cluster_sparse_solver Schur complement - how to distribute the Schur matrix?

We need to find the Schur complement matrix of a sparse matrix A of form

A₁₁    A₁₂

A₂₁    A₂₂

I.e., we want  the Schur block defined by S = A₂₂ - A₂₁ A₁₁^-1 A₁₂, which can be done by the new sparse solver update. In cluster_sparse_solver, S is stored as a dense matrix.

Our problem is, S is too large to store on a single compute node, so we would like it to be distributed across all compute nodes. We can distribute the input matrix A using the current interface, but the Schur matrix S is always returned to MPI process 0. Is there some option to make it return distributed in MKL 2018 update 2?

If there is no option, we know we can work around the problem by partitioning A₂₂ further and finding the Schur complement matrix of each section. However, this would appear to require a lot of repeated calculations. Is there some way to save intermediate calculations involving A₁₁ so that these calculations don't need repeating for subsequent Schur calculations?

Thank you,

Laura