Solving sparse linear systems is pivotal in diverse computational domains, and overall execution time of computations is heavily impacted by the efficient solution of such systems. Respective methods typically involve preconditioning an iterative solver; however, choosing optimal or sometimes even numerically stable combinations can be quite challenging. We discuss how to predict effective preconditioner and iterative solver combinations for any given sparse linear problem, through a combination of embedding and linear modelling techniques. We focus on determining useful system features and investigate different metrics to quantify the relative performance of the solvers across the SuiteSparse matrix collection on different architectures.     «

Solving sparse linear systems is pivotal in diverse computational domains, and overall execution time of computations is heavily impacted by the efficient solution of such systems. Respective methods typically involve preconditioning an iterative solver; however, choosing optimal or sometimes even numerically stable combinations can be quite challenging. We discuss how to predict effective preconditioner and iterative solver combinations for any given sparse linear problem, through a combinatio...     »