Sparse triangular solves for ILU revisited: data layout crucial to better performance

Authors:
Barry Smith;Hong Zhang
Affiliations:
Mathematics and Computer Science Division, Argonne National Laboratory, USA;Computer Science Department, Illinois Institute of Technology, USA
Venue:
International Journal of High Performance Computing Applications
Year:
2011

Citing 7
Cited 2

Achieving high sustained performance in an unstructured mesh CFD application

SC '99 Proceedings of the 1999 ACM/IEEE conference on Supercomputing
High-performacne parallel implicit CFD

Parallel Computing - Special issue on parallel computing in aerospace
A Scalable Parallel Algorithm for Incomplete Factor Preconditioning

SIAM Journal on Scientific Computing
SuperLU_DIST: A scalable distributed-memory sparse direct solver for unsymmetric linear systems

ACM Transactions on Mathematical Software (TOMS)
Iterative Methods for Sparse Linear Systems

Iterative Methods for Sparse Linear Systems
Globalized Newton-Krylov-Schwarz Algorithms and Software for Parallel Implicit CFD

International Journal of High Performance Computing Applications
Predicting memory-access cost based on data-access patterns

CLUSTER '04 Proceedings of the 2004 IEEE International Conference on Cluster Computing

Using PETSc to develop scalable applications for next-generation power grid

Proceedings of the first international workshop on High performance computing, networking and analytics for the power grid
Applications of the streamed storage format for sparse matrix operations

International Journal of High Performance Computing Applications

Quantified Score

Hi-index	0.01

Visualization

Abstract

A key to good processor utilization for sparse matrix computations is storing the data in the format that is most conducive to fast access by the memory system. In particular, for sparse matrix triangular solves the traditional compressed sparse matrix format is poor, and minor adjustments to the data structure can increase the processor utilization dramatically. Such adjustments involve storing the L and U factors separately and storing the U rows 'backwards' so that they are accessed in a simple streaming fashion during the triangular solves. Changes to the PETSc libraries to use this modified storage format resulted in over twice the floating-point rate for some matrices. This improvement can be accounted for by a decrease in the cache misses and TLB (transaction lookaside buffer) misses in the modified code.