diff options
| author |  Niels Dekker <N.Dekker@lumc.nl> | 2020-07-11 12:50:46 +0200 |
|---|---|---|
| committer | Niels Dekker <N.Dekker@lumc.nl> | 2020-07-11 13:32:28 +0200 |
| commit | 4ab32e2de2511746e2108563a43cbbeb1922fbf2 (patch) | |
| tree | a06f6a180f538ce9dfd79138bd7c2efbf5c1e65b /Eigen | |
| parent | dcf7655b3d469a399c1182f350c9009e13ad8654 (diff) | |
Allow implicit conversion from bfloat16 to float and double
Conversion from `bfloat16` to `float` and `double` is lossless. It seems natural to allow the conversion to be implicit, as the C++ language also support implicit conversion from a smaller to a larger floating point type.
Intel's OneDLL bfloat16 implementation also has an implicit `operator float()`: https://github.com/oneapi-src/oneDNN/blob/v1.5/src/common/bfloat16.hpp
Diffstat (limited to 'Eigen')
| -rw-r--r-- | Eigen/src/Core/arch/Default/BFloat16.h | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/Eigen/src/Core/arch/Default/BFloat16.h b/Eigen/src/Core/arch/Default/BFloat16.h index 561304f80..abf2ac933 100644 --- a/Eigen/src/Core/arch/Default/BFloat16.h +++ b/Eigen/src/Core/arch/Default/BFloat16.h @@ -117,10 +117,10 @@ struct bfloat16 : public bfloat16_impl::bfloat16_base { EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(unsigned long long) const { return static_cast<unsigned long long>(bfloat16_to_float(*this)); } - EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(float) const { + EIGEN_DEVICE_FUNC operator float() const { return bfloat16_impl::bfloat16_to_float(*this); } - EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(double) const { + EIGEN_DEVICE_FUNC operator double() const { return static_cast<double>(bfloat16_impl::bfloat16_to_float(*this)); } template<typename RealScalar> |