[Bug tree-optimization/121362] New: Fre sometimes does not read through aggregate copies

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121362

            Bug ID: 121362
           Summary: Fre sometimes does not read through aggregate copies
           Product: gcc
           Version: 14.2.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: pinskia at gcc dot gnu.org
  Target Milestone: ---

Take:
```
struct s1
{
  int t, t1;
};

struct s3
{
  struct s1 t;
};

struct s2
{
  struct s3 t;
};

void f(int, int);
void l();
void g(int a, int b, int *p)
{
  struct s2 c;
  {
    struct s1 tmp = {a,b};
#if WORKS
    c.t.t = tmp;
#else
    struct s3 *t = &c.t;
    t->t = tmp;
#endif
  }
  f(c.t.t.t, c.t.t.t1);
}
```

At `-O2 -fno-tree-sra` FRE1 is not able to optimize the loads of
c.t.t.t/c.t.t.t1 . But if we add -DWORKS=1 then FRE1 can look through the
aggregate copy.

Note I disable SRA just to make easier to see in .optimized that the missed
optimization has happened.

This shows up in C++ code and in highway (with turning off SRA).