Homec4science
Diffusion tamaas (maxwell_viscoelastic-branch) Browse

tamaas/src/core/loop.hh
a0869ea06429maxwell_viscoelastic…

Last Change
Raw File

loop.hh
View Options

/**
* @file
*
* @author Lucas Frérot <lucas.frerot@epfl.ch>
*
* @section LICENSE
*
* Copyright (©) 2017 EPFL (Ecole Polytechnique Fédérale de
* Lausanne) Laboratory (LSMS - Laboratoire de Simulation en Mécanique des
* Solides)
*
* Tamaas is free software: you can redistribute it and/or modify it under the
* terms of the GNU Lesser General Public License as published by the Free
* Software Foundation, either version 3 of the License, or (at your option) any
* later version.
*
* Tamaas 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 Lesser General Public License for more
* details.
*
* You should have received a copy of the GNU Lesser General Public License
* along with Tamaas. If not, see <http://www.gnu.org/licenses/>.
*
*/
/* -------------------------------------------------------------------------- */
#ifndef __LOOP_HH__
#define __LOOP_HH__
/* -------------------------------------------------------------------------- */
#include "tamaas.hh"
#include "loops/apply.hh"
#include "loops/loop_omp.hh"
#ifdef USE_CUDA
//#include "loops/loop_cuda.hh"
#endif
#include <type_traits>
#include <thrust/for_each.h>
#include <thrust/transform_reduce.h>
#include <thrust/iterator/zip_iterator.h>
#include <thrust/tuple.h>
#include <boost/preprocessor/seq.hpp>
__BEGIN_TAMAAS__
#define EXEC_CASE_MACRO(r, strided, backend) \
case backend: \
BOOST_PP_CAT(detail::exec_loop_, backend)<strided>(func, grids...); \
break;
#define REDUCE_CASE_MACRO(r, strided, backend) \
case backend: \
return BOOST_PP_CAT(detail::reduce_loop_, backend)<op, strided>(func, grids...);
#ifdef USE_CUDA
#define AVAILABLE_BACKENDS (omp)(cuda)
#else
#define AVAILABLE_BACKENDS (omp)
#endif
/**
* @brief Singleton class for automated loops using lambdas
* This class is sweet candy :) It provides abstraction of the paralelism
* paradigm used in loops and allows simple and less erro-prone loop syntax,
* with minimum boiler plate. I love it <3
*/
class Loop {
public:
/// Backends enumeration
enum backend {
omp, ///< [OpenMP](http://www.openmp.org/specifications/) backend
cuda, ///< [Cuda](http://docs.nvidia.com/cuda/index.html) backend
};
/// Helper class to count iterations within lambda-loop
struct enumerator {
constexpr enumerator() = default;
constexpr UInt getNbComponents() const { return 1.; }
constexpr UInt operator()(UInt i) const { return i; }
};
/// Initialize singleton instance
static void init(backend back);
/// Loop functor over any number of grids
template <typename Functor, typename... Grids>
static void loop(Functor&& func, Grids&&... containers) {
auto begin = thrust::make_zip_iterator(thrust::make_tuple(containers.begin()...));
auto end = thrust::make_zip_iterator(thrust::make_tuple(containers.end()...));
thrust::for_each(begin, end, detail::ApplyFunctor<Functor>(func));
}
/// Strided loop over any number of grids
template <typename Functor, typename... Grids>
static void stridedLoop(Functor&& func, Grids&&... containers) {
auto begin =
thrust::make_zip_iterator(thrust::make_tuple(containers.begin(containers.getNbComponents())...));
auto end =
thrust::make_zip_iterator(thrust::make_tuple(containers.end(containers.getNbComponents())...));
thrust::for_each(begin, end, detail::ApplyFunctor<Functor>(func));
}
/// Reduce over any number of grids
template <operation op, typename Functor, typename... Grids>
static auto reduce(Functor&& func, Grids&&... containers)
-> decltype(func(containers(0)...)) {
auto begin = thrust::make_zip_iterator(thrust::make_tuple(containers.begin()...));
auto end = thrust::make_zip_iterator(thrust::make_tuple(containers.end()...));
using reduce_type = decltype(func(containers(0)...));
return thrust::transform_reduce(begin, end, detail::ApplyFunctor<Functor, reduce_type>(func),
detail::reduction_helper<op>::template init<reduce_type>(),
detail::reduction_helper<op>());
}
/// Strided reduce over any number of grids
template <operation op, typename Functor, typename... Grids>
static auto stridedReduce(Functor&& func, Grids&&... containers)
-> decltype(func(containers(0)...)) {
auto begin = thrust::make_zip_iterator(thrust::make_tuple(containers.begin(containers.getNbComponents())...));
auto end = thrust::make_zip_iterator(thrust::make_tuple(containers.end(containers.getNbComponents())...));
using reduce_type = decltype(func(containers(0)...));
return thrust::transform_reduce(begin, end, detail::ApplyFunctor<Functor, reduce_type>(func),
detail::reduction_helper<op>::template init<reduce_type>(),
detail::reduction_helper<op>());
}
private:
/// Constructor
Loop() = default;
/// Backend type
static backend _backend;
};
__END_TAMAAS__
#undef EXEC_CASE_MACRO
#undef REDUCE_CASE_MACRO
#endif // __LOOP_HH__
c4science · Help