[Libre-soc-dev] All div pipe tests pass

Luke Kenneth Casson Leighton lkcl at lkcl.net
Fri Oct 9 18:49:22 BST 2020


On Fri, Oct 9, 2020 at 6:32 PM Cole Poirier <colepoirier at gmail.com> wrote:
>
> On Fri, Oct 9, 2020 at 10:09 AM Luke Kenneth Casson Leighton
> <lkcl at lkcl.net> wrote:
> > On Fri, Oct 9, 2020 at 6:03 PM Jacob Lifshay <programmerjake at gmail.com> wrote:
> >
> > > well, the modsw instruction has similar behavior -- sign-extending the
> > > result in all cases when the spec leaves the top 32-bits undefined.
> >
> > ahh interesting.  sigh all of these fall into "Ahmdahl nightmare
> > scenario".  i'm amazed anyone hasn't pointed that out before, or made
> > a fuss about it.
>
> I've heard of Ahmdahl's law in terms of parallel scaling before, but I
> can't quite understand the above, can you enlighten me as to why this
> sign extension behaviour falls into Ahmdahl nightmare scenario?

story i told last week.  and the week before.  and several times on
openhdl-cores.  "correct" spec behaviour but not "exactly like 360"
required amdahl to patch the OS to provide EXACT and SPEC-VIOLATING
binary-compatible behaviour due to customers complaining and refusing
to recompile existing software that they had already added
"workarounds" into when running on IBM 360 mainframes.

lesson of story that absolutely has to be hammered home even to OPF:
the "top dog" (IBM POWER9/10) is the de-facto absolute definer of the
spec, *not* the spec.

l.



More information about the Libre-soc-dev mailing list