[PATCH] Fix sibcall optimization on ppc and other arches (PR target/39240)

[Jakub Jelinek <jakub at redhat dot com>]

Hi!

On PowerPC (both -m32 and -m64) and perhaps other targets the testcase below
fails at -O2+, because it optimizes a call into tail call, eventhough the
return value differs in signedness between caller and callee and the
integral return values are passed promoted into DImode (resp. SImode)
zero/sign-extended depending on the return type.

Fixed by checking this in expand_call and disabling tail call optimization
if the signedness differ and return value is promoted.

Bootstrapped/regtested on x86_64-linux and i686-linux (with gcc_unreachable ()
instead of try_tail_call = 0) and so far bootstrapped on powerpc64-linux
(-m32 default).

Ok for trunk if testing on powerpc64-linux passes (and another
powerpc64-linux -m64 default bootstrap/regtest passes)?

2009-02-19  Jakub Jelinek  <jakub@redhat.com>

	PR target/39240
	* calls.c (expand_call): Clear try_tail_call if caller and callee
	disagree in promotion of function return value.

	* gcc.c-torture/execute/pr39240.c: New test.

--- gcc/calls.c.jj	2009-02-06 11:17:16.000000000 +0100
+++ gcc/calls.c	2009-02-19 13:36:41.000000000 +0100
@@ -2333,6 +2333,37 @@ expand_call (tree exp, rtx target, int i
       || !lang_hooks.decls.ok_for_sibcall (fndecl))
     try_tail_call = 0;
 
+  /* Check if caller and callee disagree in promotion of function
+     return value.  */
+  if (try_tail_call)
+    {
+      enum machine_mode caller_mode, caller_promoted_mode;
+      enum machine_mode callee_mode, callee_promoted_mode;
+      int caller_unsignedp, callee_unsignedp;
+      tree caller_res = DECL_RESULT (current_function_decl);
+
+      caller_unsignedp = TYPE_UNSIGNED (TREE_TYPE (caller_res));
+      caller_mode = caller_promoted_mode = DECL_MODE (caller_res);
+      callee_unsignedp = TYPE_UNSIGNED (TREE_TYPE (funtype));
+      callee_mode = callee_promoted_mode = TYPE_MODE (TREE_TYPE (funtype));
+      if (targetm.calls.promote_function_return (TREE_TYPE (current_function_decl)))
+	caller_promoted_mode
+	  = promote_mode (TREE_TYPE (caller_res), caller_mode,
+			  &caller_unsignedp, 1);
+      if (targetm.calls.promote_function_return (funtype))
+	callee_promoted_mode
+	  = promote_mode (TREE_TYPE (funtype), callee_mode,
+			  &callee_unsignedp, 1);
+      if (caller_mode != VOIDmode
+	  && (caller_promoted_mode != callee_promoted_mode
+	      || ((caller_mode != caller_promoted_mode
+		   || callee_mode != callee_promoted_mode)
+		  && (caller_unsignedp != callee_unsignedp
+		      || GET_MODE_BITSIZE (caller_mode)
+			 < GET_MODE_BITSIZE (callee_mode)))))
+	try_tail_call = 0;
+    }
+
   /* Ensure current function's preferred stack boundary is at least
      what we need.  Stack alignment may also increase preferred stack
      boundary.  */
--- gcc/testsuite/gcc.c-torture/execute/pr39240.c.jj	2009-02-19 13:38:23.000000000 +0100
+++ gcc/testsuite/gcc.c-torture/execute/pr39240.c	2009-02-19 13:39:14.000000000 +0100
@@ -0,0 +1,105 @@
+/* PR target/39240 */
+
+extern void abort (void);
+
+__attribute__ ((noinline))
+static int foo1 (int x)
+{
+  return x;
+}
+
+__attribute__ ((noinline))
+unsigned int bar1 (int x)
+{
+  return foo1 (x + 6);
+}
+
+volatile unsigned long l1 = (unsigned int) -4;
+
+__attribute__ ((noinline))
+static short int foo2 (int x)
+{
+  return x;
+}
+
+__attribute__ ((noinline))
+unsigned short int bar2 (int x)
+{
+  return foo2 (x + 6);
+}
+
+volatile unsigned long l2 = (unsigned short int) -4;
+
+__attribute__ ((noinline))
+static signed char foo3 (int x)
+{
+  return x;
+}
+
+__attribute__ ((noinline))
+unsigned char bar3 (int x)
+{
+  return foo3 (x + 6);
+}
+
+volatile unsigned long l3 = (unsigned char) -4;
+
+__attribute__ ((noinline))
+static unsigned int foo4 (int x)
+{
+  return x;
+}
+
+__attribute__ ((noinline))
+int bar4 (int x)
+{
+  return foo4 (x + 6);
+}
+
+volatile unsigned long l4 = (int) -4;
+
+__attribute__ ((noinline))
+static unsigned short int foo5 (int x)
+{
+  return x;
+}
+
+__attribute__ ((noinline))
+short int bar5 (int x)
+{
+  return foo5 (x + 6);
+}
+
+volatile unsigned long l5 = (short int) -4;
+
+__attribute__ ((noinline))
+static unsigned char foo6 (int x)
+{
+  return x;
+}
+
+__attribute__ ((noinline))
+signed char bar6 (int x)
+{
+  return foo6 (x + 6);
+}
+
+volatile unsigned long l6 = (signed char) -4;
+
+int
+main (void)
+{
+  if (bar1 (-10) != l1)
+    abort ();
+  if (bar2 (-10) != l2)
+    abort ();
+  if (bar3 (-10) != l3)
+    abort ();
+  if (bar4 (-10) != l4)
+    abort ();
+  if (bar5 (-10) != l5)
+    abort ();
+  if (bar6 (-10) != l6)
+    abort ();
+  return 0;
+}

	Jakub