[clean-list] FFT with Clean

[Siegfried Gonzi]

Gruess Gott,


so, would the following assumption hold: If Clean makes the
matrix-matrix-multiplication of an array from the one-dimension 512*512 in 50sec, I
can lay the base that Clean should take much fewer time for a FFT (Fast Fourier
transform) of an array with only 512*512 members.

The timing scale of a matrix-matrix-multiplication is N^3 (I believe). And the
timing-scale of a FFT is Nlog2(N)+c(this includes the bit-reversing).

Based on that and the knowledge that Yorick , concerning the time for
matrix-matrix-multiplication, comparable to Clean, takes for a FFT of an array with
512*512 members 4 sec or so, I tried the FFT (based on bit reversed) code, which is
included at the tutorial (Brasil...but no punish intented).

Huch, an array with 262144 (512*512) members and the FFT of that takes 90sec! 45sec
is for calculating the FFT and 45sec is for garbage collection.

So, there is something  not optimal. A first hint was, that the code takes 20sec
for the bit-reversing alone (for an expected theoretical time-scaling of 2*N or so;
my assumption).

Now I will try to implement the Cooley-Tukey code. I hope I can do it in the next
few weeks (Gonzi's Clean learning curve is only weak increased inclined).

Please could someone confrom or maybe reject my first paragraph of my message.
Someone can believe Gonzi want to get the last from a programming language, but
that is not so (I am to weak in computer sciences). But I think an expected
calulation time from 7 or 10sec compared to 90sec is worth to bother.


Regards,
S. Gonzi
[Can I study the above mentioned FFT code study with the time-profiling tool? I
tried it, but I only get a window with a row. And this row has for all values a
'0']