/// @ref gtc_noise
/// @file glm/gtc/noise.inl
///
// Based on the work of Stefan Gustavson and Ashima Arts on "webgl-noise": 
// https://github.com/ashima/webgl-noise 
// Following Stefan Gustavson's paper "Simplex noise demystified": 
// http://www.itn.liu.se/~stegu/simplexnoise/simplexnoise.pdf

namespace glm{
namespace gtc
{
	template <typename T, precision P>
	GLM_FUNC_QUALIFIER tvec4<T, P> grad4(T const & j, tvec4<T, P> const & ip)
	{
		tvec3<T, P> pXYZ = floor(fract(tvec3<T, P>(j) * tvec3<T, P>(ip)) * T(7)) * ip[2] - T(1);
		T pW = static_cast<T>(1.5) - dot(abs(pXYZ), tvec3<T, P>(1));
		tvec4<T, P> s = tvec4<T, P>(lessThan(tvec4<T, P>(pXYZ, pW), tvec4<T, P>(0.0)));
		pXYZ = pXYZ + (tvec3<T, P>(s) * T(2) - T(1)) * s.w; 
		return tvec4<T, P>(pXYZ, pW);
	}
}//namespace gtc

	// Classic Perlin noise
	template <typename T, precision P>
	GLM_FUNC_QUALIFIER T perlin(tvec2<T, P> const & Position)
	{
		tvec4<T, P> Pi = glm::floor(tvec4<T, P>(Position.x, Position.y, Position.x, Position.y)) + tvec4<T, P>(0.0, 0.0, 1.0, 1.0);
		tvec4<T, P> Pf = glm::fract(tvec4<T, P>(Position.x, Position.y, Position.x, Position.y)) - tvec4<T, P>(0.0, 0.0, 1.0, 1.0);
		Pi = mod(Pi, tvec4<T, P>(289)); // To avoid truncation effects in permutation
		tvec4<T, P> ix(Pi.x, Pi.z, Pi.x, Pi.z);
		tvec4<T, P> iy(Pi.y, Pi.y, Pi.w, Pi.w);
		tvec4<T, P> fx(Pf.x, Pf.z, Pf.x, Pf.z);
		tvec4<T, P> fy(Pf.y, Pf.y, Pf.w, Pf.w);

		tvec4<T, P> i = detail::permute(detail::permute(ix) + iy);

		tvec4<T, P> gx = static_cast<T>(2) * glm::fract(i / T(41)) - T(1);
		tvec4<T, P> gy = glm::abs(gx) - T(0.5);
		tvec4<T, P> tx = glm::floor(gx + T(0.5));
		gx = gx - tx;

		tvec2<T, P> g00(gx.x, gy.x);
		tvec2<T, P> g10(gx.y, gy.y);
		tvec2<T, P> g01(gx.z, gy.z);
		tvec2<T, P> g11(gx.w, gy.w);

		tvec4<T, P> norm = detail::taylorInvSqrt(tvec4<T, P>(dot(g00, g00), dot(g01, g01), dot(g10, g10), dot(g11, g11)));
		g00 *= norm.x;
		g01 *= norm.y;
		g10 *= norm.z;
		g11 *= norm.w;

		T n00 = dot(g00, tvec2<T, P>(fx.x, fy.x));
		T n10 = dot(g10, tvec2<T, P>(fx.y, fy.y));
		T n01 = dot(g01, tvec2<T, P>(fx.z, fy.z));
		T n11 = dot(g11, tvec2<T, P>(fx.w, fy.w));

		tvec2<T, P> fade_xy = detail::fade(tvec2<T, P>(Pf.x, Pf.y));
		tvec2<T, P> n_x = mix(tvec2<T, P>(n00, n01), tvec2<T, P>(n10, n11), fade_xy.x);
		T n_xy = mix(n_x.x, n_x.y, fade_xy.y);
		return T(2.3) * n_xy;
	}

	// Classic Perlin noise
	template <typename T, precision P>
	GLM_FUNC_QUALIFIER T perlin(tvec3<T, P> const & Position)
	{
		tvec3<T, P> Pi0 = floor(Position); // Integer part for indexing
		tvec3<T, P> Pi1 = Pi0 + T(1); // Integer part + 1
		Pi0 = detail::mod289(Pi0);
		Pi1 = detail::mod289(Pi1);
		tvec3<T, P> Pf0 = fract(Position); // Fractional part for interpolation
		tvec3<T, P> Pf1 = Pf0 - T(1); // Fractional part - 1.0
		tvec4<T, P> ix(Pi0.x, Pi1.x, Pi0.x, Pi1.x);
		tvec4<T, P> iy = tvec4<T, P>(tvec2<T, P>(Pi0.y), tvec2<T, P>(Pi1.y));
		tvec4<T, P> iz0(Pi0.z);
		tvec4<T, P> iz1(Pi1.z);

		tvec4<T, P> ixy = detail::permute(detail::permute(ix) + iy);
		tvec4<T, P> ixy0 = detail::permute(ixy + iz0);
		tvec4<T, P> ixy1 = detail::permute(ixy + iz1);

		tvec4<T, P> gx0 = ixy0 * T(1.0 / 7.0);
		tvec4<T, P> gy0 = fract(floor(gx0) * T(1.0 / 7.0)) - T(0.5);
		gx0 = fract(gx0);
		tvec4<T, P> gz0 = tvec4<T, P>(0.5) - abs(gx0) - abs(gy0);
		tvec4<T, P> sz0 = step(gz0, tvec4<T, P>(0.0));
		gx0 -= sz0 * (step(T(0), gx0) - T(0.5));
		gy0 -= sz0 * (step(T(0), gy0) - T(0.5));

		tvec4<T, P> gx1 = ixy1 * T(1.0 / 7.0);
		tvec4<T, P> gy1 = fract(floor(gx1) * T(1.0 / 7.0)) - T(0.5);
		gx1 = fract(gx1);
		tvec4<T, P> gz1 = tvec4<T, P>(0.5) - abs(gx1) - abs(gy1);
		tvec4<T, P> sz1 = step(gz1, tvec4<T, P>(0.0));
		gx1 -= sz1 * (step(T(0), gx1) - T(0.5));
		gy1 -= sz1 * (step(T(0), gy1) - T(0.5));

		tvec3<T, P> g000(gx0.x, gy0.x, gz0.x);
		tvec3<T, P> g100(gx0.y, gy0.y, gz0.y);
		tvec3<T, P> g010(gx0.z, gy0.z, gz0.z);
		tvec3<T, P> g110(gx0.w, gy0.w, gz0.w);
		tvec3<T, P> g001(gx1.x, gy1.x, gz1.x);
		tvec3<T, P> g101(gx1.y, gy1.y, gz1.y);
		tvec3<T, P> g011(gx1.z, gy1.z, gz1.z);
		tvec3<T, P> g111(gx1.w, gy1.w, gz1.w);

		tvec4<T, P> norm0 = detail::taylorInvSqrt(tvec4<T, P>(dot(g000, g000), dot(g010, g010), dot(g100, g100), dot(g110, g110)));
		g000 *= norm0.x;
		g010 *= norm0.y;
		g100 *= norm0.z;
		g110 *= norm0.w;
		tvec4<T, P> norm1 = detail::taylorInvSqrt(tvec4<T, P>(dot(g001, g001), dot(g011, g011), dot(g101, g101), dot(g111, g111)));
		g001 *= norm1.x;
		g011 *= norm1.y;
		g101 *= norm1.z;
		g111 *= norm1.w;

