"Andreas Raab" <Andreas.Raab at gmx.de> is claimed by the authorities to have written:


> The problem is even worse than that. If you use some other libraries
> these may change the FPU behavior "for you" (DirectX is known to do
> that).
Exactly this situation messed up VW some years ago. I think it started,
or at least became noticable, when that terrible Win32S chimaera
appeared. All of a sudden we found that any floating point test where we
expected to get an exception didn't. The only conclusion we felt able to
reach was that any call into almost any library would change the cpu &
fpu flags and not return them back to the previously set state. We
thought of testing on every library call but the cost was terrifying.

Even on Acorn I have to explicitly set the fpu state in orderto get the
behaviour we need. At least it actually maintains it correctly, even
across process swaps.

Guess it is yet another reminder that floating point arithmetic is not
your friend.

ti
-- 
Tim Rowledge, tim at sumeru.stanford.edu, http://sumeru.stanford.edu/tim
Nice computers don't go down.