https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127052

            Bug ID: 127052
           Summary: cond_store_replacement_limited should support more
                    than ssa name stores
           Product: gcc
           Version: 17.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: enhancement
          Priority: P3
         Component: tree-optimization
          Assignee: pinskia at gcc dot gnu.org
          Reporter: pinskia at gcc dot gnu.org
  Target Milestone: ---

Currently it does:
```
  rhs = gimple_assign_rhs1 (store_middle);
  if ((!REFERENCE_CLASS_P (lhs)
       && !DECL_P (lhs))
      || !is_gimple_reg_type (TREE_TYPE (lhs)))
    return false;
  if (TREE_CODE (rhs) != SSA_NAME)
    return false;
```
I think I was trying to reject constructors here.

Anyways simple testcase for -Ofast:
```
struct s1
{
  int t;
  s1 ();
};

struct s2 : s1
{
    s2();
    s2(int a);
};

s2::s2()
{
    if (t)
      t = 0;
}
s2::s2(int a)
{
    if (t)
      t = a;
}
```

phiopt1 can handle `= a` case but not `= 0` case.

I noticed this in std::optional::reset.

Reply via email to