		T n000 = dot(g000, Pf0);
		T n100 = dot(g100, tvec3<T, P>(Pf1.x, Pf0.y, Pf0.z));
		T n010 = dot(g010, tvec3<T, P>(Pf0.x, Pf1.y, Pf0.z));
		T n110 = dot(g110, tvec3<T, P>(Pf1.x, Pf1.y, Pf0.z));
		T n001 = dot(g001, tvec3<T, P>(Pf0.x, Pf0.y, Pf1.z));
		T n101 = dot(g101, tvec3<T, P>(Pf1.x, Pf0.y, Pf1.z));
		T n011 = dot(g011, tvec3<T, P>(Pf0.x, Pf1.y, Pf1.z));
		T n111 = dot(g111, Pf1);

		tvec3<T, P> fade_xyz = detail::fade(Pf0);
		tvec4<T, P> n_z = mix(tvec4<T, P>(n000, n100, n010, n110), tvec4<T, P>(n001, n101, n011, n111), fade_xyz.z);
		tvec2<T, P> n_yz = mix(tvec2<T, P>(n_z.x, n_z.y), tvec2<T, P>(n_z.z, n_z.w), fade_xyz.y);
		T n_xyz = mix(n_yz.x, n_yz.y, fade_xyz.x); 
		return T(2.2) * n_xyz;
	}
	/*
	// Classic Perlin noise
	template <typename T, precision P>
	GLM_FUNC_QUALIFIER T perlin(tvec3<T, P> const & P)
	{
		tvec3<T, P> Pi0 = floor(P); // Integer part for indexing
		tvec3<T, P> Pi1 = Pi0 + T(1); // Integer part + 1
		Pi0 = mod(Pi0, T(289));
		Pi1 = mod(Pi1, T(289));
		tvec3<T, P> Pf0 = fract(P); // Fractional part for interpolation
		tvec3<T, P> Pf1 = Pf0 - T(1); // Fractional part - 1.0
		tvec4<T, P> ix(Pi0.x, Pi1.x, Pi0.x, Pi1.x);
		tvec4<T, P> iy(Pi0.y, Pi0.y, Pi1.y, Pi1.y);
		tvec4<T, P> iz0(Pi0.z);
		tvec4<T, P> iz1(Pi1.z);

		tvec4<T, P> ixy = permute(permute(ix) + iy);
		tvec4<T, P> ixy0 = permute(ixy + iz0);
		tvec4<T, P> ixy1 = permute(ixy + iz1);

		tvec4<T, P> gx0 = ixy0 / T(7);
		tvec4<T, P> gy0 = fract(floor(gx0) / T(7)) - T(0.5);
		gx0 = fract(gx0);
		tvec4<T, P> gz0 = tvec4<T, P>(0.5) - abs(gx0) - abs(gy0);
		tvec4<T, P> sz0 = step(gz0, tvec4<T, P>(0.0));
		gx0 -= sz0 * (step(0.0, gx0) - T(0.5));
		gy0 -= sz0 * (step(0.0, gy0) - T(0.5));

		tvec4<T, P> gx1 = ixy1 / T(7);
		tvec4<T, P> gy1 = fract(floor(gx1) / T(7)) - T(0.5);
		gx1 = fract(gx1);
		tvec4<T, P> gz1 = tvec4<T, P>(0.5) - abs(gx1) - abs(gy1);
		tvec4<T, P> sz1 = step(gz1, tvec4<T, P>(0.0));
		gx1 -= sz1 * (step(T(0), gx1) - T(0.5));
		gy1 -= sz1 * (step(T(0), gy1) - T(0.5));

		tvec3<T, P> g000(gx0.x, gy0.x, gz0.x);
		tvec3<T, P> g100(gx0.y, gy0.y, gz0.y);
		tvec3<T, P> g010(gx0.z, gy0.z, gz0.z);
		tvec3<T, P> g110(gx0.w, gy0.w, gz0.w);
		tvec3<T, P> g001(gx1.x, gy1.x, gz1.x);
		tvec3<T, P> g101(gx1.y, gy1.y, gz1.y);
		tvec3<T, P> g011(gx1.z, gy1.z, gz1.z);
		tvec3<T, P> g111(gx1.w, gy1.w, gz1.w);

		tvec4<T, P> norm0 = taylorInvSqrt(tvec4<T, P>(dot(g000, g000), dot(g010, g010), dot(g100, g100), dot(g110, g110)));
		g000 *= norm0.x;
		g010 *= norm0.y;
		g100 *= norm0.z;
		g110 *= norm0.w;
		tvec4<T, P> norm1 = taylorInvSqrt(tvec4<T, P>(dot(g001, g001), dot(g011, g011), dot(g101, g101), dot(g111, g111)));
		g001 *= norm1.x;
		g011 *= norm1.y;
		g101 *= norm1.z;
		g111 *= norm1.w;

		T n000 = dot(g000, Pf0);
		T n100 = dot(g100, tvec3<T, P>(Pf1.x, Pf0.y, Pf0.z));
		T n010 = dot(g010, tvec3<T, P>(Pf0.x, Pf1.y, Pf0.z));
		T n110 = dot(g110, tvec3<T, P>(Pf1.x, Pf1.y, Pf0.z));
		T n001 = dot(g001, tvec3<T, P>(Pf0.x, Pf0.y, Pf1.z));
		T n101 = dot(g101, tvec3<T, P>(Pf1.x, Pf0.y, Pf1.z));
		T n011 = dot(g011, tvec3<T, P>(Pf0.x, Pf1.y, Pf1.z));
		T n111 = dot(g111, Pf1);

		tvec3<T, P> fade_xyz = fade(Pf0);
		tvec4<T, P> n_z = mix(tvec4<T, P>(n000, n100, n010, n110), tvec4<T, P>(n001, n101, n011, n111), fade_xyz.z);
		tvec2<T, P> n_yz = mix(
			tvec2<T, P>(n_z.x, n_z.y), 
			tvec2<T, P>(n_z.z, n_z.w), fade_xyz.y);
		T n_xyz = mix(n_yz.x, n_yz.y, fade_xyz.x); 
		return T(2.2) * n_xyz;
	}
	*/
	// Classic Perlin noise
	template <typename T, precision P>
	GLM_FUNC_QUALIFIER T perlin(tvec4<T, P> const & Position)
	{
		tvec4<T, P> Pi0 = floor(Position);	// Integer part for indexing
		tvec4<T, P> Pi1 = Pi0 + T(1);		// Integer part + 1
		Pi0 = mod(Pi0, tvec4<T, P>(289));
		Pi1 = mod(Pi1, tvec4<T, P>(289));
		tvec4<T, P> Pf0 = fract(Position);	// Fractional part for interpolation
		tvec4<T, P> Pf1 = Pf0 - T(1);		// Fractional part - 1.0
		tvec4<T, P> ix(Pi0.x, Pi1.x, Pi0.x, Pi1.x);
		tvec4<T, P> iy(Pi0.y, Pi0.y, Pi1.y, Pi1.y);
		tvec4<T, P> iz0(Pi0.z);
		tvec4<T, P> iz1(Pi1.z);
		tvec4<T, P> iw0(Pi0.w);
		tvec4<T, P> iw1(Pi1.w);

		tvec4<T, P> ixy = detail::permute(detail::permute(ix) + iy);
		tvec4<T, P> ixy0 = detail::permute(ixy + iz0);
		tvec4<T, P> ixy1 = detail::permute(ixy + iz1);
		tvec4<T, P> ixy00 = detail::permute(ixy0 + iw0);
		tvec4<T, P> ixy01 = detail::permute(ixy0 + iw1);
		tvec4<T, P> ixy10 = detail::permute(ixy1 + iw0);
		tvec4<T, P> ixy11 = detail::permute(ixy1 + iw1);

		tvec4<T, P> gx00 = ixy00 / T(7);
		tvec4<T, P> gy00 = floor(gx00) / T(7);
		tvec4<T, P> gz00 = floor(gy00) / T(6);
		gx00 = fract(gx00) - T(0.5);
		gy00 = fract(gy00) - T(0.5);
		gz00 = fract(gz00) - T(0.5);
		tvec4<T, P> gw00 = tvec4<T, P>(0.75) - abs(gx00) - abs(gy00) - abs(gz00);
		tvec4<T, P> sw00 = step(gw00, tvec4<T, P>(0.0));
		gx00 -= sw00 * (step(T(0), gx00) - T(0.5));
		gy00 -= sw00 * (step(T(0), gy00) - T(0.5));

		tvec4<T, P> gx01 = ixy01 / T(7);
		tvec4<T, P> gy01 = floor(gx01) / T(7);
		tvec4<T, P> gz01 = floor(gy01) / T(6);
		gx01 = fract(gx01) - T(0.5);
		gy01 = fract(gy01) - T(0.5);
		gz01 = fract(gz01) - T(0.5);
		tvec4<T, P> gw01 = tvec4<T, P>(0.75) - abs(gx01) - abs(gy01) - abs(gz01);
		tvec4<T, P> sw01 = step(gw01, tvec4<T, P>(0.0));
		gx01 -= sw01 * (step(T(0), gx01) - T(0.5));
		gy01 -= sw01 * (step(T(0), gy01) - T(0.5));

		tvec4<T, P> gx10 = ixy10 / T(7);
		tvec4<T, P> gy10 = floor(gx10) / T(7);
		tvec4<T, P> gz10 = floor(gy10) / T(6);
		gx10 = fract(gx10) - T(0.5);
		gy10 = fract(gy10) - T(0.5);
		gz10 = fract(gz10) - T(0.5);
		tvec4<T, P> gw10 = tvec4<T, P>(0.75) - abs(gx10) - abs(gy10) - abs(gz10);
		tvec4<T, P> sw10 = step(gw10, tvec4<T, P>(0));
		gx10 -= sw10 * (step(T(0), gx10) - T(0.5));
		gy10 -= sw10 * (step(T(0), gy10) - T(0.5));

		tvec4<T, P> gx11 = ixy11 / T(7);
		tvec4<T, P> gy11 = floor(gx11) / T(7);
		tvec4<T, P> gz11 = floor(gy11) / T(6);
		gx11 = fract(gx11) - T(0.5);
		gy11 = fract(gy11) - T(0.5);
		gz11 = fract(gz11) - T(0.5);
		tvec4<T, P> gw11 = tvec4<T, P>(0.75) - abs(gx11) - abs(gy11) - abs(gz11);
		tvec4<T, P> sw11 = step(gw11, tvec4<T, P>(0.0));
		gx11 -= sw11 * (step(T(0), gx11) - T(0.5));
		gy11 -= sw11 * (step(T(0), gy11) - T(0.5));

		tvec4<T, P> g0000(gx00.x, gy00.x, gz00.x, gw00.x);
		tvec4<T, P> g1000(gx00.y, gy00.y, gz00.y, gw00.y);
		tvec4<T, P> g0100(gx00.z, gy00.z, gz00.z, gw00.z);
		tvec4<T, P> g1100(gx00.w, gy00.w, gz00.w, gw00.w);
		tvec4<T, P> g0010(gx10.x, gy10.x, gz10.x, gw10.x);
		tvec4<T, P> g1010(gx10.y, gy10.y, gz10.y, gw10.y);
		tvec4<T, P> g0110(gx10.z, gy10.z, gz10.z, gw10.z);
		tvec4<T, P> g1110(gx10.w, gy10.w, gz10.w, gw10.w);
		tvec4<T, P> g0001(gx01.x, gy01.x, gz01.x, gw01.x);
		tvec4<T, P> g1001(gx01.y, gy01.y, gz01.y, gw01.y);
		tvec4<T, P> g0101(gx01.z, gy01.z, gz01.z, gw01.z);
		tvec4<T, P> g1101(gx01.w, gy01.w, gz01.w, gw01.w);
		tvec4<T, P> g0011(gx11.x, gy11.x, gz11.x, gw11.x);
		tvec4<T, P> g1011(gx11.y, gy11.y, gz11.y, gw11.y);
		tvec4<T, P> g0111(gx11.z, gy11.z, gz11.z, gw11.z);
		tvec4<T, P> g1111(gx11.w, gy11.w, gz11.w, gw11.w);

		tvec4<T, P> norm00 = detail::taylorInvSqrt(tvec4<T, P>(dot(g0000, g0000), dot(g0100, g0100), dot(g1000, g1000), dot(g1100, g1100)));
		g0000 *= norm00.x;
		g0100 *= norm00.y;
		g1000 *= norm00.z;
		g1100 *= norm00.w;

		tvec4<T, P> norm01 = detail::taylorInvSqrt(tvec4<T, P>(dot(g0001, g0001), dot(g0101, g0101), dot(g1001, g1001), dot(g1101, g1101)));
		g0001 *= norm01.x;
		g0101 *= norm01.y;
		g1001 *= norm01.z;
		g1101 *= norm01.w;

		tvec4<T, P> norm10 = detail::taylorInvSqrt(tvec4<T, P>(dot(g0010, g0010), dot(g0110, g0110), dot(g1010, g1010), dot(g1110, g1110)));
		g0010 *= norm10.x;
		g0110 *= norm10.y;
		g1010 *= norm10.z;
		g1110 *= norm10.w;

		tvec4<T, P> norm11 = detail::taylorInvSqrt(tvec4<T, P>(dot(g0011, g0011), dot(g0111, g0111), dot(g1011, g1011), dot(g1111, g1111)));
		g0011 *= norm11.x;
		g0111 *= norm11.y;
		g1011 *= norm11.z;
		g1111 *= norm11.w;

