Chris@102: /*
Chris@102:  [auto_generated]
Chris@102:  boost/numeric/odeint/external/openmp/openmp_state.hpp
Chris@102: 
Chris@102:  [begin_description]
Chris@102:  Wrappers for OpenMP.
Chris@102:  [end_description]
Chris@102: 
Chris@102:  Copyright 2013 Karsten Ahnert
Chris@102:  Copyright 2013 Mario Mulansky
Chris@102:  Copyright 2013 Pascal Germroth
Chris@102: 
Chris@102:  Distributed under the Boost Software License, Version 1.0.
Chris@102:  (See accompanying file LICENSE_1_0.txt or
Chris@102:  copy at http://www.boost.org/LICENSE_1_0.txt)
Chris@102:  */
Chris@102: 
Chris@102: 
Chris@102: #ifndef BOOST_NUMERIC_ODEINT_EXTERNAL_OPENMP_OPENMP_STATE_HPP_INCLUDED
Chris@102: #define BOOST_NUMERIC_ODEINT_EXTERNAL_OPENMP_OPENMP_STATE_HPP_INCLUDED
Chris@102: 
Chris@102: #include <omp.h>
Chris@102: #include <vector>
Chris@102: #include <algorithm>
Chris@102: #include <boost/range/adaptor/sliced.hpp>
Chris@102: #include <boost/numeric/odeint/util/copy.hpp>
Chris@102: #include <boost/numeric/odeint/util/split.hpp>
Chris@102: #include <boost/numeric/odeint/util/resize.hpp>
Chris@102: #include <boost/numeric/odeint/external/openmp/openmp_nested_algebra.hpp>
Chris@102: 
Chris@102: namespace boost {
Chris@102: namespace numeric {
Chris@102: namespace odeint {
Chris@102: 
Chris@102: /** \brief A container that is split into distinct parts, for threading.
Chris@102:  * Just a wrapper for vector<vector<T>>, use `copy` for splitting/joining.
Chris@102:  */
Chris@102: template< class T >
Chris@102: struct openmp_state : public std::vector< std::vector< T > >
Chris@102: {
Chris@102:     openmp_state() {}
Chris@102: 
Chris@102:     openmp_state(size_t n, const std::vector<T>& val = std::vector<T>())
Chris@102:     : std::vector< std::vector< T > >(n, val) {}
Chris@102: 
Chris@102:     template<class InputIterator>
Chris@102:     openmp_state(InputIterator first, InputIterator last)
Chris@102:     : std::vector< std::vector< T > >(first, last) {}
Chris@102: 
Chris@102:     openmp_state(const std::vector< std::vector< T > > &orig)
Chris@102:     : std::vector< std::vector< T > >(orig) {}
Chris@102: 
Chris@102: };
Chris@102: 
Chris@102: 
Chris@102: 
Chris@102: 
Chris@102: template< class T >
Chris@102: struct is_resizeable< openmp_state< T > > : boost::true_type { };
Chris@102: 
Chris@102: 
Chris@102: template< class T >
Chris@102: struct same_size_impl< openmp_state< T > , openmp_state< T > >
Chris@102: {
Chris@102:     static bool same_size( const openmp_state< T > &x , const openmp_state< T > &y )
Chris@102:     {
Chris@102:         if( x.size() != y.size() ) return false;
Chris@102:         for( size_t i = 0 ; i != x.size() ; i++ )
Chris@102:             if( x[i].size() != y[i].size() ) return false;
Chris@102:         return true;
Chris@102:     }
Chris@102: };
Chris@102: 
Chris@102: 
Chris@102: template< class T >
Chris@102: struct resize_impl< openmp_state< T > , openmp_state< T > >
Chris@102: {
Chris@102:     static void resize( openmp_state< T > &x , const openmp_state< T > &y )
Chris@102:     {
Chris@102:         x.resize( y.size() );
Chris@102: #       pragma omp parallel for schedule(dynamic)
Chris@102:         for(size_t i = 0 ; i < x.size() ; i++)
Chris@102:             x[i].resize( y[i].size() );
Chris@102:     }
Chris@102: };
Chris@102: 
Chris@102: 
Chris@102: /** \brief Copy data between openmp_states of same size. */
Chris@102: template< class T >
Chris@102: struct copy_impl< openmp_state< T >, openmp_state< T > >
Chris@102: {
Chris@102:     static void copy( const openmp_state< T > &from, openmp_state< T > &to )
Chris@102:     {
Chris@102: #       pragma omp parallel for schedule(dynamic)
Chris@102:         for(size_t i = 0 ; i < from.size() ; i++)
Chris@102:             std::copy( from[i].begin() , from[i].end() , to.begin() );
Chris@102:     }
Chris@102: };
Chris@102: 
Chris@102: 
Chris@102: 
Chris@102: /** \brief Copy data from some container to an openmp_state and resize it.
