SeisSol is a high-performance software framework widely used for simulating seismic wave propagation and dynamic rupture processes. The incorporation of the acoustic wave model, which is crucial for accurately representing the natural processes in SeisSol, presents new performance challenges. The acoustic wave model, consisting of four partial differential equations (PDEs), is a special case of the elastic wave model, which comprises nine PDEs. Based on this, in the current version of SeisSol, the acoustic model is implemented by setting the second Lam´e parameter to zero within the elastic model, which leads to unnecessary resource usage. We implement a standalone acoustic model for SeisSol’s proxy application, evaluate its performance improvements using the roofline model, and analyze its cache utilization. Although the performance bottlenecks have not significantly changed, the runtime and memory data volume have been reduced by approximately half under several typical test conditions, while the last-level cache (LLC) miss rate has increased by up to about 17%. Additionally, the process of building the roofline model on the Leibniz Supercomputing Center (LRZ)’s CoolMUC-2 cluster, along with its advantages and limitations, is discussed.     «

SeisSol is a high-performance software framework widely used for simulating seismic wave propagation and dynamic rupture processes. The incorporation of the acoustic wave model, which is crucial for accurately representing the natural processes in SeisSol, presents new performance challenges. The acoustic wave model, consisting of four partial differential equations (PDEs), is a special case of the elastic wave model, which comprises nine PDEs. Based on this, in the current version of Se...     »