trash_bin.hpp

/*
 * trash_bin.hpp
 *
 *  Created on: Nov 24, 2014
 *      Author: i-bird
 */

#ifndef TRASH_BIN_HPP_
#define TRASH_BIN_HPP_


template<typename T>
class E_p
{
public:
    typedef typename boost::fusion::result_of::push_front<typename T::type, size_t>::type E_a;

    typedef E_a type;

    E_a data;
};

template<typename T>
class V_p
{
public:
    typedef typename boost::fusion::result_of::push_front<typename T::type, size_t>::type V_a;

    typedef V_a type;

    V_a data;
};


#endif /* TRASH_BIN_HPP_ */


/*      template <unsigned int p>inline typename type_cpu_prop<p,T>::type & getBoostVector(grid_key_dx<dim> & v1)
{
#ifdef MEMLEAK_CHECK
    check_valid(&boost::fusion::at_c<p>(data.mem_r->operator[](g1.LinId(v1))));
#endif
    return boost::fusion::at_c<p>(data.mem_r->operator[](g1.LinId(v1)));
}*/


template<unsigned int prp>
struct red_max
{
    template<typename aggr_vect, typename encap_type>
    static void red(aggr_vect & aggr, encap_type & ec)
    {
        auto op1 = ec.template get<prp>();
        auto op2 = aggr.template get<prp>();

        aggr.template get<prp>() = (op2 < op1)?op1:op2;
    }

    template<typename red_type, typename aggr_vect, typename BlockReduce>
    static red_type red_final(red_type * rt, aggr_vect & aggr)
    {
        return BlockReduce(rt).Max(aggr.template get<prp>());
    }
};

template<unsigned int prp>
struct red_min
{
    template<typename aggr_vect, typename encap_type>
    static void red(aggr_vect & aggr, encap_type & ec)
    {
        auto op1 = ec.template get<prp>();
        auto op2 = aggr.template get<prp>();

        aggr.template get<prp>() = (op2 > op1)?op1:op2;
    }


    template<typename red_type, typename aggr_vect, typename BlockReduce>
    static red_type red_final(red_type * rt, aggr_vect & aggr)
    {
        return BlockReduce(rt).Min(aggr.template get<prp>());
    }
};

template<unsigned int prp>
struct red_sum
{
    template<typename aggr_vect, typename encap_type>
    static void red(aggr_vect & aggr, encap_type & ec)
    {
        aggr.template get<prp>() += ec.template get<prp>();
    }


    template<typename red_type, typename aggr_vect, typename BlockReduce>
    static red_type red_final(red_type * rt, aggr_vect & aggr)
    {
        return BlockReduce(rt).Sum(aggr.template get<prp>());
    }
};

template<typename aggr_vect, typename reduction_vectors>
struct reduce_op
{
    aggr_vect & red;
    reduction_vectors & input;

    __device__ __host__ inline reduce_op(aggr_vect & red, reduction_vectors & input)
    :red(red),input(input)
    {
    };

    template<typename T>
    __device__ __host__ inline void operator()(T& t) const
    {
        boost::mpl::at<reduction_vectors,boost::mpl::int_<T::value>>::red(red,input);
    }
};

template<typename aggr_vect, typename red_type, typename reduction_vectors>
struct reduce_op_final
{
    aggr_vect & red;

    red_type * red_space;

    typedef typename boost::mpl::size<reduction_vectors>::type nred;

    red_type red_final[nred::value];

    __device__ __host__ inline reduce_op_final(aggr_vect & red, red_type & red_space)
    :red(red),red_space(red_space)
    {
    };

    template<typename T>
    __device__ __host__ inline void operator()(T& t)
    {
        // fill the block
        red_space[threadIdx.x] = red.template get<T::value>();

        red_final[T::value] = boost::mpl::at<reduction_vectors,boost::mpl::int_<T::value>>::red_final(red_space);
    }
};


template<typename red_type, typename encap_type , typename reduction_vectors>
struct store_reduce_op_final
{
    typedef typename boost::mpl::size<reduction_vectors>::type nred;

    red_type (& red_final)[nred::value];

    encap_type & destination;

    __device__ __host__ inline store_reduce_op_final(encap_type & destination, red_type (& red_final)[nred::value])
    :red_final(red_final),destination(destination)
    {
    };

    template<typename T>
    __device__ __host__ inline void operator()(T& t)
    {
        typedef typename boost::mpl::at<reduction_vectors,boost::mpl::int_<T::value>>::prop prp;

        destination.template get<prp::value>() = red_final[T::value];
    }
};

template<typename vector_index_type, typename vector_index_type2, typename vector_data_type, typename reduction_operation_vector,
         typename red_type, unsigned int blockSize>
__global__ void reduce_data(vector_index_type v_offsets, vector_index_type2 v_ord, vector_data_type v_data, vector_data_type v_data_red)
{
    int offset_start = v_offsets.template get<0>(blockIdx.x);
    int offset_end = v_offsets.template get<0>(blockIdx.x+1);
    int n_block_reduce = (offset_end - offset_start) / blockDim.x;

    // Specialize BlockReduce for a 1D block
    typedef cub::BlockReduce<red_type, blockSize> BlockReduceT;

    __shared__ typename BlockReduceT::TempStorage temp_storage;

    typename vector_data_type::value_type red_v;

    typedef boost::mpl::size<reduction_operation_vector> n_reductions;

    int i = 0;
    for ( ; i < n_block_reduce ; i++)
    {
        reduce_op<typename vector_data_type::value_type,reduction_operation_vector> ro(red_v,v_data.get_o(offset_start+i*blockDim.x+threadIdx.x));

        boost::mpl::for_each_ref<boost::mpl::range_c<int,0,n_reductions::value>>(ro);
    }

    // within block reduction
    reduce_op_final<typename vector_data_type::value,red_type,reduction_operation_vector> rof(red_v,temp_storage);
    boost::mpl::for_each_ref<boost::mpl::range_c<int,0,n_reductions::value>>(rof);

    if (threadIdx.x == 0)
    {

    }
}