Add SIMD optimization for common functions

glm/detail/func_common.inl

@@ -52,7 +52,16 @@ namespace detail
 		}
 	};
 
-	template <typename T, typename U, precision P, template <class, precision> class vecType>
+	template <typename T, precision P, template <typename, precision> class vecType>
+	struct compute_abs_vector
+	{
+		GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & x)
+		{
+			return detail::functor1<T, T, P, vecType>::call(abs, x);
+		}
+	};
+
+	template <typename T, typename U, precision P, template <typename, precision> class vecType>
 	struct compute_mix_vector
 	{
 		GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & x, vecType<T, P> const & y, vecType<U, P> const & a)
@@ -63,7 +72,7 @@ namespace detail
 		}
 	};
 
-	template <typename T, precision P, template <class, precision> class vecType>
+	template <typename T, precision P, template <typename, precision> class vecType>
 	struct compute_mix_vector<T, bool, P, vecType>
 	{
 		GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & x, vecType<T, P> const & y, vecType<bool, P> const & a)
@@ -75,7 +84,7 @@ namespace detail
 		}
 	};
 
-	template <typename T, typename U, precision P, template <class, precision> class vecType>
+	template <typename T, typename U, precision P, template <typename, precision> class vecType>
 	struct compute_mix_scalar
 	{
 		GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & x, vecType<T, P> const & y, U const & a)
@@ -86,7 +95,7 @@ namespace detail
 		}
 	};
 
-	template <typename T, precision P, template <class, precision> class vecType>
+	template <typename T, precision P, template <typename, precision> class vecType>
 	struct compute_mix_scalar<T, bool, P, vecType>
 	{
 		GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & x, vecType<T, P> const & y, bool const & a)
@@ -115,7 +124,7 @@ namespace detail
 		}
 	};
 
-	template <typename T, precision P, template <class, precision> class vecType, bool isFloat = true>
+	template <typename T, precision P, template <typename, precision> class vecType, bool isFloat = true>
 	struct compute_sign
 	{
 		GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & x)
@@ -125,7 +134,7 @@ namespace detail
 	};
 
 #	if GLM_ARCH == GLM_ARCH_X86
-	template <typename T, precision P, template <class, precision> class vecType>
+	template <typename T, precision P, template <typename, precision> class vecType>
 	struct compute_sign<T, P, vecType, false>
 	{
 		GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & x)
@@ -138,10 +147,55 @@ namespace detail
 	};
 #	endif
 
-	template <typename T, precision P, template <class, precision> class vecType, typename genType, bool isFloat = true>
+	template <typename T, precision P, template <typename, precision> class vecType>
+	struct compute_floor
+	{
+		GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & x)
+		{
+			return detail::functor1<T, T, P, vecType>::call(floor, x);
+		}
+	};
+
+	template <typename T, precision P, template <typename, precision> class vecType>
+	struct compute_ceil
+	{
+		GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & x)
+		{
+			return detail::functor1<T, T, P, vecType>::call(ceil, x);
+		}
+	};
+
+	template <typename T, precision P, template <typename, precision> class vecType>
+	struct compute_fract
+	{
+		GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & x)
+		{
+			return x - floor(x);
+		}
+	};
+
+	template <typename T, precision P, template <typename, precision> class vecType>
+	struct compute_trunc
+	{
+		GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & x)
+		{
+			return detail::functor1<T, T, P, vecType>::call(trunc, x);
+		}
+	};
+
+	template <typename T, precision P, template <typename, precision> class vecType>
+	struct compute_round
+	{
+		GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & x)
+		{
+			return detail::functor1<T, T, P, vecType>::call(round, x);
+		}
+	};
+
+	template <typename T, precision P, template <typename, precision> class vecType>
 	struct compute_mod
 	{
-		GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & a, genType const & b)
+		GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & a, vecType<T, P> const & b)
 		{
 			GLM_STATIC_ASSERT(std::numeric_limits<T>::is_iec559, "'mod' only accept floating-point inputs. Include <glm/gtc/integer.hpp> for integer inputs.");
 			return a - b * floor(a / b);
@@ -166,7 +220,7 @@ namespace detail
 	template <typename T, precision P, template <typename, precision> class vecType>
 	GLM_FUNC_QUALIFIER vecType<T, P> abs(vecType<T, P> const & x)
 	{
-		return detail::functor1<T, T, P, vecType>::call(abs, x);
+		return detail::compute_abs_vector<T, P, vecType>::call(x);
 	}
 
