5 files changed, 99 insertions, 1 deletions
diff --git a/Eigen/src/Core/arch/AVX/Complex.h b/Eigen/src/Core/arch/AVX/Complex.h
index b16e0ddd4..99439c8aa 100644
--- a/Eigen/src/Core/arch/AVX/Complex.h
+++ b/Eigen/src/Core/arch/AVX/Complex.h
@@ -456,6 +456,26 @@ ptranspose(PacketBlock<Packet2cd,2>& kernel) {
  kernel.packet[0].v = tmp;
 }
 
+template<> EIGEN_STRONG_INLINE Packet4cf pinsertfirst(const Packet4cf& a, std::complex<float> b)
+{
+  return Packet4cf(_mm256_blend_ps(a.v,pset1<Packet4cf>(b).v,1|2));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd pinsertfirst(const Packet2cd& a, std::complex<double> b)
+{
+  return Packet2cd(_mm256_blend_pd(a.v,pset1<Packet2cd>(b).v,1|2));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf pinsertlast(const Packet4cf& a, std::complex<float> b)
+{
+  return Packet4cf(_mm256_blend_ps(a.v,pset1<Packet4cf>(b).v,(1<<7)|(1<<6)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd pinsertlast(const Packet2cd& a, std::complex<double> b)
+{
+  return Packet2cd(_mm256_blend_pd(a.v,pset1<Packet2cd>(b).v,(1<<3)|(1<<2)));
+}
+
 } // end namespace internal
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/arch/AVX/PacketMath.h b/Eigen/src/Core/arch/AVX/PacketMath.h
index 05b15b852..e60ef307b 100644
--- a/Eigen/src/Core/arch/AVX/PacketMath.h
+++ b/Eigen/src/Core/arch/AVX/PacketMath.h
@@ -609,6 +609,26 @@ template<> EIGEN_STRONG_INLINE Packet4d pblend(const Selector<4>& ifPacket, cons
   return _mm256_blendv_pd(thenPacket, elsePacket, false_mask);
 }
 
+template<> EIGEN_STRONG_INLINE Packet8f pinsertfirst(const Packet8f& a, float b)
+{
+  return _mm256_blend_ps(a,pset1<Packet8f>(b),1);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4d pinsertfirst(const Packet4d& a, double b)
+{
+  return _mm256_blend_pd(a,pset1<Packet4d>(b),1);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8f pinsertlast(const Packet8f& a, float b)
+{
+  return _mm256_blend_ps(a,pset1<Packet8f>(b),(1<<7));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4d pinsertlast(const Packet4d& a, double b)
+{
+  return _mm256_blend_pd(a,pset1<Packet4d>(b),(1<<3));
+}
+
 } // end namespace internal
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/arch/CUDA/PacketMathHalf.h b/Eigen/src/Core/arch/CUDA/PacketMathHalf.h
index d670f3718..ae54225f8 100644
--- a/Eigen/src/Core/arch/CUDA/PacketMathHalf.h
+++ b/Eigen/src/Core/arch/CUDA/PacketMathHalf.h
@@ -15,7 +15,7 @@ namespace Eigen {
 namespace internal {
 
 // Most of the following operations require arch >= 3.0
-#if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDACC__) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300
 
 template<> struct is_arithmetic<half2> { enum { value = true }; };
 
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 baad692e3..6f31cf12b 100755
--- a/Eigen/src/Core/arch/SSE/PacketMath.h
+++ b/Eigen/src/Core/arch/SSE/PacketMath.h
@@ -818,6 +818,44 @@ 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
+  return _mm_blend_ps(a,pset1<Packet4f>(b),(1<<3));
+#else
+  const Packet4f mask = _mm_castsi128_ps(_mm_setr_epi32(0x0,0x0,0x0,0xFFFFFFFF));
+  return _mm_or_ps(_mm_andnot_ps(mask, a), _mm_and_ps(mask, pset1<Packet4f>(b)));
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pinsertlast(const Packet2d& a, double b)
+{
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  return _mm_blend_pd(a,pset1<Packet2d>(b),(1<<1));
+#else
+  const Packet2d mask = _mm_castsi128_pd(_mm_setr_epi32(0x0,0x0,0xFFFFFFFF,0xFFFFFFFF));
+  return _mm_or_pd(_mm_andnot_pd(mask, a), _mm_and_pd(mask, pset1<Packet2d>(b)));
+#endif
+}
+
 // Scalar path for pmadd with FMA to ensure consistency with vectorized path.
 #ifdef __FMA__
 template<> EIGEN_STRONG_INLINE float pmadd(const float& a, const float& b, const float& c) {