From 8387847c42d1e7aca22f660336c0828cae37aed4 Mon Sep 17 00:00:00 2001
From: Christophe Riccio <g.truc.creation@gmail.com>
Date: Mon, 31 Jan 2011 12:38:48 +0000
Subject: [PATCH] Improved simd cast and added duplicated values function with
 smind instructions

---
 glm/gtx/simd_mat4.hpp      |  5 +++++
 glm/gtx/simd_mat4.inl      | 13 +++++++++++++
 glm/gtx/simd_vec4.hpp      | 23 ++++++++++++++++++----
 glm/gtx/simd_vec4.inl      | 40 +++++++++++++++++++++++++-------------
 test/gtx/gtx-simd-vec4.cpp |  4 ++--
 5 files changed, 65 insertions(+), 20 deletions(-)

diff --git a/glm/gtx/simd_mat4.hpp b/glm/gtx/simd_mat4.hpp
index 254fa636..f48df975 100644
--- a/glm/gtx/simd_mat4.hpp
+++ b/glm/gtx/simd_mat4.hpp
@@ -134,6 +134,11 @@ namespace glm
 	{
 		typedef detail::fmat4x4SIMD simdMat4;
 
+		//! Convert a simdMat4 to a mat4.
+		//! (From GLM_GTX_simd_mat4 extension)
+		detail::tmat4x4<float> mat4_cast(
+			detail::fmat4x4SIMD const & x);
+
 		//! Multiply matrix x by matrix y component-wise, i.e.,
 		//! result[i][j] is the scalar product of x[i][j] and y[i][j].
 		//! (From GLM_GTX_simd_mat4 extension).
diff --git a/glm/gtx/simd_mat4.inl b/glm/gtx/simd_mat4.inl
index abcb849c..59d1065e 100644
--- a/glm/gtx/simd_mat4.inl
+++ b/glm/gtx/simd_mat4.inl
@@ -237,6 +237,19 @@ namespace detail
 namespace gtx{
 namespace simd_mat4
 {
+	inline detail::tmat4x4<float> mat4_cast
+	(
+		detail::fmat4x4SIMD const & x
+	)
+	{
+		detail::tmat4x4<float> Result;
+		_mm_store_ps(&Result[0][0], x.Data[0].Data);
+		_mm_store_ps(&Result[1][0], x.Data[1].Data);
+		_mm_store_ps(&Result[2][0], x.Data[2].Data);
+		_mm_store_ps(&Result[3][0], x.Data[3].Data);
+		return Result;
+	}
+
 	inline detail::fmat4x4SIMD simdMatrixCompMult
 	(
 		detail::fmat4x4SIMD const & x,
diff --git a/glm/gtx/simd_vec4.hpp b/glm/gtx/simd_vec4.hpp
index 4a8e5313..da2736ed 100644
--- a/glm/gtx/simd_vec4.hpp
+++ b/glm/gtx/simd_vec4.hpp
@@ -51,10 +51,6 @@ namespace glm
 			fvec4SIMD(__m128 const & Data);
 			fvec4SIMD(fvec4SIMD const & v);
 
-			fvec4SIMD(tvec4<float> const & v);
-			//operator detail::tvec4<float>();
-			//operator detail::tvec4<float> const();
-
 			//////////////////////////////////////
 			// Explicit basic constructors
 
@@ -67,6 +63,8 @@ namespace glm
 				float const & y, 
 				float const & z, 
 				float const & w);
+			explicit fvec4SIMD(
+				tvec4<float> const & v);
 
 			////////////////////////////////////////
 			//// Convertion vector constructors
@@ -129,18 +127,35 @@ namespace glm
 		float simdLength(
 			detail::fvec4SIMD const & x);
 
+		//! Returns the length of x, i.e., sqrt(x * x).
+		//! (From GLM_GTX_simd_vec4 extension, geometry functions)
+		detail::fvec4SIMD simdLength4(
+			detail::fvec4SIMD const & x);
+
 		//! Returns the distance betwwen p0 and p1, i.e., length(p0 - p1).
 		//! (From GLM_GTX_simd_vec4 extension, geometry functions)
 		float simdDistance(
 			detail::fvec4SIMD const & p0,
 			detail::fvec4SIMD const & p1);
 
