FORD: 3-D FFT with FFTW MPI

Purpose

• "Ford" provides a test example of the 3-D fast Fourier transform using FFTW on a distributed system. The 3-D transform is built on a tensor product of 2-D transforms. Functions values are given as input in the real space, transformed into the frequency space (fft), and finally transformed back to the real space (inverse fft).

Specifications

• Name: Ford.
• Author: Feng Chen.
• Finishing date: 05/30/2009.
• Languages: MPI C++.
• Required libraries: FFTW (version 2.1.5 or above).

Simple Example

• Parameters: \begin{equation} N_x = N_y = N_z = 16, \quad N_{\text{cpu}}=2 \times 8. \end{equation}

Quick Start

• Compiling and running:

  cd ./Ford
  make
  mpirun -np $NPROCS a.out
  

• Output:

  The difference is: 4.1359e-25
  The working time is: 0.0540702
  ...
  

• CPU: Intel(R) Xeon(R) CPU X5630 @2.53GHz.
• OS: CentOS release 6.4 (Final).
• Compiler: mpicc/gcc 4.5.1.

References

• Fastest Fourier transform in the West (FFTW). Link.

Code Highlights

MPI_Alltoall(&temp_B[0][0], nx, type_send, &temp_trans[0][0], 1, 
 type_recv, MPI_COMM_WORLD);
fftw(p_1d_c2c,nx,&temp_trans[0][0],nx,1,NULL,nx,1);