> Is -msoft-float supposed to working on i386?  If not, why not? :)
> I have an old 386/40 with no 387 that I'm trying to use, and a 
> number of things die even with MATH_EMULATE in the kernel.

What timing, I am within a week or so of finally making good on my offer to
Perry to do something about this. I've got original Intel docs for the 38[67]
and the latest UCB softfloat package here (supports 80-bit IEEE). The scope
of the project is a complete replacement for MATH_EMULATE, complete with AMD
inspired random bit pattern tests to validate the emulator against the real
silicon.

I would be very interested in hearing bug reports against the existing code.

Todd Whitesel
toddpw @ best.com