Another set of UI messages fixes and tweaks! No functional changes.
[blender.git] / extern / Eigen3 / Eigen / src / Geometry / arch / Geometry_SSE.h
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 // Eigen is free software; you can redistribute it and/or
8 // modify it under the terms of the GNU Lesser General Public
9 // License as published by the Free Software Foundation; either
10 // version 3 of the License, or (at your option) any later version.
11 //
12 // Alternatively, you can redistribute it and/or
13 // modify it under the terms of the GNU General Public License as
14 // published by the Free Software Foundation; either version 2 of
15 // the License, or (at your option) any later version.
16 //
17 // Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
18 // WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
19 // FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
20 // GNU General Public License for more details.
21 //
22 // You should have received a copy of the GNU Lesser General Public
23 // License and a copy of the GNU General Public License along with
24 // Eigen. If not, see <http://www.gnu.org/licenses/>.
25
26 #ifndef EIGEN_GEOMETRY_SSE_H
27 #define EIGEN_GEOMETRY_SSE_H
28
29 namespace internal {
30
31 template<class Derived, class OtherDerived>
32 struct quat_product<Architecture::SSE, Derived, OtherDerived, float, Aligned>
33 {
34   inline static Quaternion<float> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b)
35   {
36     const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0,0,0,0x80000000));
37     Quaternion<float> res;
38     __m128 a = _a.coeffs().template packet<Aligned>(0);
39     __m128 b = _b.coeffs().template packet<Aligned>(0);
40     __m128 flip1 = _mm_xor_ps(_mm_mul_ps(vec4f_swizzle1(a,1,2,0,2),
41                                          vec4f_swizzle1(b,2,0,1,2)),mask);
42     __m128 flip2 = _mm_xor_ps(_mm_mul_ps(vec4f_swizzle1(a,3,3,3,1),
43                                          vec4f_swizzle1(b,0,1,2,1)),mask);
44     pstore(&res.x(),
45               _mm_add_ps(_mm_sub_ps(_mm_mul_ps(a,vec4f_swizzle1(b,3,3,3,3)),
46                                     _mm_mul_ps(vec4f_swizzle1(a,2,0,1,0),
47                                                vec4f_swizzle1(b,1,2,0,0))),
48                          _mm_add_ps(flip1,flip2)));
49     return res;
50   }
51 };
52
53 template<typename VectorLhs,typename VectorRhs>
54 struct cross3_impl<Architecture::SSE,VectorLhs,VectorRhs,float,true>
55 {
56   inline static typename plain_matrix_type<VectorLhs>::type
57   run(const VectorLhs& lhs, const VectorRhs& rhs)
58   {
59     __m128 a = lhs.template packet<VectorLhs::Flags&AlignedBit ? Aligned : Unaligned>(0);
60     __m128 b = rhs.template packet<VectorRhs::Flags&AlignedBit ? Aligned : Unaligned>(0);
61     __m128 mul1=_mm_mul_ps(vec4f_swizzle1(a,1,2,0,3),vec4f_swizzle1(b,2,0,1,3));
62     __m128 mul2=_mm_mul_ps(vec4f_swizzle1(a,2,0,1,3),vec4f_swizzle1(b,1,2,0,3));
63     typename plain_matrix_type<VectorLhs>::type res;
64     pstore(&res.x(),_mm_sub_ps(mul1,mul2));
65     return res;
66   }
67 };
68
69
70
71
72 template<class Derived, class OtherDerived>
73 struct quat_product<Architecture::SSE, Derived, OtherDerived, double, Aligned>
74 {
75   inline static Quaternion<double> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b)
76   {
77   const Packet2d mask = _mm_castsi128_pd(_mm_set_epi32(0x0,0x0,0x80000000,0x0));
78
79   Quaternion<double> res;
80
81   const double* a = _a.coeffs().data();
82   Packet2d b_xy = _b.coeffs().template packet<Aligned>(0);
83   Packet2d b_zw = _b.coeffs().template packet<Aligned>(2);
84   Packet2d a_xx = pset1<Packet2d>(a[0]);
85   Packet2d a_yy = pset1<Packet2d>(a[1]);
86   Packet2d a_zz = pset1<Packet2d>(a[2]);
87   Packet2d a_ww = pset1<Packet2d>(a[3]);
88
89   // two temporaries:
90   Packet2d t1, t2;
91
92   /*
93    * t1 = ww*xy + yy*zw
94    * t2 = zz*xy - xx*zw
95    * res.xy = t1 +/- swap(t2)
96    */
97   t1 = padd(pmul(a_ww, b_xy), pmul(a_yy, b_zw));
98   t2 = psub(pmul(a_zz, b_xy), pmul(a_xx, b_zw));
99 #ifdef __SSE3__
100   EIGEN_UNUSED_VARIABLE(mask)
101   pstore(&res.x(), _mm_addsub_pd(t1, preverse(t2)));
102 #else
103   pstore(&res.x(), padd(t1, pxor(mask,preverse(t2))));
104 #endif
105   
106   /*
107    * t1 = ww*zw - yy*xy
108    * t2 = zz*zw + xx*xy
109    * res.zw = t1 -/+ swap(t2) = swap( swap(t1) +/- t2)
110    */
111   t1 = psub(pmul(a_ww, b_zw), pmul(a_yy, b_xy));
112   t2 = padd(pmul(a_zz, b_zw), pmul(a_xx, b_xy));
113 #ifdef __SSE3__
114   EIGEN_UNUSED_VARIABLE(mask)
115   pstore(&res.z(), preverse(_mm_addsub_pd(preverse(t1), t2)));
116 #else
117   pstore(&res.z(), psub(t1, pxor(mask,preverse(t2))));
118 #endif
119
120   return res;
121 }
122 };
123
124 } // end namespace internal
125
126 #endif // EIGEN_GEOMETRY_SSE_H