[NumPy-Tickets] [NumPy] #1153: performing operations inplace to save memory allocations

[NumPy Trac]

#1153: performing operations inplace to save memory allocations
-------------------------+--------------------------------------------------
 Reporter:  lucaciti     |       Owner:  charris    
     Type:  enhancement  |      Status:  apply      
 Priority:  normal       |   Milestone:  Unscheduled
Component:  numpy.core   |     Version:  devel      
 Keywords:               |  
-------------------------+--------------------------------------------------

Comment(by m-paradox):

 This idea is pretty good, but it does have some tricky interactions with
 the new iterator ufunc implementation.  I've added an order= parameter,
 which controls the memory layout of the output, and it could be a lot of
 messy code to make sure the input chosen for reuse matches the desired
 output.  It also doesn't look portable to NumPy 2.0/libndarray, where I
 don't believe the assumption about reference counting will necessarily
 hold.  To go into NumPy, this patch would need to take into account the
 subtype/wrapping mechanism, to support masked arrays and other derived
 array types.

 The new iterator doesn't help achieve the same effect, but the deferred
 evaluation idea (longer term) would achieve better performance
 improvements than this patch.

 https://github.com/m-paradox/numpy/blob/mw_neps/doc/neps/deferred-ufunc-
 evaluation.rst

-- 
Ticket URL: <http://projects.scipy.org/numpy/ticket/1153#comment:9>
NumPy <http://projects.scipy.org/numpy>
My example project