[petsc-users] How to use f2py on a PETSc/SLEPc-based fortran code

Matthew Knepley knepley at gmail.com
Wed Mar 29 18:48:10 CDT 2017


On Wed, Mar 29, 2017 at 5:12 PM, Austin Herrema <aherrema at iastate.edu>
wrote:

> Quick update on this issue in case it brings any other thoughts/ideas to
> light. For a very simple, small problem, I am successfully able to use
> MatSeqAIJSetPreallocation in a fortran-based code compiled for python via
> f2py. I am still unsure why, in a larger code, this particular function
> call fails when the code is executed in python (on a setup that runs fine
> under pure Fortran). Does the error " nnz cannot be greater than row
> length: local row 2 value 1330400321 rowlength 37065" imply that the
> program thinks I am trying to allocate 1330400321 nonzeros in a row of
> max length 37065?
>

Yes.

  Thanks,

    Matt


> That is obviously not my intent nor what I think I have coded. I am trying
> to skip preallocation and use merely MatSetUp but, as we would expect, the
> dynamic allocation is ridiculously slow...
>
> On Wed, Mar 29, 2017 at 11:27 AM, Austin Herrema <aherrema at iastate.edu>
> wrote:
>
>> Got it--just had to link against other compiled source, as you said. I've
>> attached my makefile for doing everything (including variable definitions,
>> compiling source, and running requisite f2py commands) in case that's
>> helpful for anyone else trying to do something similar. But obviously the
>> meat of it is in what Gaetan provided.
>>
>> I am now able to successfully run simple PETSc-based fortran codes in
>> python. For a larger, more complex code, I am getting some PETSc errors
>> when running in python that I don't normally get. In particular,
>> preallocation is failing--the relevant fortran code block and PETSc error
>> is below.
>>
>>
>>     call MatCreate(PETSC_COMM_WORLD, LHS_pc, pc_ier)
>>     call MatSetSizes(LHS_pc, PETSC_DECIDE, PETSC_DECIDE, NSD*FUN%NNODE,
>> NSD*FUN%NNODE, pc_ier)
>>     call MatSetFromOptions(LHS_pc, pc_ier)
>>     call MatSeqAIJSetPreallocation(LHS_pc, 500, PETSC_NULL_INTEGER,
>> pc_ier)
>>
>>
>> [0]PETSC ERROR: --------------------- Error Message
>> --------------------------------------------------------------
>> [0]PETSC ERROR: Argument out of range
>> [0]PETSC ERROR: nnz cannot be greater than row length: local row 2 value
>> 1330400321 rowlength 37065
>> [0]PETSC ERROR: See http://www.mcs.anl.gov/petsc/documentation/faq.html
>> for trouble shooting.
>> [0]PETSC ERROR: Petsc Release Version 3.7.5, Jan, 01, 2017
>> [0]PETSC ERROR: Unknown Name on a real named
>> austin-ethernet.student.iastate.edu by Austin Wed Mar 29 10:59:33 2017
>> [0]PETSC ERROR: Configure options CC=/usr/local/bin/mpicc
>> CXX=/usr/local/bin/mpicxx F77=/usr/local/bin/mpif77
>> FC=/usr/local/bin/mpif90 --with-shared-libraries=1 --with-pthread=0
>> --with-openmp=0 --with-debugging=1 --with-ssl=0
>> --with-superlu_dist-include=/usr/local/opt/superlu_dist/include
>> --with-superlu_dist-lib="-L/usr/local/opt/superlu_dist/lib
>> -lsuperlu_dist" --with-fftw-dir=/usr/local/opt/fftw
>> --with-netcdf-dir=/usr/local/opt/netcdf --with-suitesparse-dir=/usr/local/opt/suite-sparse
>> --with-hdf5-dir=/usr/local/opt/hdf5 --with-metis-dir=/usr/local/opt/metis
>> --with-parmetis-dir=/usr/local/opt/parmetis
>> --with-scalapack-dir=/usr/local/opt/scalapack
>> --with-mumps-dir=/usr/local/opt/mumps/libexec --with-x=0
>> --prefix=/usr/local/Cellar/petsc/3.7.5/real --with-scalar-type=real
