_x_vector – Whether to zero out X before applying

• update_y_vector – Whether Y has changed since the last call to apply

• omega – The damping factor for successive over-relaxation

• numIter – How many iterations to run

Pre:

x_lhs_output_vec.extent(0) == num_cols

Pre:

y_rhs_input_vec.extent(0) == num_rows

Pre:

x_lhs_output_vec.extent(1) == y_rhs_input_vec.extent(1)

block_gauss_seidel

template<typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_>
void KokkosSparse::block_gauss_seidel_symbolic(KernelHandle *handle, typename KernelHandle::const_nnz_lno_t num_rows, typename KernelHandle::const_nnz_lno_t num_cols, typename KernelHandle::const_nnz_lno_t block_size, lno_row_view_t_ row_map, lno_nnz_view_t_ entries, bool is_graph_symmetric = true)

Block Gauss-Seidel preconditioner setup (first phase, based on sparsity pattern only)

Template Parameters:

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The matrix’s rowmap type

• lno_nnz_view_t_ – The matrix’s entries type

Parameters:

• handle – KernelHandle instance

• num_rows – Number of rows in the matrix

• num_cols – Number of columns in the matrix

• block_size – The number of degrees of freedom per block

• row_map – The matrix’s rowmap

• entries – The matrix’s entries

• is_graph_symmetric – Whether the upper-left num_rows x num_rows submatrix of A is structurally symmetric

Pre:

handle->create_gs_handle(...) has been called previously

template<KokkosSparse::SparseMatrixFormat format = KokkosSparse::SparseMatrixFormat::BSR, typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_, typename scalar_nnz_view_t_>
void KokkosSparse::block_gauss_seidel_numeric(KernelHandle *handle, typename KernelHandle::const_nnz_lno_t num_rows, typename KernelHandle::const_nnz_lno_t num_cols, typename KernelHandle::const_nnz_lno_t block_size, lno_row_view_t_ row_map, lno_nnz_view_t_ entries, scalar_nnz_view_t_ values, bool is_graph_symmetric = true)

Block Gauss-Seidel preconditioner setup (second phase, based on matrix’s numeric values)

Template Parameters:

• format – The matrix storage format, CRS or BSR

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The matrix’s rowmap type

• lno_nnz_view_t_ – The matrix’s entries type

• scalar_nnz_view_t_ – The matrix’s values type

Parameters:

• handle – handle A KokkosKernelsHandle instance

• num_rows – Number of rows in the matrix

• num_cols – Number of columns in the matrix

• block_size – The number of degrees of freedom per block

• row_map – The matrix’s rowmap

• entries – The matrix’s entries

• values – The matrix’s values

• is_graph_symmetric – Whether the upper-left num_rows x num_rows submatrix of A is structurally symmetric

template<KokkosSparse::SparseMatrixFormat format = KokkosSparse::SparseMatrixFormat::BSR, typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_, typename scalar_nnz_view_t_, typename x_scalar_view_t, typename y_scalar_view_t>
void KokkosSparse::symmetric_block_gauss_seidel_apply(KernelHandle *handle, typename KernelHandle::const_nnz_lno_t num_rows, typename KernelHandle::const_nnz_lno_t num_cols, typename KernelHandle::const_nnz_lno_t block_size, lno_row_view_t_ row_map, lno_nnz_view_t_ entries, scalar_nnz_view_t_ values, x_scalar_view_t x_lhs_output_vec, y_scalar_view_t y_rhs_input_vec, bool init_zero_x_vector, bool update_y_vector, typename KernelHandle::nnz_scalar_t omega, int numIter)

Apply symmetric (forward + backward) Block Gauss-Seidel preconditioner to system AX=Y.

Template Parameters:

• format – The matrix storage format, CRS or BSR

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The matrix’s rowmap type

• lno_nnz_view_t_ – The matrix’s entries type

• scalar_nnz_view_t_ – The matrix’s values type

• x_scalar_view_t – The type of the X (left-hand side, unknown) vector. May be rank-1 or rank-2 View.

• y_scalar_view_t – The type of the Y (right-hand side) vector. May be rank-1 or rank-2 View.

