vex::SpMat< val_t, col_t, idx_t > Class Template Reference

Sparse matrix in hybrid ELL-CSR format. More...

#include <spmat.hpp>

Public Types
typedef val_t	value_type

Public Member Functions
	SpMat ()
	Empty constructor.
	SpMat (const std::vector< cl::CommandQueue > &queue, size_t n, size_t m, const idx_t *row, const col_t *col, const val_t *val)
	Constructor. More...
void	mul (const vex::vector< val_t > &x, vex::vector< val_t > &y, val_t alpha=1, bool append=false) const
	Matrix-vector multiplication. More...
size_t	rows () const
	Number of rows.
size_t	cols () const
	Number of columns.
size_t	nonzeros () const
	Number of non-zero entries.

Static Public Member Functions
static std::string	inline_preamble (const cl::Device &device, int component, int position, kernel_generator_state &)
static std::string	inline_expression (const cl::Device &device, int component, int position, kernel_generator_state &)
static std::string	inline_parameters (const cl::Device &device, int component, int position, kernel_generator_state &)
static void	inline_arguments (cl::Kernel &kernel, uint device, size_t, uint &position, const SpMat &A, const vector< val_t > &x, kernel_generator_state &)

Detailed Description

template<typename val_t, typename col_t = size_t, typename idx_t = size_t>
class vex::SpMat< val_t, col_t, idx_t >

Sparse matrix in hybrid ELL-CSR format.

Constructor & Destructor Documentation

template<typename val_t, typename col_t = size_t, typename idx_t = size_t>

vex::SpMat< val_t, col_t, idx_t >::SpMat ( const std::vector< cl::CommandQueue > &  queue, size_t  n, size_t  m, const idx_t *  row, const col_t *  col, const val_t *  val  )

inline

Constructor.

Constructs GPU representation of the matrix. Input matrix is in CSR format. GPU matrix utilizes ELL format and is split equally across all compute devices. When there are more than one device, secondary queue can be used to perform transfer of ghost values across GPU boundaries in parallel with computation kernel.

Parameters

queue	vector of queues. Each queue represents one compute device.
n	number of rows in the matrix.
m	number of cols in the matrix.
row	row index into col and val vectors.
col	column numbers of nonzero elements of the matrix.
val	values of nonzero elements of the matrix.

References vex::is_cpu(), vex::qctx(), and vex::qdev().

Member Function Documentation

template<typename val_t, typename col_t = size_t, typename idx_t = size_t>

void vex::SpMat< val_t, col_t, idx_t >::mul ( const vex::vector< val_t > &  x, vex::vector< val_t > &  y, val_t  alpha = 1, bool  append = false  ) const

inline

Matrix-vector multiplication.

Matrix vector multiplication ( or ) is performed in parallel on all registered compute devices. Ghost values of x are transfered across GPU boundaries as needed.

Parameters

x	input vector.
y	output vector.
alpha	coefficient in front of matrix-vector product
append	if set, matrix-vector product is appended to y. Otherwise, y is replaced with matrix-vector product.

References vex::alignup(), vex::build_sources(), vex::bytes(), vex::kernel_workgroup_size(), vex::qctx(), and vex::qdev().

Referenced by vex::device_spmv_perf().

---

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

• vexcl/spmat.hpp