benchmarking Poly/ML & floating point

[Stephen Weeks]

> This is going to be in the generated C code?  If so, if you make it inline
> then it will be perfect, but if not it will be much slower (much much slower
> because it will actually pass the 2 doubles, probably on the stack).

If I make it inline I am worried that subsequent gcc optimization may reverse
the sense of the test and cause the problem you mentioned earlier.

I agree it is much slower.

Actually, I now think I need to do the same thing for Real.{<,>,<=,>=} since gcc 
may reverse the sense of those tests as well.