Phifty One Ergs Blows Up a Star
phoebus
uses submodules extensively. To make sure you get them all, clone it as
git clone --recursive git@github.com:lanl/phoebus.git
or as
git clone git@github.com:lanl/phoebus.git
cd phoebus
git submodule update --init --recursive
Here are a few tips and tricks for using and building phoebus. The first step is creating a build directory, as in-source builds are forbidden. Assuming you just cloned phoebus,
cd phoebus
mkdir bin
cd bin
You also need to make sure you have appropriate dependencies in your path. On Darwin, loading the following modules (in this order) will work:
module load gcc/7.4.0 cuda/10.2 openmpi/4.0.1-gcc_7.4.0 cmake/3.17.3
(The default build for parthenon
requires Python3 loaded. This is
not the case for phoebus
.)
If you want output, make sure hdf5
is available as well. On
Darwin
, this likely means a spack environment. On snow
, just load
the appropriate module. If you are compiling with mpi, use
hdf5-parallel
, otherwise hdf5-serial
.
If cmake
can't find your hdf5 installation, You may have luck
setting -DHDF5_ROOT=/path/to/hdf5
at configure time.
If hdf5 isn't available on your target machine, here is one way to build it, using Spack.
# download spack
git clone https://github.com/spack/spack.git
# Activate the spack bash aliases. This typically needs to be done every time you want to use spack. I have it in my bashrc
. spack/share/spack/setup-env.sh
# load your modules
module load...
# Tell spack to learn about compilers and loaded modules
spack compiler find
spack external find
# Install parallel hdf5 with high-level libraries
spack install hdf5+mpi+hl
# load it
spack load hdf5
Installing from source, or just building without hdf5 will also both probably work.
Below are some example build commands
The following will build phoebus
with MPI parallelism but no shared memory parallelism
cmake ..
make -j
The following will build phoebus
with OpenMP parallelism only
cmake -DPHOEBUS_ENABLE_MPI=Off -DPHOEBUS_ENABLE_OPENMP=ON ..
make -j
The following will build phoebus
with no MPI or OpenMP parallelism.
cmake -DPHOEBUS_ENABLE_CUDA=On -DCMAKE_CXX_COMPILER=${HOME}/phoebus/external/singularity-eos/utils/kokkos/bin/nvcc_wrapper -DKokkos_ARCH_HSW=ON -DKokkos_ARCH_VOLTA70=ON -DPHOEBUS_ENABLE_MPI=OFF ..
A few notes for this one:
- Note here the
-DCMAKE_CXX_COMPILER
flag. This is necessary. You must set the compiler tonvcc_wrapper
provided by Kokkos. - Note the
-DKokkos_ARCH_*
flags. Those set the host and device microarchitectures and are required. The choice here is on anx86_64
machine with avolta
GPU.
The build options explicitly provided by phoebus
are:
Option | Default | Comment |
---|---|---|
PHOEBUS_ENABLE_CUDA | OFF | Cuda |
PHOEBUS_ENABLE_HDF5 | ON | HDF5. Required for output and restarts. |
PHOEBUS_ENABLE_MPI | ON | MPI. Required for distributed memory parallelism. |
PHOEBUS_ENABLE_OPENMP | OFF | OpenMP. Required for shared memory parallelism. |
MACHINE_CFG | None | Machine-specific config file, optional. |
Some relevant settings from Parthenon and Kokkos you may need to play with are:
Option | Default | Comment |
---|---|---|
CMAKE_CXX_COMPILER | None | Must be set to nvcc_wrapper with cuda backend |
Kokkos_ARCH_XXXX | OFF | You must set the GPU architecture when compiling for Cuda |
You can see all the Parthenon build options here and all the Kokkos build options here
If you are proficient with cmake
You can optionally write a cmake
file that sets the configure parameters that you like on a given
machine. Both phoebus
and parthenon
can make use of it. You can point to the file with
-DMACHINE_CFG=path/to/machine/file
at config time or by setting the environment variable MACHINE_CFG
to point at it, e.g.,
export MACHINE_CFG=path/to/machine/file
An example machine file might look like
# Machine file for x86_64-volta on Darwin
message(STATUS "Loading machine configuration for Darwin x86-volta node")
message(STATUS "Assumes: module load module load gcc/7.4.0 cuda/10.2 openmpi/4.0.3-gcc_7.4.0 anaconda/Anaconda3.2019.10 cmake && spack load hdf5")
message(STATUS "Also assumes you have a valid spack installation loaded.")
set(PHOEBUS_ENABLE_CUDA ON CACHE BOOL "Cuda backend")
set(PHOEBUS_ENABLE_MPI OFF CACHE BOOL "No MPI")
set(Kokkos_ARCH_HSW ON CACHE BOOL "Haswell target")
set(Kokkos_ARCH_VOLTA70 ON CACHE BOOL "volta target")
set(CMAKE_CXX_COMPILER /home/jonahm/phoebus/external/parthenon/external/Kokkos/bin/nvcc_wrapper CACHE STRING "nvcc wrapper")
you could then configure and compile as
cmake -DMACHINE_CFG=path/to/machine/file ..
make -j
Run phoebus from the build
directory as
./src/phoebus -i path/to/input/file.pin
The input files are in phoebus/inputs/*
. There's typically one input file per problem setup file.
parthenon
asynchronous tasking and block-AMR infrastructuresingularity-eos
provides performance-portable equations of state and PTE solverssingularity-opac
provides performance-portable opacities and emissivitiesKokkos
provides performance portable shared-memory parallelism. It allows our loops to be CUDA, OpenMP, or something else. By default we use theKokkos
shipped withparthenon
.
cmake
for building
hdf5
for outputMPI
for distributed memory parallelismpython3
for visualization
We use clang format for code cleanliness. The current version of
clang-format
is pinned to clang-format-12
, however, you can try
formatting with other versions at your own risk.
You can format all files in git history automatically by calling
scripts/bash/format.sh
from anywhere within the Phoebus repo (but
outside of submodules).
You can specify which clang-format you use in the script with the CFM environment variable. For example:
CFM=clang-format-12 ./scripts/bash/format.sh
We are always happy to have users contribute to phoebus
.
To contribute, create a fork and issue a pull request against the main branch.
For more details on how to contribute to phoebus
, please see CONTRIBUTING.md.
We adhere to a code of conduct.
© 2021-2024. Triad National Security, LLC. All rights reserved. This program was produced under U.S. Government contract 89233218CNA000001 for Los Alamos National Laboratory (LANL), which is operated by Triad National Security, LLC for the U.S. Department of Energy/National Nuclear Security Administration. All rights in the program are reserved by Triad National Security, LLC, and the U.S. Department of Energy/National Nuclear Security Administration. The Government is granted for itself and others acting on its behalf a nonexclusive, paid-up, irrevocable worldwide license in this material to reproduce, prepare derivative works, distribute copies to the public, perform publicly and display publicly, and to permit others to do so.