[PATCH] D98422: [Alias] Add a ah-hoc pattern with two PHI for isKnownNonEqual

[JinGu Kang via Phabricator via llvm-commits]

jaykang10 added a comment.

In D98422#2633203 <https://reviews.llvm.org/D98422#2633203>, @jaykang10 wrote:

> I think we could check the IV of loop roughly as below.
>
>    if (Q.DT != nullptr && Q.DT->dominates(PN1->getParent(), Pred)) {
>      const SCEV *IVSCEV = nullptr;
>      if (PN1 == IV2) 
>        IVSCEV = SE.getSCEV(IV2);
>      else if (PN2 == IV1) {
>        IVSCEV = SE.getSCEV(IV1);
>   
>      if (isa<SCEVAddRecExpr>(IVSCEV))
>        continue;
>   }
>
> In this case, we could say the IV is not same among iterations... but it needs ScalarEvolution... @nikic @reames How do you think about it? I wonder whether it is acceptable to add ScalarEvolution to BasicAA and ValueTracking or not...

Any comments for SCEV in BasicAA?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D98422/new/

https://reviews.llvm.org/D98422