Re: [RFC] LRA: Carrying hard registers around

On Wed, Jun 24, 2026 at 7:34 PM Michael Matz via Gcc <[email protected]> wrote:
>
> Hello,
>
> On Wed, 24 Jun 2026, Stefan Schulze Frielinghaus wrote:
>
> > > > Due to register asm, hard register r18 is live across the division.
> > >
> > > And that's the bogosity.  The register variable r18 can live in a
> > > pseudo during that division, and _should_ live in a pseudo outside of
> > > the specific setup/teardown sequences of the two inline asms.
> >
> > This is not how register asm is implemented in GCC.  During expand
> > assignments involving register asm reduce to simple hard register
> > assignments (that is the whole point why I came up with [1] which is
> > supposed to fix at least this).  There are no pseudos involved.
>
> That's the problem then.  Register vars should expand to pseudos, and as
> you can see in this thread I was pretty sure that its done that way after
> all the years and something just broke that.  Also see the bug Alexander
> referenced, where also Segher agreed.  Consider me baffled and surprised
> that not more breaks left and right :-)

They are also "funny" on the GIMPLE side with workarounds not to
break them.

Richard.

>
> So, I don't want to distract you much further, and will get back to under
> my stone, but I think instead of trying to rectify this brokenness it
> would be better in the grander scheme of things if register vars were just
> finally expanded as pseudos, with copy-in/out around inline asms.
> (perhaps even copy-in/out around statements that aren't inline asms but
> refer to the variable, just so to keep historic broken uses like "register
> long stackp asm("sp"); return stackp;" working :-) )
>
>
> Ciao,
> Michael.