		T n0000 = dot(g0000, Pf0);
		T n1000 = dot(g1000, tvec4<T, P>(Pf1.x, Pf0.y, Pf0.z, Pf0.w));
		T n0100 = dot(g0100, tvec4<T, P>(Pf0.x, Pf1.y, Pf0.z, Pf0.w));
		T n1100 = dot(g1100, tvec4<T, P>(Pf1.x, Pf1.y, Pf0.z, Pf0.w));
		T n0010 = dot(g0010, tvec4<T, P>(Pf0.x, Pf0.y, Pf1.z, Pf0.w));
		T n1010 = dot(g1010, tvec4<T, P>(Pf1.x, Pf0.y, Pf1.z, Pf0.w));
		T n0110 = dot(g0110, tvec4<T, P>(Pf0.x, Pf1.y, Pf1.z, Pf0.w));
		T n1110 = dot(g1110, tvec4<T, P>(Pf1.x, Pf1.y, Pf1.z, Pf0.w));
		T n0001 = dot(g0001, tvec4<T, P>(Pf0.x, Pf0.y, Pf0.z, Pf1.w));
		T n1001 = dot(g1001, tvec4<T, P>(Pf1.x, Pf0.y, Pf0.z, Pf1.w));
		T n0101 = dot(g0101, tvec4<T, P>(Pf0.x, Pf1.y, Pf0.z, Pf1.w));
		T n1101 = dot(g1101, tvec4<T, P>(Pf1.x, Pf1.y, Pf0.z, Pf1.w));
		T n0011 = dot(g0011, tvec4<T, P>(Pf0.x, Pf0.y, Pf1.z, Pf1.w));
		T n1011 = dot(g1011, tvec4<T, P>(Pf1.x, Pf0.y, Pf1.z, Pf1.w));
		T n0111 = dot(g0111, tvec4<T, P>(Pf0.x, Pf1.y, Pf1.z, Pf1.w));
		T n1111 = dot(g1111, Pf1);

		tvec4<T, P> fade_xyzw = detail::fade(Pf0);
		tvec4<T, P> n_0w = mix(tvec4<T, P>(n0000, n1000, n0100, n1100), tvec4<T, P>(n0001, n1001, n0101, n1101), fade_xyzw.w);
		tvec4<T, P> n_1w = mix(tvec4<T, P>(n0010, n1010, n0110, n1110), tvec4<T, P>(n0011, n1011, n0111, n1111), fade_xyzw.w);
		tvec4<T, P> n_zw = mix(n_0w, n_1w, fade_xyzw.z);
		tvec2<T, P> n_yzw = mix(tvec2<T, P>(n_zw.x, n_zw.y), tvec2<T, P>(n_zw.z, n_zw.w), fade_xyzw.y);
		T n_xyzw = mix(n_yzw.x, n_yzw.y, fade_xyzw.x);
		return T(2.2) * n_xyzw;
	}

	// Classic Perlin noise, periodic variant
	template <typename T, precision P>
	GLM_FUNC_QUALIFIER T perlin(tvec2<T, P> const & Position, tvec2<T, P> const & rep)
	{
		tvec4<T, P> Pi = floor(tvec4<T, P>(Position.x, Position.y, Position.x, Position.y)) + tvec4<T, P>(0.0, 0.0, 1.0, 1.0);
		tvec4<T, P> Pf = fract(tvec4<T, P>(Position.x, Position.y, Position.x, Position.y)) - tvec4<T, P>(0.0, 0.0, 1.0, 1.0);
		Pi = mod(Pi, tvec4<T, P>(rep.x, rep.y, rep.x, rep.y)); // To create noise with explicit period
		Pi = mod(Pi, tvec4<T, P>(289)); // To avoid truncation effects in permutation
		tvec4<T, P> ix(Pi.x, Pi.z, Pi.x, Pi.z);
		tvec4<T, P> iy(Pi.y, Pi.y, Pi.w, Pi.w);
		tvec4<T, P> fx(Pf.x, Pf.z, Pf.x, Pf.z);
		tvec4<T, P> fy(Pf.y, Pf.y, Pf.w, Pf.w);

		tvec4<T, P> i = detail::permute(detail::permute(ix) + iy);

		tvec4<T, P> gx = static_cast<T>(2) * fract(i / T(41)) - T(1);
		tvec4<T, P> gy = abs(gx) - T(0.5);
		tvec4<T, P> tx = floor(gx + T(0.5));
		gx = gx - tx;

		tvec2<T, P> g00(gx.x, gy.x);
		tvec2<T, P> g10(gx.y, gy.y);
		tvec2<T, P> g01(gx.z, gy.z);
		tvec2<T, P> g11(gx.w, gy.w);

		tvec4<T, P> norm = detail::taylorInvSqrt(tvec4<T, P>(dot(g00, g00), dot(g01, g01), dot(g10, g10), dot(g11, g11)));
		g00 *= norm.x;
		g01 *= norm.y;
		g10 *= norm.z;
		g11 *= norm.w;

		T n00 = dot(g00, tvec2<T, P>(fx.x, fy.x));
		T n10 = dot(g10, tvec2<T, P>(fx.y, fy.y));
		T n01 = dot(g01, tvec2<T, P>(fx.z, fy.z));
		T n11 = dot(g11, tvec2<T, P>(fx.w, fy.w));

		tvec2<T, P> fade_xy = detail::fade(tvec2<T, P>(Pf.x, Pf.y));
		tvec2<T, P> n_x = mix(tvec2<T, P>(n00, n01), tvec2<T, P>(n10, n11), fade_xy.x);
		T n_xy = mix(n_x.x, n_x.y, fade_xy.y);
		return T(2.3) * n_xy;
	}

	// Classic Perlin noise, periodic variant
	template <typename T, precision P>
	GLM_FUNC_QUALIFIER T perlin(tvec3<T, P> const & Position, tvec3<T, P> const & rep)
	{
		tvec3<T, P> Pi0 = mod(floor(Position), rep); // Integer part, modulo period
		tvec3<T, P> Pi1 = mod(Pi0 + tvec3<T, P>(T(1)), rep); // Integer part + 1, mod period
		Pi0 = mod(Pi0, tvec3<T, P>(289));
		Pi1 = mod(Pi1, tvec3<T, P>(289));
		tvec3<T, P> Pf0 = fract(Position); // Fractional part for interpolation
		tvec3<T, P> Pf1 = Pf0 - tvec3<T, P>(T(1)); // Fractional part - 1.0
		tvec4<T, P> ix = tvec4<T, P>(Pi0.x, Pi1.x, Pi0.x, Pi1.x);
		tvec4<T, P> iy = tvec4<T, P>(Pi0.y, Pi0.y, Pi1.y, Pi1.y);
		tvec4<T, P> iz0(Pi0.z);
		tvec4<T, P> iz1(Pi1.z);

