CubismNova is a C++ template library used for solving Partial Differential Equations (PDEs) on structured uniform or stretched grids as well as block-structured or adaptively refined grids (AMR). The library provides data structures for point-wise and stencil operations with support for efficient halo (ghost cell) communication using the Message Passing Interface (MPI). A toolbox with vectorized kernels for common operations such as finite difference operators, WENO reconstruction, interpolation, restriction and prolongation operators as well as various data compression schemes is available to the user. Extended data structures for use with various time integration schemes is available as well.
The library is a full refactoring of its successful predecessor Cubism :cite:t:`hejazialhosseini2012a` that has later won the Gordon Bell award for a compressible multicomponent flow problem in 2013 :cite:t:`rossinelli2013a`. Further optimizations on the same code are presented in :cite:t:`hadjidoukas2015a` and :cite:t:`hadjidoukas2015b`. The refactored library offers easier access for the community by separating high performance computing (HPC) concepts from the user of the library. The library user must be concerned with the algorithm design depending on the problem that needs to be solved. The refactored library further offers integrated multigrid solvers and compression algorithms to reduce the I/O overhead at scale. Moreover, the refactored library takes into account suitable data structures for use with heterogeneous accelerators :cite:t:`wermelinger2016a`. Apart from compressible multicomponent flow simulations (:cite:t:`sukys2018a`, :cite:t:`wermelinger2018a`, :cite:t:`rasthofer2019a`), the library is also used for incompressible multi-phase flow (:cite:t:`karnakov2019a`) as well as incompressible flow with collective swimmers (:cite:t:`verma2018a`).
The software documentation is hosted at https://cubismnova.readthedocs.io
CubismNova can be downloaded from GitHub:
$ git clone --recurse-submodules https://github.com/cselab/CubismNova
$ cd CubismNovaThe library can be compiled using cmake. A working MPI implementation is
required and the mpicc and mpic++ compiler wrappers must be in the
PATH environment variable. A debug build can be generated with
$ ./cmake_init.sh debug <install path>
$ cd debug
$ make -j && make test
$ make install
$ cd .. && rm -rf debugThis assumes that your starting directory is the project root of CubismNova.
An optimized build is likewise generated with
$ ./cmake_init.sh release <install path>
$ cd release
$ make -j && make test
$ make install
$ cd .. && rm -rf releaseInstead of release you can use any other token except debug. If the
<insall path> is a system directory use sudo make install instead.
This software follows the semantic versioning specification.
BSD © 2019 ETH Zurich
.. bibliography::