aboutsummaryrefslogtreecommitdiff
path: root/Eigen/src/Core/arch/ZVector/Complex.h
diff options
context:
space:
mode:
Diffstat (limited to 'Eigen/src/Core/arch/ZVector/Complex.h')
-rw-r--r--Eigen/src/Core/arch/ZVector/Complex.h392
1 files changed, 212 insertions, 180 deletions
diff --git a/Eigen/src/Core/arch/ZVector/Complex.h b/Eigen/src/Core/arch/ZVector/Complex.h
index d39d2d105..0b9b33d99 100644
--- a/Eigen/src/Core/arch/ZVector/Complex.h
+++ b/Eigen/src/Core/arch/ZVector/Complex.h
@@ -15,6 +15,10 @@ namespace Eigen {
namespace internal {
+#if !defined(__ARCH__) || (defined(__ARCH__) && __ARCH__ >= 12)
+static Packet4ui p4ui_CONJ_XOR = { 0x00000000, 0x80000000, 0x00000000, 0x80000000 }; //vec_mergeh((Packet4ui)p4i_ZERO, (Packet4ui)p4f_MZERO);
+#endif
+
static Packet2ul p2ul_CONJ_XOR1 = (Packet2ul) vec_sld((Packet4ui) p2d_ZERO_, (Packet4ui) p2l_ZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
static Packet2ul p2ul_CONJ_XOR2 = (Packet2ul) vec_sld((Packet4ui) p2l_ZERO, (Packet4ui) p2d_ZERO_, 8);//{ 0x8000000000000000, 0x0000000000000000 };
@@ -29,10 +33,14 @@ struct Packet2cf
{
EIGEN_STRONG_INLINE Packet2cf() {}
EIGEN_STRONG_INLINE explicit Packet2cf(const Packet4f& a) : v(a) {}
+#if !defined(__ARCH__) || (defined(__ARCH__) && __ARCH__ < 12)
union {
Packet4f v;
Packet1cd cd[2];
};
+#else
+ Packet4f v;
+#endif
};
template<> struct packet_traits<std::complex<float> > : default_packet_traits
@@ -83,69 +91,33 @@ template<> struct packet_traits<std::complex<double> > : default_packet_traits
};
};
-template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2, alignment=Aligned16}; typedef Packet2cf half; };
-template<> struct unpacket_traits<Packet1cd> { typedef std::complex<double> type; enum {size=1, alignment=Aligned16}; typedef Packet1cd half; };
+template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2, alignment=Aligned16, vectorizable=true, masked_load_available=false, masked_store_available=false}; typedef Packet2cf half; };
+template<> struct unpacket_traits<Packet1cd> { typedef std::complex<double> type; enum {size=1, alignment=Aligned16, vectorizable=true, masked_load_available=false, masked_store_available=false}; typedef Packet1cd half; };
/* Forward declaration */
EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet2cf,2>& kernel);
-template<> EIGEN_STRONG_INLINE Packet2cf pload <Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cf(pload<Packet4f>((const float*)from)); }
+/* complex<double> first */
template<> EIGEN_STRONG_INLINE Packet1cd pload <Packet1cd>(const std::complex<double>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet1cd(pload<Packet2d>((const double*)from)); }
-template<> EIGEN_STRONG_INLINE Packet2cf ploadu<Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cf(ploadu<Packet4f>((const float*)from)); }
template<> EIGEN_STRONG_INLINE Packet1cd ploadu<Packet1cd>(const std::complex<double>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet1cd(ploadu<Packet2d>((const double*)from)); }
-template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((float*)to, from.v); }
template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> * to, const Packet1cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, from.v); }
-template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((float*)to, from.v); }
template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> * to, const Packet1cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, from.v); }
template<> EIGEN_STRONG_INLINE Packet1cd pset1<Packet1cd>(const std::complex<double>& from)
{ /* here we really have to use unaligned loads :( */ return ploadu<Packet1cd>(&from); }
-template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<float>& from)
-{
- Packet2cf res;
- res.cd[0] = Packet1cd(vec_ld2f((const float *)&from));
- res.cd[1] = res.cd[0];
- return res;
-}
-template<> EIGEN_DEVICE_FUNC inline Packet2cf pgather<std::complex<float>, Packet2cf>(const std::complex<float>* from, Index stride)
-{
- std::complex<float> EIGEN_ALIGN16 af[2];
- af[0] = from[0*stride];
- af[1] = from[1*stride];
- return pload<Packet2cf>(af);
-}
