grid_dist_id.hpp

#ifndef COM_UNIT_HPP
#define COM_UNIT_HPP
 
#include <vector>
#include <unordered_map>
#include "Grid/map_grid.hpp"
#include "VCluster/VCluster.hpp"
#include "Space/SpaceBox.hpp"
#include "util/mathutil.hpp"
#include "VTKWriter/VTKWriter.hpp"
#ifdef __NVCC__
#include "SparseGridGpu/SparseGridGpu.hpp"
#endif
#include "Iterators/grid_dist_id_iterator_dec.hpp"
#include "Iterators/grid_dist_id_iterator.hpp"
#include "Iterators/grid_dist_id_iterator_sub.hpp"
#include "grid_dist_key.hpp"
#include "NN/CellList/CellDecomposer.hpp"
#include "util/object_util.hpp"
#include "memory/ExtPreAlloc.hpp"
#include "Packer_Unpacker/Packer.hpp"
#include "Packer_Unpacker/Unpacker.hpp"
#include "Decomposition/CartDecomposition.hpp"
#include "data_type/aggregate.hpp"
#include "hdf5.h"
#include "grid_dist_id_comm.hpp"
#include "HDF5_wr/HDF5_wr.hpp"
#include "SparseGrid/SparseGrid.hpp"
#include "lib/pdata.hpp"
#ifdef __NVCC__
#include "cuda/grid_dist_id_kernels.cuh"
#include "Grid/cuda/grid_dist_id_iterator_gpu.cuh"
#endif
 
template<unsigned int dim>
struct offset_mv
{
    Point<dim,long int> up;
 
    Point<dim,long int> dw;
};
 
template<unsigned int dim>
struct Box_fix
{
    Box<dim,size_t> bx;
    comb<dim> cmb;
    size_t g_id;
    size_t r_sub;
};
 
#define NO_GDB_EXT_SWITCH 0x1000
 
template<bool is_sparse>
struct device_grid_copy
{
    template<typename key_type1, typename key_type2, typename spg_type, typename lg_type>
    static void assign(key_type1 & key, key_type2 & key_dst, spg_type & dg, lg_type & lg)
    {
        dg.insert_o(key_dst) = lg.get_o(key);
    }
 
    template<typename gdb_ext_type, typename loc_grid_type>
    static void pre_load(gdb_ext_type & gdb_ext_old, loc_grid_type & loc_grid_old, gdb_ext_type & gdb_ext, loc_grid_type & loc_grid) 
    {
    }
};
 
template<typename e_src, typename e_dst, typename indexT>
struct copy_all_prop_sparse
{
    const e_src & block_data_src;
    e_dst & block_data_dst;
 
    indexT local_id;
 
    __device__ __host__ inline copy_all_prop_sparse(const e_src & block_data_src, e_dst & block_data_dst, indexT local_id)
    :block_data_src(block_data_src),block_data_dst(block_data_dst),local_id(local_id)
    {
    };
 
    template<typename T>
    __device__ __host__ inline void operator()(T& t) const
    {
        block_data_dst.template get<T::value>()[local_id] = block_data_src.template get<T::value>();
    }
};
 
template<typename T>
struct variadic_caller
{};
 
template<int ... prp>
struct variadic_caller<index_tuple_sq<prp ...>>
{
    template<typename this_type, typename dec_type, typename cd_sm_type, typename loc_grid_type, typename gdb_ext_type>
    static void call(this_type * this_,
                     dec_type & dec, 
                     cd_sm_type & cd_sm,
                     loc_grid_type & loc_grid, 
                     loc_grid_type & loc_grid_old, 
                     gdb_ext_type & gdb_ext, 
                     gdb_ext_type & gdb_ext_old, 
                     gdb_ext_type & gdb_ext_global, 
                     size_t opt)
    {
        this_->template map_<prp ...>(dec,cd_sm,loc_grid,loc_grid_old,gdb_ext,gdb_ext_old,gdb_ext_global,opt);
    }
};
 
struct ids_pl
{
    size_t id;
 
        bool operator<(const ids_pl & i) const
        {
            return id < i.id;
    }
 
    bool operator==(const ids_pl & i) const
    {
        return id == i.id;
    }
};
 
