AMReX
Software framework for block structured AMR
Area: Mathematical libraries
CASS member: FASTMath
Description
AMReX is a C++-based software framework that supports the development of structured mesh algorithms for solving systems of partial differential equations, with options for adaptive mesh refinement, on machines from laptops to exascale architectures. AMReX uses an MPI+X model of hierarchical parallelism, in which X can be OpenMP for CPU-only machines, or CUDA, HIP or SYCL for NVIDIA, AMD or Intel GPUs, respectively. AMReX provides a performance portability layer for kernel launching on GPU accelerators. AMReX also provides data structures and iterators that enable developers to operate on distributed multi-dimensional arrays. Additionally, AMReX provides adaptive mesh refinement, memory management, parallel I/O, particles, complex geometries, linear solvers, and FFT. AMReX has a Python binding that provides GPU-enabled zero-copy data access for AI/ML, in situ analysis, application coupling and enables rapid, massively parallel prototyping.
Target audience
AMReX provides a unified software infrastructure with the functionality needed for a wide variety of structured grid applications to be able to effectively utilize current and future architectures. AMReX is open source and has an extensive set of online documentation, consisting of both narrative form documentation and an API reference. There are also a large number of example codes and associated documentation demonstrating how to build applications range from very simple to very complex. The power and flexibility of AMReX for enabling scientific advances is best illustrated by the range of applications that rely on it to run at the scale required by their science drivers. AMReX is being used by applications in accelerator modeling, astrophysics, atmosphere and ocean modeling, biology, combustion, cosmology, epidemiology, microelectronics, micro-fluidics, neutrino quantum kinetics, particle-laden multiphase flows, and solid mechanics, among others.