Poster 81: Performance of Devito on HPC-Optimised ARM Processors
TimeTuesday, 19 November 20198:30am - 5pm
DescriptionWe evaluate the performance of Devito, a domain specific language (DSL) for finite differences on Arm ThunderX2 processors. Experiments with two common seismic computational kernels demonstrate that Devito can apply automatic code generation and optimization across Arm and Intel platforms. The code transformations include: parallelism, and SIMD vectorization (OpenMP >=4); loop tiling (with best block shape obtained via auto-tuning); domain-specific symbolic optimisation such as common sub-expression elimination and factorisation for Flop reduction, polynomial approximations for trigonometry terms, and heuristic hoisting of time-invariant expressions. Results show that Devito can achieve performance on Arm processors which is competitive to other Intel Xeon processors.