CSR:Medium:Collaborative Research: SparseKaffe: High-Performance, Auto-Tuned, Energy-Aware Algorithms for Sparse Direct Methods on Modern Heterogeneous Architectures
Grant
Overview
Affiliation
Other
View All
Overview
abstract
The use of sparse direct methods in computational science is ubiquitous. Direct methods can be used to find solutions to many numerical algebra applications, including sparse linear systems, sparse linear least squares, and eigenvalue problems; consequently they form the backbone of a broad spectrum of large scale applications. In the widely used and actively growing University of Florida Sparse Matrix Collection, there are problems from structural engineering, computational fluid dynamics (CFD), computer graphics/vision, robotics/kinematics, theoretical and quantum chemistry, power networks, social networks, document networks, among others.The SparseKaffe project team will develop algorithms and software for high-performance parallel sparse direct methods with irregular and hierarchical structure that can exploit clusters of Hybrid Multicore Processors to achieve orders of magnitude gains in computational performance, while also paying careful attention to the energy requirements. This requires the development of novel and innovative algorithms for scheduling, energy minimization, and memory management; development of novel user-guided autotuning algorithms that exploit different hardware characteristics; and designing a common infrastructure for creating auto-tuned software.The use of sparse direct methods is extensive, with many of the relevant science and engineering application areas being pushed to run at ever higher scales. The team expects SparseKaffe solvers to be able deliver not only high performance to the applications that use them, but also the energy efficiency that they will increasingly demand. The team will also create a course, and a corresponding set of course modules, to teach students how to develop algorithms and software that deliver orders of magnitude gains in performance on clusters of hybrid multicore processors.