 	// sign
@@ -196,7 +250,8 @@ namespace detail
 	template <typename T, precision P, template <typename, precision> class vecType>
 	GLM_FUNC_QUALIFIER vecType<T, P> floor(vecType<T, P> const & x)
 	{
-		return detail::functor1<T, T, P, vecType>::call(floor, x);
+		GLM_STATIC_ASSERT(std::numeric_limits<T>::is_iec559, "'floor' only accept floating-point inputs.");
+		return detail::compute_floor<T, P, vecType>::call(x);
 	}
 
 	// trunc
@@ -215,7 +270,8 @@ namespace detail
 	template <typename T, precision P, template <typename, precision> class vecType>
 	GLM_FUNC_QUALIFIER vecType<T, P> trunc(vecType<T, P> const & x)
 	{
-		return detail::functor1<T, T, P, vecType>::call(trunc, x);
+		GLM_STATIC_ASSERT(std::numeric_limits<T>::is_iec559, "'trunc' only accept floating-point inputs");
+		return detail::compute_trunc<T, P, vecType>::call(x);
 	}
 
 	// round
@@ -234,7 +290,8 @@ namespace detail
 	template <typename T, precision P, template <typename, precision> class vecType>
 	GLM_FUNC_QUALIFIER vecType<T, P> round(vecType<T, P> const & x)
 	{
-		return detail::functor1<T, T, P, vecType>::call(round, x);
+		GLM_STATIC_ASSERT(std::numeric_limits<T>::is_iec559, "'round' only accept floating-point inputs");
+		return detail::compute_round<T, P, vecType>::call(x);
 	}
 
 /*
@@ -283,6 +340,7 @@ namespace detail
 	template <typename T, precision P, template <typename, precision> class vecType>
 	GLM_FUNC_QUALIFIER vecType<T, P> roundEven(vecType<T, P> const & x)
 	{
+		GLM_STATIC_ASSERT(std::numeric_limits<T>::is_iec559, "'roundEven' only accept floating-point inputs");
 		return detail::functor1<T, T, P, vecType>::call(roundEven, x);
 	}
 
@@ -291,22 +349,22 @@ namespace detail
 	template <typename T, precision P, template <typename, precision> class vecType>
 	GLM_FUNC_QUALIFIER vecType<T, P> ceil(vecType<T, P> const & x)
 	{
-		return detail::functor1<T, T, P, vecType>::call(ceil, x);
+		GLM_STATIC_ASSERT(std::numeric_limits<T>::is_iec559, "'ceil' only accept floating-point inputs");
+		return detail::compute_ceil<T, P, vecType>::call(x);
 	}
 
 	// fract
 	template <typename genType>
 	GLM_FUNC_QUALIFIER genType fract(genType x)
 	{
-		GLM_STATIC_ASSERT(std::numeric_limits<genType>::is_iec559, "'fract' only accept floating-point inputs");
-
 		return fract(tvec1<genType>(x)).x;
 	}
 
 	template <typename T, precision P, template <typename, precision> class vecType>
 	GLM_FUNC_QUALIFIER vecType<T, P> fract(vecType<T, P> const & x)
 	{
-		return x - floor(x);
+		GLM_STATIC_ASSERT(std::numeric_limits<T>::is_iec559, "'fract' only accept floating-point inputs");
+		return detail::compute_fract<T, P, vecType>::call(x);
 	}
 
