[llvm-dev] Potentially unsafe loop optimization

[Richard Kenner via llvm-dev]

> Had the bcmp not been there, this late run of InstructionSimplify
> wouldn't have happened. There are no other optimizations that always
> run in this part of the pipeline that would see the bad IR created
> from loop strength reduce and optimize based on it. SelectionDAG can
> do optimizations with nuw, but it only runs on a single basic block
> so it won't see the phi.

OK, I understand now.

> Nikita's patch, 835104a1141a06ae7821fe2b642b9603e00aa17b removes the
> nuw from the add on trunk.

That sounds right given that it moved the add before the branch.

> I didn't look at why llvm 10 works.

Probably because it didn't do the late run or it didn't spot the phi
would be my guess.