[Numpy-tickets] [NumPy] #432: cov segfaults when input is oriented incorrectly

NumPy numpy-tickets@scipy....
Mon Apr 9 14:32:25 CDT 2007


#432: cov segfaults when input is oriented incorrectly
-------------------------------------+--------------------------------------
 Reporter:  dhuard                   |        Owner:  somebody
     Type:  defect                   |       Status:  reopened
 Priority:  normal                   |    Milestone:          
Component:  Other                    |      Version:  devel   
 Severity:  blocker                  |   Resolution:          
 Keywords:  matrix multiply failure  |  
-------------------------------------+--------------------------------------
Changes (by charris):

  * keywords:  => matrix multiply failure
  * version:  => devel
  * severity:  normal => blocker

Comment:

 OK, I'm running on 64bit fedora and elicit an illegal instruction error.
 Here is the culprit:

 r = ones((1000,2))
 dot(r,r.T.conj())

 I suspect a strides problem with memory reference going out of bounds.
 When r = ones((n,2)) the problem begins for me when n >= 127, which is a
 bit odd -- pagesize related, perhaps? The conjugation isn't relevant.

 In [1]: r = ones((1000,2))

 In [2]: rt = ones((2,1000))

 In [3]: dot(r,rt)
 Illegal instruction

 Hmm, normal matrix multiplication fails. I'm changing the severity to
 blocker.

-- 
Ticket URL: <http://projects.scipy.org/scipy/numpy/ticket/432#comment:9>
NumPy <http://projects.scipy.org/scipy/numpy>
The fundamental package needed for scientific computing with Python.


More information about the Numpy-tickets mailing list