On 05/23/10 04:52 PM, Minh Nguyen wrote:
Hi David,
On Mon, May 24, 2010 at 12:50 AM, Dr. David Kirkby
<david.kir...@onetel.net> wrote:
If someone has a minute or two.
http://trac.sagemath.org/sage_trac/ticket/9009
The changes look harmless enough. But I wanted to test building that
updated spkg as 64-bit. It hardly ever occurred to me to build Sage as
64-bit on t2.math. In trying to do so, I did
$ export SAGE64=yes
and executed "make". Of course this didn't work because my
SAGE_FORTRAN_LIB pointed to a 32-bit Fortran library. So I figured
that I should change that variable to
$ export
SAGE_FORTRAN_LIB=/usr/local/gcc-4.4.1-sun-linker/lib/sparcv9/libgfortran.so
I then re-ran "make". Now I'm playing the waiting game to see if Sage
builds as 64-bit on t2.math, or at least your updated spkg build as
64-bit on that machine.
Sage will not build fully as 64-bit on 't2'. I've tried it on 't2' and on other
SPARCs in the past (months ago).
My experience on OpenSolaris in 64-bit mode shows me that there are some bits
which will still build 32-bit, even when SAGE64 is set to "yes". As such, they
will build 32-bit on 't2'.
You would need to run the 'file' command on the files in $SAGE_LOCAL/lib and
$SAGE_LOCAL/bin, as some might build 32-bit, but you wont discover the problems
until you try to link a library.
Others bits I know will cause a failure include:
1) sage_fortran will not add the -m64 option, so any fortran objects will be
built 32-bit.
A patch for that, written by Jaap Spies solves that, but his patch is against an
older version of the fortran package. I've rebased that against the current
package, which needs reviewing.
http://trac.sagemath.org/sage_trac/ticket/7982
Unless that is done, lapack, ATLAS etc will not build properly as 64-bit.
2) zlib will build, but you *might* find it produces a 32-bit library, rather
than a 64-bit library. I'd run the 'file' command on $SAGE_LOCAL/lib/libz*
If not, the following
http://boxen.math.washington.edu/home/kirkby/patches/zlib-1.2.5/zlib-1.2.5.spkg
should resolve it, though it needs more testing. See:
http://trac.sagemath.org/sage_trac/ticket/9008
It would be great to get a 64-bit SPARC build. But I think many of the issues
would be much quickly resolved on a machine like 'fulvia' or an OpenSolaris
machine on the x64 platform, as their better performance will allow many of the
issues to be discovered much more quickly.
Of course, there will be some issues specific to 64-bit SPARC and others
specific to 64-bit OpenSolaris on x64, but I would expect 95% of the issues
preventing 64-bit builds on both Intel and SPARC hardware will be the same on
both architectures.
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
