results with strange floating-point number bias

Hello

I'm trying to implement a FFT function with the new fftw3 library. I'm
creating a rfft fucntion first, using fftw's special real-signal
function first.

I got  the  function  running  well,  but  there  is the strangest thing
happening.   The   result  it  returns  is  summed  by  a  large  number
(455103.06358  once,  but  it  changes).

And it's only for real numbers! The imaginary part is OK

I bet it has something to do with the way octave stores the numbers, but
as  I  have  not  messed  with  octave's  code  enough,  I'm looking for
advices..

It  REALLY  looks  like  it's  mixing a pointer value with the numerical
value of the real part... But what is strange is that I could extract the
correct values once, doing a subtraction.

any ideas?

Here is a sample of what is happening:

oct:50> fft([1 0.5 0.3 -1]')
ans =

  0.80000 + 0.00000i
  0.70000 - 1.50000i
  1.80000 + 0.00000i
  0.70000 + 1.50000i
	
oct:51> fftnic([1 0.5 0.3 -1]')
ans =

  4.5509e+05 + 0.0000e+00i
  4.5509e+05 - 1.5000e+00i
  4.5509e+05 + 0.0000e+00i

oct:68> fftnic([1 0.5 0.3 -1]') - 455097
ans =

  -5.1915 + 0.0000i
  -5.2915 - 1.5000i
  -4.1915 + 0.0000i

oct:70> fftnic([1 0.5 0.3 -1]') - 455097
ans =

  6.8636 + 0.0000i
  6.7636 - 1.5000i
  7.8636 + 0.0000i

      
See, there is a constant added, the right result is there too!!



-- 
Nicolau Werneck <[email protected]>         9F99 25AB E47E 8724 2F71
http://cefala.org/~nwerneck                   EA40 DC23 42CE 6B76 B07F
"After silence, that which comes nearest to expressing the inexpressible is music."
-- Aldous Huxley