Floating point comparisons with NaN values produce wrong results

[Jeffrey A Law]

  In message < Pine.LNX.4.05.9909060849190.26619-100000@moshier.ne.mediaone.net >
you write:
  > 
  > Your fixed test case works properly on x86 and sparc.
  > Now we need a PA expert to look into the problem.
HP's have always been a little hokey in their NaN comparison handling.

I wouldn't be at all surprised if the problem stems from the necessity for
the PA backend to handle normal and reversed comparisons (we only natively
support a reversed comparison because of the rather weird HP hardware).

jeff