[SciPy-user] linalg.expm() Illegal Instruction Error?

Dylan Gorman dgorman@berkeley....
Sat Jun 13 23:46:43 CDT 2009


Thanks, Robert.

The situation is a little tricky. I installed numpy 1.3.0 from source  
from, I guess, this sourceforge link: http://sourceforge.net/project/downloading.php?group_id=1369&filename=numpy-1.3.0.tar.gz&a=58155458

I should perhaps elaborate. I have a small cluster, on which I have  
non-root access. I've installed python 2.5.2, numpy 1.3.0, and scipy  
0.7rc1 to ~. ATLAS is already configured on the system, so prior to  
installing these packages I set
export ATLAS=/usr/local/atlas
in ~/.bash_profile

I'm just aiming to run non-parallel batch jobs on each node of the  
cluster. The code works quite happily on a set of newer nodes in this  
cluster, but on the older ones we are getting this Illegal instruction  
error.

I'm wondering if there's a way to make numpy use, perhaps, an older  
version of ATLAS that will be somewhat slower, but will work on all  
the nodes. Can ATLAS be installed to ~?

Thanks,
Dylan

On Jun 13, 2009, at 3:23 PM, Robert Kern wrote:

> On Sat, Jun 13, 2009 at 10:44, Dylan Gorman<dgorman@berkeley.edu>  
> wrote:
>> Hi Folks,
>>
>> I'm having a bit of a weird problem. linalg.expm() is failing for
>> matrices larger than size (51,51), and reports "Illegal instruction"
>> and forces python to quit. I'm not sure where this is coming from. I
>> created the following routine:
>> for i in range(10,128):
>>        A = random.rand(i,i)
>>        B = linalg.expm(A)
>>        print i
>>
>> which fails with "Illegal instruction" when you get to 51.
>
> This happens when you use a numpy binary that was compiled to use an
> ATLAS library on a CPU with more advanced SSE instructions than your
> CPU. Exactly which binary did you install (please give a URL)? What is
> your CPU?
>
> -- 
> Robert Kern
>
> "I have come to believe that the whole world is an enigma, a harmless
> enigma that is made terrible by our own mad attempt to interpret it as
> though it had an underlying truth."
>  -- Umberto Eco
> _______________________________________________
> SciPy-user mailing list
> SciPy-user@scipy.org
> http://mail.scipy.org/mailman/listinfo/scipy-user



More information about the SciPy-user mailing list