[Bug tree-optimization/58727] Sub-optimal code for bit clear/set sequence

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58727

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|rtl-optimization            |tree-optimization

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
I see the issue:
clear_set produces:
  D.1381_2 = a_1(D) & 0xffbffffd;
  D.1380_3 = D.1381_2 | 2;

While clear_set_inline produces:
  D.1381_2 = a_1(D) & 0xffbfffff;
  D.1380_3 = D.1381_2 | 2;

The lower two bits make GCC optimize it easier.
I think the issue is in fold-const.c as .original is already includes the ~0x2
in the and part.