cmd/5g: make sure we normalize after unary ops on small types

We were failing ^uint16(0xffff) == 0, as we computed 0xffff0000 instead.

I could only trigger a failure for the above case, the other two tests
^uint16(0xfffe) == 1 and -uint16(0xffff) == 1 didn't seem to fail
previously.  Somehow they get MOVHUs inserted for other reasons (used
by CMP instead of TST?).  I fixed OMINUS anyway, better safe than
sorry.

Fixes #9604

Change-Id: I4c2d5bdc667742873ac029fdbe3db0cf12893c27
Reviewed-on: https://go-review.googlesource.com/2940
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Minux Ma <minux@golang.org>
diff --git a/src/cmd/5g/cgen.c b/src/cmd/5g/cgen.c
index c535cfb..87c64f6 100644
--- a/src/cmd/5g/cgen.c
+++ b/src/cmd/5g/cgen.c
@@ -236,18 +236,14 @@
 		cgen(nl, &n1);
 		nodconst(&n2, nl->type, -1);
 		gins(a, &n2, &n1);
-		gmove(&n1, res);
-		regfree(&n1);
-		goto ret;
+		goto norm;
 
 	case OMINUS:
 		regalloc(&n1, nl->type, N);
 		cgen(nl, &n1);
 		nodconst(&n2, nl->type, 0);
 		gins(optoas(OMINUS, nl->type), &n2, &n1);
-		gmove(&n1, res);
-		regfree(&n1);
-		goto ret;
+		goto norm;
 
 	// symmetric binary
 	case OAND:
@@ -483,12 +479,15 @@
 		cgen(nl, &n1);
 	}
 	gins(a, &n2, &n1);
+norm:
 	// Normalize result for types smaller than word.
 	if(n->type->width < widthptr) {
 		switch(n->op) {
 		case OADD:
 		case OSUB:
 		case OMUL:
+		case OCOM:
+		case OMINUS:
 			gins(optoas(OAS, n->type), &n1, &n1);
 			break;
 		}
diff --git a/test/fixedbugs/issue9604.go b/test/fixedbugs/issue9604.go
new file mode 100644
index 0000000..cd9e9e4
--- /dev/null
+++ b/test/fixedbugs/issue9604.go
@@ -0,0 +1,29 @@
+// run
+
+// Copyright 2015 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package main
+
+var x uint16 = 0xffff
+var y uint16 = 0xfffe
+var a uint16 = 0x7000
+var b uint16 = 0x9000
+
+func main() {
+	// Make sure we truncate to smaller-width types after evaluating expressions.
+	// This is a problem for arm where there is no 16-bit comparison op.
+	if ^x != 0 {
+		panic("^uint16(0xffff) != 0")
+	}
+	if ^y != 1 {
+		panic("^uint16(0xfffe) != 1")
+	}
+	if -x != 1 {
+		panic("-uint16(0xffff) != 1")
+	}
+	if a+b != 0 {
+		panic("0x7000+0x9000 != 0")
+	}
+}