[music-dsp] Frequncy smear? (fwd)
mister-tom at bigfoot.com
Thu Jan 11 11:58:54 EST 2001
> > Well, I'm up to 3 waveshapers (one of them is a pricey mother, but it is
> > best 12ax7 emu so far), the decimator FIR, a hum cancelling 2nd order
> > and a LP 2nd order IIR for cabinet simulation.
> You should do table lookups for the waveshapers. That's much faster, even
> for quite simple shapers.
Not always true: if you don't run the algorithm alone (i.e. it is one of the
many plugs running in a host), there is a high level of cache pollution and
probably each table lookup must wait for a cache miss. The penality for a
cache miss is much higher than the time for some math operations. Consider
also that on a PIII or on Altivect you can process 4 samples at a time if
you do not use table loockup.
But this is not all: if you have floating foint samples, you need a
float->int conversion for accessing the loockup table, and another
int->float conversion to interpolate; these operation, expecially on Intel
processors, are very slow.
For polynomial shapes on a PIII my code is 8 time faster than with the
dupswapdrop -- the music-dsp mailing list and website: subscription info,
FAQ, source code archive, list archive, book reviews, dsp links
More information about the music-dsp