AlexVlx wrote: > > We don't use `offload` at the moment, that's for HIPSPV. Of course, future, > > fancy work is more than welcome, but this merely slots into the existing > > infra and current use cases. > > I'm not a huge fan of smuggling what is essentially a Toolchain behind what > was intended as a `-mcpu` option. I guess we already somewhat support that? I > know @yaxunl was the one that made the original `--offload=` RFC / support > but I don't think it ever got 100% finished. I'm just a little concerned that > if we support it this way then we'll never be able to change it because the > argument will be that we need it for backwards compatibility. But I guess > that's already the case and this patch just simplifies it?
It's how it works today, I believe: `--offload-arch` unambiguously establishes the toolchain, and we don't have a separate toolchain for AMDGCN flavoured SPIR-V (hence the rather sneaky triple based manipulation). I don't disagree with the principle, but I think that tidying this up and, possibly, allowing things like multiple triples within a single toolchain in a less sneaky way would be a pretty big lift. On the other hand, we cannot just switch to an `--offload` based flow either, because we have lots of client apps that rely on `--offload-arch` and whose build infra incorporates it. https://github.com/llvm/llvm-project/pull/113509 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
