hmm yes this is starting to sound like something 😀

Just wondering though, does the real mt-32 use floating points for the sound generation?

Good point. I would strongly doubt a 17 year old piece of hardware would rely on floating point math. Of course, looking at my code, I stick to using intergers as much as possible after I generate the tables. I believe the only time I venture from this is in the borrowed lowpass filter code. This code is a natural target for MMX optimization (of course, for that I'll have to convert the code to use intergers only to begin with, unless I want to learn SSE and 3DNow at the same time).

Brilliant! It's getting really good.

Indeed 😀

Harekiet wrote:

Just wondering though, does the real mt-32 use floating points for the sound generation?

They would not need floating-point for these kinds of applications. Fixed point would more than suffice.

V.

Yeah, fixed point arithmatic is what I've stuck to mostly (once I generate all the lookup tables). Again, the only exception is the lowpass filter, which I'm looking to convert to interger math once I get the sound accurate. There are also several other areas I've identified as easy targets for optimization. This code should get a bit faster, again, once I'm confident I have it sounding right. I'm really looking forward to implementing a lot of the stuff using MMX. Such a change should also bring a big speed improvement.