Speaker
Dr
Ken-Ichi Ishikawa
(Hiroshima University)
Description
The most computationally demanding part of Lattice QCD simulations is
solving quark propagators.
Quark propagators are typically obtained with a linear equation solver utilizing HPC machines.
The success of Lattice QCD simulations owes much to the development of numerical algorithms and
optimization for the quark solver, and evolution of HPC machines.
The CCS QCD Benchmark is a benchmark program solving the Wilson-Clover quark propagator, and is
developed at Center of Computational Sciences (CCS), University of Tsukuba.
This is designed to be as simple as possible and is written in plain Fortran 90 so that
new algorithms or new HPC architectures can be evaluated quickly with this benchmark program.
We optimized the benchmark program for a Intel Xeon Phi (Knights Corner, KNC) system
named "COMA (PACS-IX)" at CCS Tsukuba under the Intel Parallel Computing Center program.
A single precision BiCGStab solver with the overlapped Restricted Additive Schwarz (RAS)
preconditioner was implemented using SIMD intrinsics, OpenMP and MPI in the offload-mode.
In this talk, we will show the optimization methods and the performance of the CCS QCD
benchmark on the COMA system.
Primary author
Dr
Ken-Ichi Ishikawa
(Hiroshima University)
Co-authors
Lawrence Meadows
(Intel Corporation)
Maurice Troute
(Intel Corporation)
Michael D`Mello
(Intel Corporation)
Ravi Vemuri
(Intel Corporation)
Prof.
Taisuke Boku
(CCS, U. of Tsukuba)
Prof.
Yoshinobu Kuramashi
(CCS, U. of Tsukuba)