template<>
struct device_grid_copy<true>
{
    template<typename key_type1, typename key_type2, typename spg_type, typename lg_type>
    static void assign(key_type1 & key, key_type2 & key_dst, spg_type & dg, lg_type & lg)
    {
        typename spg_type::indexT_ local_id;
        auto block_data_dst = dg.insertBlockFlush(key_dst, local_id);
        auto block_data_src = lg.get_o(key);
 
        copy_all_prop_sparse<decltype(block_data_src),decltype(block_data_dst),decltype(local_id)> cp(block_data_src, block_data_dst, local_id);
 
        boost::mpl::for_each_ref< boost::mpl::range_c<int,0,decltype(block_data_dst)::max_prop> >(cp);
    }
 
 
 
    template<typename gdb_ext_type, typename loc_grid_type>
    static void pre_load(gdb_ext_type & gdb_ext_old, loc_grid_type & loc_grid_old, gdb_ext_type & gdb_ext, loc_grid_type & loc_grid)
    {
        auto & dg = loc_grid.get(0);
        auto dlin = dg.getGrid();
        typedef typename std::remove_reference<decltype(dg)>::type sparse_grid_type;
 
        openfpm::vector<ids_pl> ids;
        openfpm::vector<grid_key_dx<sparse_grid_type::dims>> gg;
 
        size_t sz[sparse_grid_type::dims];
 
        for (int i = 0 ; i < sparse_grid_type::dims ; i++)
        {sz[i] = 2;}
 
        grid_sm<sparse_grid_type::dims,void> gvoid(sz);
 
        grid_key_dx_iterator<sparse_grid_type::dims> it(gvoid);
        while(it.isNext())
        {
            auto key = it.get();
 
            grid_key_dx<sparse_grid_type::dims> k;
            for (int i = 0 ; i < sparse_grid_type::dims ; i++)
            {
                k.set_d(i,key.get(i)*dlin.getBlockEgdeSize());
            }
 
            gg.add(k);
            ++it;
        }
 
 
        for (int i = 0 ; i < gdb_ext_old.size() ; i++)
        {
            
            auto & lg = loc_grid_old.get(i);
            auto & indexBuffer = lg.private_get_index_array();
            auto & dg = loc_grid.get(0);
 
            auto slin = loc_grid_old.get(i).getGrid();
 
            grid_key_dx<sparse_grid_type::dims> kp1 = gdb_ext.get(0).Dbox.getKP1();
 
            grid_key_dx<sparse_grid_type::dims> orig;
            for (int j = 0 ; j < sparse_grid_type::dims ; j++)
            {
                orig.set_d(j,gdb_ext_old.get(i).origin.get(j));
            }
 
                        for (int i = 0; i < indexBuffer.size() ; i++)
                        {
                for (int ex = 0; ex < gg.size() ; ex++) {
                    auto key = slin.InvLinId(indexBuffer.template get<0>(i),0);
                    grid_key_dx<sparse_grid_type::dims> key_dst;
 
                    for (int j = 0 ; j < sparse_grid_type::dims ; j++)
                    {key_dst.set_d(j,key.get(j) + orig.get(j) + kp1.get(j) + gg.get(ex).get(j));}
 
                    typename sparse_grid_type::indexT_ bid;
                    int lid;
 
                    dlin.LinId(key_dst,bid,lid);
 
                    ids.add();
                    ids.last().id = bid;
                }
            }
        }
 
        ids.sort();
        ids.unique();
 
        auto & indexBuffer = dg.private_get_index_array();
        auto & dataBuffer = dg.private_get_data_array();
        indexBuffer.reserve(ids.size()+8);
        dataBuffer.reserve(ids.size()+8);
        for (int i = 0 ; i < ids.size() ; i++) 
        {
            auto & dg = loc_grid.get(0);
            dg.insertBlockFlush(ids.get(i).id);
        }
    }
};
 
template<unsigned int dim,
         typename St,
         typename T,
         typename Decomposition = CartDecomposition<dim,St>,
         typename Memory=HeapMemory ,
         typename device_grid=grid_cpu<dim,T> >
class grid_dist_id : public grid_dist_id_comm<dim,St,T,Decomposition,Memory,device_grid>
{
    typedef grid_dist_id<dim,St,T,Decomposition,Memory,device_grid> self;
 
    Box<dim,St> domain;
 
