>Is the existing codeclib mdct being used now (which is already heavily >optimised)? Or is optimisation lower down the priority list and/or mdct not >the bottleneck?
We can't use the codeclib MDCT as Opus does not use power of 2 transforms. It
uses its own FFT based on KISSFFT. Algorithmically its probably not too bad,
but without ASM I imagine gcc makes a mess of it.
