[f2py] F2PY issue with array arguments, results in unexpected sizes

[Michelle Gill]

Dear f2py users,

I am attempting to use f2py to speed up some python calculations and am
getting some unexpected results. I made two sample fortran functions to
illustrate the issue. These two functions compile fine with f2py. When
accessed within python, the respective returned rank-2 arrays are
different sized even though their sizes should be the same.

In the case of the first function, "test1", the size of the returned
array is 3x1001, which is as expected. In the case of the second function,
"test2", the size of the returned array is 3x1, which is not correct.

This issue has arisen in more complicated function that I'm attempting
to debug and I'd appreciate any assistance in solving it. My fortran
skills are a little rusty, so it's possible there is an issue with the
way the arrays are being passed or declared. I've tried searching this
mailing list, StackOverflow, and general web searches with no luck.

I'm using gfortran 4.7.3, python 2.7.5, numpy 1.8.0, all built with
MacPorts, on a Mac running OS X 10.9.1.

If I can provide further details, let me know.

Thanks,
Michelle Gill


### Contents of file test1.f90 ###

! -*- f90 -*-
subroutine test1(simdim, v)

integer, intent(in) :: simdim
! f2py integer, intent(in) :: simdim

real, dimension(3,simdim+1), intent(out):: v(3,simdim+1)
! f2py real, dimension(3,simdim+1), intent(out) :: v(3,simdim+1)

v = spread( (/ 45.2, 31.2, 41.2 /), 2, simdim+1)

print *, "v shape ", size(v)

return
end



### Contents of file test2.f90 ###

! -*- f90 -*-
subroutine test2(simdim, r, v)

integer, intent(in) :: simdim
! f2py integer, intent(in) :: simdim

real, dimension(simdim), intent(in) :: r(simdim)
! f2py real, dimension(simdim), intent(in) :: r(simdim)

real, dimension(3,simdim+1), intent(out):: v(3,simdim+1)
! f2py real, dimension(3,simdim+1), intent(out) :: v(3,simdim+1)

v = spread( (/ 45.2, 31.2, 41.2 /), 2, simdim+1)

print *, "r shape ", size(r)
print *, "v shape ", size(v)

return
end



### Compile each with f2py ###

f2py --verbose -m test1 -c test1.f90
f2py --verbose -m test2 -c test2.f90



### Compare the functions in IPython ###


In [1]: from test1 import test1

In [2]: from test2 import test2

In [3]: import numpy as np

In [4]: simdim = 1000

In [5]: r = np.random.random(simdim)

In [6]: print r.shape
(1000,)

In [7]: v1 = test1(simdim)
v shape 3003

In [8]: v2 = test2(simdim, r)
r shape 0
v shape 3

In [9]: print v1.shape, v2.shape
(3, 1001) (3, 1)

[David Ochoa]

Dear Michelle:

I've been using f2py lately and i think that you don't need to send the
size parameters to functions created with f2py (or at least that's how I'm
coding).

So it's just change the call:
v2 = test2(r)

And it should work (in my computer it does).

in test1 is different because you are not passing a array, only creating
and sending it back.

David




On Fri, Jan 10, 2014 at 2:27 PM, Michelle Gill <

Post by Michelle Gill
Dear f2py users,
I am attempting to use f2py to speed up some python calculations and am
getting some unexpected results. I made two sample fortran functions to
illustrate the issue. These two functions compile fine with f2py. When
accessed within python, the respective returned rank-2 arrays are
different sized even though their sizes should be the same.
In the case of the first function, "test1", the size of the returned
array is 3x1001, which is as expected. In the case of the second function,
"test2", the size of the returned array is 3x1, which is not correct.
This issue has arisen in more complicated function that I'm attempting
to debug and I'd appreciate any assistance in solving it. My fortran
skills are a little rusty, so it's possible there is an issue with the
way the arrays are being passed or declared. I've tried searching this
mailing list, StackOverflow, and general web searches with no luck.
I'm using gfortran 4.7.3, python 2.7.5, numpy 1.8.0, all built with
MacPorts, on a Mac running OS X 10.9.1.
If I can provide further details, let me know.
Thanks,
Michelle Gill
Contents of file test1.f90
! -*- f90 -*-
subroutine test1(simdim, v)
integer, intent(in) :: simdim
! f2py integer, intent(in) :: simdim
real, dimension(3,simdim+1), intent(out):: v(3,simdim+1)
! f2py real, dimension(3,simdim+1), intent(out) :: v(3,simdim+1)
v = spread( (/ 45.2, 31.2, 41.2 /), 2, simdim+1)
print *, "v shape ", size(v)
return
end
Contents of file test2.f90
! -*- f90 -*-
subroutine test2(simdim, r, v)
integer, intent(in) :: simdim
! f2py integer, intent(in) :: simdim
real, dimension(simdim), intent(in) :: r(simdim)
! f2py real, dimension(simdim), intent(in) :: r(simdim)
real, dimension(3,simdim+1), intent(out):: v(3,simdim+1)
! f2py real, dimension(3,simdim+1), intent(out) :: v(3,simdim+1)
v = spread( (/ 45.2, 31.2, 41.2 /), 2, simdim+1)
print *, "r shape ", size(r)
print *, "v shape ", size(v)
return
end
Compile each with f2py
f2py --verbose -m test1 -c test1.f90
f2py --verbose -m test2 -c test2.f90
Compare the functions in IPython
In [1]: from test1 import test1
In [2]: from test2 import test2
In [3]: import numpy as np
In [4]: simdim = 1000
In [5]: r = np.random.random(simdim)
In [6]: print r.shape
(1000,)
In [7]: v1 = test1(simdim)
v shape 3003
In [8]: v2 = test2(simdim, r)
r shape 0
v shape 3
In [9]: print v1.shape, v2.shape
(3, 1001) (3, 1)
_______________________________________________
f2py-users mailing list
http://cens.ioc.ee/mailman/listinfo/f2py-users