Chris@102:  * Target container size will determine number of blocks to split into.
Chris@102:  * If it is empty, it will be resized to the maximum number of OpenMP threads.
Chris@102:  * SourceContainer must support `s::value_type`, `s::const_iterator`, `s.begin()`, `s.end()` and `s.size()`,
Chris@102:  * with Random Access Iterators; i.e. it must be a Random Access Container. */
Chris@102: template< class SourceContainer >
Chris@102: struct split_impl< SourceContainer, openmp_state< typename SourceContainer::value_type > >
Chris@102: {
Chris@102:     static void split( const SourceContainer &from, openmp_state< typename SourceContainer::value_type > &to )
Chris@102:     {
Chris@102:         if(to.size() == 0) to.resize( omp_get_max_threads() );
Chris@102:         const size_t part = from.size() / to.size();
Chris@102: #       pragma omp parallel for schedule(dynamic)
Chris@102:         for(size_t i = 0 ; i < to.size() ; i++) {
Chris@102:             typedef typename SourceContainer::const_iterator it_t;
Chris@102:             const it_t begin = from.begin() + i * part;
Chris@102:             it_t end = begin + part;
Chris@102:             // for cases where from.size() % to.size() > 0
Chris@102:             if(i + 1 == to.size() || end > from.end()) end = from.end();
Chris@102:             to[i].resize(end - begin);
Chris@102:             std::copy(begin, end, to[i].begin());
Chris@102:         }
Chris@102:     }
Chris@102: };
Chris@102: 
Chris@102: /** \brief Copy data from an openmp_state to some container and resize it.
Chris@102:  * TargetContainer must support `s::value_type`, `s::iterator`, `s.begin()` and `s.resize(n)`,
Chris@102:  * i.e. it must be a `std::vector`. */
Chris@102: template< class TargetContainer >
Chris@102: struct unsplit_impl< openmp_state< typename TargetContainer::value_type >, TargetContainer >
Chris@102: {
Chris@102:     static void unsplit( const openmp_state< typename TargetContainer::value_type > &from , TargetContainer &to )
Chris@102:     {
Chris@102:         // resize target
Chris@102:         size_t total_size = 0;
Chris@102:         for(size_t i = 0 ; i < from.size() ; i++)
Chris@102:             total_size += from[i].size();
Chris@102:         to.resize( total_size );
Chris@102:         // copy parts
Chris@102:         typename TargetContainer::iterator out = to.begin();
Chris@102:         for(size_t i = 0 ; i < from.size() ; i++)
Chris@102:             out = std::copy(from[i].begin(), from[i].end(), out);
Chris@102:     }
Chris@102: };
Chris@102: 
Chris@102: 
Chris@102: 
Chris@102: 
Chris@102: /** \brief OpenMP-parallelized algebra.
Chris@102:  * For use with openmp_state.
Chris@102:  */
Chris@102: typedef openmp_nested_algebra< range_algebra > openmp_algebra;
Chris@102: 
Chris@102: 
Chris@102: 
Chris@102: /** \brief Use `openmp_algebra` for `openmp_state`. */
Chris@102: template< class T >
Chris@102: struct algebra_dispatcher< openmp_state< T > >
Chris@102: {
Chris@102:     typedef openmp_algebra algebra_type;
Chris@102: };
Chris@102: 
Chris@102: 
Chris@102: }
Chris@102: }
Chris@102: }
Chris@102: 
Chris@102: 
Chris@102: #endif
Chris@102: