[Bug middle-end/108425] Invalid DSE

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108425

Jan Hubicka <hubicka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |REOPENED
   Last reconfirmed|                            |2023-01-16
         Resolution|INVALID                     |---
     Ever confirmed|0                           |1

--- Comment #3 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
We do not delete the stmt causing EH. Also the flag does not fix it:

jan@localhost:/tmp> gcc  -B ./ -O2 q.C -fnon-call-exceptions -S
-fdump-tree-all-details  -fno-delete-dead-exceptions
jan@localhost:/tmp> cat q.C*dse1*

;; Function test (_Z4testP1aS0_, funcdef_no=0, decl_uid=2377, cgraph_uid=1,
symbol_order=0)

  Deleted dead store: *a_2(D).e = 4;

  Deleted dead store: *a_2(D).d = 3;

  Deleted dead store: *a_2(D).c = 2;

  Deleted dead store: *a_2(D).b = 1;

  Deleted dead store: *a_2(D).a = 0;

void test (struct a * restrict a, struct a * b)
{
  <bb 2> :
  *a_2(D) = MEM[(const struct a &)b_8(D)];
  return;

}


a related testcase:

short a;
int *null;
int 
test(int val1, int val2)
{
        a=1;
        int r = val1/val2;
        a=3;
        return r;
}

this shows that the function causing EH may not be visited by the ao walk.