#include <source/sparse_matrix_simd.h>

Public Types
using	VectorizedArray = dealii::VectorizedArray< Number, simd_length >

template<typename Number2 >
using	EntryType = typename ryujin::EntryType< Number2, n_components >::type

Public Member Functions
	SparseMatrixSIMD ()

	SparseMatrixSIMD (const SparsityPatternSIMD< simd_length > &sparsity)

void	reinit (const SparsityPatternSIMD< simd_length > &sparsity)

template<typename SparseMatrix >
void	read_in (const std::array< SparseMatrix, n_components > &sparse_matrix, bool locally_indexed=true)

template<typename SparseMatrix >
void	read_in (const SparseMatrix &sparse_matrix, bool locally_indexed=true)

template<typename Number2 = Number>
EntryType< Number2 >	get_entry (const unsigned int row, const unsigned int position_within_column) const

template<typename Number2 = Number>
dealii::Tensor< 1, n_components, Number2 >	get_tensor (const unsigned int row, const unsigned int position_within_column) const

template<typename Number2 = Number>
EntryType< Number2 >	get_transposed_entry (const unsigned int row, const unsigned int position_within_column) const

template<typename Number2 = Number>
dealii::Tensor< 1, n_components, Number2 >	get_transposed_tensor (const unsigned int row, const unsigned int position_within_column) const

template<typename Number2 = Number>
void	write_entry (const Number2 entry, const unsigned int row, const unsigned int position_within_column, const bool do_streaming_store=false)

template<typename Number2 = Number>
void	write_entry (const dealii::Tensor< 1, n_components, Number2 > &entry, const unsigned int row, const unsigned int position_within_column, const bool do_streaming_store=false)

void	update_ghost_rows_start (const unsigned int communication_channel=0)

void	update_ghost_rows_finish ()

void	update_ghost_rows ()

template<typename Number2 >
DEAL_II_ALWAYS_INLINE auto	get_entry (const unsigned int row, const unsigned int position_within_column) const -> EntryType< Number2 >

template<typename Number2 >
DEAL_II_ALWAYS_INLINE dealii::Tensor< 1, n_components, Number2 >	get_tensor (const unsigned int row, const unsigned int position_within_column) const

template<typename Number2 >
DEAL_II_ALWAYS_INLINE auto	get_transposed_entry (const unsigned int row, const unsigned int position_within_column) const -> EntryType< Number2 >

template<typename Number2 >
DEAL_II_ALWAYS_INLINE dealii::Tensor< 1, n_components, Number2 >	get_transposed_tensor (const unsigned int row, const unsigned int position_within_column) const

template<typename Number2 >
DEAL_II_ALWAYS_INLINE void	write_entry (const Number2 entry, const unsigned int row, const unsigned int position_within_column, const bool do_streaming_store)

template<typename Number2 >
DEAL_II_ALWAYS_INLINE void	write_entry (const dealii::Tensor< 1, n_components, Number2 > &entry, const unsigned int row, const unsigned int position_within_column, const bool do_streaming_store)

Detailed Description

template<typename Number, int n_components, int simd_length>
class ryujin::SparseMatrixSIMD< Number, n_components, simd_length >

A specialized sparse matrix for efficient vectorized SIMD access.

In the vectorized row index region [0, n_internal_dofs) we store data as an array-of-struct-of-array type (see the documentation of class SparsityPatternSIMD for details). For the non-vectorized row index region [n_internal_dofs, n_locally_relevant_dofs) we store the matrix in CSR format (equivalent to the static dealii::SparsityPattern).

Definition at line 156 of file sparse_matrix_simd.h.

Member Typedef Documentation

◆ VectorizedArray

template<typename Number , int n_components, int simd_length>

using ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::VectorizedArray = dealii::VectorizedArray<Number, simd_length>

Definition at line 173 of file sparse_matrix_simd.h.

◆ EntryType

template<typename Number , int n_components, int simd_length>

template<typename Number2 >

using ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::EntryType = typename ryujin::EntryType<Number2, n_components>::type

The return type of get_entry and get_transposed_entry: If n_components is equal to one then EntryType reduces to Number2, otherwise the type is set to dealii::tensor<1, n_components, Number2>

Definition at line 182 of file sparse_matrix_simd.h.

Constructor & Destructor Documentation

◆ SparseMatrixSIMD() [1/2]

template<typename Number , int n_components, int simd_length>

ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::SparseMatrixSIMD

Definition at line 269 of file sparse_matrix_simd.template.h.

