lama::CSRUtilsInterface::Mult< ValueType > Struct Template Reference

Define structure for multiplication routines. More...

#include <LAMAInterface.hpp>

Public Types
typedef void(*	scaleRows )(ValueType csrValues[], const IndexType csrIA[], const IndexType numRows, const ValueType values[])
	This operation multiplies each row with an own value.
typedef void(*	normalGEMV )(ValueType result[], const ValueType alpha, const ValueType x[], const ValueType beta, const ValueType y[], const IndexType numRows, const IndexType csrIA[], const IndexType csrJA[], const ValueType csrValues[], class SyncToken *syncToken)
	result = alpha * CSR-Matrix * x + b * y.
typedef void(*	sparseGEMV )(ValueType result[], const ValueType alpha, const ValueType x[], const IndexType numNonZeroRows, const IndexType rowIndexes[], const IndexType csrIA[], const IndexType csrJA[], const ValueType csrValues[], class SyncToken *syncToken)
	result = alpha * CSR-Matrix * x, CSR matrix has only some non-zero rows
typedef void(*	gemm )(ValueType result[], const ValueType alpha, const ValueType x[], const ValueType beta, const ValueType y[], const IndexType m, const IndexType n, const IndexType p, const IndexType csrIA[], const IndexType csrJA[], const ValueType csrValues[], class SyncToken *syncToken)
	This method computes result = alpha * CSR * x + beta * y with dense result, x, y.
typedef void(*	matrixAdd )(IndexType cJA[], ValueType cValues[], const IndexType cIA[], const IndexType numRows, const IndexType numColumns, const bool diagonalProperty, const ValueType alpha, const IndexType aIA[], const IndexType aJA[], const ValueType aValues[], const ValueType beta, const IndexType bIA[], const IndexType bJA[], const ValueType bValues[])
	computes c = alpha * a + beta * b for CSR sparse matrices a, b, c
typedef void(*	matrixMultiply )(ValueType cValues[], const IndexType cIA[], const IndexType cJA[], const IndexType numRows, const ValueType alpha, const IndexType aIA[], const IndexType aJA[], const ValueType aValues[], const IndexType bIA[], const IndexType bJA[], const ValueType bValues[])
	computes c = alpha * a * b for CSR sparse matrices a, b, c

---

Detailed Description

template<typename ValueType>
struct lama::CSRUtilsInterface::Mult< ValueType >

Define structure for multiplication routines.

---

Member Typedef Documentation

template<typename ValueType >

typedef void( * lama::CSRUtilsInterface::Mult< ValueType >::gemm)(ValueType result[], const ValueType alpha, const ValueType x[], const ValueType beta, const ValueType y[], const IndexType m, const IndexType n, const IndexType p, const IndexType csrIA[], const IndexType csrJA[], const ValueType csrValues[], class SyncToken *syncToken)

This method computes result = alpha * CSR * x + beta * y with dense result, x, y.

param[out] result has size m x n param[in] alpha scaling factor param[in] x has size p x n param[in] beta scaling factor param[in] y has size m x n param[in] csrIA offset array of CSR matrix, has size m + 1 param[in] csrJA has size csrIA[m], values between 0 and p-1 param[in] csrVaues is value array of CSR matrix param[in,out] syncToken might be used for asynchronous execution

template<typename ValueType >

typedef void( * lama::CSRUtilsInterface::Mult< ValueType >::matrixAdd)(IndexType cJA[], ValueType cValues[], const IndexType cIA[], const IndexType numRows, const IndexType numColumns, const bool diagonalProperty, const ValueType alpha, const IndexType aIA[], const IndexType aJA[], const ValueType aValues[], const ValueType beta, const IndexType bIA[], const IndexType bJA[], const ValueType bValues[])

computes c = alpha * a + beta * b for CSR sparse matrices a, b, c

Parameters:

[out]	cJA,cValues	are the matrix values of output matrix
[in]	cIA	contains already computed offsets
[in]	numRows	is number of rows for matrices a, b, c
[in]	numColums	is number of columns for matrices a, b, c
[in]	diagonalProperty	if true result matrix gets diagonal property
[in]	alpha	is a scalar scaling factor for matrix a
[in]	aIA,aJA,aValues	is input matrix a in CSR format
[in]	beta	is a scalar scaling factor for matrix b
[in]	bIA,bJA,bValues	is input matrix b in CSR format

In this routine the row offsets of C must already be determined before.

template<typename ValueType >

typedef void( * lama::CSRUtilsInterface::Mult< ValueType >::matrixMultiply)(ValueType cValues[], const IndexType cIA[], const IndexType cJA[], const IndexType numRows, const ValueType alpha, const IndexType aIA[], const IndexType aJA[], const ValueType aValues[], const IndexType bIA[], const IndexType bJA[], const ValueType bValues[])

computes c = alpha * a * b for CSR sparse matrices a, b, c

Parameters:

[out]	cValues	are the matrix values of output matrix
[in]	cIA,cJA	contain structure of output matrix
[in]	numRows	is number of rows for matrix c and a
[in]	alpha	is a scalar scaling factor
[in]	aIA,aJA,aValues	is input matrix a in CSR format
[in]	bIA,bJA,bValues	is input matrix b in CSR format

In this routine the final structure of C must already be determined before. Only those values of C are computed for which cIA, cJA specify an available entry.

template<typename ValueType >

typedef void( * lama::CSRUtilsInterface::Mult< ValueType >::normalGEMV)(ValueType result[], const ValueType alpha, const ValueType x[], const ValueType beta, const ValueType y[], const IndexType numRows, const IndexType csrIA[], const IndexType csrJA[], const ValueType csrValues[], class SyncToken *syncToken)

result = alpha * CSR-Matrix * x + b * y.

Parameters:

result	is the result vector
alpha	is scaling factor for matrix x vector
x	is input vector for matrix multiplication
beta	is scaling factor for additional vector
y	is additional input vector to add
numRows	is number of elements for all vectors and rows of matrix
csrIA,csrJA,csrValues	are arrays of CSR storage
syncToken	optional, if available starts asynchronous computation

template<typename ValueType >

typedef void( * lama::CSRUtilsInterface::Mult< ValueType >::scaleRows)(ValueType csrValues[], const IndexType csrIA[], const IndexType numRows, const ValueType values[])

This operation multiplies each row with an own value.

csr[i,j] *= values[i], for i = 0, ..., numRows-1

template<typename ValueType >

typedef void( * lama::CSRUtilsInterface::Mult< ValueType >::sparseGEMV)(ValueType result[], const ValueType alpha, const ValueType x[], const IndexType numNonZeroRows, const IndexType rowIndexes[], const IndexType csrIA[], const IndexType csrJA[], const ValueType csrValues[], class SyncToken *syncToken)

result = alpha * CSR-Matrix * x, CSR matrix has only some non-zero rows

Parameters:

result	is the result vector
alpha	is scaling factor for matrix x vector
x	is input vector for matrix multiplication
numNonZeroRows	is size of rowIndexes
rowIndexes	are indexes of non-empty rows in matrix
csrIA,csrJA,csrValues	are arrays of CSR storage
syncToken	optional, if available starts asynchronous computation

Note: this routine does not provide the term 'beta * y' as it would require to run over the full result vector

---

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

• /home/brandes/workspace/LAMA/src/lama/LAMAInterface.hpp