Re: PPC64 Illegal Instruction
Hi Adrian,
Unfortunately, the issue is still not resolved. Audacious, Audacity,
and VLC (and I'm sure, many others) are still throwing "Illegal
instruction".
At first I thought it was because the baseline was changed to POWER5
instead of POWER4. Altivec or not, the PowerPC970 is still based on
POWER4, so it makes sense (at least to me) that it would still refuse
to launch.
@Michael, if I may ask (and if I have this right), if the fix was
simply a name change for a newer version, and effectively useless for
us, what was the point of changing it, and what can be done to fix our
issue?
Noah
On Mon, 2019-03-04 at 11:44 +0100, John Paul Adrian Glaubitz wrote:
> Hi!
>
> On 3/4/19 10:24 AM, Gabriel Paubert wrote:
> > No, the difference is VSX. Altivec is the same instruction set,
> > whether
> > on G4, G5, Power5 and later (IBM Power4 did not have Altivec,
> > Power5 did,
> > VSX was added to Power7).
>
> Okay, I had the assumption that AltiVec and VSX were the same. Thanks
> for
> the clarification. The amount of different POWER targets is really
> confusing.
>
> > G5 or PPC970 is actually closer to Power5 from this point of view.
>
> Yeah, if POWER4 was without AltiVec, that makes sense.
>
> > To support everything you would ideally need 3 versions (or 3
> > libraries):
> > - no vector instructions (basic Power4)
> > - Altivec (G5, P5, P6)
> > - VSX (P7 and later)
>
> We don't want to support anything newer than P5 really. POWER4
> and POWER5 is enough.
>
> > However, the big jump for performance is likely provided by
> > Altivec.
>
> Yeah, I agree.
>
> > Do you really want to support base Power4 (no Altivec) or not?
> >
> > If not, a single version (Altivec and no VSX) is probably the
> > simplest.
>
> The problem is that some people want to run the ppc64 port on
> hardware
> which doesn't support Altivec like those embedded E5550 CPUs. There
> was
> actually a large archive rebuild to lower the baseline for that
> reason
> (the discussion is in the mailing list archives for this mailing
> list).
>
> We could maybe have a poll which baseline users prefer.
>
> FWIW, the x265 bug originally reported by Noah should be fixed now.
>
> @Noah: Can you confirm the issue is resolved for you?
>
> Adrian
>
Reply to: