> > Btw, what do you mean by "run-time"?  Per command?  That strikes me as 
> > unreasonable, since FPUs don't, as a rule, come and go.
> 
> Yes, it's at shared library load time if I remember properly.
> The overload is not measurable.

This conversation struck up my curiosity.  I think in Linux, the i387 FPU
emulation is automatically handled in the kernel by trapping illegal
instructions.  If I understand this conversation correctly, it means that
NetBSD applications linked against libm will perform better on Intel
machines without FPU's because the floating point emulation is built into
the library.

But I remember seeing floating point emulation as an option in the kernel.
Is NetBSD's floating point emulation happening "per instruction" by
trapping illegal instructions?

someday I'll learn,

Ben