Lattice QCD
Parallel benchmark programs
Lattice QCD simulation programs spend most of their time in the
programs that apply the lattice Dirac operator to a given quark
field. A significant fraction of the time is also needed for the
computation of scalar products and linear combinations of such fields.
Moreover, on parallel computers the communication overhead can slow
down the simulations considerably.
Benchmark programs
The programs provided here are simple synthetic speed tests that measure the
computer time required for some frequently performed operations in a typical
lattice QCD calculation. There are two sets of programs, one testing the
network performance and the other measuring the time required for some basic
QCD programs. The latter are optimized for the common x86 processors,
but should perform well on other processors too. See the conference
report
M. Lüscher,
Lattice QCD on PCs?,
Nucl. Phys. B (Proc. Suppl.) 106 (2002) 21
(
ScienceDirect
,
arXiv.org
)
for some background information and early benchmark results.
The programs parallelize in 0,1,2,3 or 4 dimensions using standard MPI
communication routines. They are intended for measurements of the QCD
performance of workstations, SMP machines
and massively parallel machines. The only prerequisites are an ISO C89
compliant C compiler and an MPI implementation that complies with the
MPI standard 1.2.
Download
The package
QCDpbm-1.1.tar.gz
contains the benchmark programs, usage instructions and some further
documentation files. After unpacking, first read the README file in the
top directory.
License
The software may be used under the terms of the
GNU General Public Licence (GPL)
.
Last updated 20 February 2015