From a5d2a63ef340597dedd34c1e7095c7e090d80706 Mon Sep 17 00:00:00 2001
From: Christophe Riccio <mail@g-truc.net>
Date: Tue, 1 Apr 2014 01:20:03 +0200
Subject: [PATCH] - Added workaround for a CUDA compiler bug (#186, #185)

---
 glm/detail/func_exponential.inl | 14 ++++++++++++--
 glm/detail/func_geometric.inl   |  7 ++++++-
 glm/gtx/fast_square_root.inl    | 14 ++++++++++++--
 readme.txt                      |  1 +
 4 files changed, 31 insertions(+), 5 deletions(-)

diff --git a/glm/detail/func_exponential.inl b/glm/detail/func_exponential.inl
index 3a2f2379..767323f7 100644
--- a/glm/detail/func_exponential.inl
+++ b/glm/detail/func_exponential.inl
@@ -197,12 +197,22 @@ namespace detail
 	// sqrt
 	GLM_FUNC_QUALIFIER float sqrt(float x)
 	{
-		return detail::compute_sqrt<detail::tvec1, float, highp>::call(x).x;
+#		ifdef __CUDACC__ // Wordaround for a CUDA compiler bug up to CUDA6
+			detail::tvec1<float, highp> tmp(detail::compute_sqrt<detail::tvec1, float, highp>::call(x));
+			return tmp.x;
+#		else
+			return detail::compute_sqrt<detail::tvec1, float, highp>::call(x).x;
+#		endif
 	}
 
 	GLM_FUNC_QUALIFIER double sqrt(double x)
 	{
-		return detail::compute_sqrt<detail::tvec1, double, highp>::call(x).x;
+#		ifdef __CUDACC__ // Wordaround for a CUDA compiler bug up to CUDA6
+			detail::tvec1<double, highp> tmp(detail::compute_sqrt<detail::tvec1, double, highp>::call(x));
+			return tmp.x;
+#		else
+			return detail::compute_sqrt<detail::tvec1, double, highp>::call(x).x;
+#		endif
 	}
 		
 	template <typename T, precision P, template <typename, precision> class vecType>
diff --git a/glm/detail/func_geometric.inl b/glm/detail/func_geometric.inl
index 58aca38b..2f8691dd 100644
--- a/glm/detail/func_geometric.inl
+++ b/glm/detail/func_geometric.inl
@@ -43,7 +43,12 @@ namespace detail
 	{
 		GLM_FUNC_QUALIFIER static T call(detail::tvec1<T, P> const & x, detail::tvec1<T, P> const & y)
 		{
-			return detail::tvec1<T, P>(x * y).x;
+#			ifdef __CUDACC__ // Wordaround for a CUDA compiler bug up to CUDA6
+				detail::tvec1<T, P> tmp(x * y);
+				return tmp.x;
+#			else
+				return detail::tvec1<T, P>(x * y).x;
+#			endif
 		}
 	};
 
diff --git a/glm/gtx/fast_square_root.inl b/glm/gtx/fast_square_root.inl
index b3f97957..a08bdbf2 100644
--- a/glm/gtx/fast_square_root.inl
+++ b/glm/gtx/fast_square_root.inl
@@ -27,13 +27,23 @@ namespace glm
 	template <>
 	GLM_FUNC_QUALIFIER float fastInverseSqrt<float>(float const & x)
 	{
-		return detail::compute_inversesqrt<detail::tvec1, float, lowp>::call(detail::tvec1<float, lowp>(x)).x;
+#		ifdef __CUDACC__ // Wordaround for a CUDA compiler bug up to CUDA6
+			detail::tvec1<T, P> tmp(detail::compute_inversesqrt<detail::tvec1, float, lowp>::call(detail::tvec1<float, lowp>(x)));
+			return tmp.x;
+#		else
+			return detail::compute_inversesqrt<detail::tvec1, float, lowp>::call(detail::tvec1<float, lowp>(x)).x;
+#		endif
 	}
 
 	template <>
 	GLM_FUNC_QUALIFIER double fastInverseSqrt<double>(double const & x)
 	{
-		return detail::compute_inversesqrt<detail::tvec1, double, lowp>::call(detail::tvec1<double, lowp>(x)).x;
+#		ifdef __CUDACC__ // Wordaround for a CUDA compiler bug up to CUDA6
+			detail::tvec1<T, P> tmp(detail::compute_inversesqrt<detail::tvec1, double, lowp>::call(detail::tvec1<double, lowp>(x)));
+			return tmp.x;
+#		else
+			return detail::compute_inversesqrt<detail::tvec1, double, lowp>::call(detail::tvec1<double, lowp>(x)).x;
+#		endif
 	}
 
 	template <template <class, precision> class vecType, typename T, precision P>
diff --git a/readme.txt b/readme.txt
index 300c82ff..df77a668 100644
--- a/readme.txt
+++ b/readme.txt
@@ -51,6 +51,7 @@ GLM 0.9.5.3: 2014-0X-XX
 - Fixed usubBorrow (#171)
 - Fixed eulerAngle*** not consistent for right-handed coordinate system (#173)
 - Added full tests for eulerAngle*** functions (#173)
+- Added workaround for a CUDA compiler bug (#186, #185)
 
 ================================================================================
 GLM 0.9.5.2: 2014-02-08