[Bug tree-optimization/49599] New: FRE/DSE not performing well on aggregates

[Bug tree-optimization/49599] New: FRE/DSE not performing well on aggregates

[jamborm at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49599

           Summary: FRE/DSE not performing well on aggregates
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: enhancement
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: jamborm@gcc.gnu.org


In the following testcase:

struct S
{
  int i;
  unsigned short f1;
  char f2;
  unsigned short f3[2], f4;
};

int foo (struct S *p)
{
  struct S l;

  l = *p;
  l.i++;
  *p = l;
}

in which SRA does not do total scalarization because of the array in
the aggregate (we do it only for structures), the optimized dump looks
like this:

foo (struct S * p)
{
  int l$i;
  struct S l;
  int D.1967;

<bb 2>:
  l = *p_1(D);
  l$i_6 = p_1(D)->i;
  D.1967_3 = l$i_6 + 1;
  *p_1(D) = l;
  p_1(D)->i = D.1967_3;
  return;
}

The loads and stores from and to l should be eliminated, probably by
FRE and DSE.

[Bug tree-optimization/49599] FRE/DSE not performing well on aggregates

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49599

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization, TREE
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011-12-16
     Ever Confirmed|0                           |1

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> 2011-12-16 00:43:05 UTC ---
Confirmed.

[Bug tree-optimization/49599] FRE/DSE not performing well on aggregates

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49599

--- Comment #2 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-12-16 09:18:40 UTC ---
FRE does not handle removal of redundant aggregate stores, so it does not
remove the aggregate assignment

  *p_1(D) = l;

DSE does not handle redundant store removal either, and the store is
obviously not dead.

I think SRA should remove the aggregate store.