[PATCH] D63604: [Attributor] Deduce "nonnull" attribute

[Johannes Doerfert via Phabricator via llvm-commits]

jdoerfert added a comment.

In D63604#1578887 <https://reviews.llvm.org/D63604#1578887>, @xbolva00 wrote:

> In D63604#1578875 <https://reviews.llvm.org/D63604#1578875>, @jdoerfert wrote:
>
> > In D63604#1578854 <https://reviews.llvm.org/D63604#1578854>, @xbolva00 wrote:
> >
> > > Maybe you remember https://reviews.llvm.org/D53342 ? Should we do it “InstCombine”-way or some alternative one ?
> >
> >
> > I did not, thanks for reminding me. What do you mean by the "InstCombine"-way and what alternatives do you have in mind?
> >  Sorry that I'm a bit lost at the moment, I looked at D53342 <https://reviews.llvm.org/D53342> and I seemed there are still problems with that patch even though the idea is very useful and we should add it.
>
>
> I want to ask, whether it is okay to set/drop nonnull attribute in InstCombine or you would like to see this setting/dropping in Attributor.
>
> If you are OK with “InstCombine”-way, I can rebase/update my patch.


I think placing attributes for known library functions should not be part of the Attributor. So, rebase & update your patch and I'll review it.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D63604/new/

https://reviews.llvm.org/D63604