[PATCH] D76000: AMDGPU: Don't hard error on LDS globals in functions

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

In D76000#1918259 <https://reviews.llvm.org/D76000#1918259>, @yaxunl wrote:

> In D76000#1918031 <https://reviews.llvm.org/D76000#1918031>, @arsenm wrote:
>
> > In D76000#1917991 <https://reviews.llvm.org/D76000#1917991>, @yaxunl wrote:
> >
> > > So we will never be able to debug a HIP program compiled with -O0 where a device function uses shared variable since all functions are noinilne. This seems quite a restriction.
> >
> >
> > I'm strongly opposed to clang using noinline at -O0. This should be stopped
>
>
> I will see if I can change that, at least for HIP.


The thread the pulls on is that optnone artificially requires noinline, another constraint I have a problem with


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D76000/new/

https://reviews.llvm.org/D76000