Running dist_fft_test using MPICH

It appears that LAM MPI works better on the Altivec cluster; follow the instructions in Sec. A.1 instead. Here are the steps to compile and make dist_fft_test on altivec from the CVS directory c/dist_fft:

• Make sure your .bash_profile file in your home directory has the following lines:
  export PATH=/usr/local/mpich-1.2.7/ch-p4/bin:$PATH
export P4_RSHCOMMAND=rsh
  These might be in addition to your other lines needed for CVS of
  export CVS_RSH=ssh
export CVSROOT=':ext:diffmic@xray1.physics.sunysb.edu:/home/xray1/diffmic/'
  The first time you enter this in you'll have to type . .bash_profile to make it active; it will be there automatically for subsequent logins.

• Do cp Makefile.altivec Makefile to use the correct makefile. You also need to have the file mach in the directory but this should be there automatically from CVS.

• Type export CC=mpicc, and make dist_fft_test.

• Now you're ready to run! When running the program, you need to provide the $\log_{2}$ dimensionality of the calculation. The calculation will be run with this same number of points in 1D, 2D, and 3D, so if you specify 12 you will get a 1D FFT of size $2^{12}=4096$, a 2D FFT of size $2^{6}\times 2^{6}=64\times 64$, and a 3D FFT of size $2^{4}\times 2^{4}\times 2^{4}=16\times 16\times 16$. You can also specify the number of processors with the -np flag: mpirun -np 2 -machinefile mach dist_fft_test 12

• To change to Non-Interleaving mode or Double precision you must modify the file dist_fft_types.h by modifying the following lines:
  // the following options are off by default
// #define DIST_FFT_USE_INTERLEAVED_COMPLEX
// #define DIST_FFT_USE_DOUBLE