Free cookie consent management tool by TermsFeed Policy Generator

source: branches/HeuristicLab.Problems.GaussianProcessTuning/HeuristicLab.Eigen/Eigen/src/Geometry/arch/Geometry_SSE.h @ 9562

Last change on this file since 9562 was 9562, checked in by gkronber, 11 years ago

#1967 worked on Gaussian process evolution.

File size: 3.8 KB
Line 
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, Aligned>
20{
21  static inline Quaternion<float> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b)
22  {
23    const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0,0,0,0x80000000));
24    Quaternion<float> res;
25    __m128 a = _a.coeffs().template packet<Aligned>(0);
26    __m128 b = _b.coeffs().template packet<Aligned>(0);
27    __m128 flip1 = _mm_xor_ps(_mm_mul_ps(vec4f_swizzle1(a,1,2,0,2),
28                                         vec4f_swizzle1(b,2,0,1,2)),mask);
29    __m128 flip2 = _mm_xor_ps(_mm_mul_ps(vec4f_swizzle1(a,3,3,3,1),
30                                         vec4f_swizzle1(b,0,1,2,1)),mask);
31    pstore(&res.x(),
32              _mm_add_ps(_mm_sub_ps(_mm_mul_ps(a,vec4f_swizzle1(b,3,3,3,3)),
33                                    _mm_mul_ps(vec4f_swizzle1(a,2,0,1,0),
34                                               vec4f_swizzle1(b,1,2,0,0))),
35                         _mm_add_ps(flip1,flip2)));
36    return res;
37  }
38};
39
40template<typename VectorLhs,typename VectorRhs>
41struct cross3_impl<Architecture::SSE,VectorLhs,VectorRhs,float,true>
42{
43  static inline typename plain_matrix_type<VectorLhs>::type
44  run(const VectorLhs& lhs, const VectorRhs& rhs)
45  {
46    __m128 a = lhs.template packet<VectorLhs::Flags&AlignedBit ? Aligned : Unaligned>(0);
47    __m128 b = rhs.template packet<VectorRhs::Flags&AlignedBit ? Aligned : Unaligned>(0);
48    __m128 mul1=_mm_mul_ps(vec4f_swizzle1(a,1,2,0,3),vec4f_swizzle1(b,2,0,1,3));
49    __m128 mul2=_mm_mul_ps(vec4f_swizzle1(a,2,0,1,3),vec4f_swizzle1(b,1,2,0,3));
50    typename plain_matrix_type<VectorLhs>::type res;
51    pstore(&res.x(),_mm_sub_ps(mul1,mul2));
52    return res;
53  }
54};
55
56
57
58
59template<class Derived, class OtherDerived>
60struct quat_product<Architecture::SSE, Derived, OtherDerived, double, Aligned>
61{
62  static inline Quaternion<double> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b)
63  {
64  const Packet2d mask = _mm_castsi128_pd(_mm_set_epi32(0x0,0x0,0x80000000,0x0));
65
66  Quaternion<double> res;
67
68  const double* a = _a.coeffs().data();
69  Packet2d b_xy = _b.coeffs().template packet<Aligned>(0);
70  Packet2d b_zw = _b.coeffs().template packet<Aligned>(2);
71  Packet2d a_xx = pset1<Packet2d>(a[0]);
72  Packet2d a_yy = pset1<Packet2d>(a[1]);
73  Packet2d a_zz = pset1<Packet2d>(a[2]);
74  Packet2d a_ww = pset1<Packet2d>(a[3]);
75
76  // two temporaries:
77  Packet2d t1, t2;
78
79  /*
80   * t1 = ww*xy + yy*zw
81   * t2 = zz*xy - xx*zw
82   * res.xy = t1 +/- swap(t2)
83   */
84  t1 = padd(pmul(a_ww, b_xy), pmul(a_yy, b_zw));
85  t2 = psub(pmul(a_zz, b_xy), pmul(a_xx, b_zw));
86#ifdef EIGEN_VECTORIZE_SSE3
87  EIGEN_UNUSED_VARIABLE(mask)
88  pstore(&res.x(), _mm_addsub_pd(t1, preverse(t2)));
89#else
90  pstore(&res.x(), padd(t1, pxor(mask,preverse(t2))));
91#endif
92 
93  /*
94   * t1 = ww*zw - yy*xy
95   * t2 = zz*zw + xx*xy
96   * res.zw = t1 -/+ swap(t2) = swap( swap(t1) +/- t2)
97   */
98  t1 = psub(pmul(a_ww, b_zw), pmul(a_yy, b_xy));
99  t2 = padd(pmul(a_zz, b_zw), pmul(a_xx, b_xy));
100#ifdef EIGEN_VECTORIZE_SSE3
101  EIGEN_UNUSED_VARIABLE(mask)
102  pstore(&res.z(), preverse(_mm_addsub_pd(preverse(t1), t2)));
103#else
104  pstore(&res.z(), psub(t1, pxor(mask,preverse(t2))));
105#endif
106
107  return res;
108}
109};
110
111} // end namespace internal
112
113} // end namespace Eigen
114
115#endif // EIGEN_GEOMETRY_SSE_H
Note: See TracBrowser for help on using the repository browser.