Download Latest Version fftw++-2.05.tar.gz (188.1 kB) Get Updates
Home
Name Modified Size InfoDownloads / Week
README 2017-06-21 9.1 kB
fftw++-2.05.tar.gz 2017-06-21 188.1 kB
fftw++-2.04.tar.gz 2017-04-11 188.0 kB
fftw++-2.03.tar.gz 2016-07-24 187.2 kB
fftw++-2.02.tar.gz 2016-02-12 183.8 kB
fftw++-2.01.tar.gz 2016-02-06 183.6 kB
fftw++-2.00.tar.gz 2015-11-09 180.1 kB
fftw++-1.13.tar.gz 2012-10-28 105.0 kB
fftw++-1.12.tar.gz 2012-09-29 90.7 kB
fftw++-1.11.tar.gz 2012-08-29 77.3 kB
fftw++-1.10.tar.gz 2012-08-23 78.0 kB
fftw++-1.09.tar.gz 2010-12-20 144.3 kB
fftw++-1.08.tar.gz 2010-08-07 144.2 kB
fftw++-1.07.tar.gz 2010-05-24 143.7 kB
fftw++-1.06.tar.gz 2010-05-07 116.7 kB
Totals: 15 Items   2.0 MB 14
FFTW++: Fast Fourier Transform C++ Header/MPI Transpose for FFTW3 Library Copyright (C) 2004-17 John C. Bowman and Malcolm Roberts, University of Alberta http://fftwpp.sourceforge.net FFTW++ is a C++ header/MPI transpose for Version 3 of the highly optimized FFTW (http://www.fftw.org) Fourier Transform library. It provides a simple interface for 1D, 2D, and 3D complex-to-complex, real-to-complex, and complex-to-real Fast Fourier Transforms and convolutions. It takes care of the technical aspects of memory allocation, alignment, planning, wisdom, and communication on both serial and parallel (OpenMP/MPI) architectures. Wrappers for multiple 1D transforms are also provided. As with the FFTW3 library itself, both in-place and out-of-place transforms of arbitrary size are supported. Implicit dealiasing of standard and centered Hermitian convolutions is also implemented; in 2D and 3D implicit zero-padding substantially reduces memory usage and computation time. For more information, see "Efficient Dealiased Convolutions without Padding," by John C. Bowman and Malcolm Roberts, SIAM Journal on Scientific Computing, 33:1, 386-406 (2011). http://www.math.ualberta.ca/~bowman/publications/dealias.pdf "Parallel Implicitly Dealiased Convolutions on Shared Memory Architectures," by Malcolm Roberts and John C. Bowman, submitted to Journal of Scientific Computing, (2017). http://www.math.ualberta.ca/~bowman/publications/dealias2.pdf Convenient optional shift routines that place the Fourier origin in the logical center of the domain are provided for centered complex-to-real transforms in 2D and 3D; see fftw++.h for details. FFTW++ supports multithreaded transforms and convolutions. The global variable fftw::maxthreads specifies the maximum number of threads to use. The constructors invoke a short timing test to check that using multiple threads is actually beneficial for the given problem size. Multithreading requires linking with a multithreaded FFTW implementation and can be disabled by adding -DFFTWPP_SINGLE_THREAD to CFLAGS. FFTW++ can also exploit the high-performance Array class available at http://www.math.ualberta.ca/~bowman/Array (version 1.49 or higher), designed for scientific computing. The arrays in that package do memory bounds checking in debugging mode, but can be optimized by specifying the -DNDEBUG compilation option (1D arrays optimize completely to pointer operations). Detailed documentation is provided before each class in the fftw++.h header file. The included examples illustrate how easy it is to use FFTW in C++ with the FFTW++ header class. Use of the Array class is optional, but encouraged. If for some reason the Array class is not used, memory should be allocated with ComplexAlign (or doubleAlign) to ensure that the data is optimally aligned to sizeof(Complex), to enable the SIMD extensions. The optional alignment check in fftw++.h can be disabled with the -DNO_CHECK_ALIGN compiler option. ########################## MPI ########################## Hybrid OpenMP/MPI versions of the convolution routines in 2 and 3 dimensions are available in the mpi/ directory. Parallelization is accomplished using the adaptive hybrid OpenMP/MPI transpose routine described in "Adaptive Matrix Transpose Algorithms for Distributed Multicore Processors", John C. Bowman and Malcolm Roberts, Interdisciplinary Topics in Applied Mathematics, Modeling and Computational Science, Springer Proceedings in Mathematics & Statistics 117, 97-103 (2015): http://www.math.ualberta.ca/~bowman/publications/transpose.pdf Either a 1D ("slab") and 2D ("pencil") data decomposition is used for the three-dimensional convolutions, depending on the number of processors. mpi/fftw/ contains comparison code using FFTW's parallel MPI transform and explicit padding. ########################## Examples ########################## The following programs are provided in the examples directory: 1D examples using ComplexAlign allocator: example0.cc example0r.cc 1D examples using Array class: