[new-ra] DF_REF_REG_CLOBBER related bugfix

[Michael Matz]

Hi,

On 24 Oct 2003, Denis Chertykov wrote:

> I misunderstand you:
>
> > I think the condition for coalescing them there simply should be, that
> > w2's color is actually acceptable by w1->usable_regs.
>
> Why `w2's color is actually acceptable by w1->usable_regs' ?
>
> w1->usable_regs is empty. w1->usable_regs can't accept any color.

Yes, and that would then prevent the coalescing.  You want to use the
emptyness of conflict_list, I want to use the fact that the usable_regs
don't overlap at all to prevent this.  At least, that's how I understood
you: you have to prevent coalescing of spill slots for webs where one has
no usable_regs, because the conflict information isn't filled out
otherwise.  But it's not just not filled out when usable_regs is empty,
but also if both usable_regs don't intersect, so that's what should be
tested.

> > (for instance it could happen in the future, that webs have
> > usable_regs empty, but nevertheless have a conflict_list).
>
> Are you want to say that conflict_list must be even for webs with
> empty usable_regs *OR* that webs with empty usable_regs can have a
> conflict_list ?

The latter.  It doesn't happen right now.  But I can imagine that we
somewhen create the conflicts, and afterwards limit usable_regs in a way
which makes it empty.  I.e. that conflict_list is not empty doesn't mean,
that usable_regs is also not empty.  (huh, many negations in those
sentences ;-) )


Ciao,
Michael.