Installing FFTW

I have written an earlier Blog Entry regarding FFTW "Interesting Open-Source Application". This entry is how to compile FFTW to enable parallel computation.


We are assuming you have your intel Math Kernel Library installed. You can read this entry where to get the Free Non-commercial Intel Compiler Download


Step 1a: Configure for threads enablement and double-precision

# ./configure --enable-threads --enable-shared --prefix=/usr/local/fftw
# make
# make install
# make clean

Step 1b: Configure for threads enablement and single-precision

# ./configure --enable-threads --enable-shared --enable-float \
  --prefix=/usr/local/fftw
# make install 
# make clean

Step 1c: Long-Double precision

# ./configure --enable-threads --enable-shared --enable-long-double \
  --prefix=/usr/local/fftw
# make install 

* The configuration script is run 3 times because we need to build 3 sets of libraries is to create degree of floating point prevision.


Step 2: Make sure all your compute nodes have access to the FFTW and MKL.
If the compute nodes does not have access to the fftw library, there will be an error

"error while loading shared libraries: 
/opt/intel/mkl/10.0.4.023/lib/em64t/libmkl_intel_lp64.so: 
cannot open shared object file: No such file or directory"