https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128129
Bug ID: 128129
Summary: [14/15/16/17 Regression] wrong code with NaN: !(a < b)
&& a != b folded to a > b (canonicalize_bool ignores
NaNs)
Product: gcc
Version: 12.5.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: lzy20000419 at outlook dot com
Target Milestone: ---
The following program aborts at -O1 and above. It should return 0.
int sink;
__attribute__((noipa)) int f(double a, double b, int k)
{
_Bool v;
if (k) { v = a < b; sink = 1; }
else { v = a < b; sink = 2; }
if (!v)
if (a != b)
return 1;
return 0;
}
int main(void)
{
double nan = __builtin_nan ("");
if (f (nan, 1.0, 1) != 1) /* !(NaN < 1) && NaN != 1 is true */
__builtin_abort ();
return 0;
}
$ gcc -O1 t.c && ./a.out Aborted
Expected: exit 0. -O0 and -fsignaling-nans are fine, clang is fine.
The ifcombine dump says "optimizing two comparisons to a_5(D) > b_6(D)". It
should be a u> b.
It works with 7.5 and fails with 8.1 through 16.2 and current trunk (also -O2,
-O3, -Os), so it's a regression since GCC 8.
https://godbolt.org/z/P5nccxnqo (left gcc 7.5, right trunk)
Disclaimer: This test case was generated by a fuzzer, and the report was
written with LLM assistance.
