We gathered information on contemporary FFT algorithms from several
sources, including FFTW 1.2, at http://theory.lcs.mit.edu/~fftw, a package
from the Air Force Institute of Technology (AFIT), an assembly code
library from NASA Ames, Intel's Math Kernel Library, FFTPACK, and some
work by Dr. Nikos P. Pitsianis, and the Prime Factor FFTs from the
Colorado School of Mines.  One idea that stood out was the factorization
of a vector into subsections which can be handled very efficiently by
highly optimized modules.  This approach was used in FFTW 1.2, and we used
the code generator from FFTW 1.2 as an initial start, with major structural
changes to its output (call structure changes and hand-optimizations) and 
then we rewrote the framework within which the codelets were called.
That is, the underlying modules were developed using the same algorithms 
found in the FFTW 1.2 codelets.  In our case, the algorithms were entirely 
re-implemented with a different structure and more parameters to meet our 
additional needs.