Geometry_SIMD.h
Go to the documentation of this file.
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_SIMD_H
12 #define EIGEN_GEOMETRY_SIMD_H
13 
14 #include "../InternalHeaderCheck.h"
15 
16 namespace Eigen {
17 
18 namespace internal {
19 
20 template<class Derived, class OtherDerived>
21 struct quat_product<Architecture::Target, Derived, OtherDerived, float>
22 {
23  enum {
24  AAlignment = traits<Derived>::Alignment,
25  BAlignment = traits<OtherDerived>::Alignment,
26  ResAlignment = traits<Quaternion<float> >::Alignment
27  };
28  static inline Quaternion<float> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b)
29  {
30  evaluator<typename Derived::Coefficients> ae(_a.coeffs());
31  evaluator<typename OtherDerived::Coefficients> be(_b.coeffs());
32  Quaternion<float> res;
33  const float neg_zero = numext::bit_cast<float>(0x80000000u);
34  const float arr[4] = {0.f, 0.f, 0.f, neg_zero};
35  const Packet4f mask = ploadu<Packet4f>(arr);
36  Packet4f a = ae.template packet<AAlignment,Packet4f>(0);
37  Packet4f b = be.template packet<BAlignment,Packet4f>(0);
38  Packet4f s1 = pmul(vec4f_swizzle1(a,1,2,0,2),vec4f_swizzle1(b,2,0,1,2));
39  Packet4f s2 = pmul(vec4f_swizzle1(a,3,3,3,1),vec4f_swizzle1(b,0,1,2,1));
40  pstoret<float,Packet4f,ResAlignment>(
41  &res.x(),
42  padd(psub(pmul(a,vec4f_swizzle1(b,3,3,3,3)),
43  pmul(vec4f_swizzle1(a,2,0,1,0),
44  vec4f_swizzle1(b,1,2,0,0))),
45  pxor(mask,padd(s1,s2))));
46 
47  return res;
48  }
49 };
50 
51 template<class Derived>
52 struct quat_conj<Architecture::Target, Derived, float>
53 {
54  enum {
55  ResAlignment = traits<Quaternion<float> >::Alignment
56  };
57  static inline Quaternion<float> run(const QuaternionBase<Derived>& q)
58  {
59  evaluator<typename Derived::Coefficients> qe(q.coeffs());
60  Quaternion<float> res;
61  const float neg_zero = numext::bit_cast<float>(0x80000000u);
62  const float arr[4] = {neg_zero, neg_zero, neg_zero,0.f};
63  const Packet4f mask = ploadu<Packet4f>(arr);
64  pstoret<float,Packet4f,ResAlignment>(&res.x(), pxor(mask, qe.template packet<traits<Derived>::Alignment,Packet4f>(0)));
65  return res;
66  }
67 };
68 
69 
70 template<typename VectorLhs,typename VectorRhs>
71 struct cross3_impl<Architecture::Target,VectorLhs,VectorRhs,float,true>
72 {
73  enum {
74  ResAlignment = traits<typename plain_matrix_type<VectorLhs>::type>::Alignment
75  };
76  static inline typename plain_matrix_type<VectorLhs>::type
77  run(const VectorLhs& lhs, const VectorRhs& rhs)
78  {
79  evaluator<VectorLhs> lhs_eval(lhs);
80  evaluator<VectorRhs> rhs_eval(rhs);
81  Packet4f a = lhs_eval.template packet<traits<VectorLhs>::Alignment,Packet4f>(0);
82  Packet4f b = rhs_eval.template packet<traits<VectorRhs>::Alignment,Packet4f>(0);
83  Packet4f mul1 = pmul(vec4f_swizzle1(a,1,2,0,3),vec4f_swizzle1(b,2,0,1,3));
84  Packet4f mul2 = pmul(vec4f_swizzle1(a,2,0,1,3),vec4f_swizzle1(b,1,2,0,3));
85  typename plain_matrix_type<VectorLhs>::type res;
86  pstoret<float,Packet4f,ResAlignment>(&res.x(),psub(mul1,mul2));
87  return res;
88  }
89 };
90 
91 
92 
93 #if (defined EIGEN_VECTORIZE_SSE) || (EIGEN_ARCH_ARM64)
94 
95 template<class Derived, class OtherDerived>
96 struct quat_product<Architecture::Target, Derived, OtherDerived, double>
