The AMReX training session, presented by Andrew Myers, Weiqun Zhang, and Mukul Dave of LBNL, is part of the Performance Portability training series. This series, offered by OLCF, NERSC, and ALCF, features training sessions on various performance portable programming solutions to help ease developer transitions between current and emerging high-performance computing (HPC) systems, such as the NERSC Perlmutter, OLCF Frontier, and ALCF Aurora featuring NVIDIA-, AMD-, and Intel-based GPUs, respectively.
Time: 9:00 - 11:30 am (Pacific time), Thursday, March 14 (online only)
Block-structured adaptive mesh refinement (AMR) provides a natural framework to focus computing power on the most critical parts of a problem in an efficient way. AMReX supports the development of block-structured AMR algorithms for solving systems of partial differential equations (PDEs) that require structured mesh and/or particle discretizations.
We will begin with an overview of AMReX and its applications, focusing on features to solve multiphysics problems on machines from laptops to supercomputers. In particular, we will describe how one can use AMReX to develop simulation codes that will work for both CPU and GPU systems.
Hands-on exercises will include passive scalar advection with time-dependent adaptivity, and the use of native linear solvers to impose incompressibility on a flow with particles around an obstacle. Files for the hands-on session will be hosted on Perlmutter and NERSC will provide training accounts if needed to the participants.