15. Laminar Unsteady Navier-Stokes Flow on Multicore Architectures
Authors: Bahareh Mostafazadeh Davani (University of California, Irvine)Ferran Marti (University of California, Irvine)Feng Liu (University of California, Irvine)Aparna Chandramowlishwaran (University of California, Irvine)
Abstract: Stencil computations are at the core of many scientific applications, and given the fact that these computations are often memory-bound, they eventually become the limiting portion of these applications. Such limitations give rise to the increasing focus on optimizing stencil computations. A vast number of improvements have been applied to stencil kernels, but very few of these attempts have applied the optimized stencils in real applications. Our work is unique in the sense that we target an entire solver that captures the interaction between multiple stencil patterns and one that is capable of simulating real applications. Our implementation solves the fluid motion of compressible viscous flow at transonic speeds. In this work, we take advantage of the full potential of modern multicore architectures to achieve an efficient and scalable Computational Fluid Dynamics solver. Our parallel solver achieves more than 40x speedup over a baseline C++ code.
Two-page extended abstract: pdf