Add Altivec implementation of pgather/pscatter (not tested)

2 files changed, 53 insertions, 0 deletions

diff --git a/Eigen/src/Core/arch/AltiVec/Complex.h b/Eigen/src/Core/arch/AltiVec/Complex.h
index db52074f4..ec11cfaa0 100644
--- a/Eigen/src/Core/arch/AltiVec/Complex.h
+++ b/Eigen/src/Core/arch/AltiVec/Complex.h
@@ -66,6 +66,22 @@ template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<flo
   return res;
 }
 
+template<> EIGEN_DEVICE_FUNC inline Packet2cf pgather<std::complex<float>, Packet2cf>(const std::complex<float>* from, int stride)
+{
+  std::complex<float> EIGEN_ALIGN16 af[2];
+  af[0] = from[0*stride];
+  af[1] = from[1*stride];
+ return Packet2cf(vec_ld(0, af));
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet2cf>(std::complex<float>* to, const Packet2cf& from, int stride)
+{
+  std::complex<float> EIGEN_ALIGN16 af[2];
+  vec_st(from.v, 0, af);
+  to[0*stride] = af[0];
+  to[1*stride] = af[1];
+}
+
+
 template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(vec_add(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(vec_sub(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet2cf pnegate(const Packet2cf& a) { return Packet2cf(pnegate(a.v)); }
diff --git a/Eigen/src/Core/arch/AltiVec/PacketMath.h b/Eigen/src/Core/arch/AltiVec/PacketMath.h
index 16948264f..80a99a004 100755
--- a/Eigen/src/Core/arch/AltiVec/PacketMath.h
+++ b/Eigen/src/Core/arch/AltiVec/PacketMath.h
@@ -163,6 +163,43 @@ template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int&    from)   {
   return vc;
 }
 
+template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const float* from, int stride)
+{
+  float EIGEN_ALIGN16 af[4];
+  af[0] = from[0*stride];
+  af[1] = from[1*stride];
+  af[2] = from[2*stride];
+  af[3] = from[3*stride];
+ return vec_ld(0, af);
+}
+template<> EIGEN_DEVICE_FUNC inline Packet4i pgather<int, Packet4i>(const int* from, int stride)
+{
+  int EIGEN_ALIGN16 ai[4];
+  ai[0] = from[0*stride];
+  ai[1] = from[1*stride];
+  ai[2] = from[2*stride];
+  ai[3] = from[3*stride];
+ return vec_ld(0, ai);
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<float, Packet4f>(float* to, const Packet4f& from, int stride)
+{
+  float EIGEN_ALIGN16 af[4];
+  vec_st(from, 0, af);
+  to[0*stride] = af[0];
+  to[1*stride] = af[1];
+  to[2*stride] = af[2];
+  to[3*stride] = af[3];
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<int, Packet4i>(int* to, const Packet4i& from, int stride)
+{
+  int EIGEN_ALIGN16 ai[4];
+  vec_st(from, 0, ai);
+  to[0*stride] = ai[0];
+  to[1*stride] = ai[1];
+  to[2*stride] = ai[2];
+  to[3*stride] = ai[3];
+}
+
 template<> EIGEN_STRONG_INLINE Packet4f plset<float>(const float& a) { return vec_add(pset1<Packet4f>(a), p4f_COUNTDOWN); }
 template<> EIGEN_STRONG_INLINE Packet4i plset<int>(const int& a)     { return vec_add(pset1<Packet4i>(a), p4i_COUNTDOWN); }