Porting OpenMP to GPGPU to accelerate Krylov space computations in coupled geodynamics simulator

Abstract

Understanding the causality between the events leading upto and post fault slip and the earthquake recording is important for seismic design and monitoring of underground structures, bridges and reinforced concrete buildings as well as climate mitigation projects like carbon sequestration and energy technologies like enhanced geothermal systems or oilfield wastewater disposal. While the events leading upto fault slip are typically governed by poroelastostatics, the events post fault slip can easily transition into poroelastodynamics territory due to runaway fault slip velocities. An understanding of expected fault slip velocities is critical apriori, as an algorithm which can seamlessly transition from time marching in poroelastostatics realm to poroelastodynamics realm and vice-versa is extremely difficult to achieve. That being said, every effort in the direction of accelerating the computations on the flow side are a necessary step forward in rendering a fast coupled geodynamics simulator. In this document, we present a framework in which we study the porting of the OpenMP parallelism of the flow simulator to a GPGPU