Re: [RTL, ColdFire 24/63] Add support for a MODE_INDEX_REG_CLASS macro

[Jeffrey Law <law at redhat dot com>]

On Wed, 2007-01-17 at 05:56 -0800, Ian Lance Taylor wrote:
> Jeffrey Law <law@redhat.com> writes:
> 
> > On Tue, 2007-01-16 at 22:07 -0800, Ian Lance Taylor wrote:
> > > Richard Sandiford <richard@codesourcery.com> writes:
> > > 
> > > > I'd always been told that L_R_A should
> > > > only be used for optimisation, and that it was wrong to rely on it for
> > > > correctness.
> > > 
> > > I want to comment that I've heard people say that too, but I think it
> > > is, well, over optimistic.  When weird addressing modes are in play, I
> > > believe that LEGITIMIZE_RELOAD_ADDRESS is currently required to handle
> > > them.  The alternative is to teach reload about all possibilities for
> > > all chips, and that does not make sense to me.
> > As the original submitter of L_R_A, I can confirm that the original
> > intent (and it was thoroughly discussed) was for L_R_A to be purely
> > an optimization and any port which requires definition of L_R_A was
> > considered broken.
> 
> I know, but I think that it doesn't actually work that way.  There are
> too many weird addressing modes these days.
I'm not seeing anything in modern chips that we weren't already
dealing with in one way or another when I introduced L_R_A.

Jeff