Hi you all, I would like to try OpenMP sum reduction on a Eigen Vector3d.

The code is the following:

Code: Select all

Vector3d someFunction()
{
return Vector3d(1.0,2.0,3.0);
}

Vector3d val = Vector3d::Zero();
for (int i=0 ; i<8; i++)
{
    val+=someFunction();
}


A parallel OpenMP implementation should be

Code: Select all

Vector3d val = Vector3d::Zero();
#pragma omp parallel for  reduction(+:val)
for (int i=0 ; i<8; i++)
{
    val+=someFunction();
}


but I get the following g++ error (g++-4.5)

Code: Select all

error: ‘x’ has invalid type for ‘reduction’

This seems logic to me, but I want to ask you if a OpenMP reduction can in someway be applied to Eigen types and if it makes sense both in terms of speed and code readability.

Indeed, openmp supports only native types. You could emulate this by using a Matrix<double,3,Dynamic> vals(nb_threads), each thread accumulate into vals.col(threadid), and then you do val = vals.rowsize().sum().

It might be worth it only if you have to accumulate a large number of Vector3d.
As an alternative, you might simply fill a big Matrix<double,3,Dynamic, RowMajor>, and let Eigen does the reduction for you.