Docker images

These docker images use spack to build a CUDA-enabled development environment for Celeritas. There are two sets of images:

• dev (dev subdirectory) which leaves spack fully installed; and
• ci (ci subdirectory) which only copies the necessary software stack (thus requiring lower bandwidth on the CI servers).

Additionally there are two image configurations:

• rocky-cuda12: Rocky 9 with CUDA 12.
• ubuntu-rocm6: Ubuntu 24 with ROCm 6.3

Building

The included build.sh script drives the two subsequent docker builds. Its argument should be an image configuration name, or cuda or minimal as shortcuts for those.

If the final docker push fails, you may have to log in with your docker.io credentials first:

$ docker login -u sethrj

Running

The CI image is (in color) run with:

$ docker run --rm -ti -e "TERM=xterm-256color" celeritas/ci-cuda11

Note that the --rm option automatically deletes the state of the container after you exit the docker client. This means all of your work will be destroyed.

The launch-local-test script will clone an active GitHub pull request, build, and set up an image to use locally:

$ ./ci/launch-local-test.sh 123

To mount the image with your local source directory:

$ docker run --rm -ti -e "TERM=xterm-256color" \
    -v ${SOURCE}:/home/celeritas/src \
    celeritas/ci-focal-cuda11:${DATE}

where ${SOURCE} is your local Celeritas source directory and ${DATE} is the date time stamp of the desired image. If you just built locally, you can replace that last argument with the tag ci-focal-cuda11:

$ docker run --rm -ti -e "TERM=xterm-256color" -v /rnsdhpc/code/celeritas-docker:/home/celeritas/src ci-ubuntu-rocm6

After mounting, use the build scripts to configure and go:

celeritas@abcd1234:~$ cd src
celeritas@abcd1234:~/src$ ./scripts/docker/ci/run-ci.sh valgrind

Note that running as the root user requires the MPIEXEC_PREFLAGS=--allow-run-as-root to be defined for CMake: this is done by cmake-presets/ci-rocky-cuda.