OpenMP/MPI Implementation of Tile QR Factorization on T2K Open Supercomputer
Tomohiro Suzuki, Hideki Miyashita · 2013
We implement the tile QR factorization algorithm on T2K open supercomputer using an OpenMP/MPI programming model. Our implementation does not outperform the ScaLAPACK routine for single-node performance. However, as the number of CPU nodes increases, our implementation clearly outperforms ScaLAPACK and shows good scalability in the form of both strong and weak scaling. With respect to accuracy, our implementation and ScaLAPACK achieve almost the same results. In this report, we provide the technical details of our implementation.