template<> EIGEN_DEVICE_FUNC inline Packet1cd pgather<std::complex<double>, Packet1cd>(const std::complex<double>* from, Index stride EIGEN_UNUSED)
{
return pload<Packet1cd>(from);
}
-template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet2cf>(std::complex<float>* to, const Packet2cf& from, Index stride)
-{
- std::complex<float> EIGEN_ALIGN16 af[2];
- pstore<std::complex<float> >((std::complex<float> *) af, from);
- to[0*stride] = af[0];
- to[1*stride] = af[1];
-}
template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<double>, Packet1cd>(std::complex<double>* to, const Packet1cd& from, Index stride EIGEN_UNUSED)
{
pstore<std::complex<double> >(to, from);
}
-
-template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(padd<Packet4f>(a.v, b.v)); }
template<> EIGEN_STRONG_INLINE Packet1cd padd<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(a.v + b.v); }
-template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(psub<Packet4f>(a.v, b.v)); }
template<> EIGEN_STRONG_INLINE Packet1cd psub<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(a.v - b.v); }
template<> EIGEN_STRONG_INLINE Packet1cd pnegate(const Packet1cd& a) { return Packet1cd(pnegate(Packet2d(a.v))); }
-template<> EIGEN_STRONG_INLINE Packet2cf pnegate(const Packet2cf& a) { return Packet2cf(pnegate(Packet4f(a.v))); }
template<> EIGEN_STRONG_INLINE Packet1cd pconj(const Packet1cd& a) { return Packet1cd((Packet2d)vec_xor((Packet2d)a.v, (Packet2d)p2ul_CONJ_XOR2)); }
-template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a)
-{
- Packet2cf res;
- res.v.v4f[0] = pconj(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[0]))).v;
- res.v.v4f[1] = pconj(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[1]))).v;
- return res;
-}
-
template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
{
Packet2d a_re, a_im, v1, v2;
@@ -163,27 +135,17 @@ template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, con
return Packet1cd(v1 + v2);
}
-template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
-{
- Packet2cf res;
- res.v.v4f[0] = pmul(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[0])), Packet1cd(reinterpret_cast<Packet2d>(b.v.v4f[0]))).v;
- res.v.v4f[1] = pmul(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[1])), Packet1cd(reinterpret_cast<Packet2d>(b.v.v4f[1]))).v;
- return res;
-}
-
-template<> EIGEN_STRONG_INLINE Packet1cd pand <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_and(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cf pand <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pand<Packet4f>(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet1cd por <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_or(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cf por <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(por<Packet4f>(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet1cd pxor <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_xor(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cf pxor <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pxor<Packet4f>(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet1cd pandnot<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_and(a.v, vec_nor(b.v,b.v))); }
-template<> EIGEN_STRONG_INLINE Packet2cf pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pandnot<Packet4f>(a.v,b.v)); }
-
+template<> EIGEN_STRONG_INLINE Packet1cd pand <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_and(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd por <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_or(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pxor <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_xor(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pandnot <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_and(a.v, vec_nor(b.v,b.v))); }
template<> EIGEN_STRONG_INLINE Packet1cd ploaddup<Packet1cd>(const std::complex<double>* from) { return pset1<Packet1cd>(*from); }
-template<> EIGEN_STRONG_INLINE Packet2cf ploaddup<Packet2cf>(const std::complex<float>* from) { return pset1<Packet2cf>(*from); }
+template<> EIGEN_STRONG_INLINE Packet1cd pcmp_eq(const Packet1cd& a, const Packet1cd& b) {
+ Packet2d eq = vec_cmpeq (a.v, b.v);
