loop.c:get_condition broken

[Jeffrey A Law]

  In message < 19981007221501.A21728@dot.cygnus.com >you write:
  > since the fcmoveq won't respond properly to a NaN, as the cmpteq will.
  > 
  > The solution I proposed was generate the original cmov with CCmode
  > and prevent to comparisons with differing modes, namely eq:DF and
  > ne:CC, from being combined.
  > 
  > I'm rather surprised this is affecting anyone else, unless you are
  > generating comparisons that randomly have VOIDmode or some CCmode.
Hmm, I actually thought most comparisons had VOIDmode.


jeff