re PR target/17377 (cris.md bug in "return" pattern trigged by __builtin_return_address)

	PR target/17377
	* gcc.c-torture/execute/pr17377.c: New test.

From-SVN: r87250

gcc/testsuite/ChangeLog

+2004-09-09  Hans-Peter Nilsson  <hp@axis.com>
+
+	PR target/17377
+	* gcc.c-torture/execute/pr17377.c: New test.
+
 2004-09-09  Joseph S. Myers  <jsm@polyomino.org.uk>
 
 	PR c/8420

gcc/testsuite/gcc.c-torture/execute/pr17377.c

+/* PR target/17377
+   Bug in code emitted by "return" pattern on CRIS: missing pop of
+   forced return address on stack.  */
+int calls = 0;
+
+void *f (int) __attribute__ ((__noinline__));
+void *
+f (int i)
+{
+  /* The code does a little brittle song and dance to trig the "return"
+     pattern instead of the function epilogue.  This must still be a
+     leaf function for the bug to be exposed.  */
+
+  if (calls++ == 0)
+    return __builtin_return_address (0);
+
+  switch (i)
+    {
+    case 1:
+      return f;
+    case 0:
+      return __builtin_return_address (0);
+    }
+  return 0;
+}
+
+int x;
+
+void *y (int i) __attribute__ ((__noinline__));
+void *
+y (int i)
+{
+  x = 0;
+
+  /* This must not be a sibling call: the return address must appear
+     constant for different calls to this function.  Postincrementing x
+     catches otherwise unidentified multiple returns (e.g. through the
+     return-address register and then this epilogue popping the address
+     stored on stack in "f").  */
+  return (char *) f (i) + x++;
+}
+
+int
+main (void)
+{
+  void *v = y (4);
+  if (y (1) != f
+      /* Can't reasonably check the validity of the return address
+	 above, but it's not that important: the test-case will probably
+	 crash on the first call to f with the bug present, or it will
+	 run wild including returning early (in y or here), so we also
+	 try and check the number of calls.  */
+      || y (0) != v
+      || y (3) != 0
+      || y (-1) != 0
+      || calls != 5)
+    abort ();
+  exit (0);
+}