Mikael Simberg
Mikael Simberg
Exporting `NVIDIA_VISIBLE_DEVICES=all` and `NVIDIA_DRIVER_CAPABILITIES="compute,utility"` seems to be what was required to get the container images to load the correct drivers etc. and avoid ``` cudaErrorInsufficientDriver (CUDA driver version is insufficient...
All right, we're making some progress: - The GCC 12/CUDA 12 pipeline is working now. - The clang/cuda one isn't working because it uses valgrind and valgrind doesn't seem to...
The clang/cuda configuration with valgrind no longer complains about illegal instructions: good. It now reports many issues, which I don't know yet if they're real or not. I'll aim to...
This is now ready for merging for my perspective. I've updated the PR description to outline the changes.
I think we might need longer timeouts on the new valgrind CI configuration. Will try adjusting that.
Merging manually. The attempt at merging through the merge queue failed for unrelated reasons.
Ping? Any comments on this?
Hi, as before we recommend using spack for the build, especially when targeting GPUs. In your case you'll probably want to have a spec in the direction of `dla-future +rocm...
It may be worth it to consider sending results to bencher instead (icon4py has been testing it in https://github.com/C2SM/icon4py/pull/669), but it doesn't exclude reviving this performance testing.
I just merged #1260. Could you rebase this?