+		//! Returns the distance betwwen p0 and p1, i.e., length(p0 - p1).
+		//! (From GLM_GTX_simd_vec4 extension, geometry functions)
+		detail::fvec4SIMD simdDistance4(
+			detail::fvec4SIMD const & p0,
+			detail::fvec4SIMD const & p1);
+
 		//! Returns the dot product of x and y, i.e., result = x * y.
 		//! (From GLM_GTX_simd_vec4 extension, geometry functions)
 		float simdDot(
 			detail::fvec4SIMD const & x,
 			detail::fvec4SIMD const & y);
 
+		//! Returns the dot product of x and y, i.e., result = x * y.
+		//! (From GLM_GTX_simd_vec4 extension, geometry functions)
+		detail::fvec4SIMD simdDot4(
+			detail::fvec4SIMD const & x,
+			detail::fvec4SIMD const & y);
+
 		//! Returns the cross product of x and y.
 		//! (From GLM_GTX_simd_vec4 extension, geometry functions)
 		detail::fvec4SIMD simdCross(
diff --git a/glm/gtx/simd_vec4.inl b/glm/gtx/simd_vec4.inl
index de4f40a4..e38d3199 100644
--- a/glm/gtx/simd_vec4.inl
+++ b/glm/gtx/simd_vec4.inl
@@ -35,20 +35,6 @@ namespace glm
 			Data(_mm_set_ps(v.w, v.z, v.y, v.x))
 		{}
 
-		inline fvec4SIMD::operator detail::tvec4<float>()
-		{
-			detail::tvec4<float> Result;
-			_mm_store_ps(&Result[0], this->Data);
-			return Result;
-		}
-
-		//inline fvec4SIMD::operator detail::tvec4<float> const()
-		//{
-		//	detail::tvec4<float> Result;
-		//	_mm_store_ps(&Result[0], this->Data);
-		//	return Result;
-		//}
-
 		//////////////////////////////////////
 		// Explicit basic constructors
 
@@ -304,6 +290,14 @@ namespace glm
 			return Result;
 		}
 
+		inline detail::fvec4SIMD simdLength4
+		(
+			detail::fvec4SIMD const & x
+		)
+		{
+			return detail::sse_len_ps(x.Data);
+		}
+
 		inline float simdDistance
 		(
 			detail::fvec4SIMD const & p0,
@@ -315,6 +309,15 @@ namespace glm
 			return Result;
 		}
 
+		inline detail::fvec4SIMD simdDistance4
+		(
+			detail::fvec4SIMD const & p0,
+			detail::fvec4SIMD const & p1
+		)
+		{
+			return detail::sse_dst_ps(p0.Data, p1.Data);
+		}
+
 		inline float simdDot
 		(
 			detail::fvec4SIMD const & x,
@@ -326,6 +329,15 @@ namespace glm
 			return Result;
 		}
 
+		inline detail::fvec4SIMD simdDot4
+		(
+			detail::fvec4SIMD const & x,
+			detail::fvec4SIMD const & y
+		)
+		{
+			return detail::sse_dot_ss(x.Data, y.Data);
+		}
+
 		inline detail::fvec4SIMD simdCross
 		(
 			detail::fvec4SIMD const & x,
diff --git a/test/gtx/gtx-simd-vec4.cpp b/test/gtx/gtx-simd-vec4.cpp
index 56e2bd14..d5df3ada 100644
--- a/test/gtx/gtx-simd-vec4.cpp
+++ b/test/gtx/gtx-simd-vec4.cpp
@@ -17,8 +17,8 @@ int main()
 	glm::simdVec4 B1(0.4f, 0.5f, 0.6f, 0.7f);
 	glm::simdVec4 C1 = A1 + B1;
 	glm::simdVec4 D1 = A1.swizzle<glm::X, glm::Z, glm::Y, glm::W>();
-	glm::simdVec4 E1 = glm::vec4(1.0f);
-	glm::vec4 F1 = E1;
+	glm::simdVec4 E1(glm::vec4(1.0f));
+	glm::vec4 F1 = glm::vec4_cast(E1);
 	//glm::vec4 G1(E1);
 
 	//printf("A1(%2.3f, %2.3f, %2.3f, %2.3f)\n", A1.x, A1.y, A1.z, A1.w);