Re: science-optimisation (Was: About the sense of removing -march=native)
On Thu, Feb 09, 2017 at 04:41:54PM +0100, Andreas Tille wrote:
> On Thu, Feb 09, 2017 at 04:13:15PM +0100, Michael Banck wrote:
> > Probably just
> > bumping the binary package version like we do for backports or binNMUs
> > would be enough.
>
> Definitely.
>
> > A good start would be to both support DEB_BUILD_OPTIONS=custom for
> > generic performance-oriented rebuidls and
> > DEB_CFLAGS_MAINT_APPEND=-march=native for for the -march=native thing.
> >
> > Then again, just testing whether it actually makes a difference for
> > end-user applications would be useful as well.
>
> I made a mental note to suggest a GSoC project for Debian Med to some
> benchmarking between packages with and without manual optimisation. If
> another mentor from Debian Science this would be probably very helpful
> and could stretch the project to Debian Science.
I think it's useful, but for a GSoC project I would expect at least
trying to get some more general support in a couple of packages or even
to evaluate a DKMS-like rebuilder.
Benchmarking could be done for debian-med packages (I would volunteer to
co-mentor if the student looks at debichem packages as well), but for
actual changes, most of the low-hanging numerical libraries are in
debian-science (I assume), so that one should also be included.
Michael
Reply to: