[PATCH] D134704: [AAPointerInfo] Distinguish initial from unknown OffsetInfo

Johannes Doerfert via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Tue Sep 27 15:46:27 PDT 2022


jdoerfert added a comment.

In D134704#3818642 <https://reviews.llvm.org/D134704#3818642>, @sameerds wrote:

> In D134704#3818032 <https://reviews.llvm.org/D134704#3818032>, @jdoerfert wrote:
>
>> Some minor nits inlined. Also, (1) can we have a pure Attributor test for this?
>
> I'll take care of the rest, but some advise on a pure Attributor test would be useful. I did take a look at tests that were affected by previous changes to the handling of PHI nodes. But couldn't really follow how to eventually surface the effect of PointerInfo in terms of function attributes or load/store optimizations in a lit test.

How did we miscompile the AMDGPU test? I assume we thought something holds about the `%.in` load, right?  I would assume we can make that property we assumed wrongly cause us to misoptimize, no?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D134704/new/

https://reviews.llvm.org/D134704



More information about the llvm-commits mailing list