		tvec4<T, P> ixy = detail::permute(detail::permute(ix) + iy);
		tvec4<T, P> ixy0 = detail::permute(ixy + iz0);
		tvec4<T, P> ixy1 = detail::permute(ixy + iz1);

		tvec4<T, P> gx0 = ixy0 / T(7);
		tvec4<T, P> gy0 = fract(floor(gx0) / T(7)) - T(0.5);
		gx0 = fract(gx0);
		tvec4<T, P> gz0 = tvec4<T, P>(0.5) - abs(gx0) - abs(gy0);
		tvec4<T, P> sz0 = step(gz0, tvec4<T, P>(0));
		gx0 -= sz0 * (step(T(0), gx0) - T(0.5));
		gy0 -= sz0 * (step(T(0), gy0) - T(0.5));

		tvec4<T, P> gx1 = ixy1 / T(7);
		tvec4<T, P> gy1 = fract(floor(gx1) / T(7)) - T(0.5);
		gx1 = fract(gx1);
		tvec4<T, P> gz1 = tvec4<T, P>(0.5) - abs(gx1) - abs(gy1);
		tvec4<T, P> sz1 = step(gz1, tvec4<T, P>(T(0)));
		gx1 -= sz1 * (step(T(0), gx1) - T(0.5));
		gy1 -= sz1 * (step(T(0), gy1) - T(0.5));

		tvec3<T, P> g000 = tvec3<T, P>(gx0.x, gy0.x, gz0.x);
		tvec3<T, P> g100 = tvec3<T, P>(gx0.y, gy0.y, gz0.y);
		tvec3<T, P> g010 = tvec3<T, P>(gx0.z, gy0.z, gz0.z);
		tvec3<T, P> g110 = tvec3<T, P>(gx0.w, gy0.w, gz0.w);
		tvec3<T, P> g001 = tvec3<T, P>(gx1.x, gy1.x, gz1.x);
		tvec3<T, P> g101 = tvec3<T, P>(gx1.y, gy1.y, gz1.y);
		tvec3<T, P> g011 = tvec3<T, P>(gx1.z, gy1.z, gz1.z);
		tvec3<T, P> g111 = tvec3<T, P>(gx1.w, gy1.w, gz1.w);

		tvec4<T, P> norm0 = detail::taylorInvSqrt(tvec4<T, P>(dot(g000, g000), dot(g010, g010), dot(g100, g100), dot(g110, g110)));
		g000 *= norm0.x;
		g010 *= norm0.y;
		g100 *= norm0.z;
		g110 *= norm0.w;
		tvec4<T, P> norm1 = detail::taylorInvSqrt(tvec4<T, P>(dot(g001, g001), dot(g011, g011), dot(g101, g101), dot(g111, g111)));
		g001 *= norm1.x;
		g011 *= norm1.y;
		g101 *= norm1.z;
		g111 *= norm1.w;

		T n000 = dot(g000, Pf0);
		T n100 = dot(g100, tvec3<T, P>(Pf1.x, Pf0.y, Pf0.z));
		T n010 = dot(g010, tvec3<T, P>(Pf0.x, Pf1.y, Pf0.z));
		T n110 = dot(g110, tvec3<T, P>(Pf1.x, Pf1.y, Pf0.z));
		T n001 = dot(g001, tvec3<T, P>(Pf0.x, Pf0.y, Pf1.z));
		T n101 = dot(g101, tvec3<T, P>(Pf1.x, Pf0.y, Pf1.z));
		T n011 = dot(g011, tvec3<T, P>(Pf0.x, Pf1.y, Pf1.z));
		T n111 = dot(g111, Pf1);

		tvec3<T, P> fade_xyz = detail::fade(Pf0);
		tvec4<T, P> n_z = mix(tvec4<T, P>(n000, n100, n010, n110), tvec4<T, P>(n001, n101, n011, n111), fade_xyz.z);
		tvec2<T, P> n_yz = mix(tvec2<T, P>(n_z.x, n_z.y), tvec2<T, P>(n_z.z, n_z.w), fade_xyz.y);
		T n_xyz = mix(n_yz.x, n_yz.y, fade_xyz.x);
		return T(2.2) * n_xyz;
	}

	// Classic Perlin noise, periodic version
	template <typename T, precision P>
	GLM_FUNC_QUALIFIER T perlin(tvec4<T, P> const & Position, tvec4<T, P> const & rep)
	{
		tvec4<T, P> Pi0 = mod(floor(Position), rep); // Integer part modulo rep
		tvec4<T, P> Pi1 = mod(Pi0 + T(1), rep); // Integer part + 1 mod rep
		tvec4<T, P> Pf0 = fract(Position); // Fractional part for interpolation
		tvec4<T, P> Pf1 = Pf0 - T(1); // Fractional part - 1.0
		tvec4<T, P> ix = tvec4<T, P>(Pi0.x, Pi1.x, Pi0.x, Pi1.x);
		tvec4<T, P> iy = tvec4<T, P>(Pi0.y, Pi0.y, Pi1.y, Pi1.y);
		tvec4<T, P> iz0(Pi0.z);
		tvec4<T, P> iz1(Pi1.z);
		tvec4<T, P> iw0(Pi0.w);
		tvec4<T, P> iw1(Pi1.w);

		tvec4<T, P> ixy = detail::permute(detail::permute(ix) + iy);
		tvec4<T, P> ixy0 = detail::permute(ixy + iz0);
		tvec4<T, P> ixy1 = detail::permute(ixy + iz1);
		tvec4<T, P> ixy00 = detail::permute(ixy0 + iw0);
		tvec4<T, P> ixy01 = detail::permute(ixy0 + iw1);
		tvec4<T, P> ixy10 = detail::permute(ixy1 + iw0);
		tvec4<T, P> ixy11 = detail::permute(ixy1 + iw1);

		tvec4<T, P> gx00 = ixy00 / T(7);
		tvec4<T, P> gy00 = floor(gx00) / T(7);
		tvec4<T, P> gz00 = floor(gy00) / T(6);
		gx00 = fract(gx00) - T(0.5);
		gy00 = fract(gy00) - T(0.5);
		gz00 = fract(gz00) - T(0.5);
		tvec4<T, P> gw00 = tvec4<T, P>(0.75) - abs(gx00) - abs(gy00) - abs(gz00);
		tvec4<T, P> sw00 = step(gw00, tvec4<T, P>(0));
		gx00 -= sw00 * (step(T(0), gx00) - T(0.5));
		gy00 -= sw00 * (step(T(0), gy00) - T(0.5));

		tvec4<T, P> gx01 = ixy01 / T(7);
		tvec4<T, P> gy01 = floor(gx01) / T(7);
		tvec4<T, P> gz01 = floor(gy01) / T(6);
		gx01 = fract(gx01) - T(0.5);
		gy01 = fract(gy01) - T(0.5);
		gz01 = fract(gz01) - T(0.5);
		tvec4<T, P> gw01 = tvec4<T, P>(0.75) - abs(gx01) - abs(gy01) - abs(gz01);
		tvec4<T, P> sw01 = step(gw01, tvec4<T, P>(0.0));
		gx01 -= sw01 * (step(T(0), gx01) - T(0.5));
		gy01 -= sw01 * (step(T(0), gy01) - T(0.5));

