Installation

How do I install it?

Currently GPUVMEM only works in systems with Linux-based operating systems and with at least one NVIDIA GPU.

Requirements

• git
• git-lfs
• cmake >= 3.8
• cfitsio
• build-essential
• casacore v3.1.2.
• CUDA 9.0, 9.1, 9.2, 10.0 or 10.1 Update 2. Remember to add binaries and libraries to the PATH and LD_LIBRARY_PATH environment variables, respectively.
• OpenMP

Installation

In your own system

sudo apt install git git-flow -y
git clone --single-branch --branch v3.1.2 https://github.com/casacore/casacore.git
sudo apt-get install -y build-essential cmake gfortran g++ libncurses5-dev libreadline-dev flex bison libblas-dev liblapacke-dev libcfitsio-dev wcslib-dev libhdf5-serial-dev libfftw3-dev python-numpy libboost-python-dev libpython2.7-dev git-lfs
cd casacore
mkdir build
cd build
cmake ..
make
sudo make install
git clone https://github.com/miguelcarcamov/gpuvmem.git
cd gpuvmem
mkdir build
cd build
cmake ..
make -j

Using a Singularity container

We have created a recipe to use Singularity without worrying about installing CUDA and installing libraries. The container has gpuvmem compiled and all you need to work. Additionally, you can edit the recipe and add applications that you are interested in. To use the Singularity container, follow the next steps:

Install Singularity from here: http://singularity.lbl.gov/docs-installation.

Now we can use Singularity to shell into the container

mkdir container # Make a directory in your workspace
singularity shell --nv shub://miguelcarcamov/container_docker:casacore.gpuvmem.9.2 #Use this command for CUDA 9.2
# singularity shell --nv shub://miguelcarcamov/container_docker:casacore.gpuvmem.10.0 #Use this command for CUDA 10.0

The GPUVMEM binary should be in /gpuvmem/bin.