Merge pull request #407 from trofi/ia64-small-struct

ia64: fix small struct return

src/ia64/unix.S

@@ -175,7 +175,6 @@ ffi_call_unix:
 	;;
 
 .Lst_small_struct:
-	add	sp = -16, sp
 	cmp.lt	p6, p0 = 8, in3
 	cmp.lt	p7, p0 = 16, in3
 	cmp.lt	p8, p0 = 24, in3
@@ -191,6 +190,12 @@ ffi_call_unix:
 (p8)	st8	[r18] = r11
 	mov	out1 = sp
 	mov	out2 = in3
+	;;
+	// ia64 software calling convention requires
+	// top 16 bytes of stack to be scratch space
+	// PLT resolver uses that scratch space at
+	// 'memcpy' symbol reolution time
+	add	sp = -16, sp
 	br.call.sptk.many b0 = memcpy#
 	;;
 	mov	ar.pfs = loc0