Data exchange - Python and Fortran

Question

We are developing a scientific application which has the interface in python 2.7 and the computation routines written in Intel Visual Fortran. Reading the source files is done using python, then only the required data for computations has to be passed to standalone Fortran algorithms. Once the computations done, the data has to be read by python once again.

Using formatted text files seems to be taking too long and not efficient. Further, we would like to have a standard intermediate format. There can be about 20 arrays and those are huge (if written to formatted text, the file is about 500 MB).

Q1. In a similar situation where Python and Fortran data exchange is necessary. What would be recommended way of interaction? (e.g.: writing an intermediate data to be read by the other or calling Fortran from within Python or using numpy to create compatible arrays or etc.)

Q2. If writing intermediate structures is recommended, What format is good for data exchange? (We came across CDF, NETCdf, binary streaming, but didn't try any so far.)

For data, I prefer HDF5. Good performance, compression, library bindings for C, C++, Fortran, Python, Java, etc. — casey
– casey, Commented Jun 10, 2015 at 1:41
Also note that netCDF4 uses an enhanced version of HDF5 as the storage layer. Python and Fortran have good libraries to handle netCDF4. — N1B4
– N1B4, Commented Jun 10, 2015 at 20:06

rth · Accepted Answer · 2015-06-09 16:05:55Z

7

The standard way of wrapping Fortran code in Python is with f2py (included in the numpy module).

For the output of intermediary results, a number of formats could work, it really depends on your requirements.

For simple datasets, from python, just use numpy.save.
If your datasets become large, HDF5 with, for instance, PyTables in Python and libhdf5 in Fortran could be used.
Otherwise, if you don't want to link your code to an external library, custom binary files written from Fortran and parsed with numpy could work too.

edited Jun 9, 2015 at 16:05

answered Jun 9, 2015 at 15:48

rth

11.3k7 gold badges58 silver badges88 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

py_works Over a year ago

I decided to go with hdf5 format because of its self describing, fast and I managed to configure it on python and FORTRAN. Thank you!

bdforbes · Accepted Answer · 2015-06-10 06:30:53Z

3

I would interface directly between Python and Fortran. It is relatively straightforward to allocate memory using Numpy and pass a pointer through to Fortran. You use iso_c_binding to write C-compatible wrapper routines for your Fortran routines and ctypes to load the Fortran .dll and call the wrappers. If you're interested I can throw together a simple example (but I am busy right this moment).

answered Jun 10, 2015 at 6:30

bdforbes

1,5541 gold badge14 silver badges34 bronze badges

Collectives™ on Stack Overflow

Data exchange - Python and Fortran

2 Answers 2

1 Comment

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

1 Comment

Comments

Your Answer

Sign up or log in

Post as a guest

Related