Methods for compressible fluid simulation on GPUs using high-order finite differences

TL;DR

This paper presents optimized GPU methods for high-order finite-difference fluid simulations, achieving significant speedups by reducing memory bandwidth and cache requirements through innovative kernel decomposition and cache blocking techniques.

Contribution

It introduces two GPU-based high-order finite-difference methods for compressible fluid simulation, optimizing memory usage and performance with novel kernel strategies.

Findings

Achieves 343 million grid points per second with bandwidth-bound implementation.

Provides a 3.6x speedup over CPU-based hydrodynamics solver.

Demonstrates effective reduction of memory bandwidth and cache demands.

Abstract

We focus on implementing and optimizing a sixth-order finite-difference solver for simulating compressible fluids on a GPU using third-order Runge-Kutta integration. Since graphics processing units perform well in data-parallel tasks, this makes them an attractive platform for fluid simulation. However, high-order stencil computation is memory-intensive with respect to both main memory and the caches of the GPU. We present two approaches for simulating compressible fluids using 55-point and 19-point stencils. We seek to reduce the requirements for memory bandwidth and cache size in our methods by using cache blocking and decomposing a latency-bound kernel into several bandwidth-bound kernels. Our fastest implementation is bandwidth-bound and integrates $343$ million grid points per second on a Tesla K40t GPU, achieving a $3.6 \times$ speedup over a comparable hydrodynamics solver benchmarked on two Intel Xeon E5-2690v3 processors. Our alternative GPU implementation is latency-bound and achieves the rate of $168$ million updates per second.