https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96146
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |aldyh at gcc dot gnu.org
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
i_1 = ASSERT_EXPR <i_7, i_7 != POLY_INT_CST [2, 2]>;
Intersecting
int [-INF, 1] EQUIVALENCES: { i_7 } (1 elements)
so somehow we derive from i_7 != POLY_INT_CST [2, 2] that i_7 must
be [-INF, 1] - that looks wrong?
This goes via extract_range_for_var_from_comparison_expr and likely
compare_values goes off in
if (limit_vr
&& limit_vr->kind () == VR_RANGE
&& compare_values (limit_vr->min (), limit_vr->max ()) == 0)
{
min = limit_vr->min ();
max = limit_vr->max ();
}
where min/max are then POLY_INT_CST [2, 2] and the anti-range we build
from that is canonicalized in odd ways.
I think VRP all over the place assumes integer typed constants are
INTEGER_CSTs so maybe sth else is off with POLY_INT_CSTs ...
CCing VRP people.
