Coupling Static and Dynamic MPI Correctness Tools to Optimize Accuracy and Overhead - Computational Artifact

This is the computational artifact for the paper "Coupling Static and Dynamic MPI Correctness Tools to Optimize Accuracy and Overhead" submitted to the Correctness'25 workshop.

Authors: Yussur Mustafa Oraji, Simon Schwitanski, Semih Burak, Christian Bischof, Matthias S. Müller

Repository Structure

• CoVer: Contains the CoVer tool used for static analysis
• MUST: Contains the MUST tool used for dynamic analysis
• proxy-app-benchmarks: Contains the proxy application benchmarks and the results used in the paper.
• classification-quality: Contains the classification quality testing infrastructure and the results used in the paper.
• schema_v1.json: JSON Schema File for the format used in the paper

Reproducing Results

Classification Quality

See here

Proxy Application Benchmarks

The proxy application benchmarks use the JUBE framework. To run the benchmarks, navigate to the proxy-app-benchmarks directory and use the following JUBE commands to reproduce the results:

# LULESH benchmarks
cd proxy-app-benchmarks/benchmarks/LULESH
jube run LULESH.xml --tag layout

# TeaLeaf benchmarks
cd proxy-app-benchmarks/benchmarks/TeaLeaf
jube run TeaLeaf.xml --tag layout

# Stencil benchmarks
cd proxy-app-benchmarks/benchmarks/Stencil
jube run PRK_stencil.xml --tag layout S_fixed

The results can be retrieved using

jube result

All results are stored in the results directory within the proxy-app-benchmarks directory:

• LULESH results
• TeaLeaf results
• Stencil results