diff --git a/xpcom/reflect/xptcall/src/md/unix/xptcinvoke_gcc_x86_unix.cpp b/xpcom/reflect/xptcall/src/md/unix/xptcinvoke_gcc_x86_unix.cpp
index 324d7e91f77..ae90c2edc7f 100644
--- a/xpcom/reflect/xptcall/src/md/unix/xptcinvoke_gcc_x86_unix.cpp
+++ b/xpcom/reflect/xptcall/src/md/unix/xptcinvoke_gcc_x86_unix.cpp
@@ -161,8 +161,14 @@ __asm__ (
  * is assumed and required, e.g. by GCC4's -ftree-vectorize option.
  */
 	"andl  $0xfffffff0, %ecx\n\t"   /* drop(?) stack ptr to 128-bit align */
-	"subl  $8, %ecx\n\t"	        /* lower again; push/call below will re-align */
-	
+/* $esp should be aligned to a 16-byte boundary here (note we include an 
+ * additional 4 bytes in a later push instruction). This will ensure $ebp 
+ * in the function called below is aligned to a 0x8 boundary. SSE instructions 
+ * like movapd/movdqa expect memory operand to be aligned on a 16-byte
+ * boundary. The GCC compiler will generate the memory operand using $ebp
+ * with an 8-byte offset.
+ */
+	"subl  $0xc, %ecx\n\t"          /* lower again; push/call below will re-align */
 	"movl  %ecx, %esp\n\t"          /* make stack space */
 	"movl  0x14(%ebp), %edx\n\t"
 	"call  " SYMBOL_UNDERSCORE "invoke_copy_to_stack\n\t"