A Domain Specific Language for Performance Portable Molecular Dynamics Algorithms

摘要

Developers of Molecular Dynamics (MD) codes face significant challenges whenadapting existing simulation packages to new hardware. In a continuouslydiversifying hardware landscape it becomes increasingly difficult forscientists to be experts both in their own domain (physics/chemistry/biology)and specialists in the low level parallelisation and optimisation of theircodes. To address this challenge, we describe a "Separation of Concerns"approach for the development of parallel and optimised MD codes: the sciencespecialist writes code at a high abstraction level in a domain specificlanguage (DSL), which is then translated into efficient computer code by ascientific programmer. In a related context, an abstraction for the solution ofpartial differential equations with grid based methods has recently beenimplemented in the (Py)OP2 library. Inspired by this approach, we develop aPython code generation system for molecular dynamics simulations on differentparallel architectures, including massively parallel distributed memory systemsand GPUs. We demonstrate the efficiency of the auto-generated code by studyingits performance and scalability on different hardware and compare it to otherstate-of-the-art simulation packages. With growing data volumes the extractionof physically meaningful information from the simulation becomes increasinglychallenging and requires equally efficient implementations. A particularadvantage of our approach is the easy expression of such analysis algorithms.We consider two popular methods for deducing the crystalline structure of amaterial from the local environment of each atom, show how they can beexpressed in our abstraction and implement them in the code generationframework.