[Trilinos-Users] Trilinos-Users Digest, Vol 102, Issue 8

[Hoemmen, Mark]

Hi Chen -- The best thing for you to do would be to give the Teuchos dense matrix to a shared-memory parallel BLAS / LAPACK implementation, or to a library such as PLASMA or MAGMA.  If I recall correctly, at least PLASMA if not also MAGMA has parallel algorithms that would help with your problem.  It's not Kokkos' job to implement dense matrix factorizations; other libraries do a great job with that.

Thanks!
mfh

________________________________________
Message: 1
Date: Mon, 17 Feb 2014 11:53:32 +0900
From: ChenYurui <tin at mma.cs.tsukuba.ac.jp>
Subject: [Trilinos-Users] Use shared memory in Teuchos dense matrix
To: trilinos-users <trilinos-users at software.sandia.gov>
Message-ID: <530179AC.1050604 at mma.cs.tsukuba.ac.jp>
Content-Type: text/plain; charset=ISO-2022-JP

Hi everyone,

There is some reasons that I have to use a Teuchos dense matrix and
Teuchos lapack to compute orthonormal basis matrix. Due to size of that
matrix is somehow so big, I wonder that if shared memory can be applied
into the computation process to speed up. I only know that Kokkos can
deal with this problem, however it seems that the matrix must be a
Tpetra matrix.