[Michelle Gill]

Dear David,

You were correct. I forgot the python function inputs are not
necessarily identical to those in the fortran function. Upon examining
them once the module was imported, this is obvious. This also fixes my
issue with the more complicated function I have been working on. I've
spent the better part of three days fighting with this issue, so I'm
extremely appreciative of your help. Thank you so much.

Kind regards,
Michelle Gill

Subject: Re: [f2py] F2PY issue with array arguments, results in
unexpected sizes
Content-Type: text/plain; charset="iso-8859-1"
I've been using f2py lately and i think that you don't need to send
the
size parameters to functions created with f2py (or at least that's how
I'm
coding).
v2 = test2(r)
And it should work (in my computer it does).
in test1 is different because you are not passing a array, only
creating
and sending it back.
David
On Fri, Jan 10, 2014 at 2:27 PM, Michelle Gill <

Post by Michelle Gill
Dear f2py users,
I am attempting to use f2py to speed up some python calculations and
am
getting some unexpected results. I made two sample fortran functions
to
illustrate the issue. These two functions compile fine with f2py.
When
accessed within python, the respective returned rank-2 arrays are
different sized even though their sizes should be the same.
In the case of the first function, "test1", the size of the returned
array is 3x1001, which is as expected. In the case of the second
function,
"test2", the size of the returned array is 3x1, which is not correct.
This issue has arisen in more complicated function that I'm
attempting
to debug and I'd appreciate any assistance in solving it. My fortran
skills are a little rusty, so it's possible there is an issue with
the
way the arrays are being passed or declared. I've tried searching
this
mailing list, StackOverflow, and general web searches with no luck.
I'm using gfortran 4.7.3, python 2.7.5, numpy 1.8.0, all built with
MacPorts, on a Mac running OS X 10.9.1.
If I can provide further details, let me know.
Thanks,
Michelle Gill
Contents of file test1.f90
! -*- f90 -*-
subroutine test1(simdim, v)
integer, intent(in) :: simdim
! f2py integer, intent(in) :: simdim
real, dimension(3,simdim+1), intent(out):: v(3,simdim+1)
! f2py real, dimension(3,simdim+1), intent(out) :: v(3,simdim+1)
v = spread( (/ 45.2, 31.2, 41.2 /), 2, simdim+1)
print *, "v shape ", size(v)
return
end
Contents of file test2.f90
! -*- f90 -*-
subroutine test2(simdim, r, v)
integer, intent(in) :: simdim
! f2py integer, intent(in) :: simdim
real, dimension(simdim), intent(in) :: r(simdim)
! f2py real, dimension(simdim), intent(in) :: r(simdim)
real, dimension(3,simdim+1), intent(out):: v(3,simdim+1)
! f2py real, dimension(3,simdim+1), intent(out) :: v(3,simdim+1)
v = spread( (/ 45.2, 31.2, 41.2 /), 2, simdim+1)
print *, "r shape ", size(r)
print *, "v shape ", size(v)
return
end
Compile each with f2py
f2py --verbose -m test1 -c test1.f90
f2py --verbose -m test2 -c test2.f90
Compare the functions in IPython
In [1]: from test1 import test1
In [2]: from test2 import test2
In [3]: import numpy as np
In [4]: simdim = 1000
In [5]: r = np.random.random(simdim)
In [6]: print r.shape
(1000,)
In [7]: v1 = test1(simdim)
v shape 3003
In [8]: v2 = test2(simdim, r)
r shape 0
v shape 3
In [9]: print v1.shape, v2.shape
(3, 1001) (3, 1)
_______________________________________________
f2py-users mailing list
http://cens.ioc.ee/mailman/listinfo/f2py-users

-------------- next part --------------
An HTML attachment was scrubbed...
http://cens.ioc.ee/pipermail/f2py-users/attachments/20140110/6970388b/attachment-0001.htm
------------------------------
_______________________________________________
f2py-users mailing list
http://cens.ioc.ee/mailman/listinfo/f2py-users
End of f2py-users Digest, Vol 97, Issue 2
*****************************************