> As someone said to me the other day though "If you know what
> you're doing, fixed point math will always be faster anyway though".
That is just simply wrong in general for a very good reason:
FP multipliers are faster than integer multipliers (I mean it is
a deliberate design choice). Of course, in the embedded world,
this is reversed :) But I bet it is already changing.
> My reply was "Yeah, but show me the open source lib which saves me
> having to code it myself, man!" I think the topic was actually MP3
> encoders -- incredibly, there doesn't seem to be a decent fixed-point
> open source MP3 encoder out there.
Well there is a fixed point reference decode implementation of Ogg
And for MP3:
looks fast enough to me.
For the MP3 encoder:
Warning: I did not really test any of these implementations.