Skip to content

Latest commit

 

History

History
63 lines (48 loc) · 2.47 KB

README.md

File metadata and controls

63 lines (48 loc) · 2.47 KB
Raw

This container contains a build of wsclean with ASTRON IDG integrated to enable the container to utilize GPUs for acceleration.

Launching the container

The process for launching the wsclean + IDG container is a bit more involved than that for launching the other containers due to the need to ensure that the kernel modules are loaded and libraries linked in.

# Triggers the loading of all the needed NVIDIA kernel modules
# (should be replaced by use of `nvidia-modprobe` at some point but IIRC
#  `nvidia-modprobe` doesn't load all drivers with default params)
/usr/local/cuda/extras/demo_suite/bandwidthTest &> /dev/null

# Launch the container itself
singularity shell wsclean-gpu.simg

Note: that this container assume that CUDA has been installed in /usr/local/cuda and that this path has been bound into the container as this binds in a more complete set of libraries than adding the --nv parameter adds

Note: a wiki page describe the Image-Domain Gridder is available here.

When running the Image-Domain Gridder, the CUDA_DEVICE environment variable will control which GPUs are used. e.g. if you export CUDA_DEVICE=0,1 this will result in the first two GPUs being used.

The gridder uses OpenMP for parallelism on the CPU in hybrid and CPU mode, so if not on a dedicated machine you can set the OMP_NUM_THREADS environment variable to limit the number of CPU threads used.

Build instructions

First install CUDA using the packages available here and then add the following line to /etc/singularity/singularity.conf:

bind path = /usr/local/cuda

The build process for this container is slight different from the typical singularity build process as the CUDA libraries are not bound into the container at the time of the initial build. Thus first a writeable sandbox must be built, a first-run.sh script executed to build the application, and then the sandbox converted into a standard image.

The build can be done using a sequence of commands like the following:

SAND=${RANDOM}.sandbox
sudo singularity build --sandbox $SAND wsclean.def
sudo singularity exec --nv --writable $SAND /opt/build/first-run.sh
sudo singularity exec $SAND rm /opt/build/first-run.sh
sudo singularity build wsclean-gpu.simg $SAND
sudo rm -r $SAND