matrix_operations_row.hpp File Reference

Implementations of row-major dense matrix related operations, including matrix-vector products, using CUDA. More...

Go to the source code of this file.

Namespaces
	viennacl
	Main namespace in ViennaCL. Holds all the basic types such as vector, matrix, etc. and defines operations upon them.
	viennacl::linalg
	Provides all linear algebra operations which are not covered by operator overloads.
	viennacl::linalg::cuda
	Holds all CUDA compute kernels used by ViennaCL.

Functions
template<typename DestNumericT , typename SrcNumericT >
__global__ void	viennacl::linalg::cuda::convert_row_kernel (DestNumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const SrcNumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::trans_kernel (const NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_internal_size1, unsigned int A_internal_size2, unsigned int A_size1, unsigned int A_size2, unsigned int A_stride1, unsigned int A_stride2, NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_internal_size1, unsigned int B_internal_size2, unsigned int B_stride1, unsigned int B_stride2, bool data_major)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::am_row_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, NumericT fac2, unsigned int options2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::am_row_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *fac2, unsigned int options2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::ambm_row_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, NumericT fac2, unsigned int options2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2, NumericT fac3, unsigned int options3, const NumericT *C, unsigned int C_start1, unsigned int C_start2, unsigned int C_inc1, unsigned int C_inc2, unsigned int C_internal_size1, unsigned int C_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::ambm_row_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, NumericT fac2, unsigned int options2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2, const NumericT *fac3, unsigned int options3, const NumericT *C, unsigned int C_start1, unsigned int C_start2, unsigned int C_inc1, unsigned int C_inc2, unsigned int C_internal_size1, unsigned int C_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::ambm_row_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *fac2, unsigned int options2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2, NumericT fac3, unsigned int options3, const NumericT *C, unsigned int C_start1, unsigned int C_start2, unsigned int C_inc1, unsigned int C_inc2, unsigned int C_internal_size1, unsigned int C_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::ambm_row_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *fac2, unsigned int options2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2, const NumericT *fac3, unsigned int options3, const NumericT *C, unsigned int C_start1, unsigned int C_start2, unsigned int C_inc1, unsigned int C_inc2, unsigned int C_internal_size1, unsigned int C_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::ambm_m_row_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, NumericT fac2, unsigned int options2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2, NumericT fac3, unsigned int options3, const NumericT *C, unsigned int C_start1, unsigned int C_start2, unsigned int C_inc1, unsigned int C_inc2, unsigned int C_internal_size1, unsigned int C_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::ambm_m_row_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, NumericT fac2, unsigned int options2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2, const NumericT *fac3, unsigned int options3, const NumericT *C, unsigned int C_start1, unsigned int C_start2, unsigned int C_inc1, unsigned int C_inc2, unsigned int C_internal_size1, unsigned int C_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::ambm_m_row_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *fac2, unsigned int options2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2, NumericT fac3, unsigned int options3, const NumericT *C, unsigned int C_start1, unsigned int C_start2, unsigned int C_inc1, unsigned int C_inc2, unsigned int C_internal_size1, unsigned int C_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::ambm_m_row_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *fac2, unsigned int options2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2, const NumericT *fac3, unsigned int options3, const NumericT *C, unsigned int C_start1, unsigned int C_start2, unsigned int C_inc1, unsigned int C_inc2, unsigned int C_internal_size1, unsigned int C_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_assign_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, NumericT alpha)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_diagonal_assign_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, NumericT alpha)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::element_op_row_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2, const NumericT *C, unsigned int C_start1, unsigned int C_start2, unsigned int C_inc1, unsigned int C_inc2, unsigned int C_internal_size1, unsigned int C_internal_size2, unsigned int op_type)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::element_op_int_row_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2, const NumericT *C, unsigned int C_start1, unsigned int C_start2, unsigned int C_inc1, unsigned int C_inc2, unsigned int C_internal_size1, unsigned int C_internal_size2, unsigned int op_type)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_element_abs_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_element_acos_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_element_asin_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_element_atan_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_element_ceil_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_element_cos_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_element_cosh_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_element_exp_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_element_fabs_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_element_floor_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_element_log_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_element_log10_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_element_sin_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_element_sinh_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_element_sqrt_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_element_tan_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::matrix_row_element_tanh_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *B, unsigned int B_start1, unsigned int B_start2, unsigned int B_inc1, unsigned int B_inc2, unsigned int B_internal_size1, unsigned int B_internal_size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::vec_mul_row_kernel (const NumericT *A, unsigned int A_row_start, unsigned int A_col_start, unsigned int A_row_inc, unsigned int A_col_inc, unsigned int A_row_size, unsigned int A_col_size, unsigned int A_internal_rows, unsigned int A_internal_cols, const NumericT *v, unsigned int v_start, unsigned int v_inc, unsigned int v_size, NumericT *result, unsigned int result_start, unsigned int result_inc, unsigned int result_size)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::trans_vec_mul_row_kernel (const NumericT *A, unsigned int A_row_start, unsigned int A_col_start, unsigned int A_row_inc, unsigned int A_col_inc, unsigned int A_row_size, unsigned int A_col_size, unsigned int A_internal_rows, unsigned int A_internal_cols, const NumericT *v, unsigned int v_start, unsigned int v_inc, unsigned int v_size, NumericT *result, unsigned int result_start, unsigned int result_inc, unsigned int result_size)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::scaled_rank1_update_row_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, NumericT val, unsigned int options2, const NumericT *vec1, unsigned int start1, unsigned int inc1, unsigned int size1, const NumericT *vec2, unsigned int start2, unsigned int inc2, unsigned int size2)
template<typename NumericT >
__global__ void	viennacl::linalg::cuda::scaled_rank1_update_row_kernel (NumericT *A, unsigned int A_start1, unsigned int A_start2, unsigned int A_inc1, unsigned int A_inc2, unsigned int A_size1, unsigned int A_size2, unsigned int A_internal_size1, unsigned int A_internal_size2, const NumericT *val, unsigned int options2, const NumericT *vec1, unsigned int start1, unsigned int inc1, unsigned int size1, const NumericT *vec2, unsigned int start2, unsigned int inc2, unsigned int size2)

Detailed Description

Implementations of row-major dense matrix related operations, including matrix-vector products, using CUDA.

Definition in file matrix_operations_row.hpp.