https://github.com/bader commented:

> I'll emphasise that this is only a problem for things such as implicitly 
> generated globals (e.g. VTables or typeinfo for classes etc.)

I suppose usage of VTables and typeinfo is kind of restricted in GPU 
programming models. Right?

Anyway, I think it's a reasonable change for SPIR target, which fixes mapping 
from `global` address space to `1` LLVM address space.
It's less obvious for SPIR-V, which is not LLVM-based format and mapping 
between SPIR-V and LLVM addresses spaces is technically an implementation 
detail, which can be adjusted.
According to my understanding, today LLVM IR for SPIR-V target just follows 
SPIR target nomenclature.

I'd like Eli/John to give formal approval for this change, but I'm okay with 
it. 👍 Thanks!

https://github.com/llvm/llvm-project/pull/88455
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to