Parameters:

• handle – handle A KokkosKernelsHandle instance.

• num_rows – Number of rows in the matrix

• num_cols – Number of columns in the matrix

• block_size – The number of degrees of freedom per block

• row_map – The matrix’s rowmap

• entries – The matrix’s entries

• values – The matrix’s values

• x_lhs_output_vec – The X (left-hand side, unknown) vector

• y_rhs_input_vec – The Y (right-hand side) vector

• init_zero_x_vector – Whether to zero out X before applying

• update_y_vector – Whether Y has changed since the last call to apply

• omega – The damping factor for successive over-relaxation

• numIter – How many iterations to run (forward and backward counts as 1)

Pre:

x_lhs_output_vec.extent(0) == num_cols

Pre:

y_rhs_input_vec.extent(0) == num_rows

Pre:

x_lhs_output_vec.extent(1) == y_rhs_input_vec.extent(1)

template<KokkosSparse::SparseMatrixFormat format = KokkosSparse::SparseMatrixFormat::BSR, typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_, typename scalar_nnz_view_t_, typename x_scalar_view_t, typename y_scalar_view_t>
void KokkosSparse::forward_sweep_block_gauss_seidel_apply(KernelHandle *handle, typename KernelHandle::const_nnz_lno_t num_rows, typename KernelHandle::const_nnz_lno_t num_cols, typename KernelHandle::const_nnz_lno_t block_size, lno_row_view_t_ row_map, lno_nnz_view_t_ entries, scalar_nnz_view_t_ values, x_scalar_view_t x_lhs_output_vec, y_scalar_view_t y_rhs_input_vec, bool init_zero_x_vector, bool update_y_vector, typename KernelHandle::nnz_scalar_t omega, int numIter)

Apply forward Block Gauss-Seidel preconditioner to system AX=Y.

Template Parameters:

• format – The matrix storage format, CRS or BSR

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The matrix’s rowmap type

• lno_nnz_view_t_ – The matrix’s entries type

• scalar_nnz_view_t_ – The matrix’s values type

• x_scalar_view_t – The type of the X (left-hand side, unknown) vector. May be rank-1 or rank-2 View.

• y_scalar_view_t – The type of the Y (right-hand side) vector. May be rank-1 or rank-2 View.

Parameters:

• handle – KernelHandle instance

• num_rows – Number of rows in the matrix

• num_cols – Number of columns in the matrix

• block_size – The number of degrees of freedom per block

• row_map – The matrix’s rowmap

• entries – The matrix’s entries

• values – The matrix’s values

• x_lhs_output_vec – The X (left-hand side, unknown) vector

• y_rhs_input_vec – The Y (right-hand side) vector

• init_zero_x_vector – Whether to zero out X before applying

• update_y_vector – Whether Y has changed since the last call to apply

• omega – The damping factor for successive over-relaxation

• numIter – How many iterations to run

Pre:

x_lhs_output_vec.extent(0) == num_cols

Pre:

y_rhs_input_vec.extent(0) == num_rows

Pre:

x_lhs_output_vec.extent(1) == y_rhs_input_vec.extent(1)

template<KokkosSparse::SparseMatrixFormat format = KokkosSparse::SparseMatrixFormat::BSR, typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_, typename scalar_nnz_view_t_, typename x_scalar_view_t, typename y_scalar_view_t>
void KokkosSparse::backward_sweep_block_gauss_seidel_apply(KernelHandle *handle, typename KernelHandle::const_nnz_lno_t num_rows, typename KernelHandle::const_nnz_lno_t num_cols, typename KernelHandle::const_nnz_lno_t block_size, lno_row_view_t_ row_map, lno_nnz_view_t_ entries, scalar_nnz_view_t_ values, x_scalar_view_t x_lhs_output_vec, y_scalar_view_t y_rhs_input_vec, bool init_zero_x_vector, bool update_y_vector, typename KernelHandle::nnz_scalar_t omega, int numIter)

Apply backward Block Gauss-Seidel preconditioner to system AX=Y.

