https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87059
--- Comment #20 from Aldy Hernandez <aldyh at gcc dot gnu.org> --- (In reply to Aldy Hernandez from comment #18) > BTW, the suggest patch in comment 10 causes an x86-64 regression on > gcc.dg/attr-nonstring-3.c because of some incompatibility with ssizetype and > sizetype. I believe it's related to this: > > https://gcc.gnu.org/ml/gcc/2018-08/msg00164.html > > ...so it may be a rat's nest to fix. I may just make VRP work with broken > behavior it expected before. Testing the following. gcc/ PR 87059/tree-optimization * tree-vrp.c (extract_range_from_binary_expr_1): Pass sign of first argument to wide_int_range_min_max. diff --git a/gcc/tree-vrp.c b/gcc/tree-vrp.c index 735b3646e81..7373011d901 100644 --- a/gcc/tree-vrp.c +++ b/gcc/tree-vrp.c @@ -1566,6 +1566,15 @@ extract_range_from_binary_expr_1 (value_range *vr, wide_int vr1_min, vr1_max; extract_range_into_wide_ints (&vr0, sign, prec, vr0_min, vr0_max); extract_range_into_wide_ints (&vr1, sign, prec, vr1_min, vr1_max); + + /* FIXME: Currently the sign of MIN_EXPR and the sign of its + arguments are not always the same. Fixing the creators of + these faulty nodes caused other problems. For now use the + sign of the first argument as VRP was previously doing. See + PR87059. */ + if (vr0.min) + sign = TYPE_SIGN (TREE_TYPE (vr0.min)); + if (wide_int_range_min_max (wmin, wmax, code, sign, prec, vr0_min, vr0_max, vr1_min, vr1_max)) set_value_range (vr, VR_RANGE,
