Presumably one could take a similar approach to (I belive) the
	i386 port, of compiling the default libc with soft floating
	point instructions, and having a make option to enable use of
	the FPU instructions.

	Doing this, plus distributing the libc.so that uses floating point
	separately gives close to the best possible state for FPUless
	machines (the majority) at the cost of forcing FPU owners to
	recompile staticly linked programs to get the best FPU performance
	(dynamically linked programs should be the vast majority and those
	would only require dropping in the FPU using libc).

	Unless I'm missing something of course :)

                David/abs               abs@anim.dreamworks.com

.---- I've been too drunk to love ----.-- I've been too drunk to remember -.
|          too drunk to care          |    the hell of the night before    |
|  looked like death, felt like hell  |   I've been drinking myself blind  |
`------ been the worse for wear ------'-- and still I'll drink some more --'


On Fri, 11 Apr 1997, Thomas Boroske wrote:

> Hmmm, this makes it look easier to compile in-line fp code from start on, 
> perhaps using machine-dependent code file formats that incorporate an fpu and 
> integer-only version in the same file.
>