diff --git a/Eigen/src/Core/arch/CUDA/PacketMathHalf.h b/Eigen/src/Core/arch/CUDA/PacketMathHalf.h
index f761c938a..18a25de8e 100644
--- a/Eigen/src/Core/arch/CUDA/PacketMathHalf.h
+++ b/Eigen/src/Core/arch/CUDA/PacketMathHalf.h
@@ -531,6 +531,14 @@ ptranspose(PacketBlock<Packet8h,4>& kernel) {
 
 #elif defined(EIGEN_VECTORIZE_SSE) && !EIGEN_ARCH_x86_64
 
+} // end namespace internal
+} // end namespace Eigen
+
+#include <mmintrin.h>
+
+namespace Eigen {
+namespace internal {
+
 typedef struct {
   __m64 x;
 } Packet4h;
diff --git a/Eigen/src/Core/arch/CUDA/TypeCasting.h b/Eigen/src/Core/arch/CUDA/TypeCasting.h
index 07d272ea6..a06675812 100644
--- a/Eigen/src/Core/arch/CUDA/TypeCasting.h
+++ b/Eigen/src/Core/arch/CUDA/TypeCasting.h
@@ -130,6 +130,14 @@ template<> EIGEN_STRONG_INLINE Packet8h pcast<Packet8f, Packet8h>(const Packet8f
 
 #elif defined(EIGEN_VECTORIZE_SSE) && !EIGEN_ARCH_x86_64
 
+} // end namespace internal
+} // end namespace Eigen
+
+#include <mmintrin.h>
+
+namespace Eigen {
+namespace internal {
+
 template <>
 struct type_casting_traits<Eigen::half, float> {
   enum {