    Ghost<dim,St> ghost;
 
    Ghost<dim,long int> ghost_int;
 
    mutable openfpm::vector<device_grid> loc_grid;
 
    mutable openfpm::vector<device_grid> loc_grid_old;
 
    Decomposition dec;
 
    openfpm::vector<size_t> gdb_ext_markers;
 
    openfpm::vector<GBoxes<device_grid::dims>> gdb_ext;
 
    mutable openfpm::vector<GBoxes<device_grid::dims>> gdb_ext_global;
 
    openfpm::vector<GBoxes<device_grid::dims>> gdb_ext_old;
 
    size_t g_sz[dim];
 
    CellDecomposer_sm<dim,St,shift<dim,St>> cd_sm;
 
    size_t gpu_n_insert_thread;
 
    Vcluster<> & v_cl;
 
    openfpm::vector<std::string> prp_names;
 
    std::unordered_map<size_t,size_t> g_id_to_external_ghost_box;
 
    openfpm::vector<e_box_multi<dim>> eb_gid_list;
 
    openfpm::vector<std::unordered_map<size_t,size_t>> g_id_to_internal_ghost_box;
 
    openfpm::vector<size_t> recv_sz;
 
    openfpm::vector<HeapMemory> recv_mem_gg;
 
    grid_sm<dim,T> ginfo;
 
    grid_sm<dim,void> ginfo_v;
 
    openfpm::vector<Box<dim,long int>> bx_def;
 
    bool use_bx_def = false;
 
    bool init_local_i_g_box = false;
 
    bool init_local_e_g_box = false;
 
    bool init_e_g_box = false;
 
    bool init_i_g_box = false;
 
    bool init_fix_ie_g_box = false;
 
    openfpm::vector<ip_box_grid<dim>> ig_box;
 
    openfpm::vector<ep_box_grid<dim>> eg_box;
 
    openfpm::vector<i_lbox_grid<dim>> loc_ig_box;
 
    openfpm::vector<e_lbox_grid<dim>> loc_eg_box;
 
    size_t v_sub_unit_factor = 64;
 
    static void * msg_alloc_external_box(size_t msg_i ,size_t total_msg, size_t total_p, size_t i, size_t ri, void * ptr)
    {
        grid_dist_id<dim,St,T,Decomposition,Memory,device_grid> * g = static_cast<grid_dist_id<dim,St,T,Decomposition,Memory,device_grid> *>(ptr);
 
        g->recv_sz.resize(g->dec.getNNProcessors());
        g->recv_mem_gg.resize(g->dec.getNNProcessors());
 
        // Get the local processor id
        size_t lc_id = g->dec.ProctoID(i);
 
        // resize the receive buffer
        g->recv_mem_gg.get(lc_id).resize(msg_i);
        g->recv_sz.get(lc_id) = msg_i;
 
        return g->recv_mem_gg.get(lc_id).getPointer();
    }
 
    void set_for_adjustment(size_t sub_id,
                            const Box<dim,St> & sub_domain_other,
                            const comb<dim> & cmb,
                            Box<dim,long int> & ib,
                            Ghost<dim,long int> & g)
    {
        if (g.isInvalidGhost() == true || use_bx_def == true)
        {return;}
 
        Box<dim,long int> sub_domain;
 
        if (use_bx_def == false)
        {
            sub_domain = gdb_ext.get(sub_id).Dbox;
            sub_domain += gdb_ext.get(sub_id).origin;
        }
 
        // Convert from SpaceBox<dim,St> to SpaceBox<dim,long int>
        Box<dim,long int> sub_domain_other_exp = cd_sm.convertDomainSpaceIntoGridUnits(sub_domain_other,dec.periodicity());
 
        // translate sub_domain_other based on cmb
        for (size_t i = 0 ; i < dim ; i++)
        {
            if (cmb.c[i] == 1)
            {
                sub_domain_other_exp.setLow(i,sub_domain_other_exp.getLow(i) - ginfo.size(i));
                sub_domain_other_exp.setHigh(i,sub_domain_other_exp.getHigh(i) - ginfo.size(i));
            }
            else if (cmb.c[i] == -1)
            {
                sub_domain_other_exp.setLow(i,sub_domain_other_exp.getLow(i) + ginfo.size(i));
                sub_domain_other_exp.setHigh(i,sub_domain_other_exp.getHigh(i) + ginfo.size(i));
            }
        }
 
        sub_domain_other_exp.enlarge(g);
        if (sub_domain_other_exp.Intersect(sub_domain,ib) == false)
        {
            for (size_t i = 0 ; i < dim ; i++)
            {ib.setHigh(i,ib.getLow(i) - 1);}
        }
    }
 