97 {
98  enum {
99  BAlignment = traits<OtherDerived>::Alignment,
100  ResAlignment = traits<Quaternion<double> >::Alignment
101  };
102 
103  static inline Quaternion<double> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b)
104  {
105  Quaternion<double> res;
106 
107  evaluator<typename Derived::Coefficients> ae(_a.coeffs());
108  evaluator<typename OtherDerived::Coefficients> be(_b.coeffs());
109 
110  const double* a = _a.coeffs().data();
111  Packet2d b_xy = be.template packet<BAlignment,Packet2d>(0);
112  Packet2d b_zw = be.template packet<BAlignment,Packet2d>(2);
113  Packet2d a_xx = pset1<Packet2d>(a[0]);
114  Packet2d a_yy = pset1<Packet2d>(a[1]);
115  Packet2d a_zz = pset1<Packet2d>(a[2]);
116  Packet2d a_ww = pset1<Packet2d>(a[3]);
117 
118  // two temporaries:
119  Packet2d t1, t2;
120 
121  /*
122  * t1 = ww*xy + yy*zw
123  * t2 = zz*xy - xx*zw
124  * res.xy = t1 +/- swap(t2)
125  */
126  t1 = padd(pmul(a_ww, b_xy), pmul(a_yy, b_zw));
127  t2 = psub(pmul(a_zz, b_xy), pmul(a_xx, b_zw));
128  pstoret<double,Packet2d,ResAlignment>(&res.x(), paddsub(t1, preverse(t2)));
129 
130  /*
131  * t1 = ww*zw - yy*xy
132  * t2 = zz*zw + xx*xy
133  * res.zw = t1 -/+ swap(t2) = swap( swap(t1) +/- t2)
134  */
135  t1 = psub(pmul(a_ww, b_zw), pmul(a_yy, b_xy));
136  t2 = padd(pmul(a_zz, b_zw), pmul(a_xx, b_xy));
137  pstoret<double,Packet2d,ResAlignment>(&res.z(), preverse(paddsub(preverse(t1), t2)));
138 
139  return res;
140 }
141 };
142 
143 template<class Derived>
144 struct quat_conj<Architecture::Target, Derived, double>
145 {
146  enum {
147  ResAlignment = traits<Quaternion<double> >::Alignment
148  };
149  static inline Quaternion<double> run(const QuaternionBase<Derived>& q)
150  {
151  evaluator<typename Derived::Coefficients> qe(q.coeffs());
152  Quaternion<double> res;
153  const double neg_zero = numext::bit_cast<double>(0x8000000000000000ull);
154  const double arr1[2] = {neg_zero, neg_zero};
155  const double arr2[2] = {neg_zero, 0.0};
156  const Packet2d mask0 = ploadu<Packet2d>(arr1);
157  const Packet2d mask2 = ploadu<Packet2d>(arr2);
158  pstoret<double,Packet2d,ResAlignment>(&res.x(), pxor(mask0, qe.template packet<traits<Derived>::Alignment,Packet2d>(0)));
159  pstoret<double,Packet2d,ResAlignment>(&res.z(), pxor(mask2, qe.template packet<traits<Derived>::Alignment,Packet2d>(2)));
160  return res;
161  }
162 };
163 
164 #endif // end EIGEN_VECTORIZE_SSE_OR_EIGEN_ARCH_ARM64
165 
166 } // end namespace internal
167 
168 } // end namespace Eigen
169 
170 #endif // EIGEN_GEOMETRY_SIMD_H
Array< int, 3, 1 > b
cout<< "Here is the matrix m:"<< endl<< m<< endl;Matrix< ptrdiff_t, 3, 1 > res
Packet padd(const Packet &a, const Packet &b)
Packet pmul(const Packet &a, const Packet &b)
Packet4f ploadu< Packet4f >(const float *from)
Packet psub(const Packet &a, const Packet &b)
Packet2d pset1< Packet2d >(const double &from)
Packet2d ploadu< Packet2d >(const double *from)
Packet8h pxor(const Packet8h &a, const Packet8h &b)
Packet paddsub(const Packet &a, const Packet &b)
Packet2cf preverse(const Packet2cf &a)
__vector float Packet4f
Packet4f vec4f_swizzle1(const Packet4f &a, int p, int q, int r, int s)
: InteropHeaders
Definition: Core:139