Dear All
Recently, we have successfully installed Dedalus v2.2006 on Archer 2 UK National Supercomputing Service. The process of installing against the native Cray libraries was not entirely straightforward and the following instructions might be useful for future Archer 2 users.
We have not performed any 'standard' benchmarking but our own code scales almost perfectly up to 32768 cores (256 nodes).
N.B. The install was performed on Archer 2 4-cabinet system but the procedure is expected to stay the same when the full system becomes available before the end of the year(?).
Alexander
WORKDIR=/work/[project code]/[project code]/[username]/
cd $WORKDIR
export PYTHONUSERBASE=$WORKDIR/.local # Recommended in ARCHER2 docs for user-level Python package installation
module load cray-python
module load cray-fftw
# Build & install h5py, linked to Cray HDF5 libraries, compiled using Cray compiler (CC=cc)
module load cray-hdf5-parallel
CC=cc pip install --user --no-binary=h5py h5py
# Now we'll do a manual build & install of Dedalus
# First create temporary build directory
mkdir -p /tmp/$USER
cd /tmp/$USER
# Download and extract the latest (v2.2006) Dedalus source release
wget https://github.com/DedalusProject/dedalus/archive/v2.2006.tar.gz
tar xf v2.2006.tar.gz
cd dedalus-2.2006
# Patch Dedalus' setup.py to add explicit MPI library dependency
sed -i -e "/^libraries = /s/]/, 'mpi']/" setup.py
# Now build Dedalus extension libraries using GCC, linking to Cray FFTW & correct MPI libraries
export FFTW_PATH=$CRAY_FFTW_PREFIX
export MPI_PATH=$CRAY_MPICH_BASEDIR/gnu/$PE_MPICH_GENCOMPILERS_CRAY
CC=gcc python setup.py install --user