1 // This file is part of Eigen, a lightweight C++ template library 2 // for linear algebra. 3 // 4 // Copyright (C) 2009 Rohit Garg <rpg.314@gmail.com> 5 // Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr> 6 // 7 // This Source Code Form is subject to the terms of the Mozilla 8 // Public License v. 2.0. If a copy of the MPL was not distributed 9 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 10 11 #ifndef EIGEN_GEOMETRY_SSE_H 12 #define EIGEN_GEOMETRY_SSE_H 13 14 namespace Eigen { 15 16 namespace internal { 17 18 template<class Derived, class OtherDerived> 19 struct quat_product<Architecture::SSE, Derived, OtherDerived, float> 20 { 21 enum { 22 AAlignment = traits<Derived>::Alignment, 23 BAlignment = traits<OtherDerived>::Alignment, 24 ResAlignment = traits<Quaternion<float> >::Alignment 25 }; 26 static inline Quaternion<float> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b) 27 { 28 Quaternion<float> res; 29 const __m128 mask = _mm_setr_ps(0.f,0.f,0.f,-0.f); 30 __m128 a = _a.coeffs().template packet<AAlignment>(0); 31 __m128 b = _b.coeffs().template packet<BAlignment>(0); 32 __m128 s1 = _mm_mul_ps(vec4f_swizzle1(a,1,2,0,2),vec4f_swizzle1(b,2,0,1,2)); 33 __m128 s2 = _mm_mul_ps(vec4f_swizzle1(a,3,3,3,1),vec4f_swizzle1(b,0,1,2,1)); 34 pstoret<float,Packet4f,ResAlignment>( 35 &res.x(), 36 _mm_add_ps(_mm_sub_ps(_mm_mul_ps(a,vec4f_swizzle1(b,3,3,3,3)), 37 _mm_mul_ps(vec4f_swizzle1(a,2,0,1,0), 38 vec4f_swizzle1(b,1,2,0,0))), 39 _mm_xor_ps(mask,_mm_add_ps(s1,s2)))); 40 41 return res; 42 } 43 }; 44 45 template<class Derived> 46 struct quat_conj<Architecture::SSE, Derived, float> 47 { 48 enum { 49 ResAlignment = traits<Quaternion<float> >::Alignment 50 }; 51 static inline Quaternion<float> run(const QuaternionBase<Derived>& q) 52 { 53 Quaternion<float> res; 54 const __m128 mask = _mm_setr_ps(-0.f,-0.f,-0.f,0.f); 55 pstoret<float,Packet4f,ResAlignment>(&res.x(), _mm_xor_ps(mask, q.coeffs().template packet<traits<Derived>::Alignment>(0))); 56 return res; 57 } 58 }; 59 60 61 template<typename VectorLhs,typename VectorRhs> 62 struct cross3_impl<Architecture::SSE,VectorLhs,VectorRhs,float,true> 63 { 64 enum { 65 ResAlignment = traits<typename plain_matrix_type<VectorLhs>::type>::Alignment 66 }; 67 static inline typename plain_matrix_type<VectorLhs>::type 68 run(const VectorLhs& lhs, const VectorRhs& rhs) 69 { 70 __m128 a = lhs.template packet<traits<VectorLhs>::Alignment>(0); 71 __m128 b = rhs.template packet<traits<VectorRhs>::Alignment>(0); 72 __m128 mul1=_mm_mul_ps(vec4f_swizzle1(a,1,2,0,3),vec4f_swizzle1(b,2,0,1,3)); 73 __m128 mul2=_mm_mul_ps(vec4f_swizzle1(a,2,0,1,3),vec4f_swizzle1(b,1,2,0,3)); 74 typename plain_matrix_type<VectorLhs>::type res; 75 pstoret<float,Packet4f,ResAlignment>(&res.x(),_mm_sub_ps(mul1,mul2)); 76 return res; 77 } 78 }; 79 80 81 82 83 template<class Derived, class OtherDerived> 84 struct quat_product<Architecture::SSE, Derived, OtherDerived, double> 85 { 86 enum { 87 BAlignment = traits<OtherDerived>::Alignment, 88 ResAlignment = traits<Quaternion<double> >::Alignment 89 }; 90 91 static inline Quaternion<double> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b) 92 { 93 const Packet2d mask = _mm_castsi128_pd(_mm_set_epi32(0x0,0x0,0x80000000,0x0)); 94 95 Quaternion<double> res; 96 97 const double* a = _a.coeffs().data(); 98 Packet2d b_xy = _b.coeffs().template packet<BAlignment>(0); 99 Packet2d b_zw = _b.coeffs().template packet<BAlignment>(2); 100 Packet2d a_xx = pset1<Packet2d>(a[0]); 101 Packet2d a_yy = pset1<Packet2d>(a[1]); 102 Packet2d a_zz = pset1<Packet2d>(a[2]); 103 Packet2d a_ww = pset1<Packet2d>(a[3]); 104 105 // two temporaries: 106 Packet2d t1, t2; 107 108 /* 109 * t1 = ww*xy + yy*zw 110 * t2 = zz*xy - xx*zw 111 * res.xy = t1 +/- swap(t2) 112 */ 113 t1 = padd(pmul(a_ww, b_xy), pmul(a_yy, b_zw)); 114 t2 = psub(pmul(a_zz, b_xy), pmul(a_xx, b_zw)); 115 #ifdef EIGEN_VECTORIZE_SSE3 116 EIGEN_UNUSED_VARIABLE(mask) 117 pstoret<double,Packet2d,ResAlignment>(&res.x(), _mm_addsub_pd(t1, preverse(t2))); 118 #else 119 pstoret<double,Packet2d,ResAlignment>(&res.x(), padd(t1, pxor(mask,preverse(t2)))); 120 #endif 121 122 /* 123 * t1 = ww*zw - yy*xy 124 * t2 = zz*zw + xx*xy 125 * res.zw = t1 -/+ swap(t2) = swap( swap(t1) +/- t2) 126 */ 127 t1 = psub(pmul(a_ww, b_zw), pmul(a_yy, b_xy)); 128 t2 = padd(pmul(a_zz, b_zw), pmul(a_xx, b_xy)); 129 #ifdef EIGEN_VECTORIZE_SSE3 130 EIGEN_UNUSED_VARIABLE(mask) 131 pstoret<double,Packet2d,ResAlignment>(&res.z(), preverse(_mm_addsub_pd(preverse(t1), t2))); 132 #else 133 pstoret<double,Packet2d,ResAlignment>(&res.z(), psub(t1, pxor(mask,preverse(t2)))); 134 #endif 135 136 return res; 137 } 138 }; 139 140 template<class Derived> 141 struct quat_conj<Architecture::SSE, Derived, double> 142 { 143 enum { 144 ResAlignment = traits<Quaternion<double> >::Alignment 145 }; 146 static inline Quaternion<double> run(const QuaternionBase<Derived>& q) 147 { 148 Quaternion<double> res; 149 const __m128d mask0 = _mm_setr_pd(-0.,-0.); 150 const __m128d mask2 = _mm_setr_pd(-0.,0.); 151 pstoret<double,Packet2d,ResAlignment>(&res.x(), _mm_xor_pd(mask0, q.coeffs().template packet<traits<Derived>::Alignment>(0))); 152 pstoret<double,Packet2d,ResAlignment>(&res.z(), _mm_xor_pd(mask2, q.coeffs().template packet<traits<Derived>::Alignment>(2))); 153 return res; 154 } 155 }; 156 157 } // end namespace internal 158 159 } // end namespace Eigen 160 161 #endif // EIGEN_GEOMETRY_SSE_H 162