Hi Gaard, On 2025-05-05 05:13, Gard Spreemann wrote:
Tangentially: Happy to help with adding such tests for src:clblast. I've wanted to do so in the past, but I was unsure about how to deal with the limited availability of rocm-supporting hardware on the CI infrastructure.
If you mean dealing with the possibility that the autopkgtests run on DebCI nodes that do not have ROCm-supporting GPUs, then I would suggest launching your test with the rocm-test-launcher from pkg-rocm-tools. This will return a skip status if AMD hardware is not available, ensuring that the test does not cause any problems on the official DebCI. We can then add src:clblast to the scheduler for ci.rocm.debian.net, where the nodes all have various sorts of AMD GPUs and the tests will run normally. The hipsolver package is a good example of this [1].
If you mean that you're worried that the src:clblast test suite will overwhelm the ROCm continuous integration hardware that we have available, then I wouldn't worry about that yet. We're scheduling new jobs quite aggressively and we could adjust that to reduce the load. We could probably also acquire new hardware, if needed.
Sincerely, Cory Bloor[1]: https://salsa.debian.org/rocm-team/hipsolver/-/blob/debian/5.5.1-8/debian/tests/control?ref_type=tags