x86 _Decimal128 return value alignment

["Joseph S. Myers" <joseph@codesourcery.com>]

On 32-bit x86, _Decimal128 function return values are stored in a
location whose address is passed as a hidden parameter.  This location
must be 16-byte aligned.  Although GDB knows about 16-byte alignment
for parameters, it fails to achieve this for this stack slot,
resulting in problems if, for example, the function called from GDB
with a _Decimal128 return value uses an SSE store instruction to
store the return value.

This patch fixes this with an appropriate frame_align function, since
that's the setting used to control alignment of return value slots.
Tested on i686-pc-linux-gnu.  OK to commit?

2011-09-22  Joseph Myers  <joseph@codesourcery.com>

	* i386-tdep.c (i386_frame_align): New.
	(i386_gdbarch_init): Use i386_frame_align.

Index: i386-tdep.c
===================================================================
RCS file: /cvs/src/src/gdb/i386-tdep.c,v
retrieving revision 1.336
diff -u -p -r1.336 i386-tdep.c
--- i386-tdep.c	22 Jul 2011 15:31:51 -0000	1.336
+++ i386-tdep.c	22 Sep 2011 22:31:51 -0000
@@ -2228,6 +2228,15 @@ i386_dummy_id (struct gdbarch *gdbarch, 
   /* See the end of i386_push_dummy_call.  */
   return frame_id_build (fp + 8, get_frame_pc (this_frame));
 }
+
+/* _Decimal128 function return values need 16-byte alignment on the
+   stack.  */
+
+static CORE_ADDR
+i386_frame_align (struct gdbarch *gdbarch, CORE_ADDR sp)
+{
+  return sp & -(CORE_ADDR)16;
+}
 \f
 
 /* Figure out where the longjmp will land.  Slurp the args out of the
@@ -7316,6 +7325,7 @@ i386_gdbarch_init (struct gdbarch_info i
 
   /* Call dummy code.  */
   set_gdbarch_push_dummy_call (gdbarch, i386_push_dummy_call);
+  set_gdbarch_frame_align (gdbarch, i386_frame_align);
 
   set_gdbarch_convert_register_p (gdbarch, i386_convert_register_p);
   set_gdbarch_register_to_value (gdbarch,  i386_register_to_value);

-- 
Joseph S. Myers
joseph@codesourcery.com