		tvec4<T, P> gx10 = ixy10 / T(7);
		tvec4<T, P> gy10 = floor(gx10) / T(7);
		tvec4<T, P> gz10 = floor(gy10) / T(6);
		gx10 = fract(gx10) - T(0.5);
		gy10 = fract(gy10) - T(0.5);
		gz10 = fract(gz10) - T(0.5);
		tvec4<T, P> gw10 = tvec4<T, P>(0.75) - abs(gx10) - abs(gy10) - abs(gz10);
		tvec4<T, P> sw10 = step(gw10, tvec4<T, P>(0.0));
		gx10 -= sw10 * (step(T(0), gx10) - T(0.5));
		gy10 -= sw10 * (step(T(0), gy10) - T(0.5));

		tvec4<T, P> gx11 = ixy11 / T(7);
		tvec4<T, P> gy11 = floor(gx11) / T(7);
		tvec4<T, P> gz11 = floor(gy11) / T(6);
		gx11 = fract(gx11) - T(0.5);
		gy11 = fract(gy11) - T(0.5);
		gz11 = fract(gz11) - T(0.5);
		tvec4<T, P> gw11 = tvec4<T, P>(0.75) - abs(gx11) - abs(gy11) - abs(gz11);
		tvec4<T, P> sw11 = step(gw11, tvec4<T, P>(T(0)));
		gx11 -= sw11 * (step(T(0), gx11) - T(0.5));
		gy11 -= sw11 * (step(T(0), gy11) - T(0.5));

		tvec4<T, P> g0000(gx00.x, gy00.x, gz00.x, gw00.x);
		tvec4<T, P> g1000(gx00.y, gy00.y, gz00.y, gw00.y);
		tvec4<T, P> g0100(gx00.z, gy00.z, gz00.z, gw00.z);
		tvec4<T, P> g1100(gx00.w, gy00.w, gz00.w, gw00.w);
		tvec4<T, P> g0010(gx10.x, gy10.x, gz10.x, gw10.x);
		tvec4<T, P> g1010(gx10.y, gy10.y, gz10.y, gw10.y);
		tvec4<T, P> g0110(gx10.z, gy10.z, gz10.z, gw10.z);
		tvec4<T, P> g1110(gx10.w, gy10.w, gz10.w, gw10.w);
		tvec4<T, P> g0001(gx01.x, gy01.x, gz01.x, gw01.x);
		tvec4<T, P> g1001(gx01.y, gy01.y, gz01.y, gw01.y);
		tvec4<T, P> g0101(gx01.z, gy01.z, gz01.z, gw01.z);
		tvec4<T, P> g1101(gx01.w, gy01.w, gz01.w, gw01.w);
		tvec4<T, P> g0011(gx11.x, gy11.x, gz11.x, gw11.x);
		tvec4<T, P> g1011(gx11.y, gy11.y, gz11.y, gw11.y);
		tvec4<T, P> g0111(gx11.z, gy11.z, gz11.z, gw11.z);
		tvec4<T, P> g1111(gx11.w, gy11.w, gz11.w, gw11.w);

		tvec4<T, P> norm00 = detail::taylorInvSqrt(tvec4<T, P>(dot(g0000, g0000), dot(g0100, g0100), dot(g1000, g1000), dot(g1100, g1100)));
		g0000 *= norm00.x;
		g0100 *= norm00.y;
		g1000 *= norm00.z;
		g1100 *= norm00.w;

		tvec4<T, P> norm01 = detail::taylorInvSqrt(tvec4<T, P>(dot(g0001, g0001), dot(g0101, g0101), dot(g1001, g1001), dot(g1101, g1101)));
		g0001 *= norm01.x;
		g0101 *= norm01.y;
		g1001 *= norm01.z;
		g1101 *= norm01.w;

		tvec4<T, P> norm10 = detail::taylorInvSqrt(tvec4<T, P>(dot(g0010, g0010), dot(g0110, g0110), dot(g1010, g1010), dot(g1110, g1110)));
		g0010 *= norm10.x;
		g0110 *= norm10.y;
		g1010 *= norm10.z;
		g1110 *= norm10.w;

		tvec4<T, P> norm11 = detail::taylorInvSqrt(tvec4<T, P>(dot(g0011, g0011), dot(g0111, g0111), dot(g1011, g1011), dot(g1111, g1111)));
		g0011 *= norm11.x;
		g0111 *= norm11.y;
		g1011 *= norm11.z;
		g1111 *= norm11.w;

		T n0000 = dot(g0000, Pf0);
		T n1000 = dot(g1000, tvec4<T, P>(Pf1.x, Pf0.y, Pf0.z, Pf0.w));
		T n0100 = dot(g0100, tvec4<T, P>(Pf0.x, Pf1.y, Pf0.z, Pf0.w));
		T n1100 = dot(g1100, tvec4<T, P>(Pf1.x, Pf1.y, Pf0.z, Pf0.w));
		T n0010 = dot(g0010, tvec4<T, P>(Pf0.x, Pf0.y, Pf1.z, Pf0.w));
		T n1010 = dot(g1010, tvec4<T, P>(Pf1.x, Pf0.y, Pf1.z, Pf0.w));
		T n0110 = dot(g0110, tvec4<T, P>(Pf0.x, Pf1.y, Pf1.z, Pf0.w));
		T n1110 = dot(g1110, tvec4<T, P>(Pf1.x, Pf1.y, Pf1.z, Pf0.w));
		T n0001 = dot(g0001, tvec4<T, P>(Pf0.x, Pf0.y, Pf0.z, Pf1.w));
		T n1001 = dot(g1001, tvec4<T, P>(Pf1.x, Pf0.y, Pf0.z, Pf1.w));
		T n0101 = dot(g0101, tvec4<T, P>(Pf0.x, Pf1.y, Pf0.z, Pf1.w));
		T n1101 = dot(g1101, tvec4<T, P>(Pf1.x, Pf1.y, Pf0.z, Pf1.w));
		T n0011 = dot(g0011, tvec4<T, P>(Pf0.x, Pf0.y, Pf1.z, Pf1.w));
		T n1011 = dot(g1011, tvec4<T, P>(Pf1.x, Pf0.y, Pf1.z, Pf1.w));
		T n0111 = dot(g0111, tvec4<T, P>(Pf0.x, Pf1.y, Pf1.z, Pf1.w));
		T n1111 = dot(g1111, Pf1);

