Analyzing multithreaded preconditioned conjugate gradient (mtPCG) algorithm on multicore architecture

Modeling and simulation has been used in science and engineering with theoretical analysis and practical experiments. Generally, these models are represented as partial differential equations(PDEs) which can be solved numerically using meshes and sparse matrices. The solution cost of sparse linear systems, typically, dominates the solution cost of these PDEs. Consequently, an efficient sparse linear solver is essential in scientific and engineering simulations. Researches have been focused on solving large scale problems using clustered systems using MPI programming. With the advent of multicore architecture, exploring parallelism on a single computer is getting popular. However, we do not have proper scientific algorithms to take advantages of this new architecture paradigm. In this paper, we introduce a multithreaded Preconditioned Conjugate Gradient (PCG) algorithm, namely mtPCG, which explores parallelism on multicore architectures. We also analyzed the performance and scalability of mtPCG on multicore architecture.