[Numpy-discussion] Fast clip for native types, 2d version

[David Cournapeau]

Robert Kern wrote:
> David Cournapeau wrote:
>> Robert Kern wrote:
>>> David Cournapeau wrote:
>>>
>>>>   2: the old implementation does not upcast the input array. If the
>>>> input is int32, and min/max are float32, the function fails; if input is
>>>> float32, and min/max float64, the output is still float32. Again, this
>>>> seems against the expected numpy behaviour ?
>>> The latter is expected. As discussed previously here, 
>> Could you tell me where this was discussed, I think I missed it.
>
> http://projects.scipy.org/pipermail/numpy-discussion/2006-November/024402.html
Ok, thank you.
>
>> Thanks for the precision. Is this the expected behaviour for endianness, 
>> too ?
>
> What endianness behaviour?
This (assuming your machine is little endian):

In [1]: import numpy

In [2]: a = numpy.random.randn(3, 2)

In [3]: b = a.astype(a.dtype.newbyteorder('>'))

In [4]: a.clip(0., 1.).dtype.isnative
Out[4]: True

In [5]: b.clip(0., 1.).dtype.isnative
Out[5]: False

In [6]: b.clip(numpy.zeros(b.shape), 1.).dtype.isnative
Out[6]: True

If input is byteswapped, the output is also byteswapped for scalar 
min/max, but native if a native array is given as a min/max argument.

David