Performance Engineering for Exascale-Enabled Sparse Linear Algebra Building Blocks

The increasing demand for solving larger and more complex problems in computationalscience and engineering is a major driving factor to deploy computer systems withever-advancing performance capabilities. To increase the available performance, modernHPC platforms come with multiple levels of parallelism, complex memory hierarchies,heterogeneous architectures, and extreme scales. To match the need for sustainable andefficient software under these premises, special value has to be attached to the inherentchallenges like efficiency on all scales and performance portability across heterogeneousarchitectures.This work addresses the development of high-performance scientific software for sparselinear algebra, which is an important field of research and forms the foundation of manyapplications of computational science and engineering, with a special focus on sparseeigenvalue solvers on current and future supercomputers. Consequent employment ofperformance models as well as a holistic view on applications, algorithms, and hardwarearchitectures enable the creation of basic computational building blocks, custom computekernels, and optimized algorithmic formulations with provably high efficiency.To demonstrate the applicability of the developed software components, full-applicationperformance of selected sparse eigenvalue solvers for real-world problems on some ofthe world's largest supercomputers with completely different hardware architectures -including homogeneous multi-core CPU clusters, GPU-accelerated clusters, and selfhostedmany-core CPU clusters - is presented.