From 6ed8c3dbb3d64b0bc3af44cea9d1bfe7630d5f06 Mon Sep 17 00:00:00 2001
From: Christophe Riccio <christophe.riccio@unity3d.com>
Date: Sat, 4 Jan 2014 13:27:06 +0100
Subject: [PATCH] Added vectorized implementation of fast inversesqrt

---
 glm/detail/func_exponential.inl | 91 ++++++++++++++-------------------
 glm/detail/func_geometric.hpp   |  4 +-
 glm/gtx/fast_square_root.inl    | 13 +++--
 3 files changed, 49 insertions(+), 59 deletions(-)

diff --git a/glm/detail/func_exponential.inl b/glm/detail/func_exponential.inl
index 9c03129e..c6d59e62 100644
--- a/glm/detail/func_exponential.inl
+++ b/glm/detail/func_exponential.inl
@@ -153,68 +153,53 @@ namespace detail
 
 	VECTORIZE_VEC(sqrt)
 
-	template <typename genType>
-	GLM_FUNC_QUALIFIER genType inversesqrt
-	(
-		genType const & x
-	)
-	{
-		GLM_STATIC_ASSERT(
-			std::numeric_limits<genType>::is_iec559,
-			"'inversesqrt' only accept floating-point inputs");
-
-		assert(x > genType(0));
-
-		return genType(1) / std::sqrt(x);
-	}
-
-	VECTORIZE_VEC(inversesqrt)
-	
 	namespace detail
 	{
-		template <typename genType, typename genUType>
-		genType fastInversesqrt(genType const & v)
+		template <template <class, precision> class vecType, typename T, precision P>
+		struct compute_inversesqrt
 		{
-			genType tmp(v);
-			genType xhalf(tmp * genType(0.5f));
-			genUType i = *reinterpret_cast<genUType*>(const_cast<genType*>(&v));
-			i = genUType(0x5f375a86) - (i >> genUType(1));
-			tmp = *reinterpret_cast<genType*>(&i);
-			tmp = tmp * (genType(1.5f) - xhalf * tmp * tmp);
-			return tmp;
-		}
-	}
+			static vecType<T, P> call(vecType<T, P> const & x)
+			{
+				return static_cast<T>(1) / sqrt(x);
+			}
+		};
+		
+		template <template <class, precision> class vecType>
+		struct compute_inversesqrt<vecType, float, lowp>
+		{
+			static vecType<float, lowp> call(vecType<float, lowp> const & x)
+			{
+				vecType<float, lowp> tmp(x);
+				vecType<float, lowp> xhalf(tmp * 0.5f);
+				vecType<uint, lowp> i = *reinterpret_cast<vecType<uint, lowp>*>(const_cast<vecType<float, lowp>*>(&x));
+				i = vecType<uint, lowp>(0x5f375a86) - (i >> vecType<uint, lowp>(1));
+				tmp = *reinterpret_cast<vecType<float, lowp>*>(&i);
+				tmp = tmp * (1.5f - xhalf * tmp * tmp);
+				return tmp;
+			}
+		};
+	}//namespace detail
 	
-	template <>
-	GLM_FUNC_QUALIFIER lowp_vec1 inversesqrt(lowp_vec1 const & v)
+	// inversesqrt
+	GLM_FUNC_QUALIFIER float inversesqrt(float const & x)
 	{
-		assert(glm::all(glm::greaterThan(v, lowp_vec1(0))));
-
-		return detail::fastInversesqrt<lowp_vec1, uint>(v);
+		return 1.0f / sqrt(x);
 	}
-
-	template <>
-	GLM_FUNC_QUALIFIER lowp_vec2 inversesqrt(lowp_vec2 const & v)
+	
+	GLM_FUNC_QUALIFIER double inversesqrt(double const & x)
 	{
-		assert(glm::all(glm::greaterThan(v, lowp_vec2(0))));
-
-		return detail::fastInversesqrt<lowp_vec2, lowp_uvec2>(v);
+		return 1.0 / sqrt(x);
 	}
-
-	template <>
-	GLM_FUNC_QUALIFIER lowp_vec3 inversesqrt(lowp_vec3 const & v)
+	
+	template <template <class, precision> class vecType, typename T, precision P>
+	GLM_FUNC_QUALIFIER vecType<T, P> inversesqrt
+	(
+		vecType<T, P> const & x
+	)
 	{
-		assert(glm::all(glm::greaterThan(v, lowp_vec3(0))));
-
-		return detail::fastInversesqrt<lowp_vec3, lowp_uvec3>(v);
+		GLM_STATIC_ASSERT(std::numeric_limits<T>::is_iec559, "'inversesqrt' only accept floating-point inputs");
+		return detail::compute_inversesqrt<vecType, T, P>::call(x);
 	}
 
-	template <>
-	GLM_FUNC_QUALIFIER lowp_vec4 inversesqrt(lowp_vec4 const & v)
-	{
-		assert(glm::all(glm::greaterThan(v, lowp_vec4(0))));
-
-		return detail::fastInversesqrt<lowp_vec4, lowp_uvec4>(v);
-	}
-	
+	VECTORIZE_VEC(inversesqrt)
 }//namespace glm
diff --git a/glm/detail/func_geometric.hpp b/glm/detail/func_geometric.hpp
index 285735dd..d39a01a2 100644
--- a/glm/detail/func_geometric.hpp
+++ b/glm/detail/func_geometric.hpp
@@ -74,7 +74,7 @@ namespace glm
 	GLM_FUNC_DECL T dot(
 		vecType<T, P> const & x,
 		vecType<T, P> const & y);
-/*
+
 	/// Returns the dot product of x and y, i.e., result = x * y.
 	///
 	/// @tparam genType Floating-point vector types.
@@ -85,7 +85,7 @@ namespace glm
 	GLM_FUNC_DECL genType dot(
 		genType const & x,
 		genType const & y);
-*/
+
 	/// Returns the cross product of x and y.
 	///
 	/// @tparam valType Floating-point scalar types.
diff --git a/glm/gtx/fast_square_root.inl b/glm/gtx/fast_square_root.inl
index 496abe21..83c80b39 100644
--- a/glm/gtx/fast_square_root.inl
+++ b/glm/gtx/fast_square_root.inl
@@ -24,13 +24,18 @@ namespace glm
 	VECTORIZE_VEC(fastSqrt)
 
 	// fastInversesqrt
-	template <typename genType>
-	GLM_FUNC_QUALIFIER genType fastInverseSqrt
+	GLM_FUNC_QUALIFIER float fastInverseSqrt(float x)
+	{
+		return detail::compute_inversesqrt<detail::tvec1, float, lowp>::call(x).x;
+	}
+	
+	template <template <class, precision> class vecType, typename T, precision P>
+	GLM_FUNC_QUALIFIER vecType<T, P> fastInverseSqrt
 	(
-		genType const & x
+		vecType<T, P> const & x
 	)
 	{
-		return detail::fastInversesqrt<genType, uint>(x);
+		return detail::compute_inversesqrt<vecType, T, P>::call(x);
 	}
 
 	VECTORIZE_VEC(fastInverseSqrt)