Automated Software Development in Heterogeneous GPU/CPU Environments for Seismic Modeling

摘要

Recent studies in the field of seismic modeling show us that utilizing graphical processors as a part of computing clusters give significant performance benefits. There are different types of applications available for the most common cluster configuration. We introduce a research program framework called Conflux, which allows efficient utilization of the new paradigm of writing mixed GPU/CPU algorithms. Conflux provides a parallel programming framework to use CPUs and GPUs in collaboration as components of an integrated computing system. The paper shows advantages of using the framework to solve most typical tasks in the field of seismic modeling. The performance metrics were obtained from two approaches in the implementation of 2-D finite-difference acoustic modeling: native C implementation for NVidia CUDA library and Conflux based on C#. The results given allow to conclude that keeping the same level of performance we gain an advantage in the overall development efforts for writing seismic algorithms as well as flexibility of targeting different types of hardware architectures including GPU-enabled clusters.