[Numpy-discussion] Problem with numpy.linalg.eig?

[Michael McNeil Forbes]

On 16 Nov 2007, at 1:46 AM, Michael McNeil Forbes wrote:
> On 15 Nov 2007, at 8:23 PM, David Cournapeau wrote:
>
>> Could you try without atlas ? Also, how did you configure atlas when
>> building it ? It seems that atlas is definitely part of the problem
>> (everybody having the problem does use atlas), and that it involves
>> Core
>> 2 duo.
>>
>> David
>
> It seems to work fine without ATLAS, but then again, it is a somewhat
> "random" error.  I will let some code run tonight and see if I detect
> anything.

Just an update.  I am still having this problem, along with some  
additional problems where occasionally even dot returns nan's.  I  
have confirmed that without ATLAS everything seems to be fine, and  
that the problem still remains with newer versions of ATLAS, Python,  
gcc etc.

ATLAS was configured with

../configure --prefix=${BASE}/apps/${ATLAS}_${SUFFIX}\
              --with-netlib-lapack=${BASE}/src/${LAPACK}_${SUFFIX}/ 
lapack_LINUX.a\
              -A Core2Duo64SSE3\
              --cflags=-fPIC\
              -Fa alg -fPIC

and it passed all the tests.

The problem still exists with ATLAS version 3.8.1, gcc 4.3.0, and  
recent versions of numpy.

  >>> sys.version
  '2.5.2 (r252:60911, Mar 29 2008, 02:55:47) \n[GCC 4.3.0]'
  >>> numpy.version.version
  '1.0.5.dev4915'

I have managed to extract a matrix that causes this failure  
repeatedly once every two or four times eigh is called, so hopefully  
I should be able to run gdb and track down the problem...