Hi!
The following testcase ICEs since r15-8025.
tree_nop_conversion_p doesn't imply TREE_TYPE (@0) is uselessly convertible
to type, e.g. they could be INTEGER_TYPEs with the same precision but
different TYPE_SIGN.
The following patch just adds a convert so that it creates a valid IL
even in those cases.
Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
2025-03-14 Jakub Jelinek <ja...@redhat.com>
PR tree-optimization/119287
* match.pd (((X >> C1) & C2) * (1 << C1) to X & (C2 << C1)): Use
(convert @0) instead of @0 in the substitution.
* gcc.dg/pr119287.c: New test.
--- gcc/match.pd.jj 2025-03-13 14:05:09.689017389 +0100
+++ gcc/match.pd 2025-03-14 12:13:02.563169756 +0100
@@ -5284,7 +5284,7 @@ (define_operator_list SYNC_FETCH_AND_AND
|| tree_int_cst_sgn (@2) >= 0)
&& wi::to_wide (@3) == wi::set_bit_in_zero (shift, prec))
(with { auto mask = wide_int::from (wi::to_wide (@2), prec, UNSIGNED); }
- (bit_and @0 { wide_int_to_tree (type, mask << shift); }))))))
+ (bit_and (convert @0) { wide_int_to_tree (type, mask << shift); }))))))
/* ~(~X >> Y) -> X >> Y (for arithmetic shift). */
(simplify
--- gcc/testsuite/gcc.dg/pr119287.c.jj 2025-03-14 12:27:37.470247725 +0100
+++ gcc/testsuite/gcc.dg/pr119287.c 2025-03-14 12:27:04.367696980 +0100
@@ -0,0 +1,16 @@
+/* PR tree-optimization/119287 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -fwrapv" } */
+
+unsigned a;
+int b;
+signed char c, d;
+
+void
+foo (void)
+{
+ c = a >> 14 & 1;
+ for (; d;)
+ c = 1;
+ b = c << 14;
+}
Jakub
