NekCEM is a discontinous-Galerkin, spectral-element solver for Maxwell's equations and the drift-diffusion equations written in Fortran and C. It runs efficiently in parallel on a wide variety of systems, from laptops to the supercomputers at the Argonne Leadership Computing Facility (ALCF) and the Oak Ridge Leadership Computing Facility (OLCF), including Nvidia GPUs. Its core data structure is based on the computational fluid dynamics code Nek5000.
To run simulations with NekCEM you will need the following things.
- An MPI implementation.
- Python 2.7 or higher (including all versions of Python 3).
- BLAS and Lapack.
Some notes on the dependencies:
- To keep things simple, make sure the compiler wrappers
mpif77
(ormpifort
) andmpicc
are on your path. This isn't strictly necessary, but without them you will have to do more work when compiling simulations. - Python is only used in the build process.
- The system version of Python on some ALCF and OLCF systems is
2.6; use
softenv
ormodules
to switch to a more recent version. Runsoft add +python
on asoftenv
system andmodule load python
on a modules system. - Again to keep things simple, make sure you can link to BLAS and
Lapack using
-lblas
and-llapack
.
To install NekCEM run
git clone https://github.com/NekCEM/NekCEM
cd NekCEM
sudo make install
The command make install
does a couple of things.
- It copies
NekCEM/src
andNekCEM/bin
to/usr/local
. - It symlinks some scripts to
/usr/local/bin
.
Note that installing to /usr/local
is simply the default option; the
install directory can be controlled in the standard way using the
variables DESTDIR
, prefix
, and bindir
.
If you want to help develop the NekCEM source code, first fork the NekCEM repo on Github. Then do
git clone https://github.com/<github-username>/NekCEM
cd NekCEM
git remote add upstream https://github.com/NekCEM/NekCEM
sudo make install_inplace
The command sudo make install_inplace
only symlinks scripts to
/usr/local/bin
, allowing a developer to edit the source in
their local clone while still having the necessary scripts on
their path.
Setting up a simulation with NekCEM requires creating four files.
- A user file. This is a Fortran file which contains various
subroutines used to control the solvers. Its file extension should
be
usr
. - A size file. This file contains compile-time parameters. It should
be called
SIZE
. - A read file. This file contains parameters which are read at
runtime. Its file extension should be
rea
. - A map file. This file contains the mapping between processors and
elements. Its file extension should be
map
, and it must have the same stem as the read file.
A typical NekCEM simulation will be set up like this
example
├── readfile.map
├── readfile.rea
├── userfile.usr
└── SIZE
To build and run the code do the following from the example
directory.
configurenek <solver> userfile
make
mpirun -np <number-of-processors> ./nekcem readfile &> log
Let's break down what's going on.
- In the first step
configurenek
creates a makefile. The<solver>
option determines which equations the application is targeting; it should be one ofmaxwell
,drift
, orshrod
. - In the second step the makefile builds the code in the normal way;
it produces an executable
nekcem
. - In the third step the code is run in the normal way for MPI applications.
The third step can be replaced with nek readfile <number-of-processors>
. On a typical system this will do the exact
same thing as mpirun
, but on ALCF and OLCF machines it will also
queue your job correctly.
The tests can be run with bin/runtests [options]
. For a complete
list of options use the -h
flag.