		tvec4<T, P> fade_xyzw = detail::fade(Pf0);
		tvec4<T, P> n_0w = mix(tvec4<T, P>(n0000, n1000, n0100, n1100), tvec4<T, P>(n0001, n1001, n0101, n1101), fade_xyzw.w);
		tvec4<T, P> n_1w = mix(tvec4<T, P>(n0010, n1010, n0110, n1110), tvec4<T, P>(n0011, n1011, n0111, n1111), fade_xyzw.w);
		tvec4<T, P> n_zw = mix(n_0w, n_1w, fade_xyzw.z);
		tvec2<T, P> n_yzw = mix(tvec2<T, P>(n_zw.x, n_zw.y), tvec2<T, P>(n_zw.z, n_zw.w), fade_xyzw.y);
		T n_xyzw = mix(n_yzw.x, n_yzw.y, fade_xyzw.x);
		return T(2.2) * n_xyzw;
	}

	template <typename T, precision P>
	GLM_FUNC_QUALIFIER T simplex(glm::tvec2<T, P> const & v)
	{
		tvec4<T, P> const C = tvec4<T, P>(
			T( 0.211324865405187),  // (3.0 -  sqrt(3.0)) / 6.0
			T( 0.366025403784439),  //  0.5 * (sqrt(3.0)  - 1.0)
			T(-0.577350269189626),	// -1.0 + 2.0 * C.x
			T( 0.024390243902439)); //  1.0 / 41.0

		// First corner
		tvec2<T, P> i  = floor(v + dot(v, tvec2<T, P>(C[1])));
		tvec2<T, P> x0 = v -   i + dot(i, tvec2<T, P>(C[0]));

		// Other corners
		//i1.x = step( x0.y, x0.x ); // x0.x > x0.y ? 1.0 : 0.0
		//i1.y = 1.0 - i1.x;
		tvec2<T, P> i1 = (x0.x > x0.y) ? tvec2<T, P>(1, 0) : tvec2<T, P>(0, 1);
		// x0 = x0 - 0.0 + 0.0 * C.xx ;
		// x1 = x0 - i1 + 1.0 * C.xx ;
		// x2 = x0 - 1.0 + 2.0 * C.xx ;
		tvec4<T, P> x12 = tvec4<T, P>(x0.x, x0.y, x0.x, x0.y) + tvec4<T, P>(C.x, C.x, C.z, C.z);
		x12 = tvec4<T, P>(tvec2<T, P>(x12) - i1, x12.z, x12.w);

		// Permutations
		i = mod(i, tvec2<T, P>(289)); // Avoid truncation effects in permutation
		tvec3<T, P> p = detail::permute(
			detail::permute(i.y + tvec3<T, P>(T(0), i1.y, T(1)))
			+ i.x + tvec3<T, P>(T(0), i1.x, T(1)));

		tvec3<T, P> m = max(tvec3<T, P>(0.5) - tvec3<T, P>(
			dot(x0, x0),
			dot(tvec2<T, P>(x12.x, x12.y), tvec2<T, P>(x12.x, x12.y)), 
			dot(tvec2<T, P>(x12.z, x12.w), tvec2<T, P>(x12.z, x12.w))), tvec3<T, P>(0));
		m = m * m ;
		m = m * m ;

		// Gradients: 41 points uniformly over a line, mapped onto a diamond.
		// The ring size 17*17 = 289 is close to a multiple of 41 (41*7 = 287)

		tvec3<T, P> x = static_cast<T>(2) * fract(p * C.w) - T(1);
		tvec3<T, P> h = abs(x) - T(0.5);
		tvec3<T, P> ox = floor(x + T(0.5));
		tvec3<T, P> a0 = x - ox;

		// Normalise gradients implicitly by scaling m
		// Inlined for speed: m *= taylorInvSqrt( a0*a0 + h*h );
		m *= static_cast<T>(1.79284291400159) - T(0.85373472095314) * (a0 * a0 + h * h);

		// Compute final noise value at P
		tvec3<T, P> g;
		g.x  = a0.x  * x0.x  + h.x  * x0.y;
		//g.yz = a0.yz * x12.xz + h.yz * x12.yw;
		g.y = a0.y * x12.x + h.y * x12.y;
		g.z = a0.z * x12.z + h.z * x12.w;
		return T(130) * dot(m, g);
	}

	template <typename T, precision P>
	GLM_FUNC_QUALIFIER T simplex(tvec3<T, P> const & v)
	{
		tvec2<T, P> const C(1.0 / 6.0, 1.0 / 3.0);
		tvec4<T, P> const D(0.0, 0.5, 1.0, 2.0);

		// First corner
		tvec3<T, P> i(floor(v + dot(v, tvec3<T, P>(C.y))));
		tvec3<T, P> x0(v - i + dot(i, tvec3<T, P>(C.x)));

		// Other corners
		tvec3<T, P> g(step(tvec3<T, P>(x0.y, x0.z, x0.x), x0));
		tvec3<T, P> l(T(1) - g);
		tvec3<T, P> i1(min(g, tvec3<T, P>(l.z, l.x, l.y)));
		tvec3<T, P> i2(max(g, tvec3<T, P>(l.z, l.x, l.y)));

		//   x0 = x0 - 0.0 + 0.0 * C.xxx;
		//   x1 = x0 - i1  + 1.0 * C.xxx;
		//   x2 = x0 - i2  + 2.0 * C.xxx;
		//   x3 = x0 - 1.0 + 3.0 * C.xxx;
		tvec3<T, P> x1(x0 - i1 + C.x);
		tvec3<T, P> x2(x0 - i2 + C.y); // 2.0*C.x = 1/3 = C.y
		tvec3<T, P> x3(x0 - D.y);      // -1.0+3.0*C.x = -0.5 = -D.y

		// Permutations
		i = detail::mod289(i);
		tvec4<T, P> p(detail::permute(detail::permute(detail::permute(
			i.z + tvec4<T, P>(T(0), i1.z, i2.z, T(1))) +
			i.y + tvec4<T, P>(T(0), i1.y, i2.y, T(1))) +
			i.x + tvec4<T, P>(T(0), i1.x, i2.x, T(1))));

		// Gradients: 7x7 points over a square, mapped onto an octahedron.
		// The ring size 17*17 = 289 is close to a multiple of 49 (49*6 = 294)
		T n_ = static_cast<T>(0.142857142857); // 1.0/7.0
		tvec3<T, P> ns(n_ * tvec3<T, P>(D.w, D.y, D.z) - tvec3<T, P>(D.x, D.z, D.x));

		tvec4<T, P> j(p - T(49) * floor(p * ns.z * ns.z));  //  mod(p,7*7)

		tvec4<T, P> x_(floor(j * ns.z));
		tvec4<T, P> y_(floor(j - T(7) * x_));    // mod(j,N)

		tvec4<T, P> x(x_ * ns.x + ns.y);
		tvec4<T, P> y(y_ * ns.x + ns.y);
		tvec4<T, P> h(T(1) - abs(x) - abs(y));

		tvec4<T, P> b0(x.x, x.y, y.x, y.y);
		tvec4<T, P> b1(x.z, x.w, y.z, y.w);

