Abstract: Density-functional theory (DFT) has been well established as a workhorse computational method in the quantum chemical and materials community. The shift in the computational landscape toward massively parallel calculations on heterogeneous architectures introduces new challenges for DFT software developers, necessitating software packages suited to multiple architectures.
In this talk, I will present our experiences developing two such software packages for modern high-performance computing (HPC) resources. The first package is the massively parallel, real-space FHI-aims DFT code for computational materials science. Special attention will be paid to the FHI-aims domain decomposition scheme on nonuniform grids, which enables compute- and memory-parallel computing across thousands of nodes for real-space operations and has been ported to accelerator architectures.
The second package is the ELectronic Structure Infrastructure (ELSI), an open-source community effort to unify libraries solving or circumventing the O(N3) Kohn-Sham eigenvalue equation. ELSI facilitates access to efficient Kohn-Sham solvers via a unified, code-independent interface for different system size scales, system types, and HPC architectures. Comparative benchmarks up to ten thousands of atoms on leading HPC resources will be shown.