[RFA];

[Richard Earnshaw]

> Hi Richard,
> 
> > I believe this is wrong to the extent that you can't use
> > HARD_REGNO_NREGS on a pseudo.
> 
> *sigh* Yes.
>  
> > Funnily enough, I fixed exactly the same problem myself yesterday by 
> > breaking out most of the code to a new function (in much the same way that 
> > mark_set_1 does).  Here's my patch:
> 
> Great minds think (almost) alike.
> 
> OK I withdraw my patch in favor of yours.
> 
> Cheers
>         Nick
> 

Nick,

Do you have a testcase which shows up this failure, particularly on the 
branch -- if so, this is likely to constitute a regression that needs 
urgent fixing.  (I found it while working on some new back-end code, so 
can't use that as justification).

R.