+ Packet2d tmp = { eq[1], eq[0] };
+ return (Packet1cd)pand<Packet2d>(eq, tmp);
+}
-template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> * addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
template<> EIGEN_STRONG_INLINE void prefetch<std::complex<double> >(const std::complex<double> * addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
template<> EIGEN_STRONG_INLINE std::complex<double> pfirst<Packet1cd>(const Packet1cd& a)
@@ -193,157 +155,157 @@ template<> EIGEN_STRONG_INLINE std::complex<double> pfirst<Packet1cd>(const Pac
return res;
}
-template<> EIGEN_STRONG_INLINE std::complex<float> pfirst<Packet2cf>(const Packet2cf& a)
-{
- std::complex<float> EIGEN_ALIGN16 res[2];
- pstore<std::complex<float> >(res, a);
-
- return res[0];
-}
template<> EIGEN_STRONG_INLINE Packet1cd preverse(const Packet1cd& a) { return a; }
-template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a)
+template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet1cd>(const Packet1cd& a)
{
- Packet2cf res;
- res.cd[0] = a.cd[1];
- res.cd[1] = a.cd[0];
- return res;
+ return pfirst(a);
}
-
-template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet1cd>(const Packet1cd& a)
+template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet1cd>(const Packet1cd& a)
{
return pfirst(a);
}
-template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet2cf>(const Packet2cf& a)
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet1cd,Packet2d)
+
+template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
{
- std::complex<float> res;
- Packet1cd b = padd<Packet1cd>(a.cd[0], a.cd[1]);
- vec_st2f(b.v, (float*)&res);
- return res;
+ // TODO optimize it for AltiVec
+ 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)));
}
-template<> EIGEN_STRONG_INLINE Packet1cd preduxp<Packet1cd>(const Packet1cd* vecs)
+EIGEN_STRONG_INLINE Packet1cd pcplxflip/*<Packet1cd>*/(const Packet1cd& x)
{
- return vecs[0];
+ return Packet1cd(preverse(Packet2d(x.v)));
}
-template<> EIGEN_STRONG_INLINE Packet2cf preduxp<Packet2cf>(const Packet2cf* vecs)
+
+EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet1cd,2>& kernel)
{
- PacketBlock<Packet2cf,2> transpose;
- transpose.packet[0] = vecs[0];
- transpose.packet[1] = vecs[1];
- ptranspose(transpose);
+ Packet2d tmp = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_HI);
+ kernel.packet[1].v = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_LO);
+ kernel.packet[0].v = tmp;
+}
- return padd<Packet2cf>(transpose.packet[0], transpose.packet[1]);
-}
+/* complex<float> follows */
+template<> EIGEN_STRONG_INLINE Packet2cf pload <Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cf(pload<Packet4f>((const float*)from)); }
+template<> EIGEN_STRONG_INLINE Packet2cf ploadu<Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cf(ploadu<Packet4f>((const float*)from)); }
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((float*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((float*)to, from.v); }
-template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet1cd>(const Packet1cd& a)
+template<> EIGEN_STRONG_INLINE std::complex<float> pfirst<Packet2cf>(const Packet2cf& a)
{
- return pfirst(a);
+ std::complex<float> EIGEN_ALIGN16 res[2];
+ pstore<std::complex<float> >(res, a);
+
+ return res[0];
}
-template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet2cf>(const Packet2cf& a)
+
+
+#if !defined(__ARCH__) || (defined(__ARCH__) && __ARCH__ < 12)
+template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<float>& from)
{
- std::complex<float> res;
- Packet1cd b = pmul<Packet1cd>(a.cd[0], a.cd[1]);
- vec_st2f(b.v, (float*)&res);
+ Packet2cf res;
+ res.cd[0] = Packet1cd(vec_ld2f((const float *)&from));
+ res.cd[1] = res.cd[0];
return res;
}
-
-template<int Offset>
-struct palign_impl<Offset,Packet1cd>
+#else
+template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<float>& from)
{
- static EIGEN_STRONG_INLINE void run(Packet1cd& /*first*/, const Packet1cd& /*second*/)
- {
- // FIXME is it sure we never have to align a Packet1cd?
- // Even though a std::complex<double> has 16 bytes, it is not necessarily aligned on a 16 bytes boundary...
