[Bug tree-optimization/32306] [4.6/4.7/4.8 Regression] redundant && || not eliminated

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32306

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #23 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-01-10 11:26:09 UTC ---
It generally depends on the branch cost, e.g. out of #c19 testcase on x86_64 we
generate:
  D.1729 = b1 != 0;
  D.1730 = b2 != 0;
  D.1731 = D.1729 & D.1730;
  if (D.1731 != 0) goto <D.1732>; else goto <D.1727>;
  <D.1732>:
  D.1733 = b3 != 0;
  D.1734 = b4 != 0;
  D.1735 = D.1733 & D.1734;
  if (D.1735 != 0) goto <D.1736>; else goto <D.1727>;
  <D.1736>:
etc., but on other targets it could have just one comparison per conditional
jump, or on the other side 4.  While the tests don't have side-effects, turning
too many &&s into &s wouldn't result in very good code, though of course would
make it easier to perform some optimizations.  Anyway, you can write it in the
source as a series of ifs:
  array[0] = 1;
  if (!b1)
    array[0] = 0;
  else if (!b2)
    array[0] = 0;
  else if (!b3)
    array[0] = 0;
...
and we wouldn't be able to optimize it anyway.