diff options
Diffstat (limited to 'Eigen/src/Core/arch/AVX/TypeCasting.h')
-rw-r--r-- | Eigen/src/Core/arch/AVX/TypeCasting.h | 66 |
1 files changed, 65 insertions, 1 deletions
diff --git a/Eigen/src/Core/arch/AVX/TypeCasting.h b/Eigen/src/Core/arch/AVX/TypeCasting.h index 83bfdc604..d507fb67b 100644 --- a/Eigen/src/Core/arch/AVX/TypeCasting.h +++ b/Eigen/src/Core/arch/AVX/TypeCasting.h @@ -35,15 +35,79 @@ struct type_casting_traits<int, float> { }; +#ifndef EIGEN_VECTORIZE_AVX512 + +template <> +struct type_casting_traits<Eigen::half, float> { + enum { + VectorizedCast = 1, + SrcCoeffRatio = 1, + TgtCoeffRatio = 1 + }; +}; + + +template <> +struct type_casting_traits<float, Eigen::half> { + enum { + VectorizedCast = 1, + SrcCoeffRatio = 1, + TgtCoeffRatio = 1 + }; +}; + +template <> +struct type_casting_traits<bfloat16, float> { + enum { + VectorizedCast = 1, + SrcCoeffRatio = 1, + TgtCoeffRatio = 1 + }; +}; + +template <> +struct type_casting_traits<float, bfloat16> { + enum { + VectorizedCast = 1, + SrcCoeffRatio = 1, + TgtCoeffRatio = 1 + }; +}; + +#endif // EIGEN_VECTORIZE_AVX512 template<> EIGEN_STRONG_INLINE Packet8i pcast<Packet8f, Packet8i>(const Packet8f& a) { - return _mm256_cvtps_epi32(a); + return _mm256_cvttps_epi32(a); } template<> EIGEN_STRONG_INLINE Packet8f pcast<Packet8i, Packet8f>(const Packet8i& a) { return _mm256_cvtepi32_ps(a); } +template<> EIGEN_STRONG_INLINE Packet8i preinterpret<Packet8i,Packet8f>(const Packet8f& a) { + return _mm256_castps_si256(a); +} + +template<> EIGEN_STRONG_INLINE Packet8f preinterpret<Packet8f,Packet8i>(const Packet8i& a) { + return _mm256_castsi256_ps(a); +} + +template<> EIGEN_STRONG_INLINE Packet8f pcast<Packet8h, Packet8f>(const Packet8h& a) { + return half2float(a); +} + +template<> EIGEN_STRONG_INLINE Packet8f pcast<Packet8bf, Packet8f>(const Packet8bf& a) { + return Bf16ToF32(a); +} + +template<> EIGEN_STRONG_INLINE Packet8h pcast<Packet8f, Packet8h>(const Packet8f& a) { + return float2half(a); +} + +template<> EIGEN_STRONG_INLINE Packet8bf pcast<Packet8f, Packet8bf>(const Packet8f& a) { + return F32ToBf16(a); +} + } // end namespace internal } // end namespace Eigen |