GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-04-29 15:24:51 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix vectorization of complex sqrt on NEON

Browse Source
This commit is contained in:
David Tellenbach 2020-12-10 15:23:23 +00:00
parent 8eb461a431
commit 00be0a7ff3
1 changed files with 17 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
Eigen/src/Core/arch/NEON/Complex.h
View File

@ -76,6 +76,7 @@ template<> struct unpacket_traits<Packet1cf>
{ {
typedef std::complex<float> type; typedef std::complex<float> type;
typedef Packet1cf half; typedef Packet1cf half;
typedef Packet2f as_real;
enum enum
{ {
size = 1, size = 1,
@ -89,6 +90,7 @@ template<> struct unpacket_traits<Packet2cf>
{ {
typedef std::complex<float> type; typedef std::complex<float> type;
typedef Packet1cf half; typedef Packet1cf half;
typedef Packet4f as_real;
enum enum
{ {
size = 2, size = 2,
@ -475,6 +477,8 @@ template<> struct packet_traits<std::complex<double> > : default_packet_traits
template<> struct unpacket_traits<Packet1cd> template<> struct unpacket_traits<Packet1cd>
{ {
typedef std::complex<double> type; typedef std::complex<double> type;
typedef Packet1cd half;
typedef Packet2d as_real;
enum enum
{ {
size=1, size=1,
@ -483,7 +487,6 @@ template<> struct unpacket_traits<Packet1cd>
masked_load_available=false, masked_load_available=false,
masked_store_available=false masked_store_available=false
}; };
typedef Packet1cd half;
}; };
template<> EIGEN_STRONG_INLINE Packet1cd pload<Packet1cd>(const std::complex<double>* from) template<> EIGEN_STRONG_INLINE Packet1cd pload<Packet1cd>(const std::complex<double>* from)
@ -640,6 +643,19 @@ EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet1cd,2>& kernel)
kernel.packet[0].v = vcombine_f64(vget_low_f64(kernel.packet[0].v), vget_low_f64(kernel.packet[1].v)); kernel.packet[0].v = vcombine_f64(vget_low_f64(kernel.packet[0].v), vget_low_f64(kernel.packet[1].v));
kernel.packet[1].v = tmp; kernel.packet[1].v = tmp;
} }
template<> EIGEN_STRONG_INLINE Packet1cf psqrt<Packet1cf>(const Packet1cf& a) {
return psqrt_complex<Packet1cf>(a);
}
template<> EIGEN_STRONG_INLINE Packet2cf psqrt<Packet2cf>(const Packet2cf& a) {
return psqrt_complex<Packet2cf>(a);
}
template<> EIGEN_STRONG_INLINE Packet1cd psqrt<Packet1cd>(const Packet1cd& a) {
return psqrt_complex<Packet1cd>(a);
}
#endif // EIGEN_ARCH_ARM64 #endif // EIGEN_ARCH_ARM64
} // end namespace internal } // end namespace internal