Re: ARM: minor DImode logical improvements

[Richard Earnshaw <rearnsha at arm dot com>]

> 
>   In message <199910221017.LAA04033@pathia.cygnus.co.uk>you write:
>   > Hi Richard,
>   > 
>   > : It looks fine to me.  However, could we hold off on this until we
>   > : get the new ARM/Thumb back-end merged in; additions like this are
>   > : good, but will make the merge more complicated unless they are
>   > : applied to both.
>   > 
>   > Actually we are applying the changes to both back ends.
> Yup.  If you want a change for the newer backend, it's basically the same,
> except that the splitter are conditional on TARGET_ARM just like the
> DImode patterns.  I can gen that up pretty quickly once we're happy with
> the basic patch.

Ok, I withdraw my objection :-)

> 
>   > I am willing to do this if you think that the new back end is not
>   > ready for the mainstream yet.  Ideally I would like to submit the new
>   > back end right away and archive off the old back end, so that we do
>   > not have any more dual maintainance to contend with.
> I'll leave this decision to y'all.  Making a branch is fine by me if that's
> what you want to do.

I do want to get the new code installed into cvs -- it makes maintenance 
much easier.  However, I'm not sure I want the old back-end to go away 
until I've had a better look at the quality of the code generated -- at 
present ARM code seems to be marginally worse with the new back-end;  I'd 
been hoping to do this over the last couple of weeks, but was held back by 
the fact that the compiler wouldn't even bootstrap during much of that 
period.

R.