This is a short guide about installation procedures of the AdePT software stack. I installed Adept both on lxplus-gpu.cern.ch and on Riccardo's build machine (pitt-buildnode-01.cern.ch, ask Riccardo for an account if you wish), with different snags being encountered (lxplus-gpu was basically ready to use, pitt-buildnode-01 just being reconfigured on purpose.

Prerequisites

Two packages must be installed on the machine where one builds AdePT:

Xerces-C

if Xerces-C is pre-installed on the machine one is using, there is nothing to be done. if Xerces-C is not installed, the source code can be downloaded from Apache. Note that if Xerces-C is not installed in a "standard" directory (e.g. /usr/local) CMake will have to be told where to look for it, e.g.:

-DCMAKE_PREFIX_PATH=“/opt/shibboleth”

Geant4

if Geant4 is installed on the machine one is working on, the only thing to be done is to run the G4 initialisation shell which comes with the G4 installation, e.g.

 source <path_to_Geant4_initialization>/bin/geant4.sh 

 source /cvmfs/geant4.cern.ch/geant4/11.1/x86_64-centos8-gcc11-optdeb/bin/geant4.sh 

If Geant4 is not installed, the source code can be downloaded from the Geant4 web site and a local installation can be built: the G4 initialisation shell will take care of setting all environment variables accordingly. Please note that, in any case, GDML must be ON for AdePT examples to work correctly

 cmake -DGEANT4_USE_GDML=ON .

AdePT installation

The instructions one find in the AdePT GitHub repository are quite good and straight to the point but there may be a few differences here and there, depending on the machine configuration etc.

Cuda

Check where Cuda sits on your machine (e.g. /usr/local/cuda), there can be more than one installation, choose wisely! add Cuda to your PATH/LD_LIBRARY_PATH variables:

export PATH=/usr/local/cuda/bin:${PATH}
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH} 

note that by prepending the Cuda path you override any other (possible) Cuda installation on your system.

Then check the value of the Cuda architecture. if the Cuda demo suite is installed, this can be done using the deviceQuery utility

/usr/local/cuda/extras/demo_suite/deviceQuery

or, if the demo suite is not installed, by running:

nvidia-smi --query-gpu=compute_cap --format=csv

in the following I will be installing VecCore, VecGeom and Adept in the same ./install directory, of course installation can be customised.

mkdir install

VecCore

Note: you can skip the installation of VecCore if you need it for VecGeom only. See the VecGeom section for details.

Clone the VecCore repository and install all header files.

git clone https://github.com/root-project/veccore.git
cd veccore

cmake -S. -B./veccore-build -DCMAKE_INSTALL_PREFIX="../install"
cmake --build ./veccore-build --target install
cd ..

VecGeom

Clone the VecGeom repository, build and install libraries

git clone https://gitlab.cern.ch/VecGeom/VecGeom.git
cd VecGeom

cmake -S. -B./vecgeom-build   -DCMAKE_INSTALL_PREFIX=../install -DVECGEOM_ENABLE_CUDA=ON   -DVECGEOM_GDML=ON   -DBACKEND=Scalar   -DCMAKE_CUDA_ARCHITECTURES=75   -DVECGEOM_USE_NAVINDEX=ON   -DCMAKE_BUILD_TYPE=Release

cmake --build ./vecgeom-build --target install -- -j4
cd ..

NOTE: If you skipped the installation of VecCore at the previous step, you can use the VECGEOM_BUILTIN_VECCORE option to get a builtin version of it while building VecGeom:

git clone https://gitlab.cern.ch/VecGeom/VecGeom.git
cd VecGeom

cmake -S. -B./vecgeom-build  -DCMAKE_INSTALL_PREFIX=../install  -DVECGEOM_BUILTIN_VECCORE=ON -DVECGEOM_ENABLE_CUDA=ON  -DVECGEOM_GDML=ON   -DBACKEND=Scalar   -DCMAKE_CUDA_ARCHITECTURES=75   -DVECGEOM_USE_NAVINDEX=ON   -DCMAKE_BUILD_TYPE=Release 

cmake --build ./vecgeom-build --target install -- -j4
cd ..

AdePT

Clone the AdePT repository, fetch the G4HepEM library component and build everything into the examples

git clone https://github.com/apt-sim/AdePT.git
cd AdePT

don't forget to fetch the G4HepEM stuff

git submodule update --init 

configure and make

Note: Please, be sure to have correctly configured Geant4 in the shell you use to compile AdePT; otherwise, AdePT will be compiled anyway, but without the support for Geant4; also, the examples related to Geant4 (like the Example17 we use in our tests as a starting point) will not be compiled. Look for the CMake output, it should throw a warning when it does not find Geant4.

cmake -S. -B./adept-build -DCMAKE_PREFIX_PATH="../install" -DCMAKE_CUDA_ARCHITECTURES=75 -DCMAKE_BUILD_TYPE=Release

cmake --build ./adept-build -- -j4

Assuming everything was fine to here, examples can be found in

adept-build/BuildProducts/bin

Running example5

cd adept-build
CUDA_VISIBLE_DEVICES=0 BuildProducts/bin/example5 

AdePT-ATLAS

We forked the official AdePT repository into:

https://github.com/adept-atlas/AdePT

We will use this repo for all our ATLAS-related work.

Adding the new remote to an existing clone

If you already cloned a working copy from the official AdePT repository and started to modify it, you can push your changes to our repository by adding a new remote URL:

• using SSH (recommended, but you must have configured your SSH keys):

git remote add atlas git@github.com:adept-atlas/AdePT.git

• using HTTPS:

git remote add atlas https://github.com/adept-atlas/AdePT.git

After having set the new remote, you can push your changes to our ATLAS repository by running:

git add . 
git commit -m "my changes"
git push atlas master  # change 'master' to push to another branch

You can also set the 'atlas' fork as the default repository for your working copy. You can do that by running:

git branch -u atlas/master

After that, you can omit the 'atlas' repository when you want to pull changes from our fork. You only need to run:

git pull