Geometry_SSE.h
Go to the documentation of this file.
00001 // This file is part of Eigen, a lightweight C++ template library
00002 // for linear algebra.
00003 //
00004 // Copyright (C) 2009 Rohit Garg <rpg.314@gmail.com>
00005 // Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
00006 //
00007 // This Source Code Form is subject to the terms of the Mozilla
00008 // Public License v. 2.0. If a copy of the MPL was not distributed
00009 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
00010 
00011 #ifndef EIGEN_GEOMETRY_SSE_H
00012 #define EIGEN_GEOMETRY_SSE_H
00013 
00014 namespace Eigen { 
00015 
00016 namespace internal {
00017 
00018 template<class Derived, class OtherDerived>
00019 struct quat_product<Architecture::SSE, Derived, OtherDerived, float, Aligned>
00020 {
00021   static inline Quaternion<float> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b)
00022   {
00023     const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0,0,0,0x80000000));
00024     Quaternion<float> res;
00025     __m128 a = _a.coeffs().template packet<Aligned>(0);
00026     __m128 b = _b.coeffs().template packet<Aligned>(0);
00027     __m128 flip1 = _mm_xor_ps(_mm_mul_ps(vec4f_swizzle1(a,1,2,0,2),
00028                                          vec4f_swizzle1(b,2,0,1,2)),mask);
00029     __m128 flip2 = _mm_xor_ps(_mm_mul_ps(vec4f_swizzle1(a,3,3,3,1),
00030                                          vec4f_swizzle1(b,0,1,2,1)),mask);
00031     pstore(&res.x(),
00032               _mm_add_ps(_mm_sub_ps(_mm_mul_ps(a,vec4f_swizzle1(b,3,3,3,3)),
00033                                     _mm_mul_ps(vec4f_swizzle1(a,2,0,1,0),
00034                                                vec4f_swizzle1(b,1,2,0,0))),
00035                          _mm_add_ps(flip1,flip2)));
00036     return res;
00037   }
00038 };
00039 
00040 template<typename VectorLhs,typename VectorRhs>
00041 struct cross3_impl<Architecture::SSE,VectorLhs,VectorRhs,float,true>
00042 {
00043   static inline typename plain_matrix_type<VectorLhs>::type
00044   run(const VectorLhs& lhs, const VectorRhs& rhs)
00045   {
00046     __m128 a = lhs.template packet<VectorLhs::Flags&AlignedBit ? Aligned : Unaligned>(0);
00047     __m128 b = rhs.template packet<VectorRhs::Flags&AlignedBit ? Aligned : Unaligned>(0);
00048     __m128 mul1=_mm_mul_ps(vec4f_swizzle1(a,1,2,0,3),vec4f_swizzle1(b,2,0,1,3));
00049     __m128 mul2=_mm_mul_ps(vec4f_swizzle1(a,2,0,1,3),vec4f_swizzle1(b,1,2,0,3));
00050     typename plain_matrix_type<VectorLhs>::type res;
00051     pstore(&res.x(),_mm_sub_ps(mul1,mul2));
00052     return res;
00053   }
00054 };
00055 
00056 
00057 
00058 
00059 template<class Derived, class OtherDerived>
00060 struct quat_product<Architecture::SSE, Derived, OtherDerived, double, Aligned>
00061 {
00062   static inline Quaternion<double> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b)
00063   {
00064   const Packet2d mask = _mm_castsi128_pd(_mm_set_epi32(0x0,0x0,0x80000000,0x0));
00065 
00066   Quaternion<double> res;
00067 
00068   const double* a = _a.coeffs().data();
00069   Packet2d b_xy = _b.coeffs().template packet<Aligned>(0);
00070   Packet2d b_zw = _b.coeffs().template packet<Aligned>(2);
00071   Packet2d a_xx = pset1<Packet2d>(a[0]);
00072   Packet2d a_yy = pset1<Packet2d>(a[1]);
00073   Packet2d a_zz = pset1<Packet2d>(a[2]);
00074   Packet2d a_ww = pset1<Packet2d>(a[3]);
00075 
00076   // two temporaries:
00077   Packet2d t1, t2;
00078 
00079   /*
00080    * t1 = ww*xy + yy*zw
00081    * t2 = zz*xy - xx*zw
00082    * res.xy = t1 +/- swap(t2)
00083    */
00084   t1 = padd(pmul(a_ww, b_xy), pmul(a_yy, b_zw));
00085   t2 = psub(pmul(a_zz, b_xy), pmul(a_xx, b_zw));
00086 #ifdef EIGEN_VECTORIZE_SSE3
00087   EIGEN_UNUSED_VARIABLE(mask)
00088   pstore(&res.x(), _mm_addsub_pd(t1, preverse(t2)));
00089 #else
00090   pstore(&res.x(), padd(t1, pxor(mask,preverse(t2))));
00091 #endif
00092   
00093   /*
00094    * t1 = ww*zw - yy*xy
00095    * t2 = zz*zw + xx*xy
00096    * res.zw = t1 -/+ swap(t2) = swap( swap(t1) +/- t2)
00097    */
00098   t1 = psub(pmul(a_ww, b_zw), pmul(a_yy, b_xy));
00099   t2 = padd(pmul(a_zz, b_zw), pmul(a_xx, b_xy));
00100 #ifdef EIGEN_VECTORIZE_SSE3
00101   EIGEN_UNUSED_VARIABLE(mask)
00102   pstore(&res.z(), preverse(_mm_addsub_pd(preverse(t1), t2)));
00103 #else
00104   pstore(&res.z(), psub(t1, pxor(mask,preverse(t2))));
00105 #endif
00106 
00107   return res;
00108 }
00109 };
00110 
00111 } // end namespace internal
00112 
00113 } // end namespace Eigen
00114 
00115 #endif // EIGEN_GEOMETRY_SSE_H


win_eigen
Author(s): Daniel Stonier
autogenerated on Wed Sep 16 2015 07:10:40