diff --git a/Eigen/src/Core/arch/CUDA/MathFunctions.h b/Eigen/src/Core/arch/CUDA/MathFunctions.h
index e7305c01e..3bea88bea 100644
--- a/Eigen/src/Core/arch/CUDA/MathFunctions.h
+++ b/Eigen/src/Core/arch/CUDA/MathFunctions.h
@@ -18,49 +18,49 @@ namespace internal {
 // introduce conflicts between these packet_traits definitions and the ones
 // we'll use on the host side (SSE, AVX, ...)
 #if defined(__CUDACC__) && defined(EIGEN_USE_GPU)
-template<> EIGEN_STRONG_INLINE
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
 float4 plog<float4>(const float4& a)
 {
   return make_float4(logf(a.x), logf(a.y), logf(a.z), logf(a.w));
 }
 
-template<> EIGEN_STRONG_INLINE
+template<>  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
 double2 plog<double2>(const double2& a)
 {
   return make_double2(log(a.x), log(a.y));
 }
 
-template<> EIGEN_STRONG_INLINE
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
 float4 pexp<float4>(const float4& a)
 {
   return make_float4(expf(a.x), expf(a.y), expf(a.z), expf(a.w));
 }
 
-template<> EIGEN_STRONG_INLINE
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
 double2 pexp<double2>(const double2& a)
 {
   return make_double2(exp(a.x), exp(a.y));
 }
 
-template<> EIGEN_STRONG_INLINE
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
 float4 psqrt<float4>(const float4& a)
 {
   return make_float4(sqrtf(a.x), sqrtf(a.y), sqrtf(a.z), sqrtf(a.w));
 }
 
-template<> EIGEN_STRONG_INLINE
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
 double2 psqrt<double2>(const double2& a)
 {
   return make_double2(sqrt(a.x), sqrt(a.y));
 }
 
-template<> EIGEN_STRONG_INLINE
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
 float4 prsqrt<float4>(const float4& a)
 {
   return make_float4(rsqrtf(a.x), rsqrtf(a.y), rsqrtf(a.z), rsqrtf(a.w));
 }
 
-template<> EIGEN_STRONG_INLINE
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
 double2 prsqrt<double2>(const double2& a)
 {
   return make_double2(rsqrt(a.x), rsqrt(a.y));