On 08/25/10 02:01 AM, William Stein wrote:
On Tue, Aug 24, 2010 at 1:38 PM, William Stein<wst...@gmail.com>  wrote:
On Mon, Aug 23, 2010 at 8:25 PM, Jonathan Hanke<jonha...@gmail.com>  wrote:
Hi,

I'm trying to build sage 4.5.2 on a Sun Fire X4450 server (Intel Xenon
processors) running Redhat linux (Enterprise edition), and get an error when
building Atlas in the make script.  Any comments are appreciated.  Thanks,

I just tried building 4.5.3.alpha2 on *your* computer (rosemary) in
/home/wstein/build/sage-4.5.3.alpha2 and it worked fine.  Note that I
set this environment variable:

  export SAGE_FAT_BINARY=yes

This changes how ATLAS is built to use a sort of "lowest common
denominator" configuration.
You might try that.   It can potentially make numerical linear algebra
slower for you, but I doubt
you do a lot of that.

Hi,

I waited for the Sage testsuite to run on rosemary (with
sage-4.5.3.alpha2), and interestingly

What is rosemary.math? Does it happen to have an Opteron processor?

there are a huge number of failures, despite the building completing
fine.  It looks like
every failure is "ImportError: No module named scipy".  This suggests
that despite the build
of Sage exiting with an error, for some reason the exit code was still
zero, so the testsuite got run.


Checking further I see that PIL failed to build:

/usr/bin/ld: skipping incompatible /usr/lib/libjpeg.so when searching for -ljpeg
/usr/bin/ld: /usr/local/lib/libpython2.6.a(abstract.o): relocation R_X86_64_32 a
gainst `a local symbol' can not be used when making a shared object; recompile w
ith -fPIC

I've not looked at the code, but it is *often* possible to get away without compiling libraries with position independent code. But this is not reliable.

Until recently, Singular compiled without PIC code on 32-bit build builds, except on I think Opteron processors. The developers belived this was only necessary on 64-bit and IIRC on Opteron processors.

After a lot of discussion, I finally managed to convince the Singular developers that building without PIC code is unreliable. This is very clearly documented in the Sun linker manual, and a bit less so in the GCC manual. So now Singular builds PIC on every platform all the time.

There's a small performance disadvantage with this, which the Singular developers measured at around 5%.

So *IF* that library is not being built PIC, then I am not surprised it fails on some systems at some times.

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

Reply via email to