[clang] [NFCI][clang] Allow overriding any global variable address space (PR #195612)

[Sameer Sahasrabuddhe via cfe-commits]

ssahasra wrote:

> > Until then, everything is just a hack. As long as we agree on it being a hack, we can always move forward with lots of TODO and FIXME comments in place.
> 
> IMHO the named barrier type is already a hack in itself. It has been exposed to the language w/o much documentation nor concerns for how it feels like to use it and how it interacts with other language features.
> 
> I think this is why I got a bit carried away with my earlier comment. The fundamental problem is that the named barrier type was upstreamed with minimal review, so it has lots of issues, and your and other reviewers comments are entirely fair, but to me it felt a bit like I accidentally signed up for fixing all issues about this type without noticing.

That's understandable. I am okay with the current incremental improvement. I am adding my approval just to be clear that I am no longer blocked on the nature of this change, but please do make a judgement call on whether to wait for other reviewers.

https://github.com/llvm/llvm-project/pull/195612