https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99673
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Summary|[11/12 Regression] bogus |[11 Regression] bogus
|-Wstringop-overread warning |-Wstringop-overread warning
|with address sanitizer due |with address sanitizer due
|to member address |to member address
|substitution |substitution
Keywords| |needs-bisection
Known to work| |12.0
Known to fail| |11.2.1
Priority|P3 |P2
--- Comment #8 from Richard Biener <rguenth at gcc dot gnu.org> ---
On trunk the testcase no longer reproduces the issue, it's still present on the
GCC 11 branch though. I wonder what fixed it.
Note before .asan we have
<bb 2> [local count: 1073741824]:
_1 = MEM[(struct B *)pc_2(D) + 1B].i;
pd_4(D)->i = _1;
_9 = (sizetype) i_6(D);
_10 = _9 * 16;
_11 = _10 + 4;
_12 = &MEM[(struct B *)pc_2(D) + 1B];
psa_7 = _12 + _11;
f (psa_7);
and after
<bb 2> [local count: 1073741824]:
_13 = &MEM[(struct B *)pc_2(D) + 1B].i;
.ASAN_CHECK (6, _13, 4, 4);
_1 = MEM[(struct B *)pc_2(D) + 1B].i;
_14 = &pd_4(D)->i;
.ASAN_CHECK (7, _14, 4, 4);
pd_4(D)->i = _1;
_9 = (sizetype) i_6(D);
_10 = _9 * 16;
_11 = _10 + 4;
_12 = &MEM[(struct B *)pc_2(D) + 1B];
psa_7 = _12 + _11;
f (psa_7);
so I see nothing wrong. Later FRE determines that _13 and _12 are equal
(again correct but this confuses the diagnostics).
On trunk we seem to have lowered _12 to a pointer-plus and VN fails to CSE
the addresses. That's an optimization regression.
