[Slackbuilds-users] A few questions/remarks

[Eugéne Suter]

On 31 January 2010 01:07, Serban Udrea
<serban.udrea at skmail.ikp.physik.tu-darmstadt.de> wrote:
> Hello!
>
> I wrote a slackbuild for compiling ATLAS 3.8.3, one of the tuned BLAS
> libraries, and would like to submit it. Nevertheless, I cannot, at the time
> beeing, test it on a 64-bit system. Thus, I think that it would be nice if
> someone could help with testing on such a machine, before I do the
> submission.
>
> ATLAS by itself doesn't bring much, the main thing one has to do is to get
> also LAPACK compiled against it, before all other packages which depend on
> LAPACK+BLAS. Thus I took the original slackbuild written by Eugene Suter and
> modified it such that it can optionally perform the LAPACK build against
> ATLAS. But I have the same problem as with ATLAS and I also would like to
> know how to eventually proceed with the submission:  make a separate
> slackbuild out of it or share/pass maintenance with/to Eugene -- if he
> agrees and it's possible?

Hi, I'd be more than happy to work with you on an ATLAS build of BLAS
and LAPACK, after all it is preferable to simply using the un-tuned
reference implementations.
I've actually been meaning to make an ATLAS slackbuild for quite a
while but I'm busy with my end-of-semester exams until February 10th,
so after that date I don't mind dedicating some time to work on this;
also ATLAS can be a notorious PITA to build ;)

>
> c) Netlib BLAS does to my knowledge not have version 3.2.1, since it doesn't
> follow LAPACK. It is a (relatively old) reference implementation and maybe
> the version should be like with cblas the date of last modifications
> (packaging) 20070321.

I agree with you here, it is more sensible to use the latest revision
data as the version number. The reason I chose to give it the same
version as LAPACK is because the BLAS code bundled with the LAPACK
source is practically identical to the BLAS code at netlib. The only
differences I could find after diff'ing the files were minor things
like new comments.

Having an "atlas" package that contains both a tuned BLAS and LAPACK
build would eliminate this "problem" :)

>
> d) To my knowledge scipy does not depend, and thus the corresponding
> slackbuild should not require any of matplotlib, iPython and FFTW. Moreover,
> starting with version 0.7.0 the release notes of scipy state that all fft
> modules except (netlib) fftpack have been removed. Actually, I lost a lot of
> time because scipy/INSTALL and numpy/site.cfg.example still mention FFTW as
> an option but the build of scipy ignores its presence.

I'll update that SlackBuild as soon as I get a chance.

>
> Best regards,
>
> Serban Udrea
> _______________________________________________
> SlackBuilds-users mailing list
> SlackBuilds-users at slackbuilds.org
> http://lists.slackbuilds.org/mailman/listinfo/slackbuilds-users
> Archives - http://lists.slackbuilds.org/pipermail/slackbuilds-users/
> FAQ - http://slackbuilds.org/faq/
>
>

Cheers,
Eugéne