Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

Compressed basis GMRES on high-performance graphics processing units

Journal Article · · International Journal of High Performance Computing Applications
 [1];  [2];  [3];  [4];  [4]
  1. Universitat Jaume I, Castelló (Spain)
  2. Karlsruhe Inst. of Technology (KIT) (Germany); Univ. of Tennessee, Knoxville, TN (United States)
  3. Karlsruhe Inst. of Technology (KIT) (Germany)
  4. Universitat Politècnica de València (Spain)

Krylov methods provide a fast and highly parallel numerical tool for the iterative solution of many large-scale sparse linear systems. To a large extent, the performance of practical realizations of these methods is constrained by the communication bandwidth in current computer architectures, motivating the investigation of sophisticated techniques to avoid, reduce, and/or hide the message-passing costs (in distributed platforms) and the memory accesses (in all architectures). This article leverages Ginkgo’s memory accessor in order to integrate a communication-reduction strategy into the (Krylov) GMRES solver that decouples the storage format (i.e., the data representation in memory) of the orthogonal basis from the arithmetic precision that is employed during the operations with that basis. Given that the execution time of the GMRES solver is largely determined by the memory accesses, the cost of the datatype transforms can be mostly hidden, resulting in the acceleration of the iterative step via a decrease in the volume of bits being retrieved from memory. Together with the special properties of the orthonormal basis (whose elements are all bounded by 1), this paves the road toward the aggressive customization of the storage format, which includes some floating-point as well as fixed-point formats with mild impact on the convergence of the iterative process. We develop a high-performance implementation of the “compressed basis GMRES” solver in the Ginkgo sparse linear algebra library using a large set of test problems from the SuiteSparse Matrix Collection. We demonstrate robustness and performance advantages on a modern NVIDIA V100 graphics processing unit (GPU) of up to 50% over the standard GMRES solver that stores all data in IEEE double-precision.

Research Organization:
US Department of Energy (USDOE), Washington, DC (United States). Office of Science, Exascale Computing Project
Sponsoring Organization:
USDOE National Nuclear Security Administration (NNSA)
OSTI ID:
2424930
Journal Information:
International Journal of High Performance Computing Applications, Journal Name: International Journal of High Performance Computing Applications Journal Issue: 2 Vol. 37; ISSN 1094-3420
Publisher:
SAGECopyright Statement
Country of Publication:
United States
Language:
English

References (21)

Adaptive precision in block-Jacobi preconditioning for iterative sparse linear system solvers: Adaptive precision in block-Jacobi preconditioning for iterative solvers journal March 2018
Using Ginkgo's memory accessor for improving the accuracy of memory‐bound low precision BLAS journal October 2021
Asynchronous Iterative Algorithm for Computing Incomplete Factorizations on GPUs book January 2015
Iterative Sparse Triangular Solves for Preconditioning book January 2015
Solving linear least squares problems by Gram-Schmidt orthogonalization journal March 1967
Accuracy and effectiveness of the Lanczos algorithm for the symmetric eigenproblem journal December 1980
Behavior of slightly perturbed Lanczos and conjugate-gradient recurrences journal February 1989
Incomplete Sparse Approximate Inverses for Parallel Preconditioning journal January 2018
Analyzing and improving maximal attainable accuracy in the communication hiding pipelined BiCGStab method journal August 2019
The Lanczos and conjugate gradient algorithms in finite precision arithmetic journal May 2006
1.1 Computing's energy problem (and what we can do about it)
  • Horowitz, Mark
  • 2014 IEEE International Solid- State Circuits Conference (ISSCC), 2014 IEEE International Solid-State Circuits Conference Digest of Technical Papers (ISSCC) https://doi.org/10.1109/ISSCC.2014.6757323
conference February 2014
Modified Gram-Schmidt (MGS), Least Squares, and Backward Stability of MGS-GMRES journal January 2006
Iterative Methods for Sparse Linear Systems book January 2003
Accuracy and Stability of Numerical Algorithms book January 2002
Inexact Krylov Subspace Methods for Linear Systems journal January 2004
Estimating the Attainable Accuracy of Recursively Computed Residual Methods journal July 1997
Theory of Inexact Krylov Subspace Methods and Applications to Scientific Computing journal January 2003
The university of Florida sparse matrix collection journal November 2011
Load-balancing Sparse Matrix Vector Product Kernels on GPUs journal April 2020
Toward a modular precision ecosystem for high-performance computing journal May 2019
Ginkgo: A high performance numerical linear algebra library journal August 2020


Similar Records

Advances in Mixed Precision Algorithms: 2021 Edition
Technical Report · Sat Aug 21 00:00:00 EDT 2021 · OSTI ID:1814447

Advances in Mixed Precision Algorithms: 2021 Edition
Technical Report · Wed Aug 18 00:00:00 EDT 2021 · OSTI ID:1814677

Reproduced Computational Results Report for “Ginkgo: A Modern Linear Operator Algebra Framework for High Performance Computing”
Journal Article · Tue Feb 15 23:00:00 EST 2022 · ACM Transactions on Mathematical Software · OSTI ID:1860819