Add optimized version of memcpy on ia32.

src/ia32/assembler-ia32.h

Index: src/ia32/assembler-ia32.h
diff --git a/src/ia32/assembler-ia32.h b/src/ia32/assembler-ia32.h
index 9ece74432b87ff8bd86de27934a157bc3558ddd2..52d3c640003ef0d63b8977a1ac7d34350da23352 100644
--- a/src/ia32/assembler-ia32.h
+++ b/src/ia32/assembler-ia32.h
@@ -790,6 +790,15 @@ class Assembler : public Malloced {
   void pxor(XMMRegister dst, XMMRegister src);
   void ptest(XMMRegister dst, XMMRegister src);
 
+  // Parallel XMM operations.
+  void movntdqa(XMMRegister src, const Operand& dst);
+  void movntdq(const Operand& dst, XMMRegister src);
+  // Prefetch src position into cache level.
+  // Level 1, 2 or 3 specifies CPU cache level. Level 0 specifies a
+  // non-temporal
+  void prefetch(const Operand& src, int level);
+  // TODO(lrn): Need SFENCE for movnt?
+
   // Debugging
   void Print();