Leveraging the High Bandwidth of Last-Level Cache for the First-Order Reverse Time Migration
Pavel Plotnitskii · 2024
Summary In this work we present performance results of multicore wavefront diamond tiling blocking (MWD-TB) RTM and its superiority to the traditional spatial blocking-based RTM. MWD-TB RTM provides 2X speedup, while the use of MWD-TB technique in forward modeling provides 4X speedup. For future work, we plan to deploy our MWD approach for the first-order acoustic wave equation into the full RTM and FWI pipelines using the latest Intel / AMD x86 architectures and GPU hardware accelerators. Please refer to the recent publication on this research [ 4 ] for the wavefield modeling problem results on different architectures