>> --with-hypre-dir=/usr/local/opt/hypre --with-sundials-dir=/usr/local/opt/sundials
>> --with-hwloc-dir=/usr/local/opt/hwloc
>> [0]PETSC ERROR: #1 MatSeqAIJSetPreallocation_SeqAIJ() line 3598 in
>> /private/tmp/petsc-20170223-508-1xeniyc/petsc-3.7.5/src/mat/
>> impls/aij/seq/aij.c
>> [0]PETSC ERROR: #2 MatSeqAIJSetPreallocation() line 3570 in
>> /private/tmp/petsc-20170223-508-1xeniyc/petsc-3.7.5/src/mat/
>> impls/aij/seq/aij.c
>>
>>
>> Is there anything about the MatSeqAIJSetPreallocation function that would
>> make it not work correctly in Python even though everything else seems to
>> work properly? If anyone has thoughts on this that would be great. But,
>> again, I do realize I'm venturing into potentially unsupported territory.
>>
>>
>> On Tue, Mar 28, 2017 at 4:53 PM, Gaetan Kenway <gaetank at gmail.com> wrote:
>>
>>> Looks like it isn't finding your source from run_analysis.f90. You still
>>> need to compile that yourself and include in the final link. In my example,
>>> all the "original" source code was precompiled into a library from a
>>> different makefile and then this was run after-the-fact.
>>>
>>> Gaetan
>>>
>>> On Tue, Mar 28, 2017 at 2:38 PM, Austin Herrema <aherrema at iastate.edu>
>>> wrote:
>>>
>>>> Gotcha. In that case, it seems I should be good without that line. I've
>>>> gotten the compile to succeed, but upon attempting to import the module I
>>>> get the following:
>>>>
>>>> >>> import run_analysis_final
>>>> Traceback (most recent call last):
>>>>   File "<stdin>", line 1, in <module>
>>>> ImportError: dlopen(./run_analysis_final.so, 2): Symbol not found:
>>>> _run_analysis_
>>>>   Referenced from: ./run_analysis_final.so
>>>>   Expected in: flat namespace
>>>>  in ./run_analysis_final.so
>>>>
>>>> Seems I may have gotten the linking wrong somehow. Will keep searching,
>>>> but the simplified makefile that I used is attached in case anyone thinks
>>>> they might be able to spot the issue in it. That said, I do realize that
>>>> this may be starting to reach beyond the scope of this mailing list so feel
>>>> free to ignore...
>>>>
>>>> On Tue, Mar 28, 2017 at 2:31 PM, Gaetan Kenway <gaetank at gmail.com>
>>>> wrote:
>>>>
>>>>> You only get that file if you have wrapped a module explicitly in the
>>>>> .pyf file. If you haven't wrapped a module, that doesn't get created.
>>>>>
>>>>> Gaetan
>>>>>
>>>>> On Tue, Mar 28, 2017 at 12:28 PM, Austin Herrema <aherrema at iastate.edu
>>>>> > wrote:
>>>>>
>>>>>> Gaetan,
>>>>>>
>>>>>> Thank you for this. With your help, I think I am getting close to
>>>>>> getting this to work for my case. At the moment, I am hung up on the line
>>>>>> of your makefile which reads "$(FF90) $(FF90_ALL_FLAGS) -I$(MAIN_DIR)/mod
>>>>>> -c warpustruct-f2pywrappers2.f90". Am I correct that
>>>>>> warpustruct-f2pywrappers2.f90 should be generated by f2py? If so, do you
>>>>>> (or does anyone else) know the command for telling f2py to do so? At the
>>>>>> moment I am using:
>>>>>>
>>>>>> f2py run_analysis.f90 -m run_analysis -h run_analysis.pyf
>>>>>>
>>>>>> to get the requisite .pyf and .c files, but no .f90 file. If I am
>>>>>> wrong about the origin of this file, please do tell me!
>>>>>>
>>>>>> Thank you,
>>>>>> Austin
>>>>>>
>>>>>> On Mon, Mar 27, 2017 at 5:13 PM, Gaetan Kenway <gaetank at gmail.com>
>>>>>> wrote:
>>>>>>
>>>>>>> Austin
>>>>>>>
>>>>>>> Here is the full makefile for a code we use. The variables defined
>>>>>>> externally in a separate config file are:
>>>>>>> $(FF90)
>>>>>>> $(FF90_FLAGS)
