Added support for fast reciprocal square root computation.

2 files changed, 19 insertions, 4 deletions

diff --git a/Eigen/src/Core/arch/AVX/MathFunctions.h b/Eigen/src/Core/arch/AVX/MathFunctions.h
index 2810a7a0b..aecfdd6ad 100644
--- a/Eigen/src/Core/arch/AVX/MathFunctions.h
+++ b/Eigen/src/Core/arch/AVX/MathFunctions.h
@@ -300,16 +300,29 @@ psqrt<Packet8f>(const Packet8f& _x) {
   return pmul(_x, x);
 }
 #else
-template <>
-EIGEN_STRONG_INLINE Packet8f psqrt<Packet8f>(const Packet8f& x) {
+template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet8f psqrt<Packet8f>(const Packet8f& x) {
   return _mm256_sqrt_ps(x);
 }
 #endif
-template <>
-EIGEN_STRONG_INLINE Packet4d psqrt<Packet4d>(const Packet4d& x) {
+template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4d psqrt<Packet4d>(const Packet4d& x) {
   return _mm256_sqrt_pd(x);
 }
 
+template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet8f prsqrt<Packet8f>(const Packet8f& x) {
+  _EIGEN_DECLARE_CONST_Packet8f(one, 1.0f);
+  return _mm256_div_ps(p8f_one, _mm256_sqrt_ps(x));
+}
+#endif
+template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4d prsqrt<Packet4d>(const Packet4d& x) {
+  _EIGEN_DECLARE_CONST_Packet4d(one, 1.0);
+  return _mm256_div_pd(p4d_one, _mm256_sqrt_pd(x));
+}
+
+
 }  // 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 695185a49..fb20a45cc 100644
--- a/Eigen/src/Core/arch/AVX/PacketMath.h
+++ b/Eigen/src/Core/arch/AVX/PacketMath.h
@@ -65,6 +65,7 @@ template<> struct packet_traits<float>  : default_packet_traits
     HasLog  = 1,
     HasExp  = 1,
     HasSqrt = 1,
+    HasRsqrt = 1,
     HasBlend = 1
   };
 };
@@ -81,6 +82,7 @@ template<> struct packet_traits<double> : default_packet_traits
     HasDiv  = 1,
     HasExp  = 0,
     HasSqrt = 1,
+    HasRsqrt = 1,
     HasBlend = 1
   };
 };