D. J. Bernstein
Fast arithmetic

How to install djbfft

Like any other piece of software (and information generally), djbfft comes with NO WARRANTY.

System requirements

The djbfft installation procedure assumes that you have a UNIX system with gcc.


Download the djbfft package. The latest published djbfft package is djbfft-0.76.tar.gz.

Unpack the djbfft package:

     gunzip djbfft-0.76.tar
     tar -xf djbfft-0.76.tar
     cd djbfft-0.76

Decide where to install djbfft. The default installation directory is /usr/local/djbfft. To change this directory, edit conf-home now.

Compile the djbfft library and test programs:


As root, install the library and include files:

     make setup check
Now you can use djbfft in your programs by compiling with -I/usr/local/djbfft/include and linking with /usr/local/djbfft/lib/djbfft.a.


Check the rms errors for some convolutions of pseudorandom data:
Single-precision (r4 and c4) errors should be on the scale of 10^-8. Double-precision (r8 and c8) errors should be on the scale of 10^-16. If you see any double-precision results larger than 10^-12, please let me know.

Check the rms errors for a spanning set of transforms:

Note that this test takes a while to run.

Time the library:

The result is a detailed speed report.

To report success:

     ( echo 'First M. Last'; cat `cat SYSDEPS` ) \
     | mail djb-qst@cr.yp.to
Replace First M. Last with your name.

Things to do if you have more time

If you're interested in trying to extract the best possible speed out of djbfft, you can edit conf-cc and conf-ld to select a compiler, and conf-opt to select machine-specific optimizations; then recompile.