win_eigen: GenericPacketMath.h Source File

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) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
00005 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
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_GENERIC_PACKET_MATH_H
00012 #define EIGEN_GENERIC_PACKET_MATH_H
00013 
00014 namespace Eigen {
00015 
00016 namespace internal {
00017 
00026 #ifndef EIGEN_DEBUG_ALIGNED_LOAD
00027 #define EIGEN_DEBUG_ALIGNED_LOAD
00028 #endif
00029 
00030 #ifndef EIGEN_DEBUG_UNALIGNED_LOAD
00031 #define EIGEN_DEBUG_UNALIGNED_LOAD
00032 #endif
00033 
00034 #ifndef EIGEN_DEBUG_ALIGNED_STORE
00035 #define EIGEN_DEBUG_ALIGNED_STORE
00036 #endif
00037 
00038 #ifndef EIGEN_DEBUG_UNALIGNED_STORE
00039 #define EIGEN_DEBUG_UNALIGNED_STORE
00040 #endif
00041 
00042 struct default_packet_traits
00043 {
00044   enum {
00045     HasAdd    = 1,
00046     HasSub    = 1,
00047     HasMul    = 1,
00048     HasNegate = 1,
00049     HasAbs    = 1,
00050     HasAbs2   = 1,
00051     HasMin    = 1,
00052     HasMax    = 1,
00053     HasConj   = 1,
00054     HasSetLinear = 1,
00055 
00056     HasDiv    = 0,
00057     HasSqrt   = 0,
00058     HasExp    = 0,
00059     HasLog    = 0,
00060     HasPow    = 0,
00061 
00062     HasSin    = 0,
00063     HasCos    = 0,
00064     HasTan    = 0,
00065     HasASin   = 0,
00066     HasACos   = 0,
00067     HasATan   = 0
00068   };
00069 };
00070 
00071 template<typename T> struct packet_traits : default_packet_traits
00072 {
00073   typedef T type;
00074   enum {
00075     Vectorizable = 0,
00076     size = 1,
00077     AlignedOnScalar = 0
00078   };
00079   enum {
00080     HasAdd    = 0,
00081     HasSub    = 0,
00082     HasMul    = 0,
00083     HasNegate = 0,
00084     HasAbs    = 0,
00085     HasAbs2   = 0,
00086     HasMin    = 0,
00087     HasMax    = 0,
00088     HasConj   = 0,
00089     HasSetLinear = 0
00090   };
00091 };
00092 
00094 template<typename Packet> inline Packet
00095 padd(const Packet& a,
00096         const Packet& b) { return a+b; }
00097 
00099 template<typename Packet> inline Packet
00100 psub(const Packet& a,
00101         const Packet& b) { return a-b; }
00102 
00104 template<typename Packet> inline Packet
00105 pnegate(const Packet& a) { return -a; }
00106 
00108 template<typename Packet> inline Packet
00109 pconj(const Packet& a) { return conj(a); }
00110 
00112 template<typename Packet> inline Packet
00113 pmul(const Packet& a,
00114         const Packet& b) { return a*b; }
00115 
00117 template<typename Packet> inline Packet
00118 pdiv(const Packet& a,
00119         const Packet& b) { return a/b; }
00120 
00122 template<typename Packet> inline Packet
00123 pmin(const Packet& a,
00124         const Packet& b) { using std::min; return (min)(a, b); }
00125 
00127 template<typename Packet> inline Packet
00128 pmax(const Packet& a,
00129         const Packet& b) { using std::max; return (max)(a, b); }
00130 
00132 template<typename Packet> inline Packet
00133 pabs(const Packet& a) { return abs(a); }
00134 
00136 template<typename Packet> inline Packet
00137 pand(const Packet& a, const Packet& b) { return a & b; }
00138 
00140 template<typename Packet> inline Packet
00141 por(const Packet& a, const Packet& b) { return a | b; }
00142 
00144 template<typename Packet> inline Packet
00145 pxor(const Packet& a, const Packet& b) { return a ^ b; }
00146 
00148 template<typename Packet> inline Packet
00149 pandnot(const Packet& a, const Packet& b) { return a & (!b); }
00150 
00152 template<typename Packet> inline Packet
00153 pload(const typename unpacket_traits<Packet>::type* from) { return *from; }
00154 
00156 template<typename Packet> inline Packet
00157 ploadu(const typename unpacket_traits<Packet>::type* from) { return *from; }
00158 
00160 template<typename Packet> inline Packet
00161 ploaddup(const typename unpacket_traits<Packet>::type* from) { return *from; }
00162 
00164 template<typename Packet> inline Packet
00165 pset1(const typename unpacket_traits<Packet>::type& a) { return a; }
00166 
00168 template<typename Scalar> inline typename packet_traits<Scalar>::type
00169 plset(const Scalar& a) { return a; }
00170 
00172 template<typename Scalar, typename Packet> inline void pstore(Scalar* to, const Packet& from)
00173 { (*to) = from; }
00174 
00176 template<typename Scalar, typename Packet> inline void pstoreu(Scalar* to, const Packet& from)
00177 { (*to) = from; }
00178 
00180 template<typename Scalar> inline void prefetch(const Scalar* addr)
00181 {
00182 #if !defined(_MSC_VER)
00183 __builtin_prefetch(addr);
00184 #endif
00185 }
00186 