 	// mod
@@ -319,13 +377,13 @@ namespace detail
 	template <typename T, precision P, template <typename, precision> class vecType>
 	GLM_FUNC_QUALIFIER vecType<T, P> mod(vecType<T, P> const & x, T y)
 	{
-		return detail::compute_mod<T, P, vecType, T, std::numeric_limits<T>::is_iec559>::call(x, y);
+		return detail::compute_mod<T, P, vecType>::call(x, vecType<T, P>(y));
 	}
 
 	template <typename T, precision P, template <typename, precision> class vecType>
 	GLM_FUNC_QUALIFIER vecType<T, P> mod(vecType<T, P> const & x, vecType<T, P> const & y)
 	{
-		return detail::compute_mod<T, P, vecType, vecType<T, P>, std::numeric_limits<T>::is_iec559>::call(x, y);
+		return detail::compute_mod<T, P, vecType>::call(x, y);
 	}
 
 	// modf
@@ -333,7 +391,6 @@ namespace detail
 	GLM_FUNC_QUALIFIER genType modf(genType x, genType & i)
 	{
 		GLM_STATIC_ASSERT(std::numeric_limits<genType>::is_iec559, "'modf' only accept floating-point inputs");
-
 		return std::modf(x, &i);
 	}
 

glm/detail/func_common_simd.inl

@@ -10,6 +10,28 @@
 namespace glm{
 namespace detail
 {
+	template <precision P>
+	struct compute_abs_vector<float, P, tvec4>
+	{
+		GLM_FUNC_QUALIFIER static tvec4<float, P> call(tvec4<float, P> const & v)
+		{
+			tvec4<float, P> result(uninitialize);
+			result.data = glm_f32v4_abs(v.data);
+			return result;
+		}
+	};
+
+	template <precision P>
+	struct compute_abs_vector<int, P, tvec4>
+	{
+		GLM_FUNC_QUALIFIER static tvec4<int, P> call(tvec4<int, P> const & v)
+		{
+			tvec4<int, P> result(uninitialize);
+			result.data = glm_i32v4_abs(v.data);
+			return result;
+		}
+	};
+
 	template <precision P>
 	struct compute_mix_vector<float, bool, P, tvec4>
 	{
@@ -28,109 +50,60 @@ namespace detail
 		}
 	};
 
