[llvm] r219135 - [BasicAA] Revert "Revert r218714 - Make better use of zext and sign information."

[Hal Finkel]

----- Original Message -----
> From: "Nicholas White" <n.j.white at gmail.com>
> To: "Richard Barton" <richard.barton at arm.com>
> Cc: "Hal Finkel" <hfinkel at anl.gov>, "Commit Messages and Patches for LLVM" <llvm-commits at cs.uiuc.edu>, "Lang Hames"
> <lhames at gmail.com>, "James Molloy" <James.Molloy at arm.com>
> Sent: Monday, October 20, 2014 7:29:52 AM
> Subject: Re: [llvm] r219135 - [BasicAA] Revert "Revert r218714 - Make better use of zext and sign information."
> 
> 
> 
> I've added this test case and the fix to
> http://reviews.llvm.org/D5866 . Hal - I had to remove the
> allPositive loop condition you added as we should check every
> GEPVariable to see if it changes values across loop iterations
> (!isValueEqualInPotentialCycles). If any variables do change non of
> the analysis in that code block can be used, as it is only valid for
> the current loop iteration. Thanks -

I don't understand. If AllPositive becomes false, then we don't return NoAlias later; why does it matter if we do the remaining checks -- AllPositive will remain false once it has been set to false, right? 

 -Hal

> 
> Nick
> 

-- 
Hal Finkel
Assistant Computational Scientist
Leadership Computing Facility
Argonne National Laboratory