RFH: testers for ieee test

[Joern Rennecke <joernr at arc dot com>]

I'm trying to write a test to check I get all the subnormal
corner cases for ieee-754 double precision floating point division right.
The usual thing to do with a new test is to try it on a known good
platform.  Unfortunately, the x86 processor of my workstation is actually
known to get these calculations wrong.  Could someone with a fully
ieee compliant FPU try this test?

I've currently made it output lots of debug information and to continue
in the faceof errors; to get only a single result via the exit code,
just disable the DEBUG define.

It appears my machine gets about one in 3.7 of the fraction approximation
tests wrong :-(

FWIW, the random stuff is cribbed from arith-rand-ll.c.

Attachment: denorm-rand.c
Description: Text document