论文信息 - A GPU Implementation of the SIP Method

A GPU Implementation of the SIP Method

This article presents the implementation of the Strongly Implicit Procedure (SIP) to resolve penta-diagonal linear systems on a GPU in order to improve its computational efficiency. In particular, two different implementations of the SIP method on a GPU are presented and discussed. The experimental evaluation of the proposed methods demonstrates that a significant reduction on the runtime can be attained (acceleration values up to 4x ) when compared with both serial and parallel CPU implementations.

Martín Pedemonte | Pablo Ezzatti | Pablo Alfaro | Pablo Igounet

[1] Yao Zhang,et al. An Auto-tuned Method for Solving Large Tridiagonal Systems on the GPU , 2011, 2011 IEEE International Parallel & Distributed Processing Symposium.

[2] Enrique S. Quintana-Ortí,et al. Using graphics processors to accelerate the computation of the matrix inverse , 2011, The Journal of Supercomputing.

[3] Michael Garland,et al. Eﬃcient Sparse Matrix-Vector Multiplication on CUDA , 2008 .

[4] Jie Cheng,et al. Programming Massively Parallel Processors. A Hands-on Approach , 2010, Scalable Comput. Pract. Exp..

[5] Robert Strzodka,et al. Cyclic Reduction Tridiagonal Solvers on GPUs Applied to Mixed-Precision Multigrid , 2011, IEEE Transactions on Parallel and Distributed Systems.

[6] Frederica Darema,et al. The SPMD Model : Past, Present and Future , 2001, PVM/MPI.

[7] Gerhard Wellein,et al. Basic Optimisation Strategies for CFD-Codes , 2003 .

[8] H. L. Stone. ITERATIVE SOLUTION OF IMPLICIT APPROXIMATIONS OF MULTIDIMENSIONAL PARTIAL DIFFERENTIAL EQUATIONS , 1968 .

[9] Barbara Chapman,et al. Using OpenMP: Portable Shared Memory Parallel Programming (Scientific and Engineering Computation) , 2007 .