archive: Re: SETI Re: High bandwidth digitizer, FFT's, and analyzer

Re: SETI Re: High bandwidth digitizer, FFT's, and analyzer

Corey Ashford ( yeroca@rocketmail.com )
Mon, 5 Oct 1998 10:47:43 -0700 (PDT)

It's interesting to note that a lot of RISC machine's
floating point multiplies are actually faster than its
integer math. This makes the switch to integer math
for FFT's a negative performance gain!

I suspect the same thing may be true of the Intel
Merced
when it finally arrives.

Cheers

- Corey
KB7KD

---David Woolley <david@djwhome.demon.co.uk> wrote:
>
> > Well, thanks for helping clear that up. Now I
have a bandwidth limit for the
> > FFT processing of a PC. Have your figures taken
into account the vast
> > improvement in speed from changing the real mode
math to integer math (ala the
>
> The code fragment I did the instruction count on
used integer, not real
> arithmetic. The only factor that would really vary
would be the estimate
> of the number of processor cycles per instruction -
I assumed an average of
> 2, which would be very optimisitc for a 486, but I
don't have the relevant
> data for a Pentium.
>
> One thing, though, is that a 100MHz system would
probably need 64 bit
> arithmetic, which might require multiple integer
instructions.
>
> > tremendous fractal graphics program Fractint)? I
suspect that your
> > calculations are done using real mode math,
instead of integer mode.
>
> Incidentally, the book was wrong on the number of
multiplies needed; it
> gives the correct figure in a formula, but the
wrong figure in a table.
> I think the specgrm is code has the minimum number
for a 2 point butterfly.
>

_________________________________________________________
DO YOU YAHOO!?
Get your free @yahoo.com address at http://mail.yahoo.com