jhuber6 wrote:

> maybe with `__has_builtin` instead of `__nvvm_reflect`?

That would have the effect of disabling it. The nvvm_reflect thing is a big 
hack and there's been a lot of discussion around it in the past, but basically 
it turns a compile time check into a runtime one, so one IR file can 'do the 
optimal thing' on a sufficiently new GPU. AMDGCN recently added some 
`__builtin_amdgcn_is_invocable` or something which does something similar, but 
it's only slightly less hacky.

https://github.com/llvm/llvm-project/pull/198815
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to