Hi Christian,
For this, we (the Debian ROCm Team) should be absolutely be providing a Makefile that could be included into d/rules, analogous to /usr/share/dpkg/*.mk, that exports something like ROCM_SUPPORTED_GPUS and/or ROCM_<release>_SUPPORTED_GPUS or whatever. This does not just simplify other maintainer's lives, this allows us to change this value Archive-wide without having to change N source packages with hardcoded values; we just update the package with the main definition, and trigger a binNMU.
Perhaps DEB_HOST_HIP_ARCHITECTURES? That's a mashup between
Debian's naming conventions and the CMAKE_HIP_ARCHITECTURES
variable [1]. IDK. Naming things is hard.
In any case, this might also be good for GitLab runners. The ROCm packages take a long time to build, in part because they must be built for many GPU architectures in serial. It might be good to have a mechanism to reduce the target architecture set to a single GPU for running a quick check on the package.
AMD has asked if Debian could use any GitLab Runners with GPUs. I
think those would be useful—particularly if the runners had server
GPUs rarely found in packager workstations—but I think we'd need
to optimize our usage before taking them up on that.
Sincerely,
Cory Bloor
[1]:
https://cmake.org/cmake/help/latest/variable/CMAKE_HIP_ARCHITECTURES.html