ATLAS in Sage is rather old, with there being many more recent beta releases.
ATLAS development has not stopped, but I understand from Clint (the main
developer), that he has other commitments which means he will spend less time on
ATLAS for a while. So there are not going to be any new stable release coming
out soon.
There are a few specific facts.
* The current version of ATLAS in Sage takes 8.5 hours to build on 't2' as it
has no tuning parameters for the Sun T2+ processors. The latest beta fixes that
and apparently will build in under an hour on 't2'. So an update should save
more than 7.5 hours when building Sage on 't2'. That should reduce the build
time of Sage on 't2' from around 27 hours to 19.5 hours.
* The ATLAS package in Sage has additional tuning parameters for 3 sorts of
processors. I've never managed to work out how to move them to a newer release,
so they might be lost if we changed to a new beta.
* The latest beta has lots of turning parameters for new processors, so should
build more quickly, and probably run faster too on many new processors. Less
CPUs would be classed as a generic 386. I suspect there is a good chance the 3
processors in Sage would all be in the updated ATLAS beta, but I don't know for
sure.
* The Sage package has quite a few hacks to get around problems. These may or
may not be necessary if ATLAS was updated. Integrating them might be difficult,
and probably difficult to test, as some require hardware/software combinations
that are quite rarely seen.
* I don't know if there are extensive doc tests in Sage for ATLAS, but a grep
of 'atlas' in the ptestlong.log file shows nothing with 'atlas' in the name,
which make me think there might not be. Clearly if we don't test ATLAS, it is
more difficult to know of any problems if an upgrade takes place.
I would generally be reluctant to use beta software, but in this case, the
problems with building ATLAS on 't2' are very severe.
Another option, that would solve the ATLAS problem on 't2' is that the ATLAS
package is expanded with some binaries for such processors. That would add 1.6
MB to the Sage package.
I've discussed this with William a bit off-list, and we agree that if a new
ATLAS package is made, it should go in an alpha0, so gets the longest test
duration. We would then have to sort out any problems.
Thoughts?
Dave
--
To post to this group, send an email to sage-devel@googlegroups.com
To unsubscribe from this group, send an email to
sage-devel+unsubscr...@googlegroups.com
For more options, visit this group at http://groups.google.com/group/sage-devel
URL: http://www.sagemath.org
