[PATCH] D118376: [x86] try harder to scalarize a vector load with extracted integer op uses

Thu Feb 10 14:33:46 PST 2022

spatel added a comment.

In D118376#3310203 <https://reviews.llvm.org/D118376#3310203>, @dyung wrote:

> I was able to reproduce the issue with a small testcase that I have filed as issue 53695 <https://github.com/llvm/llvm-project/issues/53695>. Hope that helps.

That was very helpful - thanks! 
If I'm seeing it correctly, that example shows that the bug fix for this patch requires using 'DAG.makeEquivalentMemoryOrdering(OriginalLoad, Load)'.
But the test example can also be modified slightly to expose a bug in the DAGCombiner code that this patch copied from. Ie, there's a visible miscompile even with this patch theoretically fixed.
I haven't been able to come up with a test that exposes the bug independently of applying this patch yet though.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D118376/new/

https://reviews.llvm.org/D118376