floating point

[utzoo!decvax!ucbvax!unix-wizards]

>From decvax!utzoo!henry at Berkeley Tue Oct 27 15:55:15 1981
Human Computing Resources of Toronto has done what MBM at MIT-XX suggested,
putting the floating-point simulator in the kernel.  It's available from
them as part of a general V7-enhancements package, which also includes
a simple overlay system and a very thorough V6 simulator.  The kernel
fpsim is great, for the following reasons:

	1. You can run I/D floating point programs like awk.
	2. Each and every program no longer has to drag around fpsim.
	3. Object modules are 100% movable between FP and non-FP cpus.
	4. [the big one] Put it in the kernel and you can IGNORE THE
		WHOLE ISSUE THENCEFORTH.  You don't need to remember
		the stupid -f, don't need to comb through every incoming
		distribution to figure out which programs need -f,
		don't have to keep a list of which ones you found (and
		keep discovering ones you didn't find), etc etc.

Obviously, I think #4 is the main justification for it.  For example, when
we got Bell's "S" statistics package, we just dropped it in and it ran.  No
recompiles, no kludges to make I/D things non-I/D, no hardware mods, no new
strange system calls to fetch from I-space, and so on.  Once you have fpsim
in the kernel, the problems go away so nicely that you wonder why it took
so long for somebody to get around to doing it.

This can even coexist with the subroutine-call approach if necessary;  use
-f to trigger subroutine calls, and then things which MUST run fast can do
so and everything else runs unchanged.  I haven't bothered.

For distribution and prices, you'll have to talk to HCR.  The package is
not grossly expensive.  They are at (416)922-1937.