README.md

WRF Benchmark

This is a lightly modified version of the NCAR 2019/2020 procurement WRF benchmark. The data files and validation scripts are identical.
Download WRFv4 and configure as desired. On niagara we used v4.5.2 ¹. See configure.wrf.niagara for our configuration. In particular, we added -DRSL0_ONLY to CFLAGS_LOCAL.
./compile -j xxx em_real
Download and unpack the benchmark data scinet-lp2bm-wrf-data.tar.gz ².
cd scinet-lp2bm-wrf-data/benchmark

Link in the run files from WRF:

for f in "$WRF_HOME/run/"*; do
  bf=$(basename "$f")
  [[ -e $bf ]] || ln -s "$f" "$bf"
done

Run; we used OMP_NUM_THREADS=1 mpirun "$WRF_HOME/main/wrf.exe" in a SLURM job script.
(cd ../validation && make) (requires gfortran).
python3 -m venv venv && . venv/bin/activate && pip install netCDF4
../validation/validate.csh . | tee validate.output. You should see three "thumbs up" images; see validate.output.niagara for an example.
The benchmark metric is the value of Total Time:.

Upload the following files:

configure.wrf
namelist.output
rsl.error.0000
rsl.out.0000
validate.output
wrfout_d01_2019-05-05_22:00:00
wrfout_d01_2019-05-05_22:06:00

If you used a custom version of WRF, please upload a copy of the source, or a link if public.