This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[PATCH] Fix ICE in chrec_convert_1


Hi!

The PR34163 fix caused a regression on the testcase below, build2 (PLUS_EXPR, ...)
is called with POINTER_TYPE_P types of both arguments and the result type.

This is on a
y_N = z_N + 1;
x_N = (char *) y_N;
where [yz]_N have int type and x_N is char *.

The following patch fixes it, bootstrapped/regtested on x86_64-linux.  Ok
for trunk?

2009-12-15  Jakub Jelinek  <jakub@redhat.com>

	* tree-chrec.c (chrec_convert_1): Only fold (T2)(t +- x) to
	(T2)t +- (T2)x if only integer types are involved.

	* gcc.c-torture/compile/20091215-1.c: New test.

--- gcc/tree-chrec.c.jj	2009-11-25 16:47:35.000000000 +0100
+++ gcc/tree-chrec.c	2009-12-15 19:34:15.000000000 +0100
@@ -1293,6 +1293,8 @@ keep_cast:
   if (use_overflow_semantics
       && (TREE_CODE (chrec) == PLUS_EXPR
 	  || TREE_CODE (chrec) == MINUS_EXPR)
+      && TREE_CODE (type) == INTEGER_TYPE
+      && TREE_CODE (ct) == INTEGER_TYPE
       && TYPE_PRECISION (type) > TYPE_PRECISION (ct)
       && TYPE_OVERFLOW_UNDEFINED (ct))
     res = fold_build2 (TREE_CODE (chrec), type,
--- gcc/testsuite/gcc.c-torture/compile/20091215-1.c.jj	2009-12-15 23:45:09.000000000 +0100
+++ gcc/testsuite/gcc.c-torture/compile/20091215-1.c	2009-12-15 19:04:41.000000000 +0100
@@ -0,0 +1,15 @@
+void bar ();
+
+void
+foo (void *x, short y)
+{
+  bar (x, y + 1);
+}
+
+void
+bar (x, y)
+  void *x;
+  char *y;
+{
+  baz (y);
+}

	Jakub


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]