		// vec4 s0 = vec4(lessThan(b0,0.0))*2.0 - 1.0;
		// vec4 s1 = vec4(lessThan(b1,0.0))*2.0 - 1.0;
		tvec4<T, P> s0(floor(b0) * T(2) + T(1));
		tvec4<T, P> s1(floor(b1) * T(2) + T(1));
		tvec4<T, P> sh(-step(h, tvec4<T, P>(0.0)));

		tvec4<T, P> a0 = tvec4<T, P>(b0.x, b0.z, b0.y, b0.w) + tvec4<T, P>(s0.x, s0.z, s0.y, s0.w) * tvec4<T, P>(sh.x, sh.x, sh.y, sh.y);
		tvec4<T, P> a1 = tvec4<T, P>(b1.x, b1.z, b1.y, b1.w) + tvec4<T, P>(s1.x, s1.z, s1.y, s1.w) * tvec4<T, P>(sh.z, sh.z, sh.w, sh.w);

		tvec3<T, P> p0(a0.x, a0.y, h.x);
		tvec3<T, P> p1(a0.z, a0.w, h.y);
		tvec3<T, P> p2(a1.x, a1.y, h.z);
		tvec3<T, P> p3(a1.z, a1.w, h.w);

		// Normalise gradients
		tvec4<T, P> norm = detail::taylorInvSqrt(tvec4<T, P>(dot(p0, p0), dot(p1, p1), dot(p2, p2), dot(p3, p3)));
		p0 *= norm.x;
		p1 *= norm.y;
		p2 *= norm.z;
		p3 *= norm.w;

		// Mix final noise value
		tvec4<T, P> m = max(T(0.6) - tvec4<T, P>(dot(x0, x0), dot(x1, x1), dot(x2, x2), dot(x3, x3)), tvec4<T, P>(0));
		m = m * m;
		return T(42) * dot(m * m, tvec4<T, P>(dot(p0, x0), dot(p1, x1), dot(p2, x2), dot(p3, x3)));
	}

	template <typename T, precision P>
	GLM_FUNC_QUALIFIER T simplex(tvec4<T, P> const & v)
	{
		tvec4<T, P> const C(
			0.138196601125011,  // (5 - sqrt(5))/20  G4
			0.276393202250021,  // 2 * G4
			0.414589803375032,  // 3 * G4
			-0.447213595499958); // -1 + 4 * G4

		// (sqrt(5) - 1)/4 = F4, used once below
		T const F4 = static_cast<T>(0.309016994374947451);

		// First corner
		tvec4<T, P> i  = floor(v + dot(v, vec4(F4)));
		tvec4<T, P> x0 = v -   i + dot(i, vec4(C.x));

		// Other corners

		// Rank sorting originally contributed by Bill Licea-Kane, AMD (formerly ATI)
		tvec4<T, P> i0;
		tvec3<T, P> isX = step(tvec3<T, P>(x0.y, x0.z, x0.w), tvec3<T, P>(x0.x));
		tvec3<T, P> isYZ = step(tvec3<T, P>(x0.z, x0.w, x0.w), tvec3<T, P>(x0.y, x0.y, x0.z));
		//  i0.x = dot(isX, vec3(1.0));
		//i0.x = isX.x + isX.y + isX.z;
		//i0.yzw = static_cast<T>(1) - isX;
		i0 = tvec4<T, P>(isX.x + isX.y + isX.z, T(1) - isX);
		//  i0.y += dot(isYZ.xy, vec2(1.0));
		i0.y += isYZ.x + isYZ.y;
		//i0.zw += 1.0 - tvec2<T, P>(isYZ.x, isYZ.y);
		i0.z += static_cast<T>(1) - isYZ.x;
		i0.w += static_cast<T>(1) - isYZ.y;
		i0.z += isYZ.z;
		i0.w += static_cast<T>(1) - isYZ.z;

		// i0 now contains the unique values 0,1,2,3 in each channel
		tvec4<T, P> i3 = clamp(i0, T(0), T(1));
		tvec4<T, P> i2 = clamp(i0 - T(1), T(0), T(1));
		tvec4<T, P> i1 = clamp(i0 - T(2), T(0), T(1));

		//  x0 = x0 - 0.0 + 0.0 * C.xxxx
		//  x1 = x0 - i1  + 0.0 * C.xxxx
		//  x2 = x0 - i2  + 0.0 * C.xxxx
		//  x3 = x0 - i3  + 0.0 * C.xxxx
		//  x4 = x0 - 1.0 + 4.0 * C.xxxx
		tvec4<T, P> x1 = x0 - i1 + C.x;
		tvec4<T, P> x2 = x0 - i2 + C.y;
		tvec4<T, P> x3 = x0 - i3 + C.z;
		tvec4<T, P> x4 = x0 + C.w;

		// Permutations
		i = mod(i, tvec4<T, P>(289)); 
		T j0 = detail::permute(detail::permute(detail::permute(detail::permute(i.w) + i.z) + i.y) + i.x);
		tvec4<T, P> j1 = detail::permute(detail::permute(detail::permute(detail::permute(
			i.w + tvec4<T, P>(i1.w, i2.w, i3.w, T(1))) +
			i.z + tvec4<T, P>(i1.z, i2.z, i3.z, T(1))) +
			i.y + tvec4<T, P>(i1.y, i2.y, i3.y, T(1))) +
			i.x + tvec4<T, P>(i1.x, i2.x, i3.x, T(1)));

		// Gradients: 7x7x6 points over a cube, mapped onto a 4-cross polytope
		// 7*7*6 = 294, which is close to the ring size 17*17 = 289.
		tvec4<T, P> ip = tvec4<T, P>(T(1) / T(294), T(1) / T(49), T(1) / T(7), T(0));

		tvec4<T, P> p0 = gtc::grad4(j0,   ip);
		tvec4<T, P> p1 = gtc::grad4(j1.x, ip);
		tvec4<T, P> p2 = gtc::grad4(j1.y, ip);
		tvec4<T, P> p3 = gtc::grad4(j1.z, ip);
		tvec4<T, P> p4 = gtc::grad4(j1.w, ip);

		// Normalise gradients
		tvec4<T, P> norm = detail::taylorInvSqrt(tvec4<T, P>(dot(p0, p0), dot(p1, p1), dot(p2, p2), dot(p3, p3)));
		p0 *= norm.x;
		p1 *= norm.y;
		p2 *= norm.z;
		p3 *= norm.w;
		p4 *= detail::taylorInvSqrt(dot(p4, p4));

		// Mix contributions from the five corners
		tvec3<T, P> m0 = max(T(0.6) - tvec3<T, P>(dot(x0, x0), dot(x1, x1), dot(x2, x2)), tvec3<T, P>(0));
		tvec2<T, P> m1 = max(T(0.6) - tvec2<T, P>(dot(x3, x3), dot(x4, x4)             ), tvec2<T, P>(0));
		m0 = m0 * m0;
		m1 = m1 * m1;
		return T(49) * 
			(dot(m0 * m0, tvec3<T, P>(dot(p0, x0), dot(p1, x1), dot(p2, x2))) + 
			dot(m1 * m1, tvec2<T, P>(dot(p3, x3), dot(p4, x4))));
	}
}//namespace glm