 
 
    void create_ig_box()
    {
        // temporal vector used for computation
        openfpm::vector_std<result_box<dim>> ibv;
 
        if (init_i_g_box == true)   return;
 
        // Get the grid info
        auto g = cd_sm.getGrid();
 
        g_id_to_internal_ghost_box.resize(dec.getNNProcessors());
 
        // Get the number of near processors
        for (size_t i = 0 ; i < dec.getNNProcessors() ; i++)
        {
            ig_box.add();
            auto&& pib = ig_box.last();
 
            pib.prc = dec.IDtoProc(i);
            for (size_t j = 0 ; j < dec.getProcessorNIGhost(i) ; j++)
            {
                // Get the internal ghost boxes and transform into grid units
                ::Box<dim,St> ib_dom = dec.getProcessorIGhostBox(i,j);
                ::Box<dim,long int> ib = cd_sm.convertDomainSpaceIntoGridUnits(ib_dom,dec.periodicity());
 
                size_t sub_id = dec.getProcessorIGhostSub(i,j);
                size_t r_sub = dec.getProcessorIGhostSSub(i,j);
 
                auto & n_box = dec.getNearSubdomains(dec.IDtoProc(i));
 
                set_for_adjustment(sub_id,
                                   n_box.get(r_sub),dec.getProcessorIGhostPos(i,j),
                                   ib,ghost_int);
 
                // Here we intersect the internal ghost box with the definition boxes
                // this operation make sense when the grid is not defined in the full
                // domain and we have to intersect the internal ghost box with all the box
                // that define where the grid is defined
                bx_intersect<dim>(bx_def,use_bx_def,ib,ibv);
 
                for (size_t k = 0 ; k < ibv.size() ; k++)
                {
                    // Check if ib is valid if not it mean that the internal ghost does not contain information so skip it
                    if (ibv.get(k).bx.isValid() == false)
                    {continue;}
 
                    // save the box and the sub-domain id (it is calculated as the linearization of P1)
                    ::Box<dim,size_t> cvt = ibv.get(k).bx;
 
                    i_box_id<dim> bid_t;
                    bid_t.box = cvt;
                    bid_t.g_id = dec.getProcessorIGhostId(i,j) | (k) << 52;
 
                    bid_t.sub = convert_to_gdb_ext(dec.getProcessorIGhostSub(i,j),
                                                   ibv.get(k).id,
                                                   gdb_ext,
                                                   gdb_ext_markers);
 
                    bid_t.cmb = dec.getProcessorIGhostPos(i,j);
                    bid_t.r_sub = dec.getProcessorIGhostSSub(i,j);
                    pib.bid.add(bid_t);
 
                    g_id_to_internal_ghost_box.get(i)[bid_t.g_id | (k) << 52 ] = pib.bid.size()-1;
                }
            }
        }
 
        init_i_g_box = true;
    }
 
 
    void create_eg_box()
    {
        // Get the grid info
        auto g = cd_sm.getGrid();
 
        if (init_e_g_box == true)   return;
 
        // Here we collect all the calculated internal ghost box in the sector different from 0 that this processor has
 
        openfpm::vector<size_t> prc;
        openfpm::vector<size_t> prc_recv;
        openfpm::vector<size_t> sz_recv;
        openfpm::vector<openfpm::vector<Box_fix<dim>>> box_int_send(dec.getNNProcessors());
        openfpm::vector<openfpm::vector<Box_fix<dim>>> box_int_recv;
 
        for(size_t i = 0 ; i < dec.getNNProcessors() ; i++)
        {
            for (size_t j = 0 ; j < ig_box.get(i).bid.size() ; j++)
            {
                box_int_send.get(i).add();
                box_int_send.get(i).last().bx = ig_box.get(i).bid.get(j).box;
                box_int_send.get(i).last().g_id = ig_box.get(i).bid.get(j).g_id;
                box_int_send.get(i).last().r_sub = ig_box.get(i).bid.get(j).r_sub;
                box_int_send.get(i).last().cmb = ig_box.get(i).bid.get(j).cmb;
            }
            prc.add(dec.IDtoProc(i));
        }
 