-/*
-	static const __m128 GLM_VAR_USED zero = _mm_setzero_ps();
-	static const __m128 GLM_VAR_USED one = _mm_set_ps1(1.0f);
-	static const __m128 GLM_VAR_USED minus_one = _mm_set_ps1(-1.0f);
-	static const __m128 GLM_VAR_USED two = _mm_set_ps1(2.0f);
-	static const __m128 GLM_VAR_USED three = _mm_set_ps1(3.0f);
-	static const __m128 GLM_VAR_USED pi = _mm_set_ps1(3.1415926535897932384626433832795f);
-	static const __m128 GLM_VAR_USED hundred_eighty = _mm_set_ps1(180.f);
-	static const __m128 GLM_VAR_USED pi_over_hundred_eighty = _mm_set_ps1(0.017453292519943295769236907684886f);
-	static const __m128 GLM_VAR_USED hundred_eighty_over_pi = _mm_set_ps1(57.295779513082320876798154814105f);
-
-	static const ieee754_QNAN absMask;
-	static const __m128 GLM_VAR_USED abs4Mask = _mm_set_ps1(absMask.f);
-
-	static const __m128 GLM_VAR_USED _epi32_sign_mask = _mm_castsi128_ps(_mm_set1_epi32(static_cast<int>(0x80000000)));
-	//static const __m128 GLM_VAR_USED _epi32_inv_sign_mask = _mm_castsi128_ps(_mm_set1_epi32(0x7FFFFFFF));
-	//static const __m128 GLM_VAR_USED _epi32_mant_mask = _mm_castsi128_ps(_mm_set1_epi32(0x7F800000));
-	//static const __m128 GLM_VAR_USED _epi32_inv_mant_mask = _mm_castsi128_ps(_mm_set1_epi32(0x807FFFFF));
-	//static const __m128 GLM_VAR_USED _epi32_min_norm_pos = _mm_castsi128_ps(_mm_set1_epi32(0x00800000));
-	static const __m128 GLM_VAR_USED _epi32_0 = _mm_set_ps1(0);
-	static const __m128 GLM_VAR_USED _epi32_1 = _mm_set_ps1(1);
-	static const __m128 GLM_VAR_USED _epi32_2 = _mm_set_ps1(2);
-	static const __m128 GLM_VAR_USED _epi32_3 = _mm_set_ps1(3);
-	static const __m128 GLM_VAR_USED _epi32_4 = _mm_set_ps1(4);
-	static const __m128 GLM_VAR_USED _epi32_5 = _mm_set_ps1(5);
-	static const __m128 GLM_VAR_USED _epi32_6 = _mm_set_ps1(6);
-	static const __m128 GLM_VAR_USED _epi32_7 = _mm_set_ps1(7);
-	static const __m128 GLM_VAR_USED _epi32_8 = _mm_set_ps1(8);
-	static const __m128 GLM_VAR_USED _epi32_9 = _mm_set_ps1(9);
-	static const __m128 GLM_VAR_USED _epi32_127 = _mm_set_ps1(127);
-	//static const __m128 GLM_VAR_USED _epi32_ninf = _mm_castsi128_ps(_mm_set1_epi32(0xFF800000));
-	//static const __m128 GLM_VAR_USED _epi32_pinf = _mm_castsi128_ps(_mm_set1_epi32(0x7F800000));
-
-	static const __m128 GLM_VAR_USED _ps_1_3 = _mm_set_ps1(0.33333333333333333333333333333333f);
-	static const __m128 GLM_VAR_USED _ps_0p5 = _mm_set_ps1(0.5f);
-	static const __m128 GLM_VAR_USED _ps_1 = _mm_set_ps1(1.0f);
-	static const __m128 GLM_VAR_USED _ps_m1 = _mm_set_ps1(-1.0f);
-	static const __m128 GLM_VAR_USED _ps_2 = _mm_set_ps1(2.0f);
-	static const __m128 GLM_VAR_USED _ps_3 = _mm_set_ps1(3.0f);
-	static const __m128 GLM_VAR_USED _ps_127 = _mm_set_ps1(127.0f);
-	static const __m128 GLM_VAR_USED _ps_255 = _mm_set_ps1(255.0f);
-	static const __m128 GLM_VAR_USED _ps_2pow23 = _mm_set_ps1(8388608.0f);
+	template <precision P>
+	struct compute_floor<float, P, tvec4>
+	{
+		GLM_FUNC_QUALIFIER static tvec4<float, P> call(tvec4<float, P> const & v)
+		{
+			tvec4<float, P> result(uninitialize);
+			result.data = glm_f32v4_flr(v.data);
+			return result;
+		}
+	};
 
-	static const __m128 GLM_VAR_USED _ps_1_0_0_0 = _mm_set_ps(1.0f, 0.0f, 0.0f, 0.0f);
-	static const __m128 GLM_VAR_USED _ps_0_1_0_0 = _mm_set_ps(0.0f, 1.0f, 0.0f, 0.0f);
-	static const __m128 GLM_VAR_USED _ps_0_0_1_0 = _mm_set_ps(0.0f, 0.0f, 1.0f, 0.0f);
-	static const __m128 GLM_VAR_USED _ps_0_0_0_1 = _mm_set_ps(0.0f, 0.0f, 0.0f, 1.0f);
+	template <precision P>
+	struct compute_ceil<float, P, tvec4>
+	{
+		GLM_FUNC_QUALIFIER static tvec4<float, P> call(tvec4<float, P> const & v)
+		{
+			tvec4<float, P> result(uninitialize);
+			result.data = glm_f32v4_ceil(v.data);
+			return result;
+		}
+	};
 
