Re: [f2py] F2PY: Unable to prevent Fortran ordered array copy

On Thu, Nov 19, 2009 at 3:07 PM, Robin <[email protected]> wrote:
> On Thu, Nov 19, 2009 at 2:50 PM, Ravi Kanda <[email protected]> wrote:
>> I have been using Numpy+f2py for a couple of years, and the combination
>>  has worked well for me for many types of problems.  However, I have
>> never been able to prevent F2PY from copying arrays even though I
>> initialize them as Fortran ordered in the calling python script.  I have
>> spent a considerable time looking in the archives (both F2PY & NUMPY)
>> and found several leads to diagnosing array-copy problems (see results
>> below), but none of them seem to work for me.
>
> I didn't have time yet to read through the details of what you've
> tried - but thought I'd point out Fortran vs C order is only one of
> the things that will trigger and f2py copy. A copy is also created if
> a cast is required between the dtype of the numpy array and the
> fortran type. (I found this out the hard way moving integer code
> between 32 bit and 64 bit platforms - I couldn't understand why it was
> slower on 64 bit, but it was because the cast was causing a copy).
>
> In your code it looks like you are using real(kind=4) whereas numpy
> default is 64 bit double I think (real(kind=8)) so I suspect this is
> where the copies are coming from. Try passing numpy arrays with
> float32 dtype, or change fortran types...

I just saw in your Python script you are using float32 - so sorry for
the noise... I don't have any other ideas!

Cheers

Robin