Merge branch 'main' of code.ornl.gov:76a/olcfbaseimages into main

# Centos 8, Cuda 11.0.3, tensorflow 2.5.0, JAX

FROM code.ornl.gov:4567/76a/olcfbaseimages/mpiimage-centos-cuda:latest

ARG mpi_root

RUN dnf install -y dnf-plugins-core \

	&& dnf config-manager --set-enabled powertools

#RUN pip install horovod

#ENV LD_LIBRARY_PATH=/sw/peak/spack-envs/base/opt/linux-rhel8-ppc64le/gcc-9.1.0/spectrum-mpi-10.4.0.3-20210112-6jbupg3thjwhsabgevk6xmwhd2bbyxdc/lib:/usr/local/cuda/lib64:/usr/local/cuda/compat:/usr/local/nvidia/lib:/usr/local/nvidia/lib64

#RUN dnf install -y --allowerasing openmpi 

ENV PATH=$PATH:/usr/mpi/gcc/openmpi-4.0.3rc4/bin

ENV CPATH=$CPATH:/usr/mpi/gcc/openmpi-4.0.3rc4/include

ENV LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/mpi/gcc/openmpi-4.0.3rc4/lib64

RUN HOROVOD_WITH_MPI=1 CXX=/usr/bin/g++ CC=/usr/bin/gcc pip install --no-cache-dir --no-binary horovod horovod

RUN HOROVOD_WITH_MPI=1 CC=$mpi_root/bin/mpicc CXX=$mpi_root/bin/mpicxx pip install --no-cache-dir --no-binary horovod horovod

#

#ENV LD_LIBRARY_PATH="/opt/conda/lib:$LD_LIBRARY_PATH"

#!/bin/bash

module purge

module load DefApps

module load  gcc/9.1.0

export XDG_CACHE_HOME=/gpfs/alpine/stf007/world-shared/subil/tmp/podman-cache;

# this is needed because a dnf install step during the container build will fail saying that it can't create a file. 

# Said file might have been created in previous build attempts and so needs to be cleared. 

rm -rf /var/cache/dnf;

ulimit -n 262144;

podman build --build-arg mpi_root=$MPI_ROOT -v $MPI_ROOT:$MPI_ROOT,/gpfs/alpine/stf007/world-shared/subil/tmp:/tmp --ulimit nofile=262144:262144 -f Dockerfile.mpiimagebase_peak -t code.ornl.gov:4567/76a/olcfbaseimages/tensorflowimage-mpiimagebase-centos-cuda:latest . ;

podman save -o tensorflow_v2.5.0_mpiimagebase.tar code.ornl.gov:4567/76a/olcfbaseimages/tensorflowimage-mpiimagebase-centos-cuda:latest;

singularity build tensorflow_v2.5.0_mpiimagebase.sif docker-archive://tensorflow_v2.5.0_mpiimagebase.tar;