This is the Gkeyll code. The name is pronounced as in the book "The Strange Case of Dr. Jekyll and Mr. Hyde" which is required reading for all members of the Gkeyll Team. Gkeyll is written in a combination of LuaJIT and C/C++. Gkeyll is developed at Princeton Plasma Physics Laboratory (PPPL) and is copyrighted 2016-2023 by Ammar Hakim and the Gkeyll Team.
Documentation for the code is available at http://gkeyll.rtfd.io.
Building gkyl requires
- a modern C/C++ compiler.
- Python 3 (for use in the
wafbuild system and post-processing). - NVIDIA Cuda compiler (nvcc) if using (NVIDIA) GPUs.
- The NCCL multi-GPU communication library.
- gkylzero (https://github.com/ammarhakim/gkylzero) compiled and installed.
The following instructions assume that these tools are present.
Installing gkyl consists of three steps: building or indicating, dependencies, configuring, and building the executable. If we have built gkyl in the computer of interest before, we likely saved scripts (machine files) that simplify this process. If we haven't, you'll need to either build new machine files or perform each of the installation steps manually.
We provide a set of "machine files" to ease the build process.
These are stored in the machines/ directory. For example, to
build on Perlmutter please run
./machines/mkdeps.perlmutter.sh
./machines/configure.perlmutter.sh
The first of these will install whichever dependencies are needed
(e.g. LuaJIT). The default is for these installations take place
in $HOME/gkylsoft; if a different install directory is desired,
specify it via the --prefix argument. The second of these
steps tells our waf build system where to find the
dependencies and where to place the gkyl executable
($HOME/gkylsoft by default, but a non-default directory can
be specified by changing GKYLSOFT). These two steps only need to be
done once, unless one wishes to change the dependencies.
Next, manually load the modules listed at the top of the
machines/configure.<machine name>.sh file. For example,
for Perlmutter do:
module load PrgEnv-gnu/8.3.3
module load cray-mpich/8.1.22
module load python/3.9-anaconda-2021.11
module load cudatoolkit/11.7
module load nccl/2.15.5-ofi
module unload darshan
Finally, build the gkyl executable using
./waf build install
The result will be a gkyl executable located in the
$HOME/gkylsoft/gkyl/bin/ directory.
For systems that do not already have corresponding files in the
machines/ directory, we encourage you to author machine files
for your machine following the existing ones as guides.
Instructions can be found in machines/README.md.
As a preliminary test, just to make sure the gkyl executable is
ok, you can do
$HOME/gkylsoft/gkyl/bin/gkyl -v
This will print some version information and the libraries gkyl
was built with. Since gkyl is a parallel code, and some clusters
don't allow simply calling the gkyl executable (especially on the
login node), you may have to use mpirun, mpiexec or srun
(see your cluster's documentation) to run gkyl with, for example,
srun -n 1 $HOME/gkylsoft/gkyl/bin/gkyl -v
You can run a regression test as a first simulation. For example, to run the Vlasov-Maxwell 2x2v Weibel regression test on a CPU, do
cd Regression/vm-weibel/
srun -n 1 $HOME/gkylsoft/gkyl/bin/gkyl rt-weibel-2x2v-p2.lua
and to run it on a GPU you may use
srun -n 1 $HOME/gkylsoft/gkyl/bin/gkyl -g rt-weibel-2x2v-p2.lua
You can run the full suite of unit tests using
cd Regression/
$HOME/gkylsoft/gkyl/bin/gkyl runregression config
$HOME/gkylsoft/gkyl/bin/gkyl runregression rununit
The postgkyl python package has been developed for plotting diagnostic
files from Gkeyll. It can be installed via conda using
conda install -c gkyl -c conda-forge postgkyl
For more information about postgkyl and how to use it, please see
https://gkeyll.readthedocs.io/en/latest/postgkyl/main.html.
All contributions to the code that improve the code via new functionality and/or refactoring of existing functionality are welcomes. Please strive for excellence in your programming and follow carefully the rest of the code structure while doing so.
Formatting guidelines given below are meant to reduce the thought given to minor (but asthetically important) issues. There are as many opionions on how to format code as there are developers. Hence, in Gkeyll these guidelines have been determined by the lead developer of the code and are not open for further discussion.
-
Do not modify existing code alignment or comments unless the code is wrong or the comment is incorrect, or if the formatting is egregiously bad.
-
Do not align multiple consecutive statements with = signs.
-
Do not mix tabs and spaces. Uses spaces consistently
-
Leave single space between LHS and RHS expressions.
-
You may or may not leave spaces between operators.
-
You may or may not leave spaces after a comma in a function call.
-
Do not comment obvious pieces of code.
-
Comment function call signatures for user-facing functions.
Gkeyll can be used freely for research at universities, national laboratories and other research institutions. If you want to use Gkeyll in a commercial environment, please ask us first.
We follow an open-source but closed development model. Even though read access to the code is available to everyone, write access to the source-code repository is restricted to those who need to modify the code. In practice, this means researchers at PPPL and our partner institutions. In particular, this means that for write access you either need to have jointly funded projects or jointly supervised graduate students or postdocs with Princeton University/PPPL.
In general, we allow users to "fork" the code to make their own modifications. However, we would appreciate if you would work with us to merge your features back into the main-line (if those features are useful to the larger Gkeyll team). You can submit a "pull request" and we will try our best to merge your changes into the mainline. Contributed code should compile and have sufficient unit/regression tests.
Gkeyll is developed at the Princeton Plasma Physics Laboratory (PPPL), a Department of Energy (DOE) national lab, managed by Princeton University. Funding for the code comes from Department of Energy, Airforce Office of Scientific Research, Advanced Projects Agency - Energy, National Science Foundation and NASA.
The institutions involved in Gkeyll development are PPPL, Princeton University, Virginia Tech, University of Maryland and MIT.
The CEO and Algorithm Alchemist of the project is Ammar Hakim.
The lead physicists for the project are Greg Hammett, Jason TenBarge and Ammar Hakim.
The major contributors to the code are: Noah Mandell, Manaure (Mana) Francisquez, Petr Cagas, James (Jimmy) Juno, Liang Wang and Tess Bernard.
sed -i '' -e "s/[[:space:]]* =/ =/g"