        v_cl.SSendRecv(box_int_send,box_int_recv,prc,prc_recv,sz_recv);
 
        eg_box.resize(dec.getNNProcessors());
 
        for (size_t i = 0 ; i < eg_box.size() ; i++)
        {eg_box.get(i).prc = dec.IDtoProc(i);}
 
        for (size_t i = 0 ; i < box_int_recv.size() ; i++)
        {
            size_t pib_cnt = 0;
            size_t p_id = dec.ProctoID(prc_recv.get(i));
            auto&& pib = eg_box.get(p_id);
            pib.prc = prc_recv.get(i);
 
            eg_box.get(p_id).recv_pnt = 0;
            eg_box.get(p_id).n_r_box = box_int_recv.get(i).size();
 
            // For each received internal ghost box
            for (size_t j = 0 ; j < box_int_recv.get(i).size() ; j++)
            {
                size_t vol_recv = box_int_recv.get(i).get(j).bx.getVolumeKey();
 
                eg_box.get(p_id).recv_pnt += vol_recv;
                size_t send_list_id = box_int_recv.get(i).get(j).r_sub;
 
                if (use_bx_def == true)
                {
                    // First we understand if the internal ghost box sent intersect
                    // some local extended sub-domain.
 
                    // eb_gid_list, for an explanation check the declaration
                    eb_gid_list.add();
                    eb_gid_list.last().e_id = p_id;
 
                    // Now we have to check if a received external ghost box intersect one
                    // or more sub-grids
                    for (size_t k = 0 ; k < gdb_ext.size() ; k++)
                    {
                        Box<dim,long int> bx = gdb_ext.get(k).GDbox;
                        bx += gdb_ext.get(k).origin;
 
                        Box<dim,long int> output;
                        Box<dim,long int> flp_i = flip_box(box_int_recv.get(i).get(j).bx,box_int_recv.get(i).get(j).cmb,ginfo);
 
                        // it intersect one sub-grid
                        if (bx.Intersect(flp_i,output))
                        {
                            // link
 
                            size_t g_id = box_int_recv.get(i).get(j).g_id;
                            add_eg_box<dim>(k,box_int_recv.get(i).get(j).cmb,output,
                                    g_id,
                                    gdb_ext.get(k).origin,
                                    box_int_recv.get(i).get(j).bx.getP1(),
                                    pib.bid);
 
                            eb_gid_list.last().eb_list.add(pib.bid.size() - 1);
 
                            g_id_to_external_ghost_box[g_id] = eb_gid_list.size() - 1;
                        }
                    }
 
                    // now we check if exist a full match across the full intersected
                    // ghost parts
 
                    bool no_match = true;
                    for (size_t k = 0 ; k < eb_gid_list.last().eb_list.size() ; k++)
                    {
                        size_t eb_id = eb_gid_list.last().eb_list.get(k);
 
                        if (pib.bid.get(eb_id).g_e_box == box_int_recv.get(i).get(j).bx)
                        {
                            // full match found
 
                            eb_gid_list.last().full_match = eb_id;
                            no_match = false;
 
                            break;
                        }
                    }
 
                    // This is the case where a full match has not been found. In this case we
                    // generate an additional gdb_ext and local grid with only external ghost
 
                    if (no_match == true)
                    {
                        // Create a grid with the same size of the external ghost
 
                        size_t sz[dim];
                        for (size_t s = 0 ; s < dim ; s++)
                        {sz[s] = box_int_recv.get(i).get(j).bx.getHigh(s) - box_int_recv.get(i).get(j).bx.getLow(s) + 1;}
 
                        // Add an unlinked gdb_ext
                        // An unlinked gdb_ext is an empty domain with only a external ghost
                        // part
                        Box<dim,long int> output = flip_box(box_int_recv.get(i).get(j).bx,box_int_recv.get(i).get(j).cmb,ginfo);
 
