Fixing missing inlines on device functions for newer CUDA cards

1 files changed, 16 insertions, 16 deletions

diff --git a/Eigen/src/Core/arch/CUDA/Half.h b/Eigen/src/Core/arch/CUDA/Half.h
index ee24e615a..666b23c6e 100644
--- a/Eigen/src/Core/arch/CUDA/Half.h
+++ b/Eigen/src/Core/arch/CUDA/Half.h
@@ -154,55 +154,55 @@ namespace half_impl {
 // versions to get the ALU speed increased), but you do save the
 // conversion steps back and forth.
 
-__device__ half operator + (const half& a, const half& b) {
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator + (const half& a, const half& b) {
   return __hadd(a, b);
 }
-__device__ half operator * (const half& a, const half& b) {
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator * (const half& a, const half& b) {
   return __hmul(a, b);
 }
-__device__ half operator - (const half& a, const half& b) {
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator - (const half& a, const half& b) {
   return __hsub(a, b);
 }
-__device__ half operator / (const half& a, const half& b) {
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator / (const half& a, const half& b) {
   float num = __half2float(a);
   float denom = __half2float(b);
   return __float2half(num / denom);
 }
-__device__ half operator - (const half& a) {
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator - (const half& a) {
   return __hneg(a);
 }
-__device__ half& operator += (half& a, const half& b) {
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator += (half& a, const half& b) {
   a = a + b;
   return a;
 }
-__device__ half& operator *= (half& a, const half& b) {
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator *= (half& a, const half& b) {
   a = a * b;
   return a;
 }
-__device__ half& operator -= (half& a, const half& b) {
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator -= (half& a, const half& b) {
   a = a - b;
   return a;
 }
-__device__ half& operator /= (half& a, const half& b) {
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator /= (half& a, const half& b) {
   a = a / b;
   return a;
 }
-__device__ bool operator == (const half& a, const half& b) {
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator == (const half& a, const half& b) {
   return __heq(a, b);
 }
-__device__ bool operator != (const half& a, const half& b) {
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator != (const half& a, const half& b) {
   return __hne(a, b);
 }
-__device__ bool operator < (const half& a, const half& b) {
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator < (const half& a, const half& b) {
   return __hlt(a, b);
 }
-__device__ bool operator <= (const half& a, const half& b) {
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator <= (const half& a, const half& b) {
   return __hle(a, b);
 }
-__device__ bool operator > (const half& a, const half& b) {
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator > (const half& a, const half& b) {
   return __hgt(a, b);
 }
-__device__ bool operator >= (const half& a, const half& b) {
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator >= (const half& a, const half& b) {
   return __hge(a, b);
 }
 
@@ -619,7 +619,7 @@ struct hash<Eigen::half> {
 
 // Add the missing shfl_xor intrinsic
 #if defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 300
-__device__ EIGEN_STRONG_INLINE Eigen::half __shfl_xor(Eigen::half var, int laneMask, int width=warpSize) {
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Eigen::half __shfl_xor(Eigen::half var, int laneMask, int width=warpSize) {
   #if EIGEN_CUDACC_VER < 90000
   return static_cast<Eigen::half>(__shfl_xor(static_cast<float>(var), laneMask, width));
   #else