Opened 6 years ago

Closed 6 years ago

Last modified 6 years ago

#14753 closed enhancement (fixed)

Revert ATLAS package

Reported by: jdemeyer Owned by: jdemeyer
Priority: blocker Milestone: sage-5.10
Component: packages: standard Keywords:
Cc: Merged in: sage-5.10
Authors: Jeroen Demeyer Reviewers: Jeroen Demeyer
Report Upstream: N/A Work issues:
Branch: Commit:
Dependencies: Stopgaps:

Description (last modified by jdemeyer)

There are some problems with generic binaries (SAGE_FAT_BINARY=yes) in the new ATLAS spkg. In order not to delay the release of Sage 5.10 any longer, I will revert #10508.

Fixing the issues below (and upgrading ATLAS to 3.10.x) is #14754.

On cicero (Fedora 16 i686), the build fails: http://build.sagemath.org/sage/builders/Skynet%20cicero%20%28Fedora%2016%20i686%29%20bin/builds/90/steps/shell_5/logs/atlas

Note the use of -DATL_3DNow and lots of "Illegal instruction" messages, proving that this likely is not a generic binary, unlike what it should be.

On arando (Ubuntu 13.04 i686), the build works but there is doctest failure:

sage -t devel/sage/sage/matrix/matrix_double_dense.pyx
**********************************************************************
File "devel/sage/sage/matrix/matrix_double_dense.pyx", line 3823, in sage.matrix.matrix_double_dense.Matrix_double_dense.is_positive_definite
Failed example:
    H.is_positive_definite()
Expected:
    True
Got:
    False
**********************************************************************

Attachments (2)

14753_revert_atlas.patch (3.2 KB) - added by jdemeyer 6 years ago.
14753_revert_atlas_docs.patch (4.5 KB) - added by jdemeyer 6 years ago.

Download all attachments as: .zip

Change History (10)

Changed 6 years ago by jdemeyer

Changed 6 years ago by jdemeyer

comment:1 Changed 6 years ago by jdemeyer

  • Description modified (diff)
  • Milestone set to sage-5.10

comment:2 Changed 6 years ago by jdemeyer

  • Description modified (diff)

comment:3 Changed 6 years ago by jdemeyer

  • Description modified (diff)

comment:4 Changed 6 years ago by jdemeyer

  • Authors set to Jeroen Demeyer
  • Merged in set to sage-5.10
  • Resolution set to fixed
  • Reviewers set to Jeroen Demeyer
  • Status changed from new to closed

comment:5 follow-up: Changed 6 years ago by jpflori

The fact that 3DNow is used was discussed at #13706.

Of course we can remove that.

Jeroen: could you try to modify spkg-install: replace

isa_ext = ('3DNow',)

by

isa_ext = ()

on line 404 (in some version of the p0 spkg I have access to right now.) It's in the configure_base function, the part about "Generic configuration on Intel i386 compatible CPUs"; and then try to rebuild on cicero? (I have no access to it)

comment:6 Changed 6 years ago by jdemeyer

  • Description modified (diff)

comment:7 Changed 6 years ago by jdemeyer

  • Description modified (diff)

comment:8 in reply to: ↑ 5 Changed 6 years ago by jdemeyer

Replying to jpflori:

The fact that 3DNow is used was discussed at #13706.

Of course we can remove that.

Jeroen: could you try to modify spkg-install: replace

isa_ext = ('3DNow',)

by

isa_ext = ()

on line 404 (in some version of the p0 spkg I have access to right now.) It's in the configure_base function, the part about "Generic configuration on Intel i386 compatible CPUs"; and then try to rebuild on cicero? (I have no access to it)

I posted a spkg at #14754 doing that, I'll try it on cicero (a single-core Pentium 4 machine).

In any case, this will not change my decision to revert #10508. I want to release Sage 5.10 as soon as possible.

Note: See TracTickets for help on using tickets.