Vorbis dev - Aug 2007 - Faster implementation of res2_inverse() ?

Hi,

Is there a faster implementation of the res2_inverse() function found
in Tremor (res012.c) and in the Vorbis reference implementation
(lib/res0.c) ?

The comments on top of res0.c say it's slow, and profiling the decoder
confirms this (#2 most CPU-intensive function, after mdct_backward).

I need a faster implementation to acheive good performance on a
portable music player I'm designing for my company, and before I dig
into the code to reimplement a compatible function, I'd like to know
if someone else did it before.

Regards,

Sebastien