00188 template<typename Packet> inline typename unpacket_traits<Packet>::type pfirst(const Packet& a)
00189 { return a; }
00190 
00192 template<typename Packet> inline Packet
00193 preduxp(const Packet* vecs) { return vecs[0]; }
00194 
00196 template<typename Packet> inline typename unpacket_traits<Packet>::type predux(const Packet& a)
00197 { return a; }
00198 
00200 template<typename Packet> inline typename unpacket_traits<Packet>::type predux_mul(const Packet& a)
00201 { return a; }
00202 
00204 template<typename Packet> inline typename unpacket_traits<Packet>::type predux_min(const Packet& a)
00205 { return a; }
00206 
00208 template<typename Packet> inline typename unpacket_traits<Packet>::type predux_max(const Packet& a)
00209 { return a; }
00210 
00212 template<typename Packet> inline Packet preverse(const Packet& a)
00213 { return a; }
00214 
00215 
00217 template<typename Packet> inline Packet pcplxflip(const Packet& a)
00218 { return Packet(imag(a),real(a)); }
00219 
00220 /**************************
00221 * Special math functions
00222 ***************************/
00223 
00225 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
00226 Packet psin(const Packet& a) { return sin(a); }
00227 
00229 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
00230 Packet pcos(const Packet& a) { return cos(a); }
00231 
00233 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
00234 Packet ptan(const Packet& a) { return tan(a); }
00235 
00237 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
00238 Packet pasin(const Packet& a) { return asin(a); }
00239 
00241 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
00242 Packet pacos(const Packet& a) { return acos(a); }
00243 
00245 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
00246 Packet pexp(const Packet& a) { return exp(a); }
00247 
00249 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
00250 Packet plog(const Packet& a) { return log(a); }
00251 
00253 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
00254 Packet psqrt(const Packet& a) { return sqrt(a); }
00255 
00256 /***************************************************************************
00257 * The following functions might not have to be overwritten for vectorized types
00258 ***************************************************************************/
00259 
00261 // NOTE: this function must really be templated on the packet type (think about different packet types for the same scalar type)
00262 template<typename Packet>
00263 inline void pstore1(typename unpacket_traits<Packet>::type* to, const typename unpacket_traits<Packet>::type& a)
00264 {
00265   pstore(to, pset1<Packet>(a));
00266 }
00267 
00269 template<typename Packet> inline Packet
00270 pmadd(const Packet&  a,
00271          const Packet&  b,
00272          const Packet&  c)
00273 { return padd(pmul(a, b),c); }
00274 
00277 template<typename Packet, int LoadMode>
00278 inline Packet ploadt(const typename unpacket_traits<Packet>::type* from)
00279 {
00280   if(LoadMode == Aligned)
00281     return pload<Packet>(from);
00282   else
00283     return ploadu<Packet>(from);
00284 }
00285 
00288 template<typename Scalar, typename Packet, int LoadMode>
00289 inline void pstoret(Scalar* to, const Packet& from)
00290 {
00291   if(LoadMode == Aligned)
00292     pstore(to, from);
00293   else
00294     pstoreu(to, from);
00295 }
00296 
00298 template<int Offset,typename PacketType>
00299 struct palign_impl
00300 {
00301   // by default data are aligned, so there is nothing to be done :)
00302   static inline void run(PacketType&, const PacketType&) {}
00303 };
00304 
00307 template<int Offset,typename PacketType>
00308 inline void palign(PacketType& first, const PacketType& second)
00309 {
00310   palign_impl<Offset,PacketType>::run(first,second);
00311 }
00312 
00313 /***************************************************************************
00314 * Fast complex products (GCC generates a function call which is very slow)
00315 ***************************************************************************/
00316 
00317 template<> inline std::complex<float> pmul(const std::complex<float>& a, const std::complex<float>& b)
00318 { return std::complex<float>(real(a)*real(b) - imag(a)*imag(b), imag(a)*real(b) + real(a)*imag(b)); }
00319 
00320 template<> inline std::complex<double> pmul(const std::complex<double>& a, const std::complex<double>& b)
00321 { return std::complex<double>(real(a)*real(b) - imag(a)*imag(b), imag(a)*real(b) + real(a)*imag(b)); }
00322 
00323 } // end namespace internal
00324 
00325 } // end namespace Eigen
00326 
00327 #endif // EIGEN_GENERIC_PACKET_MATH_H
00328