Add masked_store_available to unpacket_traits

This commit is contained in:
Eugene Zhulenev
2019-05-02 14:52:58 -07:00
parent 96e30e936a
commit e9f0eb8a5e
20 changed files with 53 additions and 49 deletions

View File

@@ -47,7 +47,7 @@ template<> struct packet_traits<std::complex<float> > : default_packet_traits
};
#endif
template<> struct unpacket_traits<Packet4cf> { typedef std::complex<float> type; enum {size=4, alignment=Aligned32, vectorizable=true, masked_load_available=false}; typedef Packet2cf half; };
template<> struct unpacket_traits<Packet4cf> { typedef std::complex<float> type; enum {size=4, alignment=Aligned32, vectorizable=true, masked_load_available=false, masked_store_available=false}; typedef Packet2cf half; };
template<> EIGEN_STRONG_INLINE Packet4cf padd<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_add_ps(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet4cf psub<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_sub_ps(a.v,b.v)); }
@@ -263,7 +263,7 @@ template<> struct packet_traits<std::complex<double> > : default_packet_traits
};
#endif
template<> struct unpacket_traits<Packet2cd> { typedef std::complex<double> type; enum {size=2, alignment=Aligned32, vectorizable=true, masked_load_available=false}; typedef Packet1cd half; };
template<> struct unpacket_traits<Packet2cd> { typedef std::complex<double> type; enum {size=2, alignment=Aligned32, vectorizable=true, masked_load_available=false, masked_store_available=false}; typedef Packet1cd half; };
template<> EIGEN_STRONG_INLINE Packet2cd padd<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_add_pd(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cd psub<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_sub_pd(a.v,b.v)); }

View File

@@ -118,14 +118,14 @@ template<> struct unpacket_traits<Packet8f> {
typedef Packet4f half;
typedef Packet8i integer_packet;
typedef uint8_t mask_t;
enum {size=8, alignment=Aligned32, vectorizable=true, masked_load_available=true};
enum {size=8, alignment=Aligned32, vectorizable=true, masked_load_available=true, masked_store_available=true};
};
template<> struct unpacket_traits<Packet4d> {
typedef double type;
typedef Packet2d half;
enum {size=4, alignment=Aligned32, vectorizable=true, masked_load_available=false};
enum {size=4, alignment=Aligned32, vectorizable=true, masked_load_available=false, masked_store_available=false};
};
template<> struct unpacket_traits<Packet8i> { typedef int type; typedef Packet4i half; enum {size=8, alignment=Aligned32, vectorizable=false, masked_load_available=false}; };
template<> struct unpacket_traits<Packet8i> { typedef int type; typedef Packet4i half; enum {size=8, alignment=Aligned32, vectorizable=false, masked_load_available=false, masked_store_available=false}; };
template<> EIGEN_STRONG_INLINE Packet8f pset1<Packet8f>(const float& from) { return _mm256_set1_ps(from); }
template<> EIGEN_STRONG_INLINE Packet4d pset1<Packet4d>(const double& from) { return _mm256_set1_pd(from); }