Subject: Re: Floating point in the kernel
To: Andreas Gustafsson <gson@araneus.fi>
From: Jukka Marin <jmarin@pyy.jmp.fi>
List: tech-kern
Date: 09/21/1998 10:03:59
On Mon, Sep 21, 1998 at 09:23:27AM +0300, Andreas Gustafsson wrote:
> > If the signal processing algorithm is appropriately scaled, then all of the
> > multiplications can be treated as multiplications by a factor of two.  What
> > does that sound like?  Full marks if you said "arithmetic shift left"! :)
> 
> If this were true, it would be quite a revolution for the field of
> digital signal processing.  Why not take it a step further and say
> that "If the signal processing algorithm is appropriately scaled, then
> all of the multiplications can be treated as multiplications by a
> factor of one", and you could eliminate the shifts, too.

It _is_ possible to design a FIR filter with all the multipliers being 2^N
(which allows the programmer to change all multiplications into bit shifts),
but this is almost always compromises the filter performance.  Besides, on
some CPU's, bit shifting isn't all that fast.. (well, I guess on moderns
CPU's it is).

  -jm (who knows nothing about digital signal processing :)