-	static const __m128 GLM_VAR_USED _ps_pi = _mm_set_ps1(3.1415926535897932384626433832795f);
-	static const __m128 GLM_VAR_USED _ps_pi2 = _mm_set_ps1(6.283185307179586476925286766560f);
-	static const __m128 GLM_VAR_USED _ps_2_pi = _mm_set_ps1(0.63661977236758134307553505349006f);
-	static const __m128 GLM_VAR_USED _ps_pi_2 = _mm_set_ps1(1.5707963267948966192313216916398f);
-	static const __m128 GLM_VAR_USED _ps_4_pi = _mm_set_ps1(1.2732395447351626861510701069801f);
-	static const __m128 GLM_VAR_USED _ps_pi_4 = _mm_set_ps1(0.78539816339744830961566084581988f);
+	template <precision P>
+	struct compute_fract<float, P, tvec4>
+	{
+		GLM_FUNC_QUALIFIER static tvec4<float, P> call(tvec4<float, P> const & v)
+		{
+			tvec4<float, P> result(uninitialize);
+			result.data = glm_f32v4_frc(v.data);
+			return result;
+		}
+	};
 
-	static const __m128 GLM_VAR_USED _ps_sincos_p0 = _mm_set_ps1(0.15707963267948963959e1f);
-	static const __m128 GLM_VAR_USED _ps_sincos_p1 = _mm_set_ps1(-0.64596409750621907082e0f);
-	static const __m128 GLM_VAR_USED _ps_sincos_p2 = _mm_set_ps1(0.7969262624561800806e-1f);
-	static const __m128 GLM_VAR_USED _ps_sincos_p3 = _mm_set_ps1(-0.468175413106023168e-2f);
-	static const __m128 GLM_VAR_USED _ps_tan_p0 = _mm_set_ps1(-1.79565251976484877988e7f);
-	static const __m128 GLM_VAR_USED _ps_tan_p1 = _mm_set_ps1(1.15351664838587416140e6f);
-	static const __m128 GLM_VAR_USED _ps_tan_p2 = _mm_set_ps1(-1.30936939181383777646e4f);
-	static const __m128 GLM_VAR_USED _ps_tan_q0 = _mm_set_ps1(-5.38695755929454629881e7f);
-	static const __m128 GLM_VAR_USED _ps_tan_q1 = _mm_set_ps1(2.50083801823357915839e7f);
-	static const __m128 GLM_VAR_USED _ps_tan_q2 = _mm_set_ps1(-1.32089234440210967447e6f);
-	static const __m128 GLM_VAR_USED _ps_tan_q3 = _mm_set_ps1(1.36812963470692954678e4f);
-	static const __m128 GLM_VAR_USED _ps_tan_poleval = _mm_set_ps1(3.68935e19f);
-	static const __m128 GLM_VAR_USED _ps_atan_t0 = _mm_set_ps1(-0.91646118527267623468e-1f);
-	static const __m128 GLM_VAR_USED _ps_atan_t1 = _mm_set_ps1(-0.13956945682312098640e1f);
-	static const __m128 GLM_VAR_USED _ps_atan_t2 = _mm_set_ps1(-0.94393926122725531747e2f);
-	static const __m128 GLM_VAR_USED _ps_atan_t3 = _mm_set_ps1(0.12888383034157279340e2f);
-	static const __m128 GLM_VAR_USED _ps_atan_s0 = _mm_set_ps1(0.12797564625607904396e1f);
-	static const __m128 GLM_VAR_USED _ps_atan_s1 = _mm_set_ps1(0.21972168858277355914e1f);
-	static const __m128 GLM_VAR_USED _ps_atan_s2 = _mm_set_ps1(0.68193064729268275701e1f);
-	static const __m128 GLM_VAR_USED _ps_atan_s3 = _mm_set_ps1(0.28205206687035841409e2f);
+	template <precision P>
+	struct compute_round<float, P, tvec4>
+	{
+		GLM_FUNC_QUALIFIER static tvec4<float, P> call(tvec4<float, P> const & v)
+		{
+			tvec4<float, P> result(uninitialize);
+			result.data = glm_f32v4_rnd(v.data);
+			return result;
+		}
+	};
 
