Add pinsertfirst function and implement pinsertlast for complex on SSE/AVX.

2 files changed, 38 insertions, 0 deletions

diff --git a/Eigen/src/Core/arch/SSE/Complex.h b/Eigen/src/Core/arch/SSE/Complex.h
index fd7f4d740..5607fe0ab 100644
--- a/Eigen/src/Core/arch/SSE/Complex.h
+++ b/Eigen/src/Core/arch/SSE/Complex.h
@@ -476,6 +476,26 @@ template<>  EIGEN_STRONG_INLINE Packet2cf pblend(const Selector<2>& ifPacket, co
   return Packet2cf(_mm_castpd_ps(result));
 }
 
+template<> EIGEN_STRONG_INLINE Packet2cf pinsertfirst(const Packet2cf& a, std::complex<float> b)
+{
+  return Packet2cf(_mm_loadl_pi(a.v, reinterpret_cast<const __m64*>(&b)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd pinsertfirst(const Packet1cd&, std::complex<double> b)
+{
+  return pset1<Packet1cd>(b);
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cf pinsertlast(const Packet2cf& a, std::complex<float> b)
+{
+  return Packet2cf(_mm_loadh_pi(a.v, reinterpret_cast<const __m64*>(&b)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd pinsertlast(const Packet1cd&, std::complex<double> b)
+{
+  return pset1<Packet1cd>(b);
+}
+
 } // end namespace internal
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/arch/SSE/PacketMath.h b/Eigen/src/Core/arch/SSE/PacketMath.h
index b519e3748..6f31cf12b 100755
--- a/Eigen/src/Core/arch/SSE/PacketMath.h
+++ b/Eigen/src/Core/arch/SSE/PacketMath.h
@@ -818,6 +818,24 @@ template<> EIGEN_STRONG_INLINE Packet2d pblend(const Selector<2>& ifPacket, cons
 #endif
 }
 
+template<> EIGEN_STRONG_INLINE Packet4f pinsertfirst(const Packet4f& a, float b)
+{
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  return _mm_blend_ps(a,pset1<Packet4f>(b),1);
+#else
+  return _mm_move_ss(a, _mm_load_ss(&b));
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pinsertfirst(const Packet2d& a, double b)
+{
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  return _mm_blend_pd(a,pset1<Packet2d>(b),1);
+#else
+  return _mm_move_sd(a, _mm_load_sd(&b));
+#endif
+}
+
 template<> EIGEN_STRONG_INLINE Packet4f pinsertlast(const Packet4f& a, float b)
 {
 #ifdef EIGEN_VECTORIZE_SSE4_1