[llvm] cfd594f - [SROA] `isVectorPromotionViable()`: memory intrinsics operate on vectors of bytes (take 3)

[Roman Lebedev via llvm-commits]

On Thu, Jan 5, 2023 at 10:54 PM Tom Stellard <tstellar at redhat.com> wrote:
>
> On 1/5/23 11:50, Roman Lebedev wrote:
> > As of this point in time, the patch is not in relanded state.
> > Roman
> >
>
> I'm not sure exactly what this means.  Was the patch reverted?
That feedback was posted after the patch was reverted the last time.
The patch has not been recommitted since then.

> -Tom
Roman

> > On Thu, Jan 5, 2023 at 10:45 PM Tom Stellard <tstellar at redhat.com> wrote:
> >>
> >> On 12/20/22 06:15, Jeremy Morse via llvm-commits wrote:
> >>> Hi Roman,
> >>>
> >>> In addition to what Nikita mentioned, before reapplying could you take
> >>> a closer look at the debug-info tests you've updated: with this patch
> >>> as-is, they no longer cover the code paths they are intended for. For
> >>> example, memcpy.ll has the memcpys being tested completely elided, but
> >>> it looks like adding some nondeterministic stores followed by a
> >>> volatile load keeps the intended behaviour (SROA splitting the alloca
> >>> into two parts). Something similar happens in store.ll, and the others
> >>> where the number of CHECK lines have reduced.
> >>>
> >>> Without these tests we can't detect debug-info regressions in the future.
> >>>
> >>
> >> Have these issues been fixed in trunk either by reverting cfd594f
> >> or with a follow up patch?
> >>
> >> -Tom
> >>
> >>> --
> >>> Thanks,
> >>> Jeremy
> >>> _______________________________________________
> >>> llvm-commits mailing list
> >>> llvm-commits at lists.llvm.org
> >>> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
> >>
> >
>