BNLNPPS/esi-shell - Githubissues

esi-shell

The goal of this project is to provide a stable containerized environment for reproducible simulation jobs levereging on the Geant4 and NVIDIA OptiX ray tracing capabilities.

Prerequisites

Before starting, make sure you have the following prerequisites available and installed:

A CUDA-capable NVIDIA GPU
Docker Engine
NVIDIA container toolkit (installation guide)

Quick start

The installer script for the esi-shell container is available directly at bnlnpps.github.io/esi-shell/. It can be downloaded and then made executable:

curl -Os https://bnlnpps.github.io/esi-shell/esi-shell && chmod u+x esi-shell

The esi-shell environment can be used interactively by running the script:

./esi-shell

Once the container is up, you can execute the code relying on GPU functionality, e.g. run the available tests:

opticks-full-prepare
opticks-t

It is also possible to run any container command non-interactively:

./esi-shell "opticks-full-prepare && opticks-t"

Use the -h/--help option to get a quick summary of available options and to learn how to pass arguments to the underlying container, e.g.:

./esi-shell --help
./esi-shell -- -v $HOME/out:/tmp/results

For developers

If you plan to develop the code utilizing GPU capabilities, you will likely need to install NVIDIA OptiX. Place the downloaded file on the same path where you cloned github.com/BNLNPPS/esi-shell:

cd esi-shell
ls
... NVIDIA-OptiX-SDK-7.6.0-linux64-x86_64.sh ...

Now, the esi-shell image can be built locally

docker build -t esi-shell .

For local development with OptiX, install it on your host system. We recommend installing OptiX in /usr/local/optix but any other path will be as good:

export OPTIX_DIR=$HOME/optix
mkdir -p $OPTIX_DIR
./NVIDIA-OptiX-SDK-7.6.0-linux64-x86_64.sh --prefix=$OPTIX_DIR

When running esi-shell, make sure that the environment variable OPTIX_DIR is configured to point to the directory where OptiX is installed. If not set, the default path OPTIX_DIR=/usr/local/optix will be mounted insdie the container at runtime.

Docker

If preferred, you can pull a tagged release from the registry and work with the images directly. The list of all tagged releases can be found here. Run the tagged image with the local NVIDIA OptiX installation, e.g.:

docker run --rm -it --gpus all -v /usr/local/optix:$OPTIX_DIR ghcr.io/bnlnpps/esi-shell:<tag>

Opticks

One can get familiar with Opticks by running provided tests and examining the produced output. For example, in the properly setup environment do:

opticks/g4cx/tests/G4CXTest_raindrop.sh
python -i opticks/g4cx/tests/G4CXOpticks_setGeometry_Test.py

import plotly.graph_objects as go

tri = cf.sim.stree.mesh.G4_WATER_solid.tri
vtx = cf.sim.stree.mesh.G4_WATER_solid.vtx
m=go.Mesh3d(x=vtx.T[0], y=vtx.T[1], z=vtx.T[2], i=tri.T[0], j=tri.T[1], k=tri.T[2], color='green', opacity=0.2)
fig = go.Figure(data=[m])
fig.show()