PFFT is a software library for computing massively parallel, fast Fourier transformations on distributed memory architectures. PFFT can be understood as a generalization of FFTW-MPI to multidimensional data decomposition. The library is written in C and MPI. A Fortran interface is also available. Support for hybrid parallelization based on OpenMP and MPI is under development.

At first, you need an install of FFTW-3.3 with enabled MPI support. We highly recommend to install the latest release of FFTW since the MPI code got several bugfixes in the last releases. Since some fixes are still pending until release FFTW-3.3.5. we offer some helpful scripts for installing FFTW-3.3.4 together with two patches (one bugfix and one performance improvement) at our sofware page

https://www-user.tu-chemnitz.de/~potts/workgroup/pippig/software.php#scripts

The install of PFFT follows the typical steps:

./bootstrap.sh
./configure
make
make install

Optionally, a bunch of test programs can be built with:

make check

The bootstrap step can be skipped if you delivered a PFFT tarball, i.e., the file configure was already generated. Make sure that configure can find a working install of FFTW-3.3 with MPI support, i.e., set:

CPPFLAGS=$PATHTOFFTW/include

and:

LDFLAGS=$PATHTOFFTW/lib64 or LDFLAGS=$PATHTOFFTW/lib

depending on your system architecture.

PFFT tarballs include a detailed user guide at:

doc/manual.pdf

If you have cloned the PFFT repository, the user manual is automatically built during make. Of course this requires a working LaTeX enviroment. Alternatively, you can download a recent version of the user manual at

http://www.tu-chemnitz.de/~potts/workgroup/pippig/software.php.en

Note that using PFFT is very similar to FFTW. The interface is as close to the fftw_mpi interface as possible. Therefore, it is a good start to read FFTW manual:

http://www.fftw.org/fftw3_doc

At least you should understand how FFTW deals with distributed memory FFTs:

http://www.fftw.org/fftw3_doc/Distributed_002dmemory-FFTW-with-MPI.html#Distributed_002dmemory-FFTW-with-MPI

Next, you can have a look at the test programs in directory 'tests' to learn the details of the PFFT interface.

For installation instructions, you can also refer to the file INSTALL in this directory.

For an theoretical introduction, please read the paper

''PFFT - An Extension of FFTW to Massively Parallel Architectures''

available at

http://www.tu-chemnitz.de/~potts/workgroup/pippig/publikationen.php.en

This is the most current general paper, and the one that we recommend if you wish to cite PFFT.

A python interface is available at a distinct repository

https://github.com/rainwoodman/pfft-python

Many thanks to Yu Feng for his great work.

Your comments are welcome! This is the first version of the library and may not be as robust or well documented as it should be. Please keep track of bugs or missing/confusing instructions and report them to

Michael Pippig <[email protected]>

If you find PFFT useful, we would be delighted to hear about what application you are using PFFT for!