In the current state, the benchmarks are very basic. A few common SMASH run scenarios are tested. More scenarios as well as microbenchmarks might be added in the future.
Before running, prepare a cleaned and (re)compiled build
directory with the version you want to benchmark.
The benchmarks are run by executing the shell script.
First the benchmark collects system information and then runs the different SMASH setups. The measurement is done with perf
and performed a few times to obtain an average and measure system fluctuations. The measurement is written to a markdown formatted file named bm-results-SMASH-VERSION.md
.
For comparison, results from previous tagged versions are attached to the corresponding github releases.
A utility script to easily compare the benchmark output is available and can be used specifying the result files of the version to be compared, e.g.
You may add other common SMASH scenarios. First add the configs to the respective directory and then modify the shell script accordingly.