[PATCH] D67332: [InstSimplify] simplifyUnsignedRangeCheck(): if we know that X != 0, handle more cases (PR43246)

[Roman Lebedev via Phabricator via llvm-commits]

lebedev.ri added a comment.

In D67332#1662491 <https://reviews.llvm.org/D67332#1662491>, @lebedev.ri wrote:

> In D67332#1662486 <https://reviews.llvm.org/D67332#1662486>, @spatel wrote:
>
> > Sorry - I just looked over the comments in:
> >  https://bugs.llvm.org/show_bug.cgi?id=43246
> >
> > Does this allow removing a more specific (offset is constant) blob of code from instcombine?
>
>
> Can you please point me to it?
>  I suspect we get it via multiple folds that happen because
>  those instructions have no extra uses and constant on RHS.




  declare void @use64(i64)
  declare void @use1(i1)
  define i1 @test(i64 %offset) {
    %1 = add i64 %offset, 100
    call void @use64(i64 %1)
    %2 = icmp ne i64 %1, 0
    %3 = icmp ugt i64 %1, 99
    %r = and i1 %2, %3
    ret i1 %r
  }

gets folded by

  // X < Y && Y != 0  -->  X < Y
  // X < Y || Y != 0  -->  Y != 0
  if (UnsignedPred == ICmpInst::ICMP_ULT && EqPred == ICmpInst::ICMP_NE)
    return IsAnd ? UnsignedICmp : ZeroICmp;

in `simplifyUnsignedRangeCheck()` i don't know if there is anything to delete.
Proceeding to lading the patch.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D67332/new/

https://reviews.llvm.org/D67332