◆ SparseMatrixSIMD() [2/2]

template<typename Number , int n_components, int simd_length>

ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::SparseMatrixSIMD ( const SparsityPatternSIMD< simd_length > & sparsity )

Definition at line 276 of file sparse_matrix_simd.template.h.

References ryujin::SparsityPatternSIMD< simd_length >::n_nonzero_elements().

Member Function Documentation

◆ reinit()

template<typename Number , int n_components, int simd_length>

void ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::reinit ( const SparsityPatternSIMD< simd_length > & sparsity )

Definition at line 285 of file sparse_matrix_simd.template.h.

References ryujin::SparsityPatternSIMD< simd_length >::n_nonzero_elements().

◆ read_in() [1/2]

template<typename Number , int n_components, int simd_length>

template<typename SparseMatrix >

void ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::read_in	(	const std::array< SparseMatrix, n_components > &	sparse_matrix,
		bool	locally_indexed = `true`
	)

Definition at line 295 of file sparse_matrix_simd.template.h.

References RYUJIN_OMP_FOR, RYUJIN_PARALLEL_REGION_BEGIN, RYUJIN_PARALLEL_REGION_END, and ryujin::write_entry().

◆ read_in() [2/2]

template<typename Number , int n_components, int simd_length>

template<typename SparseMatrix >

void ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::read_in	(	const SparseMatrix &	sparse_matrix,
		bool	locally_indexed = `true`
	)

Definition at line 357 of file sparse_matrix_simd.template.h.

References RYUJIN_OMP_FOR, RYUJIN_PARALLEL_REGION_BEGIN, RYUJIN_PARALLEL_REGION_END, and ryujin::write_entry().

◆ get_entry() [1/2]

template<typename Number , int n_components, int simd_length>

template<typename Number2 = Number>

EntryType< Number2 > ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::get_entry	(	const unsigned int	row,
		const unsigned int	position_within_column
	)		const

Return the entry indexed by row and position_within_column. The return type will either be scalar if n_components is equal to 1, or a tensor if n_components is greater than 1.

Note: If the template argument Number2 is a vetorized array a specialized, faster access will be performed. In this case the index row must be within the interval [0, n_internal_dofs) and must be divisible by simd_length.

◆ get_tensor() [1/2]

template<typename Number , int n_components, int simd_length>

template<typename Number2 = Number>

dealii::Tensor< 1, n_components, Number2 > ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::get_tensor	(	const unsigned int	row,
		const unsigned int	position_within_column
	)		const

Return the tensor-valued entry indexed by row and position_within_column. This function performs the same operation as get_entry() except that it always returns the entry as a tensor (even if it is effectively a scalar entry).

Note: If the template argument Number2 is a vetorized array a specialized, faster access will be performed. In this case the index row must be within the interval [0, n_internal_dofs) and must be divisible by simd_length.

◆ get_transposed_entry() [1/2]

template<typename Number , int n_components, int simd_length>

template<typename Number2 = Number>

EntryType< Number2 > ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::get_transposed_entry	(	const unsigned int	row,
		const unsigned int	position_within_column
	)		const

Return the transposed entry indexed by row and position_within_column. The return type will either be scalar if n_components is equal to 1, or a tensor if n_components is greater than 1.

Note: If the template argument Number2 is a vetorized array a specialized, faster access will be performed. In this case the index row must be within the interval [0, n_internal_dofs) and must be divisible by simd_length.

◆ get_transposed_tensor() [1/2]

template<typename Number , int n_components, int simd_length>

template<typename Number2 = Number>

dealii::Tensor< 1, n_components, Number2 > ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::get_transposed_tensor	(	const unsigned int	row,
		const unsigned int	position_within_column
	)		const

Return the transposed tensor-valued entry indexed by row and position_within_column. This function performs the same operation as get_entry() except that it always returns the entry as a tensor (even if it is effectively a scalar entry).

Note: If the template argument Number2 is a vetorized array a specialized, faster access will be performed. In this case the index row must be within the interval [0, n_internal_dofs) and must be divisible by simd_length.

◆ write_entry() [1/4]

template<typename Number , int n_components, int simd_length>

template<typename Number2 = Number>

void ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::write_entry	(	const Number2	entry,
		const unsigned int	row,
		const unsigned int	position_within_column,
		const bool	do_streaming_store = `false`
	)

Write a (scalar valued) entry to the matrix indexed by row and position_within_column.

Note: If the template argument Number2 is a vetorized array a specialized, faster access will be performed. In this case the index row must be within the interval [0, n_internal_dofs) and must be divisible by simd_length.; This function is only available if n_components is equal to 1.

