I need to compute element-wise products of all pairwise combinations of rows of two arrays (one row from each array). There are two problems:

1. The matrices are generated by R, which is column-major always (I think), so I am taking big hits on memory access. It isn't obvious to me how best to convert. Making a copy in row-major fashion is the apparent choice.

2. I also would like to parallelize this operation and want to optimize for cache by using chunking. Since this is a ubiquitous technique, I am sure Eigen uses it, so I would like to know whether there is generic code in Eigen I can use for chunking.

Thanks a lot.

As for 1: "If the storage order is not specified, then Eigen normally defaults to storing the entry in column-major order."
// http://eigen.tuxfamily.org/dox-devel/To ... rders.html

Thanks for the reply.

Like I said, the two arrays are passed in from R, another program, which isn't as flexible in storage order as Eigen.

you can keep your column major storage and do scalar * vector products,

for(int k=0; k<mat.rows();++k)
#pragma omp parallel for
for(int j=0; j<mat.cols();++j)
res[k].col(j) = mat(k,j) * mat.col(j);