https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108384
--- Comment #11 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #10)
> If it is __builtin_clear_padding only, I'll have a look...
After the __builtin_clear_padding is exposed and before the IPA passes, the IR
looks like:
l_26[1].f0 = 4;
MEM <long int> [(struct S0[2] *)&l_26 + 16B] = 25770065925;
func_23 (l_26[1], func_2_l_27);
So here is another testcase which you can reproduce it with -O3
-fno-strict-aliasing (because I voilate aliasing rules to get a similar enough
IR):
```
struct S0 {
int f0;
short f1;
unsigned f2 : 7;
short f3;
} func_2_l_27;
int *g_389;
int safe_sub_func_int16_t_s_s(void);
void safe_lshift_func_uint8_t_u_s(int);
void func_23(struct S0 p_24, struct S0 p_25) {
int *l_1051 = g_389;
if (safe_sub_func_int16_t_s_s())
for (;;)
safe_lshift_func_uint8_t_u_s(p_24.f1);
*l_1051 = p_25.f0;
}
void func_2(void) {
struct S0 l_26[2];
l_26[1].f0 = 4;
((long long*)&l_26)[2] = 25770065925;
func_23(l_26[1], func_2_l_27);
}
```
