Fedora 22

Dan HorĂ¡k dan at danny.cz
Tue Jan 6 12:45:53 UTC 2015


On Tue, 16 Dec 2014 14:22:14 -0600
Michael J Wolf <mjwolf at us.ibm.com> wrote:

> > > Subject:
> > >
> > > Re: Fedora 22
> > >
> > > Sent by:
> > >
> > > ppc-bounces at lists.fedoraproject.org
> > >
> > > >> Hi
> > > >>   We briefly discussed priorities for Fedora 22 and I had
> > > >> taken an action item to start an email conversation about
> > > >> this.  So here is what I would like to see for Fedora 22.
> > > >>
> > > >> 1)  Get the -mcpu and -mtune flags set properly for the LE
> > > >> build. Should be -mcpu=power7 -mtune=power8
> > > >
> > > > done, all packages that honour the Fedora system wide compiler
> > > > flags use them, if they don't it's a packaging bug
> > > >
> > > >> 2) Have a cloud image available
> > > >> 3) For BE I would like another subarch.  Same packages as the
> current
> > > >> one but tuned for P8.
> > > >
> > > > you mean in addition to ppc64p7? can't we just switch ppc64p7
> > > > from -mcpu=power7 -mtune=power7 to -mcpu=power7 -mtune=power8?
> > >
> > > This makes sense to me as it then mirrors what we have in ppc64le
> > > and it saves having more targets.
> >
> > The disadvantage of this would be to cut off users that have Power 7
> systems
> > and optimized code.  So why would people want to optimize from
> > Fedora 21
> to
> > Fedora 22.  You would be taking a big step back in performance.  I
> > don't want to suggest keeping a subarch for each type of POWER
> > system out
> there.  I
> > was thinking of keeping two.  So when the next POWER arch that
> > comes out, the Power 7 subarch goes away and you would have Power 8
> > and the new Power arch.
> 
> The above comment is not correct I was assuming -mcpu=power8 but your
> comment states
> -mcpu=power7 so yes that is reasonable and keeps things consistent
> across BE and LE.
> The only drawback here is that you would be leaving some performance
> behind as you
> would not be using all the new Power 8 instructions.

I think it (as not using P8 instructions) would be acceptable
compromise for a binary distribution. But it would be interesting to see
some numbers what the difference between the various combinations of
-mcpu and -mtune and used hardware is.

Also introducing a new subarch means updating (patching) our build and
release engineering tools starting with rpm itself.


		Dan


More information about the ppc mailing list