[llvm-commits] Patch to add support for WidenVecRes_MERGE_VALUES

Villmow, Micah Micah.Villmow at amd.com
Tue Aug 30 13:06:16 PDT 2011


Eli,
 So for the other cases, it would just be call DecomposeMERGE_VALUES and then call the correct
legalization function on the return value, correct?

I'll send a second patch with those later.

Micah

> -----Original Message-----
> From: Eli Friedman [mailto:eli.friedman at gmail.com]
> Sent: Tuesday, August 30, 2011 12:56 PM
> To: Tobias Grosser
> Cc: Villmow, Micah; llvm-commits
> Subject: Re: [llvm-commits] Patch to add support for
> WidenVecRes_MERGE_VALUES
> 
> On Tue, Aug 30, 2011 at 11:35 AM, Tobias Grosser <tobias at grosser.es>
> wrote:
> > On 08/30/2011 01:04 PM, Villmow, Micah wrote:
> >>
> >> Thanks for the feedback, new revision attached.
> >
> > There seems one unneeded white space change. (Can be fixed on commit)
> >
> >> SDValue MakeLibCall(RTLIB::Libcall LC, EVT RetVT,
> >>                    const SDValue *Ops, unsigned NumOps, bool
> isSigned,
> >>                    DebugLoc dl);
> >> +
> >>        std::pair<SDValue, SDValue>
> ExpandChainLibCall(RTLIB::Libcall
> >
> > Otherwise, it looks good. Let's see if Eli has any additional
> comments.
> 
> Looks fine.
> 
> Note that it would be nice to make sure we implement MERGE_VALUES for
> all the other cases in type legalization... (integer promotion, vector
> scalarization, etc.)
> 
> -Eli






More information about the llvm-commits mailing list