[NumPy-Tickets] [NumPy] #1982: In-place arithmetic operations: wrong calculation order?

NumPy Trac numpy-tickets@scipy....
Fri Nov 18 12:50:13 CST 2011


#1982: In-place arithmetic operations: wrong calculation order?
------------------------+---------------------------------------------------
 Reporter:  ling        |       Owner:  somebody   
     Type:  defect      |      Status:  new        
 Priority:  normal      |   Milestone:  Unscheduled
Component:  numpy.core  |     Version:  1.6.0      
 Keywords:              |  
------------------------+---------------------------------------------------

Comment(by rkern):

 I think we have always left the order in which we iterate over arrays in
 ufuncs officially undefined. We have reserved the right to reorder the
 computations such that we can make memory access more efficient. We have
 always done this for transposed axes, for example. I think either 1.5 or
 1.6 introduced this for reversed arrays, too.

 I have a vague recollection that the last time this came up, we considered
 introducing a check if the output array overlapped one of the input
 arrays, then we would not reorder the iteration any. I don't remember if
 we ever instituted this policy (in which case, the current behavior is
 breaking said policy) or not.

-- 
Ticket URL: <http://projects.scipy.org/numpy/ticket/1982#comment:1>
NumPy <http://projects.scipy.org/numpy>
My example project


More information about the NumPy-Tickets mailing list