rsmd

OMG! Molecular Dynamics with Rust!

Benchmarks

LJ Direct Summation

• To benchmark the rust LJ direct summation with criterion, do the following steps:

• set the environment variable to make use of THP and compile in release mode MALLOC_CONF="thp:always,metadata_thp:always" cargo build --release
• run the benchmarks with cargo bench

• To compare the current performance of the Rsmd implementation with the one of yamd, just let the plot.py -p plot the comparison of the previously measured Rust benchmark with the yamd one.

• To run the benchmark on the C++ yamd, please use the repo https://github.com/Heavypilgrim/MscMolecDynCPPBenchmark.

Verlet loop over ndarray positions and velocities

• To implement the iteration over all positions and velocities simultaneously and with the best performance, the following chart depicts the performance of different loop variations:
• The best performing one uses Zip to combine the two iterators for positions and velocities. (named loop_zip in chart)

            Zip::from(&mut self.positions)
                .and(&self.velocities)
                .for_each(|position, &velocity| {
                *position += velocity * 0.0001;
            });

• For detailed insights of the benchmark either open the index.html in a browser to interactively explore the individual graphs including probability density functions... benchmark_verlet_folder