On Mon, 11 Dec 2023, Richard Sandiford wrote:
> > It all seems a bit hackish. I don't think ports have had much success
> > using 'p' through the decades. I think I generally ended up having to
> > go with distinct constraints rather than relying on 'p'.
> >
> > OK for the trunk, but ewww.
>
> Thanks, pushed. And yeah, eww is fair. I'd be happy for this to become
> an unconditional VOIDmode once reload is removed.
Hmm, LRA seems unable currently to work with indexed address modes, such
as with these address load machine instructions:
movaq 0x12345678[%r1],%r2
movaq (%r0)[%r1],%r2
movaq 0x12345678(%r0)[%r1],%r2
movaq *0x12345678[%r1],%r2
movaq *(%r0)[%r1],%r2
movaq *0x12345678(%r0)[%r1],%r2
(where R1 is scaled according to the width of data the address refers to
before adding to the direct or indirect address component worked out from
base+displacement, by 8 in this example, suitably for DImode or DFmode) so
who knows what we'll end up with once the VAX port has been converted.
Maciej
