[Numpy-discussion] Need help for implementing a fast clip in numpy (was slow clip)

Francesc Altet faltet at carabos.com
Sat Jan 13 09:20:15 EST 2007


El ds 13 de 01 del 2007 a les 14:42 +0900, en/na David Cournapeau va
escriure:
> > No, only if it's contiguous (I'm not sure about alignment)
> According to numpy ebook, you need alignement for deferencing
> pointers. But I guess non aligned arrays are not really common.

Unaligned arrays can be quite common if you use recarrays often:

In [18]:a=numpy.empty(3, dtype="S1,i4,f8")
In [19]:a.dtype.fields.items()
Out[19]:
[('f0', (dtype('|S1'), 0)),
 ('f1', (dtype('int32'), 1)),
 ('f2', (dtype('float64'), 5))]
In [20]:f1=a['f1']
In [21]:f1.flags
Out[21]:
  C_CONTIGUOUS : False
  F_CONTIGUOUS : False
  OWNDATA : False
  WRITEABLE : True
  ALIGNED : False
  UPDATEIFCOPY : False

And dereferencing unaligned pointers do cause a crash in general, but
not in all architectures (i386 being one of them).

Cheers,

-- 
Francesc Altet    |  Be careful about using the following code --
Carabos Coop. V.  |  I've only proven that it works, 
www.carabos.com   |  I haven't tested it. -- Donald Knuth




More information about the NumPy-Discussion mailing list