shared_y3_cluster_cpp_on_wc.md

Shared y3_cluster_cpp base environment on Wilson Cluster

Context

I want several people to be able to use a share-as-much-as-possible environment on the Wilson Cluster, to help make sure we're all really working with the same y3_cluster_cpp code.

• Since we are doing development of y3_cluster_cpp, each person doing development will need a copy of y3_cluster_cpp itself.

• Since gpuintegration is regularly being updated, each person should have a copy of gpuintegration. This allows us each to update when it is personally convenient. Note that you can choose to share my installation of gpuintegration, but then you're subject to getting a new version of gpuintegration whenever I modify code, rather than when you want to and are ready to update.

• The remainder of the software stack is not being modified regularly. Thus a single build can be shared.

Setting up your y3_cluster_cpp build environment

How the base environment was created.

My general plan is to use as much system-installed software as feasible, and to install with conda what ever else is feasible. pip installation into the conda environment is only used for Python modules not available through conda.

Modules and base conda installation.

module load tmux    # you can leave this out if you don't use tmux
module load cuda11
module load condaforge
conda activate /work1/numint/paterno/anaconda3/envs/cosmosis
# This sets CUDA_HOME but does not export it. That makes the
#  nvcc driver script in the conda environment fail.
export CUDA_HOME    # because CUDA_HOME is set, but not exported