[PATCH] Move CodeGenPrepare into lib/CodeGen

Tom Stellard tom at stellard.net
Thu Feb 20 06:45:10 PST 2014


On Wed, Feb 19, 2014 at 10:56:58PM -0600, Hal Finkel wrote:
> ----- Original Message -----
> > From: "Quentin Colombet" <qcolombet at apple.com>
> > To: "Lang Hames" <lhames at gmail.com>
> > Cc: "Commit Messages and Patches for LLVM" <llvm-commits at cs.uiuc.edu>
> > Sent: Wednesday, February 19, 2014 10:50:50 PM
> > Subject: Re: [PATCH] Move CodeGenPrepare into lib/CodeGen
> > 
> > 
> > 
> > 
> > 
> > Envoyé de mon iPhone
> > 
> > Le 19 févr. 2014 à 20:00, Lang Hames < lhames at gmail.com > a écrit :
> > 
> > 
> > 
> > 
> > Hey All,
> > 
> > 
> > 
> > To keep going into that direction, I'd love a tool (opt or other)
> > that would
> > 
> > 
> > handle IR to IR, IR to MIR (format not defined), and MIR to MIR. My
> > point is
> > 
> > 
> > making the distinction between code gen-ish passes and transform
> > passes does
> > 
> > 
> > not make much sense to me.
> > 
> > Do you mean a single tool? Yes, ultimately :).
> > 
> > 
> > 
> > 
> > I don't have any objection to that, but the
> > new tool would still have to solve the problem of invoking single IR
> > -> IR passes without running any MIR passes, which seems like the
> > same
> > problem we're now facing in llc?
> > Okay, I think I miss something.
> > What is the problem we are facing with IR to IR passes in llc?
> > Couldn't we just run them into opt?
> > 
> > 
> > 
> > 
> > If that's true, then I'd be inclined
> > to tackle it within llc for now (borrowing ideas/infrastructure from
> > opt where appropriate). Later, if we decide it's not worth keeping
> > the
> > tools separate, we could merge llc and opt.
> > I am missing something here too. Is Running llc on just specific
> > passes supposed to work?
> > 
> > 
> > It seems like I completely miss the underlying problem.
> > The goal of this patch was to solve a dependency issue between two
> > libraries.
> > If we want more, we need to go back to the discussion without any
> > patch in mind.
> 
> I agree, I don't see why we're conflating all of these different things. Can we not move the pass into CodeGen, and then deal with adding IR-pass-driving features into llc as a follow-up matter?
> 

I agree too, I really don't want any llc modifications or redesigns to hold
up this patch.  It is a prerequisite for a regression fix in the R600
backend.

-Tom

>  -Hal
> 
> > 
> > 
> > Thanks,
> > -Quentin
> > 
> > 
> > 
> > 
> > - Lang.
> > 
> > _______________________________________________
> > 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
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits




More information about the llvm-commits mailing list