[Clang] Convergent Attribute

Ettore Speziale via cfe-commits cfe-commits at lists.llvm.org
Tue May 17 08:22:03 PDT 2016


>>> CUDA? In any case, I don't see how the restriction helps users, and the attribute at the IR level has a well-defined meaning regardless. If a user were to have a use case, they'd simply find the restriction arbitrary and frustrating.
>> Yes, CUDA was already considered as well. I just think that compilers should help to reduce amount of erroneous or meaningless use cases. That's one of the reasons to have language options for the attributes. But I don't feel strongly about this particular case anyways. So let's make it language independent then. ;)
> Is the patch OK now or you guys want to apply some other modifications?
> <convergent.diff>

Sorry guys, but on a second tough this patch might be completely useless. I guess I’d should be marking all function calls convergent at CodeGen time, and then rely on LLMV’s FunctionAttrs to remove the unnecessary ones.

Thank you for the good review and sorry for wasting your times,
Ettore Speziale

Ettore Speziale — Compiler Engineer
speziale.ettore at gmail.com
espeziale at apple.com

More information about the cfe-commits mailing list