Template Parameters:

• format – The matrix storage format, CRS or BSR

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The matrix’s rowmap type

• lno_nnz_view_t_ – The matrix’s entries type

• scalar_nnz_view_t_ – The matrix’s values type

• x_scalar_view_t – The type of the X (left-hand side, unknown) vector. May be rank-1 or rank-2 View.

• y_scalar_view_t – The type of the Y (right-hand side) vector. May be rank-1 or rank-2 View.

Parameters:

• handle – KernelHandle instance

• num_rows – Number of rows in the matrix

• num_cols – Number of columns in the matrix

• block_size – The number of degrees of freedom per block

• row_map – The matrix’s rowmap

• entries – The matrix’s entries

• values – The matrix’s values

• x_lhs_output_vec – The X (left-hand side, unknown) vector

• y_rhs_input_vec – The Y (right-hand side) vector

• init_zero_x_vector – Whether to zero out X before applying

• update_y_vector – Whether Y has changed since the last call to apply

• omega – The damping factor for successive over-relaxation

• numIter – How many iterations to run

Pre:

x_lhs_output_vec.extent(0) == num_cols

Pre:

y_rhs_input_vec.extent(0) == num_rows

Pre:

x_lhs_output_vec.extent(1) == y_rhs_input_vec.extent(1)

par_ilut

template<typename KernelHandle, typename ARowMapType, typename AEntriesType, typename LRowMapType, typename URowMapType>
void KokkosSparse::Experimental::par_ilut_symbolic(KernelHandle *handle, ARowMapType &A_rowmap, AEntriesType &A_entries, LRowMapType &L_rowmap, URowMapType &U_rowmap)

Performs the symbolic phase of par_ilut. This is a non-blocking function.

The sparsity pattern of A will be analyzed and L_rowmap and U_rowmap will be populated with the L (lower triangular) and U (upper triagular) non-zero counts respectively. Having a separate symbolic phase allows for reuse when dealing with multiple matrices with the same sparsity pattern. This routine will set some values on handle for symbolic info (row count, nnz counts).

Template Parameters:

• KernelHandle – Template for the KernelHandle type

• ARowMapType – Template for A_rowmap type

• AEntriesType – Template for A_entries type

• LRowMapType – Template for L_rowmap type

• URowMapType – Template for U_rowmap type

Parameters:

• handle – The kernel handle. It is expected that create_par_ilut_handle has been called on it

• A_rowmap – The row map (row nnz offsets) for the A CSR (Input)

• A_entries – The entries (column ids) for the A CSR (Input)

• L_rowmap – The row map for the L CSR, should already be sized correctly (numRows+1) (Output)

• U_rowmap – The row map for the U CSR, should already be sized correctly (numRows+1) (Output)

template<typename KernelHandle, typename ARowMapType, typename AEntriesType, typename AValuesType, typename LRowMapType, typename LEntriesType, typename LValuesType, typename URowMapType, typename UEntriesType, typename UValuesType>
void KokkosSparse::Experimental::par_ilut_numeric(KernelHandle *handle, ARowMapType &A_rowmap, AEntriesType &A_entries, AValuesType &A_values, LRowMapType &L_rowmap, LEntriesType &L_entries, LValuesType &L_values, URowMapType &U_rowmap, UEntriesType &U_entries, UValuesType &U_values)

Performs the numeric phase (for specific CSRs, not reusable) of the par_ilut algorithm (described in the header). This is a non-blocking functions. It is expected that par_ilut_symbolic has already been called for the.

Template Parameters:

• KernelHandle – Template for the handle type

• ARowMapType – Template for the A_rowmap type

• AEntriesType – Template for the A_entries type

• AValuesType – Template for the A_values type

• LRowMapType – Template for the L_rowmap type

• LEntriesType – Template for the L_entries type

• LValuesType – Template for the L_values type

• URowMapType – Template for the U_rowmap type

• UEntriesType – Template for the U_entries type

• UValuesType – Template for the U_values type

Parameters:

• handle – The kernel handle. It is expected that create_par_ilut_handle has been called on it

• A_rowmap – The row map (row nnz offsets) for the A CSR (Input)