- }
-};
+ Packet2cf res;
+ if((std::ptrdiff_t(&from) % 16) == 0)
+ res.v = pload<Packet4f>((const float *)&from);
+ else
+ res.v = ploadu<Packet4f>((const float *)&from);
+ res.v = vec_perm(res.v, res.v, p16uc_PSET64_HI);
+ return res;
+}
+#endif
-template<int Offset>
-struct palign_impl<Offset,Packet2cf>
+template<> EIGEN_DEVICE_FUNC inline Packet2cf pgather<std::complex<float>, Packet2cf>(const std::complex<float>* from, Index stride)
{
- static EIGEN_STRONG_INLINE void run(Packet2cf& first, const Packet2cf& second)
- {
- if (Offset == 1) {
- first.cd[0] = first.cd[1];
- first.cd[1] = second.cd[0];
- }
- }
-};
-
-template<> struct conj_helper<Packet1cd, Packet1cd, false,true>
+ std::complex<float> EIGEN_ALIGN16 af[2];
+ af[0] = from[0*stride];
+ af[1] = from[1*stride];
+ return pload<Packet2cf>(af);
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet2cf>(std::complex<float>* to, const Packet2cf& from, Index stride)
{
- EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
- { return padd(pmul(x,y),c); }
+ std::complex<float> EIGEN_ALIGN16 af[2];
+ pstore<std::complex<float> >((std::complex<float> *) af, from);
+ to[0*stride] = af[0];
+ to[1*stride] = af[1];
+}
- EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
- {
- return internal::pmul(a, pconj(b));
- }
-};
+template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(padd<Packet4f>(a.v, b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(psub<Packet4f>(a.v, b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pnegate(const Packet2cf& a) { return Packet2cf(pnegate(Packet4f(a.v))); }
-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); }
+template<> EIGEN_STRONG_INLINE Packet2cf pand <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pand<Packet4f>(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf por <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(por<Packet4f>(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pxor <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pxor<Packet4f>(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pandnot<Packet4f>(a.v,b.v)); }
- EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
- {
- return internal::pmul(pconj(a), b);
- }
-};
+template<> EIGEN_STRONG_INLINE Packet2cf ploaddup<Packet2cf>(const std::complex<float>* from) { return pset1<Packet2cf>(*from); }
-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); }
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> * addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
- EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
- {
- return pconj(internal::pmul(a, b));
- }
-};
-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); }
+#if !defined(__ARCH__) || (defined(__ARCH__) && __ARCH__ < 12)
- EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
- {
- return internal::pmul(a, pconj(b));
- }
-};
+template<> EIGEN_STRONG_INLINE Packet2cf pcmp_eq(const Packet2cf& a, const Packet2cf& b) {
+ Packet4f eq = pcmp_eq<Packet4f> (a.v, b.v);
+ Packet2cf res;
+ Packet2d tmp1 = { eq.v4f[0][1], eq.v4f[0][0] };
+ Packet2d tmp2 = { eq.v4f[1][1], eq.v4f[1][0] };
+ res.v.v4f[0] = pand<Packet2d>(eq.v4f[0], tmp1);
+ res.v.v4f[1] = pand<Packet2d>(eq.v4f[1], tmp2);
+ return res;
+}
-template<> struct conj_helper<Packet2cf, Packet2cf, true,false>
+template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a)
{
- EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
- { return padd(pmul(x,y),c); }
+ Packet2cf res;
+ res.v.v4f[0] = pconj(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[0]))).v;
+ res.v.v4f[1] = pconj(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[1]))).v;
+ return res;
+}
- EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
- {
- return internal::pmul(pconj(a), b);
- }
-};
+template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
+{
+ Packet2cf res;
+ res.v.v4f[0] = pmul(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[0])), Packet1cd(reinterpret_cast<Packet2d>(b.v.v4f[0]))).v;
+ res.v.v4f[1] = pmul(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[1])), Packet1cd(reinterpret_cast<Packet2d>(b.v.v4f[1]))).v;
+ return res;
+}
-template<> struct conj_helper<Packet2cf, Packet2cf, true,true>
+template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a)
{
- EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
- { return padd(pmul(x,y),c); }
+ Packet2cf res;
+ res.cd[0] = a.cd[1];
+ res.cd[1] = a.cd[0];
+ return res;
+}
- EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
- {
- return pconj(internal::pmul(a, b));
- }
-};
+template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet2cf>(const Packet2cf& a)
+{
+ std::complex<float> res;
+ Packet1cd b = padd<Packet1cd>(a.cd[0], a.cd[1]);
+ vec_st2f(b.v, (float*)&res);
+ return res;
+}
-template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet2cf>(const Packet2cf& a)
{
- // TODO optimize it for AltiVec
- Packet1cd res = conj_helper<Packet1cd,Packet1cd,false,true>().pmul(a,b);
- Packet2d s = vec_madd(b.v, b.v, p2d_ZERO_);
- return Packet1cd(pdiv(res.v, s + vec_perm(s, s, p16uc_REVERSE64)));
+ std::complex<float> res;
+ Packet1cd b = pmul<Packet1cd>(a.cd[0], a.cd[1]);
+ vec_st2f(b.v, (float*)&res);
+ return res;
}
+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
@@ -353,11 +315,6 @@ template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, con
return res;
}
-EIGEN_STRONG_INLINE Packet1cd pcplxflip/*<Packet1cd>*/(const Packet1cd& x)
-{
- return Packet1cd(preverse(Packet2d(x.v)));
-}
-
EIGEN_STRONG_INLINE Packet2cf pcplxflip/*<Packet2cf>*/(const Packet2cf& x)
{
Packet2cf res;
@@ -366,13 +323,6 @@ EIGEN_STRONG_INLINE Packet2cf pcplxflip/*<Packet2cf>*/(const Packet2cf& x)
return res;
}
-EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet1cd,2>& kernel)
-{
- Packet2d tmp = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_HI);
- kernel.packet[1].v = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_LO);
- kernel.packet[0].v = tmp;
-}
-
EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet2cf,2>& kernel)
{
Packet1cd tmp = kernel.packet[0].cd[1];
@@ -386,6 +336,88 @@ template<> EIGEN_STRONG_INLINE Packet2cf pblend(const Selector<2>& ifPacket, con
result.v = pblend<Packet4f>(ifPacket4, thenPacket.v, elsePacket.v);
return result;
}
+#else
+template<> EIGEN_STRONG_INLINE Packet2cf pcmp_eq(const Packet2cf& a, const Packet2cf& b) {
+ Packet4f eq = vec_cmpeq (a.v, b.v);
+ Packet4f tmp = { eq[1], eq[0], eq[3], eq[2] };
+ return (Packet2cf)pand<Packet4f>(eq, tmp);
+}
+template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a) { return Packet2cf(pxor<Packet4f>(a.v, reinterpret_cast<Packet4f>(p4ui_CONJ_XOR))); }
+template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
+{
+ Packet4f a_re, a_im, prod, prod_im;
+
+ // Permute and multiply the real parts of a and b
+ a_re = vec_perm(a.v, a.v, p16uc_PSET32_WODD);
+
+ // Get the imaginary parts of a
+ a_im = vec_perm(a.v, a.v, p16uc_PSET32_WEVEN);
+
+ // multiply a_im * b and get the conjugate result
+ prod_im = a_im * b.v;
+ prod_im = pxor<Packet4f>(prod_im, reinterpret_cast<Packet4f>(p4ui_CONJ_XOR));
+ // permute back to a proper order
+ prod_im = vec_perm(prod_im, prod_im, p16uc_COMPLEX32_REV);
+
+ // multiply a_re * b, add prod_im
+ prod = pmadd<Packet4f>(a_re, b.v, prod_im);
+
+ return Packet2cf(prod);
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a)
+{
+ Packet4f rev_a;
+ rev_a = vec_perm(a.v, a.v, p16uc_COMPLEX32_REV2);
+ return Packet2cf(rev_a);
+}
+
+template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet2cf>(const Packet2cf& a)
+{
+ Packet4f b;
+ b = vec_sld(a.v, a.v, 8);
+ b = padd<Packet4f>(a.v, b);
+ return pfirst<Packet2cf>(Packet2cf(b));
+}
+
+template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet2cf>(const Packet2cf& a)
+{
+ Packet4f b;
+ Packet2cf prod;
+ b = vec_sld(a.v, a.v, 8);
+ prod = pmul<Packet2cf>(a, Packet2cf(b));
+
+ return pfirst<Packet2cf>(prod);
+}
+
+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 = 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))));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cf pcplxflip<Packet2cf>(const Packet2cf& x)
+{
+ return Packet2cf(vec_perm(x.v, x.v, p16uc_COMPLEX32_REV));
+}
+
+EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet2cf,2>& kernel)
+{
+ Packet4f tmp = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_HI);
+ kernel.packet[1].v = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_LO);
+ kernel.packet[0].v = tmp;
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cf pblend(const Selector<2>& ifPacket, const Packet2cf& thenPacket, const Packet2cf& elsePacket) {
+ Packet2cf result;
+ result.v = reinterpret_cast<Packet4f>(pblend<Packet2d>(ifPacket, reinterpret_cast<Packet2d>(thenPacket.v), reinterpret_cast<Packet2d>(elsePacket.v)));
+ return result;
+}
+#endif
} // end namespace internal
generated by cgit v1.2.3 (git 2.25.1) at 2024-05-17 23:33:13 +0000