Bug 20825 - Do we really want and need libatlas-opt*?
Summary: Do we really want and need libatlas-opt*?
Status: RESOLVED OLD
Alias: None
Product: Mageia
Classification: Unclassified
Component: RPM Packages (show other bugs)
Version: Cauldron
Hardware: All Linux
Priority: Normal normal
Target Milestone: ---
Assignee: Shlomi Fish
QA Contact:
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2017-05-10 22:59 CEST by Chris Denice
Modified: 2021-09-12 19:02 CEST (History)
1 user (show)

See Also:
Source RPM: atlas-3.10.2-6.mga6.src.rpm
CVE:
Status comment:


Attachments

Description Chris Denice 2017-05-10 22:59:42 CEST
Hi,
I had a look to upgrade atlas to last version 3.10.3, in an attempt to fix autobuild failure for mga6:

http://pkgsubmit.mageia.org/autobuild/history.php?package=atlas

Then, I opened the "SPEC" file. What the hell is going on here? :)

It is full of junk concerning ppc and other arch we do not support, source files without references and origin (ARMV732NEON) and a bunch of sed everywhere to force compiling the library either with sse2 or sse3 only.

I don't see the point. Atlas is supposed to be a optimized lapack/blas and upstream now provide AVX2 instructions that we seem to disable on purpose. 

It seems to me that the only point of having atlas would precisely be to provide a AVX optimized library for the few machines that support it, otherwise any other program could be linked to lapack or blas directly. And the library compiled with -sse2 and -sse3 already breaks our policy of portability to pentium, so why not having only one -avx2 build? Who needs -sse2 -sse3 today?

For ARM, I don't know, do people use them for computationally intensive linear algebra calculations? 

And not a lot of packages are built against it:
urpmq --whatrequires lib64atlas3-x86_64
lib64atlas3-x86_64
libatlas-x86_64-devel
ocaml-gpr
ocaml-gsl
psfex
python-pymc
python3-pymc
sextractor

So it seems that we are pulling out a lot of efforts for nothing :-/

I am happy to have a look to the package, but would like to have some feedback before nuking everything!

Cheers,
Chris.
Comment 1 Marja Van Waes 2017-05-13 16:56:35 CEST
(In reply to Chris Denice from comment #0)

......
> 
> So it seems that we are pulling out a lot of efforts for nothing :-/
> 
> I am happy to have a look to the package, but would like to have some
> feedback before nuking everything!
> 
> Cheers,
> Chris.

Aassigning to the registered maintainer, Shlomi, who'll answer you if he sees this :-)

CC: (none) => marja11
Assignee: bugsquad => shlomif

Comment 2 Shlomi Fish 2017-05-13 18:04:33 CEST
Thanks, Marja! Chris, I'm personally am fine with you trying to clean up the mess somehow. I don't know much about libatlas.
Comment 3 David Walser 2017-05-14 02:54:35 CEST
Please do be careful.  This package is a nightmare to update, maintain, or get to build.  We have it at least somewhat synced with Fedora to try to make it easier, so please don't make things more difficult by making us more out of sync with them unnecessarily.
Comment 4 Chris Denice 2017-05-14 21:27:15 CEST
Yep, I agree, but it does not build anymore. And syncing with Fedora when Fedora provides such a mess does not seem to me a good idea. Anyway, I'll play locally and report here before making any push to be sure.

I remember the atlas mess from https://bugs.mageia.org/show_bug.cgi?id=11619, so I really consider we could get ride of it.

thanks for the answers guys,
Cheers.
Comment 5 Marja Van Waes 2021-09-12 19:02:04 CEST
@ Chris

Atlas (still with lib64atlas3-x86_64) was pushed 14 times since you wrote this report, of which only two by umeabot. The others were pterjan, tv, you and joequant. If you still think it should be nuked, then please start a RFC thread on dev ml.

Closing this report as OLD, because it saw no action since over four years ago.

Resolution: (none) => OLD
Status: NEW => RESOLVED


Note You need to log in before you can comment on or make changes to this bug.