Installation of gan-phsp
Author: Tobias Stäuble
The reference installation was done on a Windows machine using WSL 1 with an Ubuntu 18.04 distribution. You should however be able to follow these instruction for most Unix systems.
Preparation
sudo apt-get update
Optionally, install cmake >= 3.12 to allow building in parallel, which will speed things up significantly.
Dependencies
sudo apt-get install git dpkg-dev cmake g++ gcc binutils libx11-dev libxpm-dev libxft-dev libxext-dev libpng-dev liblzma-dev libjpeg-dev python python-dev mesa-utils libtool automake build-essential libssl-dev
To check if you have a CUDA capable device: lspci | grep -i nvidia.
Install kernel headers and development packages for CUDA:
sudo apt-get install linux-headers-$(uname -r)
Head to https://developer.nvidia.com/cuda-downloads and download the runfile for the target system with e.g.
wget http://developer.download.nvidia.com/compute/cuda/11.0.1/local_installers/cuda_11.0.1_450.36.06_linux.run
Then run the file with sudo sh cuda_11.0.1_450.36.06_linux.run.
For libtorch 1.3.1 -> use cuda_10.1.105_418.39_linux.run
Download libcudnn from the NVIDIA site (an account may be necessary). Extract the archive into the install directory.
!! choose correct version here !!
sudo cp cuda/include/cudnn*.h /usr/local/cuda/include
sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64
sudo chmod a+r /usr/local/cuda/include/cudnn*.h /usr/local/cuda/lib64/libcudnn*
Root
cd install
curl https://root.cern/download/root_v6.20.04.source.tar.gz | tar xvz
mkdir builds
cd builds && mkdir root
cd root
cmake ../../root-6.20.04/
cmake --build .
Update path with: source bin/thisroot.sh
Geant4
cd install
curl http://geant4-data.web.cern.ch/geant4-data/releases/geant4.10.06.p02.tar.gz | tar xvz
cd builds && mkdir geant4 && geant4-build
cd geant4-build
cmake -DCMAKE_INSTALL_PREFIX=<PATH TO geant4 DIR IN BUILDS DIR> ../../geant4.10.06.p02/
make -j4
sudo make install
Geant4 is now installed at <PATH TO geant4 DIR IN BUILDS DIR>.
Update path with: source <installdir>/bin/geant4.sh
Libtorch
On the pytorch website, select (Stable, Linux, LibTorch, C++/Java, None or CUDA depending on whether a GPU is available) and download the cxx11 ABI file.
Note: I downloaded 1.3.1 and not the newest version as it seems to have compatiblity issues with gcc.
Extract the ZIP contents to install/libtorch/ folder.
Gate
Download and extract Gate 9.0 to the install folder.
cd Gate-9.0
sudo apt-get install cmake-curses-gui qt5-default libxmu-dev
cd builds && mkdir gate-9-build && mkdir gate-9
cd gate-9-build
ccmake ../../Gate-9.0 -DCMAKE_INSTALL_PREFIX=/mnt/c/Users/mlsostt/Documents/gan-phsp/install/builds/gate-9
Quick and dirty fix to install with CUDNN 8+:
Follow these steps: https://github.com/pytorch/pytorch/issues/40965
Hit c to configure, then enable GATE_USE_TORCH. Hit c again. Now an error shows up, but if you go to the variable definitions again, you can now configure Torch_DIR.
Set Torch_DIR to <PATH TO installdir>/libtorch/share/cmake/Torch. Hit c again. If it succeeds, hit g.
make -j4
sudo make install
export PATH=/<PATH TO gate-9 dir in install folder>/bin:$PATH
GAN 4 Gate (GAGA)
sudo apt-get install python3-pip
pip3 install gaga-phsp
pip3 install gatetools
pip3 install torch===1.5.1 torchvision===0.6.1 -f https://download.pytorch.org/whl/torch_stable.html
Note: libtorch = 1.3.0 / pytorch = 1.5.1
To be continued.
For building phsp_iaea_to_root: Set USE_Torch Flag to Off and specifically build only the target and not Gate
libtorch 1.3.1
with CUDA 10.1
and with cudnn cudnn-10.1-linux-x64-v7.6.5.32