Re: [Numpy-discussion] problem with lapack_lite / BLAS

[Karol L. <kar...@kn...>]

On Saturday 07 of October 2006 20:32, Pauli Virtanen wrote:
> la, 2006-10-07 kello 20:01 +0200, Karol Langner kirjoitti:
> > I still get a floating point exception when running the numpy tests,
> > though. I haven't checked out numpy for some time, so I don't now if it's
> > a bug, or if it's my setup. The same thing happens when I use my manually
> > built atlas/lapack and the built-in debian atlas/lapack libraries. I'd be
> > grateful
> >
> > for a comment on if this is just me:
> > >>numpy.test(10,10)
> >
> > [output]
> > Check reading the top fields of a nested array ... ok
> > Check reading the nested fields of a nested array (1st level) ... ok
> > Check access nested descriptors of a nested array (1st level) ... ok
> > Check reading the nested fields of a nested array (2nd level) ... ok
> > Check access nested descriptors of a nested array (2nd level) ... ok
> > check_access_fields
> > (numpy.core.tests.test_numerictypes.test_read_values_plain_multiple) ...
> > ok check_access_fields
> > (numpy.core.tests.test_numerictypes.test_read_values_plain_single) ... ok
> > check_cdouble (numpy.tests.test_linalg.test_det)Floating point exception
>
> If you are using Debian stable (sarge), you might need to read
>
> http://www.its.caltech.edu/~astraw/coding.html#libc-patched-for-debian-sarg
>e-to-fix-floating-point-exceptions-on-sse2
>
> In short, libc in Debian stable has a bug that makes programs crash with
> SIGFPE when SSE instructions are invoked. The solution is to recompile
> libc from patched sources, and replace libm.so.6. At least for me this
> fixed crashes in numpy.
>
> 	Pauli Virtanen

Big thanks for that link. I took the leap and it does fix that SIGFPE. As a 
note, after building the glibc sources, you only have to install the libc6 
binary deb. There should be a link to that page somewhere on the wiki.

Karol

-- 
written by Karol Langner
Tue Oct 10 08:07:04 CEST 2006