diff options
Diffstat (limited to 'Eigen/src/Core/arch/CUDA/TypeCasting.h')
-rw-r--r-- | Eigen/src/Core/arch/CUDA/TypeCasting.h | 25 |
1 files changed, 2 insertions, 23 deletions
diff --git a/Eigen/src/Core/arch/CUDA/TypeCasting.h b/Eigen/src/Core/arch/CUDA/TypeCasting.h index b2a9724de..396b38eaf 100644 --- a/Eigen/src/Core/arch/CUDA/TypeCasting.h +++ b/Eigen/src/Core/arch/CUDA/TypeCasting.h @@ -71,6 +71,7 @@ struct functor_traits<scalar_cast_op<half, float> > +#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 530 template <> struct type_casting_traits<half, float> { @@ -82,22 +83,9 @@ struct type_casting_traits<half, float> { }; template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pcast<half2, float4>(const half2& a, const half2& b) { -#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300 float2 r1 = __half22float2(a); float2 r2 = __half22float2(b); return make_float4(r1.x, r1.y, r2.x, r2.y); -#else - half r1; - r1.x = a.x & 0xFFFF; - half r2; - r2.x = (a.x & 0xFFFF0000) >> 16; - half r3; - r3.x = b.x & 0xFFFF; - half r4; - r4.x = (b.x & 0xFFFF0000) >> 16; - return make_float4(static_cast<float>(r1), static_cast<float>(r2), - static_cast<float>(r3), static_cast<float>(r4)); -#endif } template <> @@ -111,20 +99,11 @@ struct type_casting_traits<float, half> { template<> EIGEN_STRONG_INLINE half2 pcast<float4, half2>(const float4& a) { // Simply discard the second half of the input -#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300 return __float22half2_rn(make_float2(a.x, a.y)); -#else - half r1 = static_cast<half>(a.x); - half r2 = static_cast<half>(a.y); - half2 r; - r.x = 0; - r.x |= r1.x; - r.x |= (static_cast<unsigned int>(r2.x) << 16); - return r; -#endif } #endif +#endif } // end namespace internal |