-	static const __m128 GLM_VAR_USED _ps_exp_hi = _mm_set_ps1(88.3762626647949f);
-	static const __m128 GLM_VAR_USED _ps_exp_lo = _mm_set_ps1(-88.3762626647949f);
-	static const __m128 GLM_VAR_USED _ps_exp_rln2 = _mm_set_ps1(1.4426950408889634073599f);
-	static const __m128 GLM_VAR_USED _ps_exp_p0 = _mm_set_ps1(1.26177193074810590878e-4f);
-	static const __m128 GLM_VAR_USED _ps_exp_p1 = _mm_set_ps1(3.02994407707441961300e-2f);
-	static const __m128 GLM_VAR_USED _ps_exp_q0 = _mm_set_ps1(3.00198505138664455042e-6f);
-	static const __m128 GLM_VAR_USED _ps_exp_q1 = _mm_set_ps1(2.52448340349684104192e-3f);
-	static const __m128 GLM_VAR_USED _ps_exp_q2 = _mm_set_ps1(2.27265548208155028766e-1f);
-	static const __m128 GLM_VAR_USED _ps_exp_q3 = _mm_set_ps1(2.00000000000000000009e0f);
-	static const __m128 GLM_VAR_USED _ps_exp_c1 = _mm_set_ps1(6.93145751953125e-1f);
-	static const __m128 GLM_VAR_USED _ps_exp_c2 = _mm_set_ps1(1.42860682030941723212e-6f);
-	static const __m128 GLM_VAR_USED _ps_exp2_hi = _mm_set_ps1(127.4999961853f);
-	static const __m128 GLM_VAR_USED _ps_exp2_lo = _mm_set_ps1(-127.4999961853f);
-	static const __m128 GLM_VAR_USED _ps_exp2_p0 = _mm_set_ps1(2.30933477057345225087e-2f);
-	static const __m128 GLM_VAR_USED _ps_exp2_p1 = _mm_set_ps1(2.02020656693165307700e1f);
-	static const __m128 GLM_VAR_USED _ps_exp2_p2 = _mm_set_ps1(1.51390680115615096133e3f);
-	static const __m128 GLM_VAR_USED _ps_exp2_q0 = _mm_set_ps1(2.33184211722314911771e2f);
-	static const __m128 GLM_VAR_USED _ps_exp2_q1 = _mm_set_ps1(4.36821166879210612817e3f);
-	static const __m128 GLM_VAR_USED _ps_log_p0 = _mm_set_ps1(-7.89580278884799154124e-1f);
-	static const __m128 GLM_VAR_USED _ps_log_p1 = _mm_set_ps1(1.63866645699558079767e1f);
-	static const __m128 GLM_VAR_USED _ps_log_p2 = _mm_set_ps1(-6.41409952958715622951e1f);
-	static const __m128 GLM_VAR_USED _ps_log_q0 = _mm_set_ps1(-3.56722798256324312549e1f);
-	static const __m128 GLM_VAR_USED _ps_log_q1 = _mm_set_ps1(3.12093766372244180303e2f);
-	static const __m128 GLM_VAR_USED _ps_log_q2 = _mm_set_ps1(-7.69691943550460008604e2f);
-	static const __m128 GLM_VAR_USED _ps_log_c0 = _mm_set_ps1(0.693147180559945f);
-	static const __m128 GLM_VAR_USED _ps_log2_c0 = _mm_set_ps1(1.44269504088896340735992f);
-*/
+	template <precision P>
+	struct compute_mod<float, P, tvec4>
+	{
+		GLM_FUNC_QUALIFIER static tvec4<float, P> call(tvec4<float, P> const & x, tvec4<float, P> const & y)
+		{
+			tvec4<float, P> result(uninitialize);
+			result.data = glm_f32v4_mod(x.data, y.data);
+			return result;
+		}
+	};
 
