https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114552
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Target Milestone|--- |13.3
Last reconfirmed| |2024-04-01
Summary|wrong code at -O1 and above |[13/14 Regression] wrong
|on x86_64-linux-gnu |code at -O1 and above on
| |x86_64-linux-gnu
Component|tree-optimization |middle-end
Status|UNCONFIRMED |NEW
Target| |x86_64-linux-gnu
Keywords| |needs-bisection, wrong-code
Ever confirmed|0 |1
--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Better reduced testcase, removing the pragma and putting the packed on the
struct that is causing the issue:
```
struct a {
short b;
int c;
}__attribute__((packed));
struct d {
struct a b;
int e;
};
static const struct d k = {{1,0},0};
[[gnu::noinline]]
void p() { asm("":::"memory"); }
[[gnu::noinline]]
void q(struct a n) {
p();
}
int main() {
q(k.b);
return 0;
}
```
The bug is in the middle-end expanding the call `q(k.b);` such that k.b is on
the stack but it messes up and uses the wrong size or something.
