Hi Paul, On 2025-02-15 10:00, Paul Gevers wrote: > Sorry for not replying earlier, we (at least I) received the original > mail too, but ENOENERGY.
thank you for your reply and sorry if I created pressure, it was not my intention. It was just rather untypical to not receive a reply and I've seen mails to team-maintained <package>@p.d.o get "lost" for other packages, so I assumed that this happened here, too. > We're having related questions inside autopkgtest/debci too: https:// > salsa.debian.org/ci-team/debci/-/merge_requests/285#note_583193 Thank you for pointing me to this, I added my feedback. > We should have an answer, but currently I don't. Helmut (in CC) > suggested to discuss this in real life in Hamburg. I think that might be > easiest, but that means delaying an answer. I'm happy to contribute in any way if I can be useful. Since there is no immediate clear solution, I hope the following would be not be objectionable to the maintainers: I leave the ROCm backends as-is for now, and I'll adjust pkg-rocm-tools to whatever consensus emerges eventually, also in light of the incus issue above. The reason is simple: the backends are really needed for any developer wishing to add autopkgtests to packages making use of AMD GPUs. This has come up again and again, and (I expect) this will only increase the more AI/ML work enters Debian. The performance differences to CPUs can be substantial, see [1] for example. Best, Christian [1]: https://lists.debian.org/debian-ai/2025/01/msg00136.html
