[Numpy-discussion] numpy.linalg.eigvals crashes whn calling lapack_lite.pyd

Simon simonpy2008@gmail....
Wed Jan 9 01:24:50 CST 2008


Charles R Harris <charlesr.harris <at> gmail.com> writes:

> 
> 
> On Jan 8, 2008 6:49 PM, Simon <simonpy2008 <at> gmail.com> wrote:
> Newbie here. Trying to generate eigenvalues from a matrix using:print
numpy.linalg.eigvals(matrix)This works with small matrices, say 5 x 5, but
causes python to crash on largermatrices, say 136 x 136, which is not really
very large.
> Setup:Win XP SP2Python 2.5.1 (from .msi)numpy 1.0.4 (from .msi)pywin32-210
(from .exe installer)When running from either the command line or the Pythonwin
IDE, python.execrashes. The info in the microsoft error reporting thingy is:
> AppName: python.exeModName: lapack_lite.pydOffset: 000b7434Stepping through
linalg.py using Pythonwin, I get as far as line 418 (in theeigvals function):  
     results = lapack_routine('N', 'N', n, a, n, wr, wi,
>                                  dummy, 1, dummy, 1, work, lwork, 0)and then
python.exe crashes.That's the extent of my troubleshooting skills at this stage.
I haven't workedout if there is a specific matrix size where this starts
occurring. Where to now?
> 
> 
> 
> Probably just a different execution path depending on matrix size. But I am
not that familiar with lapack_lite. 
> 
> 
> I can send the actual data for the matrix if need be, but as it's very large
Ithought it would mess up the list if I posted it here.
> 
> 
> This sounds like a compiler and/or architecture incompatibility since ATLAS
doesn't seem to be part of the mix. What is your hardware?
> Chuck
> 
> 
> 
> 
> 
> _______________________________________________
> Numpy-discussion mailing list
> Numpy-discussion <at> scipy.org
> http://projects.scipy.org/mailman/listinfo/numpy-discussion
> 


Xeon 5150 in Dell 490.

I've also tried it on a couple of other boxes with different hardware since my
first post, but don't have the details in front of me.

Might try a compile from source on a unix box tomorrow, unless someone comes up
with a solution before then.

Simon


Simon



More information about the Numpy-discussion mailing list