[PATCH] Fix ICEs with bogus computed goto (PR tree-optimization/58164)

[Jakub Jelinek]

Hi!

gimple_goto_dest is is_gimple_val, so can be ADDR_EXPR (though just for bad
testcases), and in that case we weren't walking it in some cases.

I've tried to reject ADDR_EXPRs in gimple_goto_dest, but that turned out to
be much larger patch and still incomplete.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk/4.8?

2013-08-15  Jakub Jelinek  <ja...@redhat.com>

        PR tree-optimization/58164
        * gimple.c (walk_stmt_load_store_addr_ops): For visit_addr
        walk gimple_goto_dest of GIMPLE_GOTO.

        * gcc.c-torture/compile/pr58164.c: New test.

--- gcc/gimple.c.jj     2013-05-13 09:44:53.000000000 +0200
+++ gcc/gimple.c        2013-08-15 15:22:06.745236769 +0200
@@ -4049,6 +4049,13 @@ walk_stmt_load_store_addr_ops (gimple st
            ret |= visit_addr (stmt, TREE_OPERAND (op, 0), data);
        }
     }
+  else if (visit_addr
+          && gimple_code (stmt) == GIMPLE_GOTO)
+    {
+      tree op = gimple_goto_dest (stmt);
+      if (TREE_CODE (op) == ADDR_EXPR)
+       ret |= visit_addr (stmt, TREE_OPERAND (op, 0), data);
+    }
 
   return ret;
 }
--- gcc/testsuite/gcc.c-torture/compile/pr58164.c.jj    2013-08-15 
15:24:04.117313781 +0200
+++ gcc/testsuite/gcc.c-torture/compile/pr58164.c       2013-08-15 
15:23:47.000000000 +0200
@@ -0,0 +1,8 @@
+/* PR tree-optimization/58164 */
+
+int
+foo (void)
+{
+  int x = 0;
+  goto *&x;
+}

        Jakub