This repository documents the software stack and the experimental setup of the manuscript: Automatic Code Generation for High-Performance Discontinuous Galerkin Methods on Modern Architectures.
It provides the technical details of how the results of the manuscript were achieved, but reproduction might require adapting its scripts and configuration files to your computing environment.
In general the procedure is:
- Clone this repository recursively
- Run ./patch.sh
- Run ./build_{haswell,skylake}.sh
- Go into the individual subfolders of build/dune-codegen-paper and apply this procedure to produce a plot:
- Run the execution script (like
donkey.sbatch
orskylake.sh
) (watch out for requirements on the working directory...) - If there, run
process.sh
to distill the data - Copy the resulting
csv
files into the source and commit them throughgit-lfs
- Run
plot.py
in a suitable env (say throughrun-in-dune-env
) to generate pdfs.
- Run the execution script (like