 }//namespace detail
 }//namespace glm

glm/simd/common.h

@@ -60,29 +60,37 @@ GLM_FUNC_QUALIFIER __m128 glm_f32v4_sgn(__m128 x)
 //round
 GLM_FUNC_QUALIFIER __m128 glm_f32v4_rnd(__m128 x)
 {
-	__m128 const sgn0 = _mm_castsi128_ps(_mm_set1_epi32(0x80000000));
-	__m128 const and0 = _mm_and_ps(sgn0, x);
-	__m128 const or0 = _mm_or_ps(and0, _mm_set_ps1(8388608.0f));
-	__m128 const add0 = _mm_add_ps(x, or0);
-	__m128 const sub0 = _mm_sub_ps(add0, or0);
-	return sub0;
+#	if GLM_ARCH & GLM_ARCH_SSE41_BIT
+		return _mm_round_ps(x, _MM_FROUND_TO_NEAREST_INT);
+#	else
+		__m128 const sgn0 = _mm_castsi128_ps(_mm_set1_epi32(0x80000000));
+		__m128 const and0 = _mm_and_ps(sgn0, x);
+		__m128 const or0 = _mm_or_ps(and0, _mm_set_ps1(8388608.0f));
+		__m128 const add0 = _mm_add_ps(x, or0);
+		__m128 const sub0 = _mm_sub_ps(add0, or0);
+		return sub0;
+#	endif
 }
 
 //floor
 GLM_FUNC_QUALIFIER __m128 glm_f32v4_flr(__m128 x)
 {
-	__m128 const rnd0 = glm_f32v4_rnd(x);
-	__m128 const cmp0 = _mm_cmplt_ps(x, rnd0);
-	__m128 const and0 = _mm_and_ps(cmp0, _mm_set1_ps(1.0f));
-	__m128 const sub0 = _mm_sub_ps(rnd0, and0);
-	return sub0;
+#	if GLM_ARCH & GLM_ARCH_SSE41_BIT
+		return _mm_floor_ps(x);
+#	else
+		__m128 const rnd0 = glm_f32v4_rnd(x);
+		__m128 const cmp0 = _mm_cmplt_ps(x, rnd0);
+		__m128 const and0 = _mm_and_ps(cmp0, _mm_set1_ps(1.0f));
+		__m128 const sub0 = _mm_sub_ps(rnd0, and0);
+		return sub0;
+#	endif
 }
 
 //trunc
-//GLM_FUNC_QUALIFIER __m128 _mm_trc_ps(__m128 v)
-//{
-//	return __m128();
-//}
+GLM_FUNC_QUALIFIER __m128 glm_f32v4_trc(__m128 x)
+{
+	return __m128();
+}
 
 //roundEven
 GLM_FUNC_QUALIFIER __m128 glm_f32v4_rde(__m128 x)
@@ -97,11 +105,15 @@ GLM_FUNC_QUALIFIER __m128 glm_f32v4_rde(__m128 x)
 
 GLM_FUNC_QUALIFIER __m128 glm_f32v4_ceil(__m128 x)
 {
-	__m128 const rnd0 = glm_f32v4_rnd(x);
-	__m128 const cmp0 = _mm_cmpgt_ps(x, rnd0);
-	__m128 const and0 = _mm_and_ps(cmp0, _mm_set1_ps(1.0f));
-	__m128 const add0 = _mm_add_ps(rnd0, and0);
-	return add0;
+#	if GLM_ARCH & GLM_ARCH_SSE41_BIT
+		return _mm_ceil_ps(x);
+#	else
+		__m128 const rnd0 = glm_f32v4_rnd(x);
+		__m128 const cmp0 = _mm_cmpgt_ps(x, rnd0);
+		__m128 const and0 = _mm_and_ps(cmp0, _mm_set1_ps(1.0f));
+		__m128 const add0 = _mm_add_ps(rnd0, and0);
+		return add0;
+#	endif
 }
 
 GLM_FUNC_QUALIFIER __m128 glm_f32v4_frc(__m128 x)