> The Shark can comfortably do this with a fixed-point based player;
> someone for the Acorn RiscOS machines wrote such a thing. I think, too,
> with FP you'll be pushing it. We can get a factor of 2x on the current
> l;ibrary, but that's about it I think.

The best way to speed up IEEE is to get rid of NaN propagation, denormals,
and options for rounding & signalling (just whack SIGFPE if the number looks
yucky). The actual storage format is not so bad, it's all the cheesy special
cases that bloat and bog down the code.

Todd Whitesel
toddpw @ best.com