Tpetra parallel linear algebra  Version of the Day
Public Types | Public Member Functions | Static Public Attributes | List of all members
Tpetra::Experimental::LittleVector< Scalar, LO > Class Template Reference

Nonowning view of a set of degrees of freedom corresponding to a mesh point in a block vector or multivector. More...

#include <Tpetra_Experimental_BlockView.hpp>

Public Types

typedef Kokkos::LayoutRight array_layout
 Data layout (in the same sense as Kokkos::View). More...
 

Public Member Functions

 LittleVector (Scalar *const A, const LO blockSize, const LO strideX)
 Constructor. More...
 
template<class T >
 LittleVector (T *const A, const LO blockSize, const LO strideX, typename std::enable_if< !std::is_same< Scalar, T >::value &&std::is_convertible< Scalar, T >::value &&sizeof(Scalar)==sizeof(T), int * >::type ignoreMe=NULL)
 Constructor that takes an impl_scalar_type pointer. More...
 
Scalar * getRawPtr () const
 Pointer to the vector's entries. More...
 
Scalar * ptr_on_device () const
 Pointer to the vector's entries. More...
 
LO getBlockSize () const
 The block size (number of degrees of freedom per mesh point). More...
 
LO dimension_0 () const
 Number of entries in the vector. More...
 
LO size () const
 Number of entries in the vector. More...
 
LO getStride () const
 Stride between consecutive entries. More...
 
template<class IntegerType >
void stride (IntegerType *const s) const
 Stride between consecutive entries. More...
 
impl_scalar_type & operator() (const LO i) const
 Reference to entry (i) of the vector. More...
 

Static Public Attributes

static const int rank = 1
 Number of dimensions. More...
 

Detailed Description

template<class Scalar, class LO = int>
class Tpetra::Experimental::LittleVector< Scalar, LO >

Nonowning view of a set of degrees of freedom corresponding to a mesh point in a block vector or multivector.

Template Parameters
ScalarThe type of entries.
LOThe type of local indices. See the documentation of the first template parameter of Map for requirements.

"Little" means local (not distributed over multiple MPI processes; stored to maximize locality) and small (think length 3, not length 1000).

The Scalar template parameter may be const or nonconst. This is one reason why instance methods below that take a LittleVector accept it as a template parameter: that lets you add a const LittleVector (e.g., LittleVector<const double, int>) to a nonconst LittleVector (e.g., LittleVector<double, int>).

Definition at line 1259 of file Tpetra_Experimental_BlockView.hpp.

Member Typedef Documentation

template<class Scalar, class LO = int>
typedef Kokkos::LayoutRight Tpetra::Experimental::LittleVector< Scalar, LO >::array_layout

Data layout (in the same sense as Kokkos::View).

Definition at line 1272 of file Tpetra_Experimental_BlockView.hpp.

Constructor & Destructor Documentation

template<class Scalar, class LO = int>
Tpetra::Experimental::LittleVector< Scalar, LO >::LittleVector ( Scalar *const  A,
const LO  blockSize,
const LO  strideX 
)
inline

Constructor.

Parameters
A[in] Pointer to the vector's entries
blockSize[in] Dimension of the vector
strideX[in] Stride between consecutive entries

Definition at line 1278 of file Tpetra_Experimental_BlockView.hpp.

template<class Scalar, class LO = int>
template<class T >
Tpetra::Experimental::LittleVector< Scalar, LO >::LittleVector ( T *const  A,
const LO  blockSize,
const LO  strideX,
typename std::enable_if< !std::is_same< Scalar, T >::value &&std::is_convertible< Scalar, T >::value &&sizeof(Scalar)==sizeof(T), int * >::type  ignoreMe = NULL 
)
inline

Constructor that takes an impl_scalar_type pointer.

Parameters
A[in] Pointer to the vector's entries, as impl_scalar_type* rather than Scalar*
blockSize[in] Dimension of the vector
strideX[in] Stride between consecutive entries

While this constructor is templated on a type T, the intent is that T == impl_scalar_type. (We must template on T rather than using impl_scalar_type directly, because of how std::enable_if works.) The long, complicated std::enable_if expression ensures that this constructor only exists if Scalar differs from impl_scalar_type, but the two types are mutually compatible and have the same size. (They must be bitwise compatible, so that reinterpret_cast makes sense between them.)

Definition at line 1301 of file Tpetra_Experimental_BlockView.hpp.

Member Function Documentation

template<class Scalar, class LO = int>
Scalar* Tpetra::Experimental::LittleVector< Scalar, LO >::getRawPtr ( ) const
inline

Pointer to the vector's entries.

Definition at line 1315 of file Tpetra_Experimental_BlockView.hpp.

template<class Scalar, class LO = int>
Scalar* Tpetra::Experimental::LittleVector< Scalar, LO >::ptr_on_device ( ) const
inline

Pointer to the vector's entries.

Definition at line 1320 of file Tpetra_Experimental_BlockView.hpp.

template<class Scalar, class LO = int>
LO Tpetra::Experimental::LittleVector< Scalar, LO >::getBlockSize ( ) const
inline

The block size (number of degrees of freedom per mesh point).

Definition at line 1325 of file Tpetra_Experimental_BlockView.hpp.

template<class Scalar, class LO = int>
LO Tpetra::Experimental::LittleVector< Scalar, LO >::dimension_0 ( ) const
inline

Number of entries in the vector.

Definition at line 1330 of file Tpetra_Experimental_BlockView.hpp.

template<class Scalar, class LO = int>
LO Tpetra::Experimental::LittleVector< Scalar, LO >::size ( ) const
inline

Number of entries in the vector.

Definition at line 1335 of file Tpetra_Experimental_BlockView.hpp.

template<class Scalar, class LO = int>
LO Tpetra::Experimental::LittleVector< Scalar, LO >::getStride ( ) const
inline

Stride between consecutive entries.

Definition at line 1340 of file Tpetra_Experimental_BlockView.hpp.

template<class Scalar, class LO = int>
template<class IntegerType >
void Tpetra::Experimental::LittleVector< Scalar, LO >::stride ( IntegerType *const  s) const
inline

Stride between consecutive entries.

This exists for compatibility with Kokkos::View.

Definition at line 1348 of file Tpetra_Experimental_BlockView.hpp.

template<class Scalar, class LO = int>
impl_scalar_type& Tpetra::Experimental::LittleVector< Scalar, LO >::operator() ( const LO  i) const
inline

Reference to entry (i) of the vector.

Note
To Tpetra developers: This is returned as impl_scalar_type and not as Scalar, in order to avoid a lot of reinterpret_cast calls in the inner loop of the sparse matrix-vector multiply kernel of Tpetra::Experimental::BlockCrsMatrix. Any pair of types impl_scalar_type, Scalar used here should always be convertible in either direction, so the return type should not pose any issues in practice.

Definition at line 1362 of file Tpetra_Experimental_BlockView.hpp.

Member Data Documentation

template<class Scalar, class LO = int>
const int Tpetra::Experimental::LittleVector< Scalar, LO >::rank = 1
static

Number of dimensions.

Definition at line 1269 of file Tpetra_Experimental_BlockView.hpp.


The documentation for this class was generated from the following file: