[patch] Fix pr14893

[Hal Finkel]

----- Original Message -----
> From: "Rafael Espíndola" <rafael.espindola at gmail.com>
> To: "llvm-commits" <llvm-commits at cs.uiuc.edu>
> Sent: Sunday, January 26, 2014 9:33:33 PM
> Subject: [patch] Fix pr14893
> 
> The root problem is that we were speculating a load and not dropping
> its range metadata. We know it is safe to speculate, but once we do
> we
> cannot say anything about the value we get out of the load.
> 
> I noticed that there is not a convenient facility to drop all
> non-debug metadata. Is that intentional? From the old discussions, if
> I remember correctly, passes should drop metadata they know nothing
> about. Is that still the case or have we moved to a model where
> metadata is a proper part of the IR and passes have to know about it?

Passes are still responsible for dropping it. I agree, however, that we need more-convenient facilities for this.

 -Hal

> 
> Cheers,
> Rafael
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
> 

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