The FFTPACK Fortran-library uses REAL variables, not DOUBLE PRECISION.
Since the rest of Tela uses double, you have to tell the Fortran compiler
to convert reals to double when compiling FFTPACK. On SGI-Iris and on some
other machines, the relevant option is f77 -r8. On IBM-AIX it is -qautodbl=dbl.
On Sun-Solaris-2.x it is not needed since it is the default. On Cray/UNICOS
it is irrelevant since everything is 64-bit anyway. Etc. etc.

Failing to take this into account will produce a incorrect answers from the
FFT function. When in doubt, check that the Tela demo 'Power law spectrum'
produces a nice noiselike plot and that the numerical values are in a
reasonable range. -PJ 2.11.1994.