Support ROCm 6.0

[fwyzard]

ROCm 6.0 was release a couple of days ago: https://rocm.docs.amd.com/en/docs-6.0.0/about/release-notes.html .

[fwyzard]

Can we add ROCm 5.6.1, 5.7.1 and 6.0.0 to the CI ?

Do the CI job run on bare metal or in a container ?
If it's a container, what version of ROCm is installed on the host OS ? What kernel driver does it use ?

According to https://rocm.docs.amd.com/projects/install-on-linux/en/latest/reference/user-kernel-space-compat-matrix.html , ROCm is a lot less flexible than CUDA with respect to backward and forward compatibility.

[psychocoderHPC]

We build on own container build on top of the original ROCm container.
The reason is that we need pre-build some dependencies to reduce the CI time.

I opened an issue in the container repo that we need ROCm 6.

https://codebase.helmholtz.cloud/crp/alpaka-group-container/-/issues/36

[psychocoderHPC]

If it's a container, what version of ROCm is installed on the host OS ? What kernel driver does it use ?

That's hard to say because we do not maintain the host OS. We ask from time to time that the admins should update the driver.
I will do it next year, if I trigger this week an update it could end in a broken CI for the next weeks.

[SimeonEhrig]

The containers should be not a problem. We use it only for caching. If a container with a specific ROCm version is not available, the CI will use another container and install ROCm each time, if the test is started.

Actual you need only add the version numbers to the version.py and the CI will generate the test. In the case of ROCm, there is something special. You need to add the clang version of each SDK version here:

alpaka/script/job_generator/generate_job_yaml.py

Line 335 in 8d04265

if job[DEVICE_COMPILER][NAME] == HIPCC:

I will open a PR and add the tests. But if a test fails, I have not time to fix it at the moment.

[SimeonEhrig]

The PR is open: #2207