Re: [gentoo-soc] Interest in Improve packaging of scientific software

[Andrew Savchenko]

On Mon, 26 Mar 2018 08:03:32 +0900 Benda Xu wrote:
> Hi Andrew,
>
> Andrew Savchenko <bircoph@g.o> writes:
>
> >> I’ll then start the most challenge part, porting and packaging
> >> Intel Tools.
> >
> > While Intel stuff may rightfully be a part of your project, I do not
> > recommend to focus on them too much, since this is a proprietary
> > software and GSoC is all about Free/Libre software.
> >
> >> I’m also interested to port Intel’s python distribution
> >
> > I've discussed this project with Intel devs on one of the
> > conferences. There is nothing special about it: it is a normal
> > Python linked with Intel libraries and with some math libs replaced
> > with more optimized free software solutions. So everyone can do the
> > same with Intel MKL without need to obtain Intel Python. They
> > created this project mostly due to marketing issues, since python
> > is a popular language and management want to establish Intel's
> > presence in this area.
> >
> > If you want to pursue this task, I recommend to build on FLOSS
> > solutions as described above, packaging Intel Python itself is
> > quite useless.
>
> +1
>
> With a more rubost blas/lapack framework/eclass, an optimized
> scipy/numpy linked with OpenBLAS or Intel MKL will be on par if not
> overtake the Intel python binaries.

[OT]
I made some tests with linpack (HPL) with MKL vs OpenBLAS on
Gentoo-based HPC cluster at my previous job. OpenBLAS was beating
MKL by 0.5 ± 0.1%. Not large, but statistically significant result.
But of course MKL is not just BLAS/LAPACK...
[/OT]

Best regards,
Andrew Savchenko