Maintainable Development of Performant User-Friendly Seismic Processing

This work presents a scalable and hardware-portable software framework designed to reconcile the conflicting goals of high-performance computing (HPC) optimization, software usability, and flexibility in seismic processing workflows. Graph-flow-based software offers users the ability to customize algorithms to specific data and geological needs, but complicates HPC development due to unpredictable data dependencies. To address this, the proposed framework decouples geophysical logic from data access by introducing a centralized buffering mechanism exposed through an abstract I/O interface. This enables complex algorithms—like RTM and KDM—to function seamlessly both within and outside of graph-based workflows, using optimized data movement strategies such as remote memory access (RMA). By making all data accessible through file-like abstractions, the framework simplifies tool development, supports backward compatibility, and enhances performance across diverse hardware environments.