example1.cc example1r.cc 2D examples using Array class: example2.cc example2r.cc 3D examples using Array class: example3.cc example3r.cc Examples of implicitly dealiased convolutions on complex non-centered data in 1, 2, and 3 dimensions: examplecconv.cc, examplecconv2.cc, examplecconv3.cc Examples of implicitly dealiased convolutions on complex Hermitian-symmetric centered data in 1, 2, and 3 dimensions: exampleconv.cc, exampleconv2.cc, exampleconv3.cc Local transpose (in-place or out-of-place): exampletranspose.cc More general types of convolutions (for example, autoconvolutions) can be performed by defining a custom multiplier or realmultiplier function pointer. ########################## Wrappers ########################## Wrappers for the convolution routines are available for C, Fortran, and Python. Examples are given in the wrappers/ directory. The C wrapper may be found in cfftw++.h and cfftw++.cc, the Fortran wrapper in fftwpp.f90, and the Python wrapper in fftwpp.py. A unit-testing script, test.py, is also available. Results for the given input data are checked with a simple hash. Compilation uses the environment variables CPLUS_INCLUDE_PATH to tell the compiler where to find fftw3.h, and FORTRAN_INCLUDE_PATH to indicate to the compiler the location of fftw3.f03 from FFTW. The following programs are available in the wrappers directory: Using C to call multi-threaded 1D, 2D, and 3D binary convolutions and 1D and 2D ternary convolutions, with and without passing work arrays, where the operation in physical space may correspond to either a scalar multiplication (M=1) or a dot product (M > 1): cexample.c Using Fortran to call multi-threaded 1D, 2D, and 3D binary convolutions, with and without passing work arrays, where the operation in physical space may correspond to either a scalar multiplication (mm=1) or a dot product (mm > 1): fexample.f90 Using Python to call multi-threaded 1D, 2D, and 3D binary convolutions (for scalar multiplication (M=1) and with work arrays created by the constructor): pexample.py ########################## MPI ########################## Hybrid OpenMP/MPI versions of the convolution routines in 2 and 3 dimensions are available in the mpi directory. cconv2.cc and cconv3.cc are examples of two- and three-dimensional complex non-centered convolutions. conv2.cc and conv3.cc are examples of two- and three-dimensional Hermitian-symmetric complex centered convolutions. fft2.cc and fft2r are examples of two-dimensional hybrid MPI/OpenMP FFTs using a 1D data decomposition, for complex and real data, respectively. fft3.cc and fft3r are examples of three-dimensional hybrid MPI/OpenMP FFTs using a 1D (slab) or 2D (pencil) data decomposition (depending on the number of MPI processes), for complex and real data, respectively. timing.py is a script which performs timing tests for mpi-based convolutions. The directory mpi/explicit contains comparison code using FFTW's parallel MPI transform and explicit padding. ########################## Test Programs ########################## The following programs are provided in tests/, along with various timing and error analysis scripts. Asymptote (http://asymptote.sourceforge.net/) scripts are provided for visualizing the output. 1D complex convolution test: cconv.cc 1D Hermitian convolution test: conv.cc 1D Hermitian ternary convolution test: tconv.cc 2D complex convolution test: cconv2.cc 2D Hermitian convolution test: conv2.cc 2D Hermitian ternary convolution test: tconv2.cc 3D complex convolution test: cconv3.cc 3D Hermitian convolution test: conv3.cc 1D FFT: fft1.cc 1D real FFT: fft1r.cc 1D multiple FFT: mfft1.cc 1D multiple real FFT: mfft1r.cc 2D FFT: fft2.cc 2D real FFT: fft2r.cc 3D FFT: fft3.cc 3D real FFT: fft3r.cc ######################## Availability and License ######################## To compile from Git developmental source code: git clone https://github.com/dealias/fftwpp All source files in the FFTW++ project, unless explicitly noted otherwise, are released under version 3 (or later) of the GNU Lesser General Public License (see the files LICENSE.LESSER and LICENSE in the top-level source directory). ======================================================================== This program is free software; you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License as published by the Free Software Foundation; either version 3 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more details. You should have received a copy of the GNU Lesser General Public License along with this program; if not, write to the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. ========================================================================
Source: README, updated 2017-06-21