Introduction

This repository is a research artefact for the paper ``Partitioning In-Place on Massively Parallel Architectures"

Validating correctness

The script check.sh validates the results on various inputs and code sanitizers.

Replicating Results

To replicate the results on the system used by the authors, run

./bench_all_3070ti.sh 10

on a laptop with a 3070 Ti, and

./bench_all_a30.sh 10

on our SLURM-administrated cluster.

Reproducing Results

Running the experiments on different hardware will give different numbers, but the trend should be the same. If the test is done on a system using SLURM, the #SBATCH statements in bench.sh have to be modified accordingly.

You can then make a copy of bench_all_3070ti.sh (no SLURM), or bench_all_a30.sh (SLURM), and modify the smallest and largest problem sizes.