• A_entries – The entries (column ids) for the A CSR (Input)

• A_values – The values (non-zero matrix values) for the A CSR (Input)

• L_rowmap – The row map (row nnz offsets) for the L CSR (Input/Output)

• L_entries – The entries (column ids) for the L CSR (Output)

• L_values – The values (non-zero matrix values) for the L CSR (Output)

• U_rowmap – The row map (row nnz offsets) for the U CSR (Input/Output)

• U_entries – The entries (column ids) for the U CSR (Output)

• U_values – The values (non-zero matrix values) for the U CSR (Output)

template<class size_type_, class lno_t_, class scalar_t_, class ExecutionSpace, class TemporaryMemorySpace, class PersistentMemorySpace>
class PAR_ILUTHandle

Handle for par_ilut. Contains useful types, par_ilut configuration settings, symbolic settings and scalar output info.

For more info, see KokkosSparse_par_ilut.hpp doxygen

Public Functions

inline PAR_ILUTHandle(const size_type max_iter_, const float_t residual_norm_delta_stop_, const float_t fill_in_limit_, const bool async_update_, const bool verbose_)

The A - LU residual norm at the time the algorithm finished

gmres

template<typename KernelHandle, typename AMatrix, typename BType, typename XType>
void KokkosSparse::Experimental::gmres(KernelHandle *handle, AMatrix &A, BType &B, XType &X, Preconditioner<AMatrix> *precond = nullptr)

Template Parameters:

• KernelHandle –

• AMatrix –

• BType –

• XType –

Parameters:

• handle –

• A –

• B –

• X –

• precond –

sptrsv

template<typename ExecutionSpace, typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_>
void KokkosSparse::sptrsv_symbolic(const ExecutionSpace &space, KernelHandle *handle, lno_row_view_t_ rowmap, lno_nnz_view_t_ entries)

sptrsv symbolic phase for linear system Ax=b

Template Parameters:

• ExecutionSpace – This kernels execution space type

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The CRS matrix’s (A) rowmap type

• lno_nnz_view_t_ – The CRS matrix’s (A) entries type

Parameters:

• space – The execution space instance this kernel will run on

• handle – KernelHandle instance

• rowmap – The CRS matrix’s (A) rowmap

• entries – The CRS matrix’s (A) entries

template<typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_>
void KokkosSparse::sptrsv_symbolic(KernelHandle *handle, lno_row_view_t_ rowmap, lno_nnz_view_t_ entries)

sptrsv symbolic phase for linear system Ax=b

Template Parameters:

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The CRS matrix’s (A) rowmap type

• lno_nnz_view_t_ – The CRS matrix’s (A) entries type

Parameters:

• handle – KernelHandle instance

• rowmap – The CRS matrix’s (A) rowmap

• entries – The CRS matrix’s (A) entries

template<typename ExecutionSpace, typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_, typename scalar_nnz_view_t_>
void KokkosSparse::sptrsv_symbolic(ExecutionSpace &space, KernelHandle *handle, lno_row_view_t_ rowmap, lno_nnz_view_t_ entries, scalar_nnz_view_t_ values)

sptrsv symbolic phase for linear system Ax=b

Template Parameters:

• ExecutionSpace – This kernels execution space type

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The CRS matrix’s (A) rowmap type

• lno_nnz_view_t_ – The CRS matrix’s (A) entries type

Parameters:

• space – The execution space instance this kernel will run on

• handle – KernelHandle instance

• rowmap – The CRS matrix’s (A) rowmap

• entries – The CRS matrix’s (A) entries

• values – The CRS matrix’s (A) values

template<typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_, typename scalar_nnz_view_t_>
void KokkosSparse::sptrsv_symbolic(KernelHandle *handle, lno_row_view_t_ rowmap, lno_nnz_view_t_ entries, scalar_nnz_view_t_ values)

sptrsv symbolic phase for linear system Ax=b

Template Parameters:

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The CRS matrix’s (A) rowmap type

• lno_nnz_view_t_ – The CRS matrix’s (A) entries type

Parameters:

