[Numpy-discussion] Strange crash in numpy.linalg.lstsq.

Jonathan Taylor jonathan.taylor@utoronto...
Tue Aug 18 14:14:55 CDT 2009


Right... So I was able to get everything working finally.  I am not
100% sure how or why it works though so I am going to outline what I
did here for reference.

I first tried just using LAPACK 3.1.1 (since it seemed set up for g77
instead of gfortran which I do not have).  I compiled this to yield
the associated lapack and blas libraries using the fortran compiler
settings detailed on the scipy install web page.  This actually gave
me a numpy with the same problems (hanging on numpy.test()).  Thus I
realized the problem was LAPACK and not with ATLAS.

Eventually I got numpy with LAPACK to work when I used the default
settings in the example config file of LAPACK instead of using the
suggested settings on the numpy web site.  I then compiled ATLAS and
this worked as well.  It still seems a little bit weird that these
settings can break the software though.

Thanks for the suggestions and I hope this helps someone.
Jonathan.

On Mon, Aug 17, 2009 at 5:34 PM, Keith Goodman<kwgoodman@gmail.com> wrote:
> On Mon, Aug 17, 2009 at 2:27 PM, Jonathan
> Taylor<jonathan.taylor@utoronto.ca> wrote:
>> It seems to automatically detect it though.  Specifically
>> lapack_lite.so always seems to reference libatlas.
>>
>> On Mon, Aug 17, 2009 at 5:21 PM, Keith Goodman<kwgoodman@gmail.com> wrote:
>>> On Mon, Aug 17, 2009 at 2:13 PM, Jonathan
>>> Taylor<jonathan.taylor@utoronto.ca> wrote:
>>>> Is there some alternative to get a working numpy going?  How might I
>>>> go about compiling numpy without ATLAS?  I really got to get at least
>>>> something working temporarily.
>>>
>>> Just build numpy again but skip the ATLAS steps.
>
> Yes, sorry. The only way I've tried doing it is uninstalling the
> unbuntu ATLAS binary. I don't know how to ignore it if it is
> installed.
> _______________________________________________
> NumPy-Discussion mailing list
> NumPy-Discussion@scipy.org
> http://mail.scipy.org/mailman/listinfo/numpy-discussion
>


More information about the NumPy-Discussion mailing list