CUDA bugfix does not compile in CUDA9

[hschwane]

Compiling an Application using glm with CUDA9 gives me the following error:

/usr/include/glm/gtx/fast_square_root.inl(26): error: identifier "T" is undefined

/usr/include/glm/gtx/fast_square_root.inl(26): error: identifier "Q" is undefined

/usr/include/glm/gtx/fast_square_root.inl(26): error: argument list for alias template "glm::tvec1" is missing

Changeing /usr/include/glm/gtx/fast_square_root.inl
from

	// fastInversesqrt
	template<typename genType>
	GLM_FUNC_QUALIFIER genType fastInverseSqrt(genType x)
	{
#		ifdef __CUDACC__ // Wordaround for a CUDA compiler bug up to CUDA6
			vec<1, T, Q> tmp(detail::compute_inversesqrt<tvec1, genType, lowp, detail::is_aligned<lowp>::value>::call(vec<1, genType, lowp>(x)));
			return tmp.x;
#		else
			return detail::compute_inversesqrt<1, genType, lowp, detail::is_aligned<lowp>::value>::call(vec<1, genType, lowp>(x)).x;
#		endif
	}

to

	// fastInversesqrt
	template<typename genType>
	GLM_FUNC_QUALIFIER genType fastInverseSqrt(genType x)
	{
//#		ifdef __CUDACC__ // Wordaround for a CUDA compiler bug up to CUDA6
//			vec<1, T, Q> tmp(detail::compute_inversesqrt<tvec1, genType, lowp, detail::is_aligned<lowp>::value>::call(vec<1, genType, lowp>(x)));
//			return tmp.x;
//#		else
			return detail::compute_inversesqrt<1, genType, lowp, detail::is_aligned<lowp>::value>::call(vec<1, genType, lowp>(x)).x;
//#		endif
	}

resolves the issue, however i suspect this will break CUDA6 support...

[Groovounet]

This issue should be fixed in master branch for GLM 0.9.9.6 release

Thanks for reporting,
Christophe