When strict enums are in use, we don't always get ranges reduced
sufficiently thru casts and such. Ranger vrp has been missing the odd
case because I didn't notice that gimple_stmt_nonnegative_warnv_p() use
to be called on all integral results that were varying, not just calls
like it was doing. this corrects that.
Bootstrapped on x86_64-pc-linux-gnu with no regressions. Pushed.
Andrew
commit a87819b8f1b890d36a3f05bd9de80be20e9525dd
Author: Andrew MacLeod <amacl...@redhat.com>
Date: Tue Oct 25 15:16:47 2022 -0400
Check if varying may also be non-negative.
When using strict enums, we can sometimes turn varying into a better
range.
* gimple-range-fold.cc (fold_using_range::fold_stmt): Check if
stmt is non-negative and adjust the range.
diff --git a/gcc/gimple-range-fold.cc b/gcc/gimple-range-fold.cc
index f91923782dc..a899d8260b3 100644
--- a/gcc/gimple-range-fold.cc
+++ b/gcc/gimple-range-fold.cc
@@ -494,6 +494,14 @@ fold_using_range::fold_stmt (vrange &r, gimple *s,
fur_source &src, tree name)
else if (is_a<gassign *> (s) && gimple_assign_rhs_code (s) == COND_EXPR)
res = range_of_cond_expr (r, as_a<gassign *> (s), src);
+ // If the result is varying, check for basic nonnegativeness.
+ // Specifically this helps for now with strict enum in cases like
+ // g++.dg/warn/pr33738.C.
+ bool so_p;
+ if (res && r.varying_p () && INTEGRAL_TYPE_P (r.type ())
+ && gimple_stmt_nonnegative_warnv_p (s, &so_p))
+ r.set_nonnegative (r.type ());
+
if (!res)
{
// If no name specified or range is unsupported, bail.