[music-dsp] faster Hermite? - or not..
xk at myrealbox.com
Tue Apr 9 18:56:39 EDT 2002
> A hybrid algorithm that does all computations as integers before
> storing to the float buffer should be significantly faster since the
> compiler should be able to turn all of the computations a, b
> and c into shifts, adds and subtracts. Better yet would be to use an
> integer mix buffer to avoid the overhead of the int->float conversions.
> I'd guess you could get it below 60 in the former case and below 40 in
> the latter case. An assembly language SIMD implementation could probably
> get it below 20.
I'm not sure if it's worth the pain. You can do the int->float conversion
one time at startup in the case of a sampler. But even if you read data from
disk, this conversion it's 1 operation out of 40. So it's not signifiant.
And with int's you have to check for overflow.
And it's cool to be able to go higher than 0 dB with floats, because you can
limit/normalize at the last stage, so you don't have to constantly watch and
tweak the volume.
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