>>>>>>> $(LIBDIR)
>>>>>>> $(PETSC_LINKER_FLAGS)
>>>>>>> $(LINKER_FLAGS)
>>>>>>> $(CGNS_LINKER_FLAGS)
>>>>>>>
>>>>>>> $(PYTHON)
>>>>>>> $(PYTHON-CONIFG)
>>>>>>> $(F2PY)
>>>>>>> (These are usually use python, python-config and f2py. You can
>>>>>>> overwrite as necessary)
>>>>>>>
>>>>>>> $(CC)
>>>>>>> $(CC_ALL_FLAGS)
>>>>>>>
>>>>>>> This essentially just mimics what f2py does automatically but we
>>>>>>> found it easier to control exactly what is going on. Essentially you are
>>>>>>> just compiling exactly as you normally an executable, but instead make a
>>>>>>> .so (with the -shared option) and including the additional .o files
>>>>>>> generated by compiling the f2py-generated wrappers.
>>>>>>>
>>>>>>> Hope this helps,
>>>>>>> Gaetan
>>>>>>>
>>>>>>> On Sat, Mar 25, 2017 at 5:38 AM, Lisandro Dalcin <dalcinl at gmail.com>
>>>>>>> wrote:
>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> On 22 March 2017 at 20:29, Barry Smith <bsmith at mcs.anl.gov> wrote:
>>>>>>>>
>>>>>>>>>
>>>>>>>>>   Lisandro,
>>>>>>>>>
>>>>>>>>>     We've had a couple questions similar to this with f2py; is
>>>>>>>>> there a way we could add to the PETSc/SLEPc makefile rules something to
>>>>>>>>> allow people to trivially use f2py without having to make their own (often
>>>>>>>>> incorrect) manual command lines?
>>>>>>>>>
>>>>>>>>>    Thanks
>>>>>>>>>
>>>>>>>>>
>>>>>>>> Barry, it is quite hard and hacky to get f2py working in the
>>>>>>>> general case. I think the email from Gaetan in this thread proves my point.
>>>>>>>>
>>>>>>>> IMHO, it is easier to write a small Fortran source exposing the API
>>>>>>>> to call using ISO_C_BINDINGS, then wrap that code with the more traditional
>>>>>>>> C-based "static" tools (SWIG, Cython) or even "dynamically" with ctypes or
>>>>>>>> cffi (which use dlopen'ing).
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> --
>>>>>>>> Lisandro Dalcin
>>>>>>>> ============
>>>>>>>> Research Scientist
>>>>>>>> Computer, Electrical and Mathematical Sciences & Engineering (CEMSE)
>>>>>>>> Extreme Computing Research Center (ECRC)
>>>>>>>> King Abdullah University of Science and Technology (KAUST)
>>>>>>>> http://ecrc.kaust.edu.sa/
>>>>>>>>
>>>>>>>> 4700 King Abdullah University of Science and Technology
>>>>>>>> al-Khawarizmi Bldg (Bldg 1), Office # 0109
>>>>>>>> Thuwal 23955-6900, Kingdom of Saudi Arabia
>>>>>>>> http://www.kaust.edu.sa
>>>>>>>>
>>>>>>>> Office Phone: +966 12 808-0459 <+966%2012%20808%200459>
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> *Austin Herrema*
>>>>>> PhD Student | Graduate Research Assistant | Iowa State University
>>>>>> Wind Energy Science, Engineering, and Policy | Mechanical Engineering
>>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>> --
>>>> *Austin Herrema*
>>>> PhD Student | Graduate Research Assistant | Iowa State University
>>>> Wind Energy Science, Engineering, and Policy | Mechanical Engineering
>>>>
>>>
>>>
>>
>>
>> --
>> *Austin Herrema*
>> PhD Student | Graduate Research Assistant | Iowa State University
>> Wind Energy Science, Engineering, and Policy | Mechanical Engineering
>>
>
>
>
> --
> *Austin Herrema*
> PhD Student | Graduate Research Assistant | Iowa State University
> Wind Energy Science, Engineering, and Policy | Mechanical Engineering
>



-- 
What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which their
experiments lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20170329/fff8333e/attachment.html>


More information about the petsc-users mailing list