juliagpu / gitlab-ci Goto Github PK

The Julia tarball and depot are stored in the source repo to make it possible to cache them, but if you're working with Git in the job they can get in the way a bit (for example SciML/SciMLTutorials.jl@6e5a466). We can help by creating a gitignore file:

.julia/
downloads/

And setting it as the global gitignore:

git config --global core.excludesfile /tmp/gitignore

With CI resources being fairly limited, we should configure test jobs to get canceled if it's redundant: https://docs.gitlab.com/ee/ci/yaml/#interruptible

Similar to Travis's julia: 1, it would be nice to set and forget, while still always getting the latest stable Julia. It would be simple enough to hardcode the most recent release in this repo and update it every few months rather than hunting down all the repos still using .julia:1.4 6 months from now.

We now cache very selectively, but I wonder if it wouldn't be better to cache all of .julia, speeding up registry check-out and package installation. Or would that hurt reproducibility?

That's what I get from here:

How can I force it to submit to codecov.io?

First of all, thank you very much for your effort to set up testing via gitlab!

I would like to add Gitlab CI to CLBlast.jl, see https://github.com/JuliaGPU/CLBlast.jl/blob/gitlab/.gitlab-ci.yml. However, my attempts fail since OpenCL.jl is unable to create_some_context():

ERROR: LoadError: CLError(code=-1001, CL_PLATFORM_NOT_FOUND_KHR)
Stacktrace:
 [1] macro expansion at /builds/JuliaGPU/CLBlast.jl/.julia/packages/OpenCL/aAtsW/src/macros.jl:6 [inlined]
 [2] platforms() at /builds/JuliaGPU/CLBlast.jl/.julia/packages/OpenCL/aAtsW/src/platform.jl:21
 [3] create_some_context() at /builds/JuliaGPU/CLBlast.jl/.julia/packages/OpenCL/aAtsW/src/context.jl:287
 [4] create_compute_context() at /builds/JuliaGPU/CLBlast.jl/.julia/packages/OpenCL/aAtsW/src/util.jl:2
 [5] top-level scope at none:0
 [6] include at ./boot.jl:317 [inlined]
 [7] include_relative(::Module, ::String) at ./loading.jl:1038
 [8] include(::Module, ::String) at ./sysimg.jl:29
 [9] include(::String) at ./client.jl:398
 [10] top-level scope at none:0
in expression starting at /builds/JuliaGPU/CLBlast.jl/test/runtests.jl:19

The pipeline can be found here: https://gitlab.com/JuliaGPU/CLBlast.jl/pipelines/28956178. I have used the templates from this repository and https://github.com/JuliaGPU/CLArrays.jl/blob/master/.gitlab-ci.yml for tests using Julia v0.6.

Could you please help me with setting up gitlab correctly?

For the AMDGPU stack, it would be really helpful to have a few extra CI runs on Julia with LLVM 7 and 8 (and eventually future versions of LLVM).

Currently, there is an error that display :0 can't be opened!

Hi guys, I'm looking for a CI for GPU. I am wondering if I can use your CI for testing? I'm developing CUDA support for QuantumBFS/Yao.jl#114

We could use the podman one, which is supposed to be a drop-in replacement for Docker: https://github.com/jonasbb/podman-gitlab-runner

Using custom executors can have quite some advantages:

• select a suitable GPU (with --gpus instead of the runtime hack) based on tags (multigpu, big-gpu, recent-gpu), memory usage, or other policies (entire gpu for a job, for example)
• cache artifacts without zipping (currently accounting for significant run time)
• set JULIA_NUM_THREADS per-runner
• etc

Is it time to move to the official v1.3?