• handle – KernelHandle instance

• rowmap – The CRS matrix’s (A) rowmap

• entries – The CRS matrix’s (A) entries

• values – The CRS matrix’s (A) values

template<typename ExecutionSpace, typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_, typename scalar_nnz_view_t_, class BType, class XType>
void KokkosSparse::sptrsv_solve(ExecutionSpace &space, KernelHandle *handle, lno_row_view_t_ rowmap, lno_nnz_view_t_ entries, scalar_nnz_view_t_ values, BType b, XType x)

sptrsv solve phase of x for linear system Ax=b

Template Parameters:

• ExecutionSpace – This kernels execution space

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The CRS matrix’s (A) rowmap type

• lno_nnz_view_t_ – The CRS matrix’s (A) entries type

• scalar_nnz_view_t_ – The CRS matrix’s (A) values type

• BType – The b vector type

• XType – The x vector type

Parameters:

• space – The execution space instance this kernel will be run on

• handle – KernelHandle instance

• rowmap – The CRS matrix’s (A) rowmap

• entries – The CRS matrix’s (A) entries

• values – The CRS matrix’s (A) values

• b – The b vector

• x – The x vector

template<typename KernelHandle, typename lno_row_view_t_, typename lno_nnz_view_t_, typename scalar_nnz_view_t_, class BType, class XType>
void KokkosSparse::sptrsv_solve(KernelHandle *handle, lno_row_view_t_ rowmap, lno_nnz_view_t_ entries, scalar_nnz_view_t_ values, BType b, XType x)

sptrsv solve phase of x for linear system Ax=b

Template Parameters:

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• lno_row_view_t_ – The CRS matrix’s (A) rowmap type

• lno_nnz_view_t_ – The CRS matrix’s (A) entries type

• scalar_nnz_view_t_ – The CRS matrix’s (A) values type

• BType – The b vector type

• XType – The x vector type

Parameters:

• handle – KernelHandle instance

• rowmap – The CRS matrix’s (A) rowmap

• entries – The CRS matrix’s (A) entries

• values – The CRS matrix’s (A) values

• b – The b vector

• x – The x vector

template<typename ExecutionSpace, typename KernelHandle, class XType>
void KokkosSparse::Experimental::sptrsv_solve(ExecutionSpace &space, KernelHandle *handle, XType x, XType b)

Supernodal sptrsv solve phase of x for linear system Ax=b.

Template Parameters:

• ExecutionSpace – This kernels execution space

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• XType – The x and b vector type

Parameters:

• space – The execution space instance this kernel will run on

• handle – KernelHandle instance

• x – The x vector

• b – The b vector

template<typename KernelHandle, class XType>
void KokkosSparse::Experimental::sptrsv_solve(KernelHandle *handle, XType x, XType b)

Supernodal sptrsv solve phase of x for linear system Ax=b.

Template Parameters:

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• XType – The x and b vector type

Parameters:

• handle – KernelHandle instance

• x – The x vector

• b – The b vector

template<typename ExecutionSpace, typename KernelHandle, class XType>
void KokkosSparse::Experimental::sptrsv_solve(ExecutionSpace &space, KernelHandle *handleL, KernelHandle *handleU, XType x, XType b)

Supernodal sptrsv solve phase of x for linear system Ax=b.

Template Parameters:

• ExecutionSpace – This kernels execution space

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• XType – The x and b vector type

Parameters:

• space – The execution space instance this kernel will run on

• handleL – KernelHandle instance for lower triangular matrix

• handleU – KernelHandle instance for upper triangular matrix

• x – The x vector

• b – The b vector

template<typename KernelHandle, class XType>
void KokkosSparse::Experimental::sptrsv_solve(KernelHandle *handleL, KernelHandle *handleU, XType x, XType b)

Supernodal sptrsv solve phase of x for linear system Ax=b.

Template Parameters:

• KernelHandle – A specialization of KokkosKernels::Experimental::KokkosKernelsHandle

• XType – The x and b vector type

Parameters:

• handleL – KernelHandle instance for lower triangular matrix

• handleU – KernelHandle instance for upper triangular matrix

• x – The x vector

• b – The b vector