[HPC]^3: Workshop on High Performance Computing and Hybrid Programming Concepts for Hyperbolic PDE Codes
The King Abdullah University of Science and Technology is hosting a focused workshop on solver technology for the solution of hyperbolic PDEs. The [HPC]^3 workshop is focused on combining modern solution techniques for hyperbolic problems with high performance computing tools to enable efficient computation, including massive distributed-memory parallelism and many-core acceleration, while enabling advanced algorithmic technologies like adaptive mesh refinement and implicit solution techniques, using the Python scripting language. Workshop sessions will primarily be devoted to producing useful software by extending, adapting, and developing interoperable tools.
Participants will collaboratively develop modular software tools for the solution of hyperbolic equations on massively parallel hybrid architectures, including both distributed memory decompositions ( using PETSc+MPI) and on-node acceleration techniques (using OpenCL or CUDA). We will strive to sustain or create interoperability between Clawpack-related tools and other software projects (PetClaw, GeoClaw, SharpClaw, PyWENO/PyBLAW). In order to achieve good performance, we plan to integrate code generation techniques into the framework and automatically produce efficient kernels for hardware accelerators such as GPUs.
Additional development will focus on applications of existing code for the solution of wave equations in geophysics and in periodic materials. We will also explore the integration of scalable adaptive mesh refinement techniques through the leveraged use of parallel libraries such as p4est and CHOMBO.