◆ write_entry() [2/4]

template<typename Number , int n_components, int simd_length>

template<typename Number2 = Number>

void ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::write_entry	(	const dealii::Tensor< 1, n_components, Number2 > &	entry,
		const unsigned int	row,
		const unsigned int	position_within_column,
		const bool	do_streaming_store = `false`
	)

Write a tensor-valued entry to the matrix indexed by row and position_within_column.

Note: If the template argument Number2 is a vetorized array a specialized, faster access will be performed. In this case the index row must be within the interval [0, n_internal_dofs) and must be divisible by simd_length.

◆ update_ghost_rows_start()

template<typename Number , int n_components, int simd_length>

void ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::update_ghost_rows_start ( const unsigned int communication_channel = 0 )

inline

Definition at line 646 of file sparse_matrix_simd.h.

◆ update_ghost_rows_finish()

template<typename Number , int n_components, int simd_length>

void ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::update_ghost_rows_finish

inline

Definition at line 750 of file sparse_matrix_simd.h.

◆ update_ghost_rows()

template<typename Number , int n_components, int simd_length>

void ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::update_ghost_rows

inline

Definition at line 763 of file sparse_matrix_simd.h.

◆ get_entry() [2/2]

template<typename Number , int n_components, int simd_length>

template<typename Number2 >

DEAL_II_ALWAYS_INLINE auto ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::get_entry	(	const unsigned int	row,
		const unsigned int	position_within_column
	)		const -> EntryType<Number2>

inline

Definition at line 369 of file sparse_matrix_simd.h.

◆ get_tensor() [2/2]

template<typename Number , int n_components, int simd_length>

template<typename Number2 >

DEAL_II_ALWAYS_INLINE dealii::Tensor< 1, n_components, Number2 > ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::get_tensor	(	const unsigned int	row,
		const unsigned int	position_within_column
	)		const

inline

Definition at line 384 of file sparse_matrix_simd.h.

◆ get_transposed_entry() [2/2]

template<typename Number , int n_components, int simd_length>

template<typename Number2 >

DEAL_II_ALWAYS_INLINE auto ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::get_transposed_entry	(	const unsigned int	row,
		const unsigned int	position_within_column
	)		const -> EntryType<Number2>

inline

Definition at line 449 of file sparse_matrix_simd.h.

◆ get_transposed_tensor() [2/2]

template<typename Number , int n_components, int simd_length>

template<typename Number2 >

DEAL_II_ALWAYS_INLINE dealii::Tensor< 1, n_components, Number2 > ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::get_transposed_tensor	(	const unsigned int	row,
		const unsigned int	position_within_column
	)		const

inline

Definition at line 465 of file sparse_matrix_simd.h.

◆ write_entry() [3/4]

template<typename Number , int n_components, int simd_length>

template<typename Number2 >

DEAL_II_ALWAYS_INLINE void ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::write_entry	(	const Number2	entry,
		const unsigned int	row,
		const unsigned int	position_within_column,
		const bool	do_streaming_store
	)

inline

Definition at line 555 of file sparse_matrix_simd.h.

◆ write_entry() [4/4]

template<typename Number , int n_components, int simd_length>

template<typename Number2 >

DEAL_II_ALWAYS_INLINE void ryujin::SparseMatrixSIMD< Number, n_components, simd_length >::write_entry	(	const dealii::Tensor< 1, n_components, Number2 > &	entry,
		const unsigned int	row,
		const unsigned int	position_within_column,
		const bool	do_streaming_store
	)

inline

Definition at line 580 of file sparse_matrix_simd.h.

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

source/sparse_matrix_simd.h
source/sparse_matrix_simd.template.h

Public Types

Public Member Functions

Detailed Description

Member Typedef Documentation

◆ VectorizedArray

◆ EntryType

Constructor & Destructor Documentation

◆ SparseMatrixSIMD() [1/2]

◆ SparseMatrixSIMD() [2/2]

Member Function Documentation

◆ reinit()

◆ read_in() [1/2]

◆ read_in() [2/2]

◆ get_entry() [1/2]

◆ get_tensor() [1/2]

◆ get_transposed_entry() [1/2]

◆ get_transposed_tensor() [1/2]

◆ write_entry() [1/4]

◆ write_entry() [2/4]

◆ update_ghost_rows_start()

◆ update_ghost_rows_finish()

◆ update_ghost_rows()

◆ get_entry() [2/2]

◆ get_tensor() [2/2]

◆ get_transposed_entry() [2/2]

◆ get_transposed_tensor() [2/2]

◆ write_entry() [3/4]

◆ write_entry() [4/4]