Matrix-Matrix Operations

High-performance matrix multiplications (GEMM). More...

Functions
template<typename T_Scalar>
void	culite::ops::mult (T_Scalar alpha, op_t opA, const dns::XxMatrix< T_Scalar > &A, op_t opB, const dns::XxMatrix< T_Scalar > &B, T_Scalar beta, dns::XxMatrix< T_Scalar > &C, CuBlasHandler &cuBlasHandler=globalCuBlasHandler())
	Updates a general matrix with a matrix-matrix product.
template<typename T_Int, typename T_Scalar>
void	culite::ops::mult (T_Scalar alpha, op_t opA, const csr::XxMatrix< T_Int, T_Scalar > &A, const dns::XxMatrix< T_Scalar > &B, T_Scalar beta, dns::XxMatrix< T_Scalar > &C, CuSparseHandler &cuSparseHandler=globalCuSparseHandler())
	Updates a general dense matrix with a sparse-dense matrix-matrix product.
template<typename T_Int, typename T_Scalar>
void	culite::ops::mult (T_Scalar alpha, op_t opA, const csc::XxMatrix< T_Int, T_Scalar > &A, const dns::XxMatrix< T_Scalar > &B, T_Scalar beta, dns::XxMatrix< T_Scalar > &C, CuSparseHandler &cuSparseHandler=globalCuSparseHandler())
	Updates a general dense matrix with a sparse-dense matrix-matrix product.

Detailed Description

High-performance matrix multiplications (GEMM).

Function Documentation

mult() [1/3]

template<typename T_Scalar>

void culite::ops::mult	(	T_Scalar	alpha,
		op_t	opA,
		const dns::XxMatrix< T_Scalar > &	A,
		op_t	opB,
		const dns::XxMatrix< T_Scalar > &	B,
		T_Scalar	beta,
		dns::XxMatrix< T_Scalar > &	C,
		CuBlasHandler &	cuBlasHandler = globalCuBlasHandler() )

Updates a general matrix with a matrix-matrix product.

Performs the operation \( C = \beta \cdot C + \alpha \cdot op_A(A) \cdot op_B(B) \). C can be General/Symmetric/Hermitian. If C is Symmetric/Hermitian, it is assumed that the outcome of the operation \( op_A(A) \cdot op_B(B) \) is also Symmetric/Hermitian and only the corresponding part (upper/lower) will be calculated. Valid combinations are the following:

A	B	opA	opB	C
General	General	unconstrained	unconstrained	General
Symmetric	General	N	unconstrained	General
Hermitian	General	N	unconstrained	General
General	General	N or T	T or N	Symmetric
General	General	N or C	C or N	Hermitian

Template Parameters

T_Scalar

The scalar type (e.g., float, double, complex).

Parameters

[in]	alpha	The scaling coefficient.
[in]	opA	The operation to be performed for matrix A.
[in]	A	The input matrix.
[in]	opB	The operation to be performed for matrix B.
[in]	B	The input matrix.
[in]	beta	The scaling coefficient for C.
[in,out]	C	The matrix to be updated.
[in]	cuBlasHandler	The cuBLAS handler for GPU operations (defaults to global handler).

Examples

ex02m_dense_matrix_algebra_vmult.cpp, ex02n_dense_matrix_algebra_mmult.cpp, ex02o_dense_matrix_algebra_mtmult.cpp, ex06m_sparse_matrix_algebra_vmult.cpp, and ex06n_sparse_matrix_algebra_mmult.cpp.

mult() [2/3]

template<typename T_Int, typename T_Scalar>

void culite::ops::mult	(	T_Scalar	alpha,
		op_t	opA,
		const csr::XxMatrix< T_Int, T_Scalar > &	A,
		const dns::XxMatrix< T_Scalar > &	B,
		T_Scalar	beta,
		dns::XxMatrix< T_Scalar > &	C,
		CuSparseHandler &	cuSparseHandler = globalCuSparseHandler() )

Updates a general dense matrix with a sparse-dense matrix-matrix product.

Performs the operation \( C = \beta \cdot C + \alpha \cdot op_A(A) \cdot B \). Valid combinations are the following:

A	B	opA	opB	C
General	General	unconstrained	N	General
Symmetric	General	N	N	General
Hermitian	General	N	N	General

Template Parameters

T_Int	The integer type for indexing.
T_Scalar	The scalar type (e.g., float, double, complex).

Parameters

[in]	alpha	The scaling coefficient.
[in]	opA	The operation to be performed for matrix A.
[in]	A	The input sparse matrix.
[in]	B	The input dense matrix.
[in]	beta	The scaling coefficient for C.
[in,out]	C	The dense matrix to be updated.
[in]	cuSparseHandler	The cuSPARSE handler for GPU operations (defaults to global handler).

mult() [3/3]

template<typename T_Int, typename T_Scalar>

void culite::ops::mult	(	T_Scalar	alpha,
		op_t	opA,
		const csc::XxMatrix< T_Int, T_Scalar > &	A,
		const dns::XxMatrix< T_Scalar > &	B,
		T_Scalar	beta,
		dns::XxMatrix< T_Scalar > &	C,
		CuSparseHandler &	cuSparseHandler = globalCuSparseHandler() )

Updates a general dense matrix with a sparse-dense matrix-matrix product.

Performs the operation \( C = \beta \cdot C + \alpha \cdot op_A(A) \cdot B \). Valid combinations are the following:

A	B	opA	opB	C
General	General	unconstrained	N	General
Symmetric	General	N	N	General
Hermitian	General	N	N	General

Template Parameters

T_Int	The integer type for indexing.
T_Scalar	The scalar type (e.g., float, double, complex).

Parameters

[in]	alpha	The scaling coefficient.
[in]	opA	The operation to be performed for matrix A.
[in]	A	The input sparse matrix.
[in]	B	The input dense matrix.
[in]	beta	The scaling coefficient for C.
[in,out]	C	The dense matrix to be updated.
[in]	cuSparseHandler	The cuSPARSE handler for GPU operations (defaults to global handler).