ICL Research Profile

Overview

Batched sparse linear algebra solvers form the new frontier for algorithmic development and performance engineering. Many applications require simultaneous solutions of small linear systems of structurally sparse equations. In order to move towards higher hardware utilization, it is essential to provide these applications with appropriate interfaces to efficient batched sparse solvers running on modern hardware accelerators.

The Batched Sparse Linear Algebra effort developed batched kernels for small tensor operations for unassembled matrix-free iterative solvers, batched solvers for partially assembled operators, and batched solvers with support for various sparse formats. Participants included representatives from ECP’s numerical libraries, scientific applications, and hardware vendors. The project milestones were sets of interfaces to batched sparse linear algebra solvers running on hardware accelerators for use in ECP libraries and applications. Other noteworthy outcomes included the development of sparse batched kernels, solvers/preconditioners, and the creation of interoperability in xSDK libraries with sparse and dense batched functions to benefit ECP applications.