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

Hi,

Your scripts seemed to work for me with no copies reported. Being on a
Mac I had to disable -DREPORT_AT_EXIT and numpy.testing.memusage() is
unavailable so I didn't get those figures... but the copy warning
works and I didn't get it.

I am using gfortran 4.2, f2py 2_7618, numpy 1.4.0.dev7618 (mac)

You are using quite an old version of numpy/f2py, so it would probably
worth upgrading, at least to the latest release to see if the problem
persists.

Robn

On Thu, Nov 19, 2009 at 10:01 PM, Ravi Kanda <[email protected]> wrote:
> Thank you, Robin & Labrach, for your comments.
>
> The problem I am experiencing boils down to why the F2PY-module is
> unable to identify a legal NUMPY array to be in the correct order, or
> even recognize legal Numpy array objects!  And the problem persists
> irrespective of the F90 compiler I use (Intel, PG, or GNU).  The crux of
> this problem may be illustrated with the following two examples from the
> python interpreter ('pymm' is the f2py-wrapped module):
> --------------------------------------------------------
>  >>> # Example 1. ROW-MAJOR-ORDERED ARRAY, 'a':
> ...
>  >>> a = reshape(arange(15),(3,5))
>  >>> a
> array([[ 0,  1,  2,  3,  4],
>          [ 5,  6,  7,  8,  9],
>          [10, 11, 12, 13, 14]])
>  >>> isfortran(a)
> False                                   --------->  OK
>  >>> a.flags['F_CONTIGUOUS']
> False                                   --------->  OK
>  >>>
>  >>> pymm.has_column_major_storage(a)
> 0                                       --------->  OK
>  >>> a = pymm.as_column_major_storage(a)
> Traceback (most recent call last):
>     File "<stdin>", line 1, in <module>
> pymm.error: expected array object       --------->  (!)
>  >>>
> --------------------------------------------------------
> ...
> --------------------------------------------------------
>  >>> # Example 2. COLUMN-MAJOR-ORDERED ARRAY, 'b':
> ...
>  >>> b = reshape(arange(15),(3,5),order='f')
>  >>> b
> array([[ 0,  3,  6,  9, 12],
>        [ 1,  4,  7, 10, 13],
>        [ 2,  5,  8, 11, 14]])
>  >>> isfortran(b)
> True                                    --------->  OK
>  >>> b.flags['F_CONTIGUOUS']
> True                                    --------->  OK
>  >>> pymm.has_column_major_storage(b)
> 0                                       --------->  (!)
>  >>> c = pymm.as_column_major_storage(b)
> Traceback (most recent call last):
>   File "<stdin>", line 1, in <module>
> pymm.error: expected array object       --------->  (!)
>  >>>
> --------------------------------------------------------
>
> So, 'pymm' does not seem to recognize the legal NUMPY arrays, 'a' OR
> 'b', as valid array objects!  As a result, the NUMPY array seems to be
> cast into a 'suitable' form by F2PY resulting in array copying when I
> invoke the function 'pymm.mmfunc.mm' [if compiled with
> 'intent(in,out,overwrite)' OR 'intent(in,out)' OR
> 'intent(in,out,overwrite,c)', and passing the appropriate
> F-/C-CONTIGUOUS arrays]:
> "copied an array using PyArray_CopyFromObject: size=1440000, elsize=4
>  copied an array using PyArray_CopyFromObject: size=1440000, elsize=4
>  copied an array using PyArray_CopyFromObject: size=1440000, elsize=4"
>
> --------------------------------------------------------
> However, when I compile with 'intent(inout)', OR 'intent(inplace)', OR
> 'intent(inout,c)' with appropriate F-/C-CONTIGUOUS arrays, then I get
> the following error:
> "pymm.error: failed in converting 4th argument `a' of pymm.mmfunc.mm to
> C/Fortran array"
>
> --------------------------------------------------------
> So, I want to know whether:
>
> 1) This could be due to a bug in the F2PY version I am using
> [2.45.241_1926] that was later fixed?  Or a problem with my installation
> of F2PY itself.
>
> 2) I am using a wrong flag (or not using a specific flag) during F2PY
> compilation (please see CSH & LOG files in TGZ below)?
>
> 3) I am making some subtle mistake while calling the F2PY-module (please
> see 'CALL_MM.PY' & 'MM.F90' file in TGZ)?
>
> Again, all the relevant text files (7 in all) are available at:
> ftp://ftp.gps.caltech.edu/pub/rkanda/MemoryDeallocTest.tgz
>
> I don't know where else to look to try to resolve this problem.  So, it
> will be much appreciated if someone can take a look at the F90, PY, PYF,
> CSH, & LOG files, or actually compile/run the module to see whether  I'm
> experiencing a "unique" problem here.
>
> Thank you very much for your time!
> Ravi.
> --------------------------------------------
>
> labrach wrote:
>> It seems ok for me with the followings :
>>
>> (but I don't know how to check if the copy appeared or not, since the
>> -DF2PY_REPORT_ATEXIT returns an error dealing with undefine reference
>> to 'on_exit')
>>
>> I run that on a Windows with mingw 5.1.4
>>
> <SNIP>
>
> _______________________________________________
> f2py-users mailing list
> f2py-users-Y4l6ocDipWCuvFJfX82//[email protected]
> http://cens.ioc.ee/mailman/listinfo/f2py-users
>