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
14namespace Eigen {
15
16namespace internal {
17
18template<class Derived, class OtherDerived>
19struct quat_product<Architecture::SSE, Derived, OtherDerived, float, Aligned16>
20{
21  static inline Quaternion<float> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b)
22  {
23    Quaternion<float> res;
24    const __m128 mask = _mm_setr_ps(0.f,0.f,0.f,-0.f);
25    __m128 a = _a.coeffs().template packet<Aligned16>(0);
26    __m128 b = _b.coeffs().template packet<Aligned16>(0);
27    __m128 s1 = _mm_mul_ps(vec4f_swizzle1(a,1,2,0,2),vec4f_swizzle1(b,2,0,1,2));
28    __m128 s2 = _mm_mul_ps(vec4f_swizzle1(a,3,3,3,1),vec4f_swizzle1(b,0,1,2,1));
29    pstore(&res.x(),
30              _mm_add_ps(_mm_sub_ps(_mm_mul_ps(a,vec4f_swizzle1(b,3,3,3,3)),
31                                    _mm_mul_ps(vec4f_swizzle1(a,2,0,1,0),
32                                               vec4f_swizzle1(b,1,2,0,0))),
33                         _mm_xor_ps(mask,_mm_add_ps(s1,s2))));
34
35    return res;
36  }
37};
38
39template<class Derived, int Alignment>
40struct quat_conj<Architecture::SSE, Derived, float, Alignment>
41{
42  static inline Quaternion<float> run(const QuaternionBase<Derived>& q)
43  {
44    Quaternion<float> res;
45    const __m128 mask = _mm_setr_ps(-0.f,-0.f,-0.f,0.f);
46    pstore(&res.x(), _mm_xor_ps(mask, q.coeffs().template packet<Alignment>(0)));
47    return res;
48  }
49};
50
51
52template<typename VectorLhs,typename VectorRhs>
53struct cross3_impl<Architecture::SSE,VectorLhs,VectorRhs,float,true>
54{
55  static inline typename plain_matrix_type<VectorLhs>::type
56  run(const VectorLhs& lhs, const VectorRhs& rhs)
57  {
58    __m128 a = lhs.template packet<traits<VectorLhs>::Alignment>(0);
59    __m128 b = rhs.template packet<traits<VectorRhs>::Alignment>(0);
60    __m128 mul1=_mm_mul_ps(vec4f_swizzle1(a,1,2,0,3),vec4f_swizzle1(b,2,0,1,3));
61    __m128 mul2=_mm_mul_ps(vec4f_swizzle1(a,2,0,1,3),vec4f_swizzle1(b,1,2,0,3));
62    typename plain_matrix_type<VectorLhs>::type res;
63    pstore(&res.x(),_mm_sub_ps(mul1,mul2));
64    return res;
65  }
66};
67
68
69
70
71template<class Derived, class OtherDerived, int Alignment>
72struct quat_product<Architecture::SSE, Derived, OtherDerived, double, Alignment>
73{
74  static inline Quaternion<double> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b)
75  {
76  const Packet2d mask = _mm_castsi128_pd(_mm_set_epi32(0x0,0x0,0x80000000,0x0));
77
78  Quaternion<double> res;
79
80  const double* a = _a.coeffs().data();
81  Packet2d b_xy = _b.coeffs().template packet<Alignment>(0);
82  Packet2d b_zw = _b.coeffs().template packet<Alignment>(2);
83  Packet2d a_xx = pset1<Packet2d>(a[0]);
84  Packet2d a_yy = pset1<Packet2d>(a[1]);
85  Packet2d a_zz = pset1<Packet2d>(a[2]);
86  Packet2d a_ww = pset1<Packet2d>(a[3]);
87
88  // two temporaries:
89  Packet2d t1, t2;
90
91  /*
92   * t1 = ww*xy + yy*zw
93   * t2 = zz*xy - xx*zw
94   * res.xy = t1 +/- swap(t2)
95   */
96  t1 = padd(pmul(a_ww, b_xy), pmul(a_yy, b_zw));
97  t2 = psub(pmul(a_zz, b_xy), pmul(a_xx, b_zw));
98#ifdef EIGEN_VECTORIZE_SSE3
99  EIGEN_UNUSED_VARIABLE(mask)
100  pstore(&res.x(), _mm_addsub_pd(t1, preverse(t2)));
101#else
102  pstore(&res.x(), padd(t1, pxor(mask,preverse(t2))));
103#endif
104
105  /*
106   * t1 = ww*zw - yy*xy
107   * t2 = zz*zw + xx*xy
108   * res.zw = t1 -/+ swap(t2) = swap( swap(t1) +/- t2)
109   */
110  t1 = psub(pmul(a_ww, b_zw), pmul(a_yy, b_xy));
111  t2 = padd(pmul(a_zz, b_zw), pmul(a_xx, b_xy));
112#ifdef EIGEN_VECTORIZE_SSE3
113  EIGEN_UNUSED_VARIABLE(mask)
114  pstore(&res.z(), preverse(_mm_addsub_pd(preverse(t1), t2)));
115#else
116  pstore(&res.z(), psub(t1, pxor(mask,preverse(t2))));
117#endif
118
119  return res;
120}
121};
122
123template<class Derived, int Alignment>
124struct quat_conj<Architecture::SSE, Derived, double, Alignment>
125{
126  static inline Quaternion<double> run(const QuaternionBase<Derived>& q)
127  {
128    Quaternion<double> res;
129    const __m128d mask0 = _mm_setr_pd(-0.,-0.);
130    const __m128d mask2 = _mm_setr_pd(-0.,0.);
131    pstore(&res.x(), _mm_xor_pd(mask0, q.coeffs().template packet<Alignment>(0)));
132    pstore(&res.z(), _mm_xor_pd(mask2, q.coeffs().template packet<Alignment>(2)));
133    return res;
134  }
135};
136
137} // end namespace internal
138
139} // end namespace Eigen
140
141#endif // EIGEN_GEOMETRY_SSE_H
142