Efficient scaling of a hydrodynamics simulation using compiler-based accelerator technology
Jordan A. Bradshaw, P. Moore, Behzad Torkian · IEEE International Conference on High Performance Computing, Data, and Analytics · 2015
In the realm of numerical modeling, there is often a requirement to run simulations with higher spatial and temporal resolutions. Increasing resolution can improve the accuracy of simulations with a corresponding increase in the run time. These run times can become impractical for conventional sequentially coded simulations.Parallelizing simulation code offers great possibilities for improved run-time performance, but it can be very difficult to achieve the necessary speedups when software development and debugging time are considered. This is especially the case when extensive modifications to legacy source code are needed to incorporate parallel processing capabilities.To address this problem, we used a directive-driven compiler to parallelize a hydrodynamics simulation targeting GPU hardware. This compiler-level parallel development environment enabled us to parallelize an existing legacy program with minimal alterations to the original source code.Using compiler directives and testing methodologies, we were able to achieve more than a 15x speedup and reduce the run time of high resolution simulations from months to days while verifying that results were equivalent to the sequential legacy version of the simulation.