devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fixed contractions of 16 bit floats

Browse Source
This commit is contained in:
Benoit Steiner
2016-03-22 09:30:23 -07:00
parent 8ef3181f15
commit f9ad25e4d8
2 changed files with 15 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
Eigen/src/Core/arch/CUDA/Half.h
View File

@@ -73,8 +73,6 @@ struct half : public __half {
: __half(internal::raw_uint16_to_half(b ? 0x3c00 : 0)) {}
EIGEN_DEVICE_FUNC half(const __half& h) : __half(h) {}
EIGEN_DEVICE_FUNC half(const half& h) : __half(h) {}
EIGEN_DEVICE_FUNC half(const volatile half& h)
: __half(internal::raw_uint16_to_half(h.x)) {}
EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(float) const {
return internal::half_to_float(*this);
@@ -87,14 +85,6 @@ struct half : public __half {
x = other.x;
return *this;
}
EIGEN_DEVICE_FUNC half& operator=(const volatile half& other) {
x = other.x;
return *this;
}
EIGEN_DEVICE_FUNC volatile half& operator=(const half& other) volatile {
x = other.x;
return *this;
}
};
#if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 530
@@ -341,4 +331,14 @@ static inline EIGEN_DEVICE_FUNC Eigen::half log(const Eigen::half& a) {
} // end namespace std
// Add the missing shfl_xor intrinsic
#if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300
__device__ inline Eigen::half __shfl_xor(Eigen::half var, int laneMask, int width=warpSize) {
return static_cast<Eigen::half>(__shfl_xor(static_cast<float>(var), laneMask, width));
}
#endif
#endif // EIGEN_HALF_CUDA_H