[cfe-commits] [PATCH] Asm multiple alternative constraints revisited

[John Thompson]

Eli,

So it seems you are saying that the back-end needs to fixed to choose the
multiple constraints in parallel before you will let the front end fix
through.

Here's a revised patch that cuts off the constraint at the first comma, done
in SimplifyConstraint.  I'll look into the back end fix, bouncing it first
off the llvm list as you mentioned previously.

May I check this in?

-John


On Mon, Aug 9, 2010 at 3:50 PM, Eli Friedman <eli.friedman at gmail.com> wrote:

> On Mon, Aug 9, 2010 at 3:02 PM, John Thompson
> <john.thompson.jtsoftware at gmail.com> wrote:
> > May I have permission to check this in?
>
> The stopgap should really be to cut off the constraints just before
> the first comma rather than ignore the comma... otherwise, it would be
> possible to end up with very strange errors in the assembler with more
> complicated multi-alternative constraint usages.  Would that be much
> more difficult?
>
> -Eli
>
> > On Fri, Aug 6, 2010 at 9:28 AM, John Thompson
> > <john.thompson.jtsoftware at gmail.com> wrote:
> >>
> >> Thanks, Eli.
> >>
> >> Then do I have permission to check in the front end changes, leaving the
> >> back end as-is for now, treating any further work there as a separate
> issue
> >> to be discussed first on llvmdev?
> >>
> >> Basically, without these front-end changes, there will be an assert in
> the
> >> front end code gen (-emit-llvm) if multi-alt constraints are seen,
> because
> >> the extra commas are not expected.
> >>
> >> -John
> >>
> >> On Thu, Aug 5, 2010 at 8:14 PM, Eli Friedman <eli.friedman at gmail.com>
> >> wrote:
> >>>
> >>> On Thu, Aug 5, 2010 at 7:54 PM, John Thompson
> >>> <john.thompson.jtsoftware at gmail.com> wrote:
> >>> > Eli,
> >>> >
> >>> > So, then this is a problem with the back-end, right?  Do you agree
> that
> >>> > the
> >>> > back end is a better place to try to choose constraints, given that
> it
> >>> > knows
> >>> > better which things are already in registers and so forth?
> >>>
> >>> Yes, it would fit more naturally there... but I'm not sure a proposal
> >>> to make inline asm in the backend more complicated will be greeted
> >>> with much enthusiasm.  If you're really interested in pursuing that
> >>> path, though, ask on llvmdev first.
> >>>
> >>> -Eli
> >>
> >>
> >>
> >> --
> >> John Thompson
> >> John.Thompson.JTSoftware at gmail.com
> >>
> >
> >
> >
> > --
> > John Thompson
> > John.Thompson.JTSoftware at gmail.com
> >
> >
> > _______________________________________________
> > cfe-commits mailing list
> > cfe-commits at cs.uiuc.edu
> > http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
> >
> >
>



-- 
John Thompson
John.Thompson.JTSoftware at gmail.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20100810/804921e2/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: multaltconstraints3.patch
Type: application/octet-stream
Size: 3918 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20100810/804921e2/attachment.obj>