ParILUT - A Parallel Threshold ILU for GPUs

被引：7

作者：

Anzt, Hartwig ^{[1
,2
]}

Ribizel, Tobias ^{[1
]}

Flegar, Goran ^{[3
]}

Chow, Edmond ^{[4
]}

Dongarra, Jack ^{[2
,5
,6
]}

机构：

[1] Karlsruhe Inst Technol, Steinbuch Ctr Comp, Karlsruhe, Germany

[2] Univ Tennessee, Innovat Comp Lab ICL, Knoxville, TN 37996 USA

[3] Univ Jaume I Castellon, Dept Ingn & Ciencia Comp, Castellon De La Plana, Spain

[4] Georgia Inst Technol, Sch Computat Sci & Engn, Atlanta, GA 30332 USA

[5] Univ Manchester, Manchester, Lancs, England

[6] Oak Ridge Natl Lab ORNL, Oak Ridge, TN USA

来源：

2019 IEEE 33RD INTERNATIONAL PARALLEL AND DISTRIBUTED PROCESSING SYMPOSIUM (IPDPS 2019) | 2019年

关键词：

ParILUT; parallel threshold ILU; incomplete factorization preconditioners; parallel selection; GPU;

D O I：

10.1109/IPDPS.2019.00033

中图分类号：

TP3 [计算技术、计算机技术];

学科分类号：

0812 ;

摘要：

In this paper, we present the first algorithm for computing threshold ILU factorizations on GPU architectures. The proposed ParILUT-GPU algorithm is based on interleaving parallel fixed-point iterations that approximate the incomplete factors for an existing nonzero pattern with a strategy that dynamically adapts the nonzero pattern to the problem characteristics. This requires the efficient selection of thresholds that separate the values to be dropped from the incomplete factors, and we design a novel selection algorithm tailored towards GPUs. All components of the ParILUT-GPU algorithm make heavy use of the features available in the latest NVIDIA GPU generations, and outperform existing multithreaded CPU implementations.

引用

页码：231 / 241

页数：11

共 21 条

[1]

Adinets A., OPTIMIZED FILTERING

[2]

Anzt H., 2017, ACCELERATING CONJUGA, P35

[3] PARILUT-A NEW PARALLEL THRESHOLD ILU FACTORIZATION [J].