Mixed-Precision Orthogonalization Scheme and Adaptive Step Size for Improving the Stability and Performance of CA-GMRES on GPUs

DOI10.1007/978-3-319-17353-5_2MaRDI QIDQ6487401zbMATH OpenFDO

Authors Ichitaro Yamazaki, Stanimire Tomov, Jack Dongarra

Publication date 9 December 2022

Full work available at URL https://doi.org/10.1007/978-3-319-17353-5_2

Mathematics Subject Classification ID

Numerical linear algebra (65Fxx) Computer aspects of numerical algorithms (65Yxx)

Recommendations

• A numerically stable communication-avoiding s-step GMRES algorithm
• Mixed-Precision Cholesky QR Factorization and Its Case Studies on Multicore CPU with Multiple GPUs
• Low synchronization Gram–Schmidt and generalized minimal residual algorithms
• Varying the \(s\) in your \(s\)-step GMRES
• A Communication Optimization Scheme for Basis Computation of Krylov Subspace Methods on Multi-GPUs

Cites work

• A Block Orthogonalization Procedure with Constant Synchronization Requirements
• A Newton basis GMRES implementation
• GMRES: A Generalized Minimal Residual Algorithm for Solving Nonsymmetric Linear Systems
• Mixed-Precision Cholesky QR Factorization and Its Case Studies on Multicore CPU with Multiple GPUs

Cited in

(8)

• Benefits from using mixed precision computations in the ELPA-AEO and ESSEX-II eigensolver projects
• scientific article; zbMATH DE number 1305038 (Why is no real title available?)
• Mixed-Precision Cholesky QR Factorization and Its Case Studies on Multicore CPU with Multiple GPUs
• Randomized Gram-Schmidt process with application to GMRES
• Mixed precision algorithms in numerical linear algebra
• Efficient GMRES+AMG on GPUs: composite smoothers and mixed V-cycles
• A numerically stable communication-avoiding s-step GMRES algorithm
• Distributed-parallel proper orthogonal/dynamic mode decompositions of large flow data

This page was built for publication: Mixed-Precision Orthogonalization Scheme and Adaptive Step Size for Improving the Stability and Performance of CA-GMRES on GPUs

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6487401)