Artem-B wrote: @cgmb
> I would suggest that we should either (a) change the default GPU target to > native and make the failure to detect the user’s GPU into a hard compiler > error, or (b) change the default GPU target to SPIR-V so that it works on > every machine. The thing is that the sensible GPU target default does not exist for the GPUs in principle, IMO. We're just arguing about the least terrible variant to choose. * any fixed GPU version -- wrong for most users. It kind of forces everyone to explicitly specify what they need. Or they presumably don't care. * `native` -- correct for many users doing local builds, wrong for the production builds that are often done on machines w/o GPU, or with some GPU that's irrelevant for the build. * generic one-target-suits-all. Kind of functional, but likely a suboptimal choice for everyone. I am personally biased towards the first variant, as it's consistently and predictably bad for everyone. Other options tend to create an illusion of "just working", but are prone to creating unwanted surprises. spir-v would probably be an OK alternative. It's not great (probably leases some performance on the table vs the native build), but the binary will presumably work on other machines. https://github.com/llvm/llvm-project/pull/139281 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
