This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| AMReX | |
|---|---|
| Name | AMReX |
| Developer | Exascale Computing Project; Lawrence Berkeley National Laboratory; Argonne National Laboratory |
| Programming language | C++; Fortran |
| Operating system | Linux; macOS; Microsoft Windows |
| License | BSD license |
AMReX AMReX is an open-source software framework for block-structured adaptive mesh refinement used in high-performance computing for large-scale simulation codes. It provides foundations for writing, deploying, and optimizing stencil-based and particle-in-cell applications on diverse hardware, enabling researchers at institutions such as Lawrence Berkeley National Laboratory, Argonne National Laboratory, and projects under the Exascale Computing Project to build scalable multiphysics solvers. The framework interfaces with compilers and runtimes from vendors including NVIDIA, Intel, and AMD to exploit accelerators and multicore architectures.
AMReX offers building blocks for finite-volume, finite-difference, and finite-element solvers used by projects at Oak Ridge National Laboratory, Sandia National Laboratories, and universities like Massachusetts Institute of Technology and University of California, Berkeley. It supplies data structures for hierarchies of grids, parallel distribution strategies compatible with MPI, and GPU offload mechanisms aligned with CUDA and HIP. The framework complements ecosystem tools such as HDF5, netCDF, and performance tools from NVIDIA Nsight and Intel VTune.
Development traces to teams at Lawrence Berkeley National Laboratory and collaboration with the Exascale Computing Project as part of efforts involving Department of Energy laboratories. Early work built on experiences from projects at Princeton University and Stanford University related to adaptive mesh techniques originating from research by groups associated with University of Chicago and University of Illinois Urbana-Champaign. Contributions have come from national labs including Argonne National Laboratory, Los Alamos National Laboratory, and Sandia National Laboratories, with funding and coordination involving agencies such as US Department of Energy programs and partnerships with vendors like NVIDIA and Intel Corporation.
The design centers on block-structured adaptive mesh refinement concepts related to algorithms developed at institutions like NASA Ames Research Center and laboratories influenced by work from Courant Institute of Mathematical Sciences. Core components include hierarchical grid containers, ghost cell management, and load-balancing strategies interoperable with libraries such as ParMETIS and Zoltan. The runtime model leverages MPI-based domain decomposition and integrates tasking approaches influenced by research from Barcelona Supercomputing Center and efforts represented at conferences like SC and International Conference for High Performance Computing, Networking, Storage, and Analysis.
AMReX supports multilevel mesh refinement, embedded boundaries akin to methodologies from Lawrence Livermore National Laboratory, and particle methods comparable to implementations in WarpX and Gkeyll. It provides parallel I/O support compatible with HDF5 and visualization workflows with ParaView and VisIt. Advanced capabilities include support for radiation-hydrodynamics used in codes at Los Alamos National Laboratory and magnetohydrodynamics relevant to groups at Princeton Plasma Physics Laboratory.
The codebase is primarily written in C++ with Fortran interoperability patterns common to projects from CERN and Fermilab. It integrates with compilers from GNU Project, Clang/LLVM, Intel Corporation, and vendor toolchains from NVIDIA and AMD. GPU support is provided through backends compatible with CUDA, HIP, and standards influenced by Kokkos and OpenMP offload directives championed at Oak Ridge National Laboratory.
AMReX underpins simulation codes in astrophysics at groups associated with Harvard University and Princeton University, accelerator modeling in projects like WarpX developed with Lawrence Berkeley National Laboratory and SLAC National Accelerator Laboratory, and fusion research at Princeton Plasma Physics Laboratory. Other domains include climate modeling collaborations with National Center for Atmospheric Research, combustion research at Sandia National Laboratories, and geoscience simulations involving teams at Columbia University.
Performance engineering collaborations with centers such as NERSC and OLCF focus on strong and weak scaling to hundreds of thousands of cores and accelerators used at facilities like Argonne Leadership Computing Facility. Benchmarks typically involve comparisons with frameworks used at Los Alamos National Laboratory and optimizations guided by profiling suites from NVIDIA Nsight and Intel VTune. Techniques include cache-aware tiling, communication-avoiding algorithms discussed at IEEE International Parallel and Distributed Processing Symposium, and hybrid MPI+X parallelism patterns.
The project follows open-source governance models similar to large scientific software collaborations at Kitware and community practices from Apache Software Foundation-style projects. Contributors hail from national labs, universities, and industry partners such as NVIDIA and Intel Corporation, participating through mailing lists, workshops at conferences like SC and SIAM meetings, and code reviews hosted on platforms aligned with services used by GitHub. The governance fosters collaboration with teams developing application codes including WarpX, Nyx, and other solver projects across the Exascale Computing Project community.
Category:Scientific computing software