[Bug middle-end/31631] New: Folding of A & (1 << B) pessimizes FRE

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

void
foo (int control2, unsigned long *state, int size)
{
  int i;

  for(i=0; i<size; i++)
    {
       if (state[i] & ((unsigned long) 1 << control2))
         state[i] ^= ((unsigned long) 1 << control2);
    }
}

FRE does not recognize the ((unsigned long) 1 << control2) redundancy
because it looks like

<L0>:;
  D.1989_4 = (long unsigned int) i_1;
  D.1990_5 = D.1989_4 * 8;
  D.1991_6 = (long unsigned int *) D.1990_5;
  D.1992_8 = D.1991_6 + state_7(D);
  # VUSE <SMT.4_28>
  D.1993_9 = *D.1992_8;
  D.1994_11 = D.1993_9 >> control2_10(D);
  D.1995_12 = (int) D.1994_11;
  D.1996_13 = D.1995_12 & 1;
  if (D.1996_13 != 0) goto <L1>; else goto <L2>;

<L1>:;
  D.1989_19 = D.1989_4;
  D.1990_20 = D.1990_5;
  D.1991_21 = D.1991_6;
  D.1992_22 = D.1992_8;
  D.1993_23 = D.1993_9;
  D.1998_24 = 1 << control2_10(D);
  D.1999_25 = D.1993_23 ^ D.1998_24;
  # SMT.4_30 = VDEF <SMT.4_28>
  *D.1992_22 = D.1999_25;

This is related to PR29789 where this transformation causes other
pessimization.


-- 
           Summary: Folding of  A & (1 << B) pessimizes FRE
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: enhancement
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: rguenth at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31631