https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89698
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Priority|P3 |P2
Status|UNCONFIRMED |NEW
Known to work| |6.5.0
Keywords| |wrong-code
Last reconfirmed| |2019-03-13
Ever confirmed|0 |1
Summary|Run-time error due to |[7/8/9 Regression] Run-time
|optimization of field |error due to optimization
|access after cast at |of field access after cast
|-Os/-O2 and higher |at -Os/-O2 and higher
Target Milestone|--- |7.5
Known to fail| |7.1.0, 9.0
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed. GCC 6 seems to work. The FE already folds the test to
if ((long unsigned int) BIT_FIELD_REF <*(struct A *) vp, 64, 64> ==
8589934593)
rewriting the testcase to
int test(void) {
int res;
B b(2);
A* bp = &b;
void* vp = dynamic_cast<void*>(bp);
if (((A*)vp)->x == 1)
if (((B*)vp)->y == 2)
return 1; // PASS
return 0; // FAIL
}
makes it PASS. So I suspect this BIT_FIELD_REF doesn't tell the truth.
Note the IL looks odd as well, possibly from the dynamic_cast which looks
fishy to me.
