Get rid of code duplication for conj_helper. For packets where LhsType=RhsType a single generic implementation suffices. For scalars, the generic implementation of pconj automatically forwards to numext::conj, so much of the existing specialization can be avoided. For mixed types we still need specializations.

This commit is contained in:
Rasmus Munk Larsen
2021-06-24 15:47:48 -07:00
parent 4ad30a73fc
commit 52a5f98212
9 changed files with 105 additions and 635 deletions

View File

@@ -165,45 +165,12 @@ template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet1cd>(const
{
return pfirst(a);
}
template<> struct conj_helper<Packet1cd, Packet1cd, false,true>
{
EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
{ return padd(pmul(x,y),c); }
EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
{
return internal::pmul(a, pconj(b));
}
};
template<> struct conj_helper<Packet1cd, Packet1cd, true,false>
{
EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
{ return padd(pmul(x,y),c); }
EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
{
return internal::pmul(pconj(a), b);
}
};
template<> struct conj_helper<Packet1cd, Packet1cd, true,true>
{
EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
{ return padd(pmul(x,y),c); }
EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
{
return pconj(internal::pmul(a, b));
}
};
EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet1cd,Packet2d)
template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
{
// TODO optimize it for AltiVec
Packet1cd res = conj_helper<Packet1cd,Packet1cd,false,true>().pmul(a,b);
Packet1cd res = pmul(a,pconj(b));
Packet2d s = vec_madd(b.v, b.v, p2d_ZERO_);
return Packet1cd(pdiv(res.v, s + vec_perm(s, s, p16uc_REVERSE64)));
}
@@ -337,39 +304,6 @@ template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet2cf>(const P
return res;
}
template<> struct conj_helper<Packet2cf, Packet2cf, false,true>
{
EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
{ return padd(pmul(x,y),c); }
EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
{
return internal::pmul(a, pconj(b));
}
};
template<> struct conj_helper<Packet2cf, Packet2cf, true,false>
{
EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
{ return padd(pmul(x,y),c); }
EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
{
return internal::pmul(pconj(a), b);
}
};
template<> struct conj_helper<Packet2cf, Packet2cf, true,true>
{
EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
{ return padd(pmul(x,y),c); }
EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
{
return pconj(internal::pmul(a, b));
}
};
EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cf,Packet4f)
template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
@@ -456,45 +390,12 @@ template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet2cf>(const P
return pfirst<Packet2cf>(prod);
}
template<> struct conj_helper<Packet2cf, Packet2cf, false,true>
{
EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
{ return padd(pmul(x,y),c); }
EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
{
return internal::pmul(a, pconj(b));
}
};
template<> struct conj_helper<Packet2cf, Packet2cf, true,false>
{
EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
{ return padd(pmul(x,y),c); }
EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
{
return internal::pmul(pconj(a), b);
}
};
template<> struct conj_helper<Packet2cf, Packet2cf, true,true>
{
EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
{ return padd(pmul(x,y),c); }
EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
{
return pconj(internal::pmul(a, b));
}
};
EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cf,Packet4f)
template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
{
// TODO optimize it for AltiVec
Packet2cf res = conj_helper<Packet2cf,Packet2cf,false,true>().pmul(a, b);
Packet2cf res = pmul(a, pconj(b));
Packet4f s = pmul<Packet4f>(b.v, b.v);
return Packet2cf(pdiv(res.v, padd<Packet4f>(s, vec_perm(s, s, p16uc_COMPLEX32_REV))));
}