https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120048
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
I guess the mask/value comes from the c | 1 parameter passed to f from main.
# RANGE [irange] int [-INF, -1][1, +INF] MASK 0xfffffffe VALUE 0x1
_2 = c.0_1 | 1;
But f is
void f (int e)
{
int e_3(D) = e;
short int _1;
int _5;
int _6;
<bb 2> [local count: 1073741824]:
if (e_3(D) == 0)
goto <bb 3>; [50.00%]
else
goto <bb 6>; [50.00%]
<bb 3> [local count: 536870912]:
_1 = (short int) e_3(D);
if (_1 == 0)
goto <bb 4>; [50.00%]
else
goto <bb 5>; [50.00%]
<bb 4> [local count: 268435456]:
# USE = nonlocal escaped
_6 = d.part.0 (_1);
and so shouldn't call d.part.0 at all if e_3 is not zero. Perhaps ipa-vrp just
sees that something based on e_3(D) is passed to d.part.0 and doesn't actually
do the path specific checks? Given the # RANGE [irange] int [-INF, -1][1,
+INF] MASK 0xfffffffe VALUE 0x1 and # RANGE [irange] short [0, 0] at the same
time the d.part.0 call from f is impossible and so UNDEFINED range is
reasonable.
