/******************************************************************************* * * TRIQS: a Toolbox for Research in Interacting Quantum Systems * * Copyright (C) 2012 by O. Parcollet * * TRIQS is free software: you can redistribute it and/or modify it under the * terms of the GNU General Public License as published by the Free Software * Foundation, either version 3 of the License, or (at your option) any later * version. * * TRIQS is distributed in the hope that it will be useful, but WITHOUT ANY * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS * FOR A PARTICULAR PURPOSE. See the GNU General Public License for more * details. * * You should have received a copy of the GNU General Public License along with * TRIQS. If not, see <http://www.gnu.org/licenses/>. * ******************************************************************************/ #ifndef TRIQS_ARRAYS_BLAS_LAPACK_DOT_H #define TRIQS_ARRAYS_BLAS_LAPACK_DOT_H #include "./tools.hpp" //#include "./qcache.hpp" namespace triqs { namespace arrays { namespace blas { namespace f77 { // overload extern "C" { double TRIQS_FORTRAN_MANGLING(ddot)(const int & N, const double * x, const int& incx, const double * y, const int & incy); double TRIQS_FORTRAN_MANGLING(zdotc)(const int & N, const std::complex<double> * x, const int& incx, const std::complex<double> * y, const int& incy); double TRIQS_FORTRAN_MANGLING(zdotu)(const int & N, const std::complex<double> * x, const int& incx, const std::complex<double> * y, const int& incy); } inline double dot (const int & M, const double* x, const int & incx, const double* Y, const int & incy) { return TRIQS_FORTRAN_MANGLING(ddot)(M, x, incx, Y, incy); } //inline std::complex<double> dot (const int & M, const std::complex<double>* x, const int & incx, const std::complex<double>* Y, const int & incy) { // return TRIQS_FORTRAN_MANGLING(zdotu)(M, x, incx, Y, incy); // } } template <typename VTX, typename VTY> struct dispatch { static constexpr bool are_both_value_view = is_amv_value_or_view_class<VTX>::value && is_amv_value_or_view_class<VTY>::value; static constexpr bool are_both_double = std::is_same<typename VTX::value_type,double>::value && std::is_same<typename VTY::value_type,double>::value; static constexpr int value = (are_both_value_view ? (are_both_double ? 0 : 1) : 2); typedef decltype(std::declval<VTX>()(0)* std::declval<VTY>()(0)) result_type; }; /** * Calls dot product of 2 vectors. * Takes care of making temporary copies if necessary */ template<bool Star, typename VTX, typename VTY> typename std::enable_if< dispatch<VTX,VTY>::value==0, typename dispatch<VTX,VTY>::result_type>::type dot (VTX const & X, VTY const & Y) { //std::cerr << "dot : blas call"<< std::endl ; static_assert( is_amv_value_or_view_class<VTX>::value, "blas1 bindings only take vector and vector_view"); static_assert( is_amv_value_or_view_class<VTY>::value, "blas1 bindings only take vector and vector_view"); if (( X.size() != Y.size()) ) TRIQS_RUNTIME_ERROR << "Dimension mismatch in dot size are : X : "<<X.size()<<" and Y : "<<Y.size(); //const_qcache<VTX> Cx(X); // mettre la condition a la main //const_qcache<VTY> Cy(Y); // mettre la condition a la main return f77::dot(X.size(), X.data_start(), X.stride(), Y.data_start(), Y.stride()); //return f77::dot(X.size(), Cx().data_start(), Cx().stride(), Cy().data_start(), Cy().stride()); } template< bool Star, typename T> typename std::enable_if<boost::is_complex<T>::value && Star,T>::type _conj(T && x) { return conj(std::forward<T>(x));} template< bool Star, typename T> typename std::enable_if<!( boost::is_complex<T>::value && Star),T>::type _conj(T && x) { return std::forward<T>(x);} /** * Calls dot product of 2 vectors. * Takes care of making temporary copies if necessary * general case. Also for complex since there is a bug on some machines (os X, weiss...) for zdotu, zdotc... * a transcription from netlib zdotu */ template< bool Star, typename VTX, typename VTY> typename std::enable_if< dispatch<VTX,VTY>::value==1, typename dispatch<VTX,VTY>::result_type>::type dot (VTX const & X, VTY const & Y) { //std::cerr << "dot: transcription of blas for non uniform type with restrict"<< std::endl ; if (( X.size() != Y.size()) ) TRIQS_RUNTIME_ERROR << "Dimension mismatch in dot size are : X : "<<X.size()<<" and Y : "<<Y.size(); //const_qcache<VTX> Cx(X); // mettre la condition a la main //const_qcache<VTY> Cy(Y); // mettre la condition a la main size_t N= X.size(), incx = X.stride(), incy = Y.stride(); decltype(X(0)*Y(0)) res = 0; // This only works for object with data (ISP), not only from the concept... auto * restrict X_ = X.data_start(); auto * restrict Y_ = Y.data_start(); if ((incx==1) && (incy==1)) { for (size_t i=0; i<N; ++i) res += _conj<Star>(X_[i]) * Y_[i]; } else { // code for unequal increments or equal increments not equal to 1 for (size_t i=0, ix=0, iy=0; i<N; ++i, ix += incx, iy +=incy) {res += _conj<Star>(X_[ix]) * Y_[iy]; } } // general code for the concept. Is it really slower ? cf below return res; } /** * Generic case */ template< bool Star, typename VTX, typename VTY> typename std::enable_if< dispatch<VTX,VTY>::value==2, typename dispatch<VTX,VTY>::result_type>::type dot (VTX const & X, VTY const & Y) { //std::cerr << "dot : generic case"<< std::endl ; if (( X.size() != Y.size()) ) TRIQS_RUNTIME_ERROR << "Dimension mismatch in dot size are : X : "<<X.size()<<" and Y : "<<Y.size(); size_t N= X.size(); decltype(X(0)*Y(0)) res = 0; for (size_t i=0; i<N; ++i) res += _conj<Star>(X(i)) * Y(i); return res; } } template <typename VTX,typename VTY> auto dot (VTX const & X, VTY const & Y) DECL_AND_RETURN( blas::dot<false>(X,Y)); template <typename VTX,typename VTY> auto dotc (VTX const & X, VTY const & Y) DECL_AND_RETURN( blas::dot<true>(X,Y)); }}// namespace #endif