A Distributed Virtual Screening Miniapp for Performance Portability and Cross-Architecture Benchmarking

Drug discovery is a long and costly process with a low success probability. One of the challenges is the sheer size of the chemical space of drug candidates to explore. To mitigate this problem, we make use of large-scale virtual screening campaigns running on HPC centers to suggest which molecules should be tested in-vitro. For this reason, there is a recent trend that aims to rewrite scientific applications to improve computational efficiency. In this work, we target muDock, a single node virtual screening miniapp designed to experiment and benchmark virtual screening implementations for performance portability on different architectures. In particular, we enhance muDock with the ability to distribute the computation using a static partition, enabling scale-out across the available nodes of an HPC center. Experimental results show that muDock scales well, due to the embarrassingly parallel nature of the problem.