AthenaPK

AthenaPK: a performance portable version based on Athena++, Parthenon and Kokkos.

Current state of the code

For this reason, it is highly recommended to only use AthenaPK with the Kokkos and Parthenon versions that are provided by the submodules (see building) and to build everything (AthenaPK, Parthenon, and Kokkos) together from source. Neither other versions or nor using preinstalled Parthenon/Kokkos libraries have been tested.

Current features include

first, second, and third order (magneto)hydrodynamics with
- RK1, RK2, RK3, VL2 integrators
- piecewise constant (DC), piecewise linear (PLM), piecewise parabolic (PPM), WENO3, LimO3, and WENOZ reconstruction
- HLLE (hydro and MHD), HLLC (hydro), and HLLD (MHD) Riemann solvers
- adiabatic equation of state
- MHD based on hyperbolic divergence cleaning following Dedner+ 2002
- anisotropic thermal conduction
- optically thin cooling based on tabulated cooling tables with either Townsend 2009 exact integration or operator-split subcycling
static and adaptive mesh refinement
problem generators for
- linear waves
- circularly polarized Alfven wave
- blast wave
- Kelvin-Helmholtz instability
- field loop advection
- Orszag Tang vortex
- cloud-in-wind/cloud crushing
- turbulence (with stochastic forcing via an Ornstein-Uhlenbeck process)

Latest performance results for various methods on a single Nvidia Ampere A100 can be found here.

Getting in touch

If you

encounter a bug or problem,
have a feature request,
would like to contribute, or
have a general question or comment

please either

open an issue/merge request, or
contact us in the AthenaPK channel on matrix.org #AthenaPK:matrix.org

Getting started

Installation

Dependencies

Required

CMake 3.13 or greater
C++17 compatible compiler
Parthenon (using the submodule version provided by AthenaPK)
Kokkos (using the submodule version provided by AthenaPK)

Optional

MPI
OpenMP (for host parallelism. Note that MPI is the recommended option for on-node parallelism.)
HDF5 (for outputs)
Python3 (for regressions tests with numpy, scipy, matplotlib, unyt, and h5py modules)
Ascent (for in situ visualization and analysis)

Building AthenaPK

Obtain all (AthenaPK, Parthenon, and Kokkos) sources

git clone https://github.com/parthenon-hpc-lab/athenapk.git athenapk
cd athenapk

# get submodules (mainly Kokkos and Parthenon)
git submodule init
git submodule update

Most of the general build instructions and options for Parthenon (see here) also apply to AthenaPK. The following examples are a few standard cases.

Most simple configuration (only CPU, no MPI, no HDF5). The Kokkos_ARCH_... parameter should be adjusted to match the target machine where AthenaPK will be executed. A full list of architecture keywords is available on the Kokkos wiki.

# configure with enabling Broadwell architecture (AVX2) instructions
cmake -S. -Bbuild-host -DKokkos_ARCH_BDW=ON -DPARTHENON_DISABLE_MPI=ON -DPARTHENON_DISABLE_HDF5=ON
# now build with
cd build-host && make
# or alternatively
cmake --build build-host

An Intel Skylake system (AVX512 instructions) with NVidia Volta V100 GPUs and with MPI and HDF5 enabled (the latter is the default option, so they don't need to be specified)

cmake -S. -Bbuild-gpu -DKokkos_ARCH_SKX=ON -DKokkos_ENABLE_CUDA=ON -DKokkos_ARCH_VOLTA70=ON
# now build with
cd build-gpu && make
# or alternatively build with
cmake --build build-gpu

Run AthenaPK

Some example input files are provided in the inputs folder.

# for a simple linear wave test run
./bin/athenaPK -i ../inputs/linear_wave3d.in

# to run a convergence test:
for M in 16 32 64 128; do
  export N=$M;
  ./bin/athenaPK -i ../inputs/linear_wave3d.in parthenon/meshblock/nx1=$((2*N)) parthenon/meshblock/nx2=$N parthenon/meshblock/nx3=$N parthenon/mesh/nx1=$((2*M)) parthenon/mesh/nx2=$M parthenon/mesh/nx3=$M
done

# and check the resulting errors
cat linearwave-errors.dat

Data Analysis

There exit several options to read/process data written by AthenaPK -- specifically in the file_type = hdf5 format, see Parthenon doc:

With ParaView and VisIt. In ParaView, select the "XDMF Reader" when prompted.
With yt -- though currently through a custom frontend that is not yet part of the main yt branch and, thus, has to be installed manually, e.g., as follows:

cd ~/src # or any other folder of choice
git clone https://github.com/forrestglines/yt.git
cd yt
git checkout parthenon-frontend

# If you're using conda or virtualenv
pip install -e .
# OR alternatively, if you using the plain Python environment
pip install --user -e .

Afterwards, *.phdf files can be read as usual with yt.load().

Using Ascent (for in situ visualization and analysis). This requires Ascent to be installed/available at compile time of AthenaPK. To enable set PARTHENON_ENABLE_ASCENT=ON.
(Not recommended) Using the integrated Python script called "phdf" provided by Parthenon, i.e., the either install parthenon_tools (located in external/parthenon/scripts/python/packages/parthenon/tools) or add that directory to your Python path. Afterwards data can be read, e.g., as follows

data_file = phdf.phdf(data_filename)
prim = data_file.Get("prim")

see also an internal regression test that uses this interface here.

jisuoqing/athenapk