Re: [patch] m68k: Use ASSEMBLER_DIALECT

[Kazu Hirata <kazu at codesourcery dot com>]

Hi Roman,

> > I assigned option 0 to the Motorola syntax because the motorola syntax
> > adds characters to the MIT syntax.  This way, we can say
> >
> >   j{b}eq
>
> Some time ago I posted a patch, which replaced all of these simply with 
> jeq, but Jeff had some objections, which I still don't understand. The 
> point here is that this has nothing to do with the official MOTOROLA 
> syntax - both variants jeq and jbeq are assembler pseudo ops and the only 
> supported assembler left is gas, so it would be far simpler and easier to 
> support to clean this up properly and just use one of these pseudo ops.

OK.  I'd like to take the path of least resistance.  If Jeff objects to 
removing 'b' from "jbeq", I'd keep 'b', which my patch does.

> >    else
> >      {
> >        if (store_p)
> > -	return MOTOROLA ? "movm.l %1,%a0" : "moveml %1,%a0";
> > +	return "mov{m.|em}l %1,%a0";
> >        else
> > -	return MOTOROLA ? "movm.l %a0,%1" : "moveml %a0,%1";
> > +	return "mov{m.|em}l %a0,%1";
> >      }
>
> Here I would suggest to just replace it with movem%.l, movm is not 
> MOTOROLA syntax.

OK.

> > -	return MOTOROLA ? "lea (%1,%2.l),%0" : "lea %1@(0,%2:l),%0";
> > -      return MOTOROLA ? "lea (%c2,%1),%0" : "lea %1@(%c2),%0";
> > +	return "lea {(%1,%2.l)|%1@(0,%2:l)},%0";
> > +      return "lea {(%c2,%1)|%1@(%c2)},%0";
> >      }
> >    if (GET_CODE (operands[2]) == CONST_INT)
> >      {
>
> These things should be cleaned up in the long term by splitting them and 
> using a different pattern, so it's a simple "lea %1,%0" at some point.

Yeah, I'd like to see address expressions handled in a single place so that 
we don't have to bother syntax issues everywhere.  Maybe I'll get to that 
after this patch.

Kazu Hirata