------- Comment #6 from ghazi at gcc dot gnu dot org 2008-04-13 16:27 -------
So did we decide to fix this or not?
FWIW, I could fix the generic isinf transformation using:
isinf(x) -> isless(x,DBL_MAX) ? -1 : (isgreater(x,DBL_MAX))
and just always take the runtime penalty calculating for the -inf case when the
user says "if (isinf(x))" and doesn't care about the sign. Maybe
__builtin_expect preferring zero would help here. (?)
I have an preliminary patch to do the ?: transformation, however the obtabs
isinf versions will still fail to honor sign. This makes trouble in the
testcases if some platforms get -1 and others get 1. So switching it would
need to be coordinated.
Thoughts?
--
ghazi at gcc dot gnu dot org changed:
What |Removed |Added
----------------------------------------------------------------------------
Last reconfirmed|2008-03-08 10:40:32 |2008-04-13 16:27:57
date| |
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35509
