@@ -1,3 +1,8 @@
+2015-09-21 Jeff Law <law@redhat.com>
+
+ * config/h8300/h8300.md (andsi3_ashift_n_lower): Avoid undefined
+ behavior.
+
2015-09-21 Ulrich Weigand <Ulrich.Weigand@de.ibm.com>
* config/spu/spu.c (spu_expand_insv): Avoid undefined behavior.
@@ -3914,7 +3914,8 @@
(clobber (match_scratch:QI 4 "=X,&r"))]
"(TARGET_H8300H || TARGET_H8300S)
&& INTVAL (operands[2]) <= 15
- && INTVAL (operands[3]) == ((-1 << INTVAL (operands[2])) & 0xffff)"
+ && UINTVAL (operands[3]) == ((HOST_WIDE_INT_M1U << INTVAL (operands[2]))
+ & 0xffff)"
"#"
"&& reload_completed"
[(parallel [(set (match_dup 5)
@@ -1,3 +1,7 @@
+2015-09-21 Jeff Law <law@redhat.com>
+
+ * gcc.target/h8300/andsi3_ashift_n_lower.c: New test.
+
2015-09-21 Ville Voutilainen <ville.voutilainen@gmail.com>
Complete the implementation of N4230, Nested namespace definition.
new file mode 100644
@@ -0,0 +1,7 @@
+/* { dg-do compile } */
+/* { dg-options "-mh -O2 -fomit-frame-pointer" } */
+/* { dg-final { scan-assembler-times "extu" 1 } } */
+
+unsigned long foo(unsigned long a)
+ { return (a << 4) & 0xffff; }
+
With the patch this time... -------- Forwarded Message -------- Subject: [PATCH] Fix undefined behavior in h8300 backend Date: Mon, 21 Sep 2015 11:10:04 -0600 From: Jeff Law <law@redhat.com> To: gcc-patches@gcc.gnu.org This was found when building all configurations using the trunk compiler. A run-of-the-mill problem with left shifting -1. To make sure I didn't muck anything up badly, I went back to Kazu's introduction of this pattern (2003), took his pseudo-code and turned it into a trivial h8300 specific test. I also verified newlib built with the new compiler. newlib never triggers this pattern though. I briefly pondered adding tests for the many other patterns Kazu added in 2003 as well as generalizing andsi3_ashift_n_lower, but I just couldn't justify that much time on the H8 port :-) Installed on the trunk. Jeff