[Bug tree-optimization/52448] [4.4/4.5/4.6/4.7 Regression] cselim broken with calls

["matz at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52448

--- Comment #4 from Michael Matz <matz at gcc dot gnu.org> 2012-03-01 14:10:08 UTC ---
Bah, no.  That won't generally work.  In particular it might be calls that
don't necessarily dominate the "non-trapping" stmt that make it trapping
again:

*X = a;
if (cond)
  free_everything();
if (bla)
  *X = b;

The call doesn't dominate the second store, but the latter still might
trap (when cond is true).  Hmm, this needs a different approach.  If there's
any path from the dominating access to the access in question, which contains
a (problematic) call, it makes it trapping again.  We can ignore back edges,
hence the path between both accesses are acyclic, so we might still get away
with a non-iterating algorithm.  I'm still working on it.