Task-Based Programming on Emerging Parallel Architectures for Finite-Differences Seismic Numerical Kernel

In recent years, heterogeneous hardware have generalized in almost all supercomputer nodes, requiring a profound shift on the way numerical applications are implemented. This paper, illustrates the design and implementation of a seismic wave propagation simulator, based on the finite-differences numerical scheme, and specifically tailored for such massively parallel hardware infrastructures. The application data-flow is built on top of PaRSEC, a generic task-based runtime system. The numerical kernels, designed for maximizing data reuse can efficiently leverage large SIMD units available in modern CPU cores. A strong scalability study on a cluster of Intel KNL processors illustrates the application performances.