                        GBoxes<dim> tmp;
                        tmp.GDbox = box_int_recv.get(i).get(j).bx;
                        tmp.GDbox -= tmp.GDbox.getP1();
                        tmp.origin = output.getP1();
                        for (size_t s = 0 ; s < dim ; s++)
                        {
                            // we set an invalid box, there is no-domain
                            tmp.Dbox.setLow(s,0);
                            tmp.Dbox.setHigh(s,-1);
                        }
                        tmp.k = (size_t)-1;
                        gdb_ext.add(tmp);
 
                        // create the local grid
 
                        loc_grid.add();
                        loc_grid.last().resize(sz);
 
                        // Add an external ghost box
 
                        size_t g_id = box_int_recv.get(i).get(j).g_id;
                        add_eg_box<dim>(gdb_ext.size()-1,box_int_recv.get(i).get(j).cmb,output,
                                g_id,
                                gdb_ext.get(gdb_ext.size()-1).origin,
                                box_int_recv.get(i).get(j).bx.getP1(),
                                pib.bid);
 
                        // now we map the received ghost box to the information of the
                        // external ghost box created
                        eb_gid_list.last().full_match = pib.bid.size() - 1;
                        eb_gid_list.last().eb_list.add(pib.bid.size() - 1);
                        g_id_to_external_ghost_box[g_id] = eb_gid_list.size() - 1;
                    }
 
                    // now we create lr_e_box
 
                    size_t fm = eb_gid_list.last().full_match;
                    size_t sub_id = pib.bid.get(fm).sub;
 
                    for ( ; pib_cnt < pib.bid.size() ; pib_cnt++)
                    {
                        pib.bid.get(pib_cnt).lr_e_box -= gdb_ext.get(sub_id).origin;
                    }
 
                }
                else
                {
                    // Get the list of the sent sub-domains
                    // and recover the id of the sub-domain from
                    // the sent list
                    const openfpm::vector<size_t> & s_sub = dec.getSentSubdomains(p_id);
 
                    size_t sub_id = s_sub.get(send_list_id);
 
                    e_box_id<dim> bid_t;
                    bid_t.sub = sub_id;
                    bid_t.cmb = box_int_recv.get(i).get(j).cmb;
                    bid_t.cmb.sign_flip();
                    ::Box<dim,long int> ib = flip_box(box_int_recv.get(i).get(j).bx,box_int_recv.get(i).get(j).cmb,ginfo);
                    bid_t.g_e_box = ib;
                    bid_t.g_id = box_int_recv.get(i).get(j).g_id;
                    // Translate in local coordinate
                    Box<dim,long int> tb = ib;
                    tb -= gdb_ext.get(sub_id).origin;
                    bid_t.l_e_box = tb;
 
                    pib.bid.add(bid_t);
                    eb_gid_list.add();
                    eb_gid_list.last().eb_list.add(pib.bid.size()-1);
                    eb_gid_list.last().e_id = p_id;
                    eb_gid_list.last().full_match = pib.bid.size()-1;
 
                    g_id_to_external_ghost_box[bid_t.g_id] = eb_gid_list.size()-1;
                }
            }
        }
 
        init_e_g_box = true;
    }
 
    void create_local_ig_box()
    {
        openfpm::vector_std<result_box<dim>> ibv;
 
        // Get the grid info
        auto g = cd_sm.getGrid();
 
        if (init_local_i_g_box == true) return;
 
        // Get the number of sub-domains
        for (size_t i = 0 ; i < dec.getNSubDomain() ; i++)
        {
            loc_ig_box.add();
            auto&& pib = loc_ig_box.last();
 
            for (size_t j = 0 ; j < dec.getLocalNIGhost(i) ; j++)
            {
                if (use_bx_def == true)
                {
                    // Get the internal ghost boxes and transform into grid units
                    ::Box<dim,St> ib_dom = dec.getLocalIGhostBox(i,j);
                    ::Box<dim,long int> ib = cd_sm.convertDomainSpaceIntoGridUnits(ib_dom,dec.periodicity());
 
                    // Here we intersect the internal ghost box with the definition boxes
                    // this operation make sense when the grid is not defined in the full
                    // domain and we have to intersect the internal ghost box with all the box
                    // that define where the grid is defined
                    bx_intersect<dim>(bx_def,use_bx_def,ib,ibv);
 
                    for (size_t k = 0 ; k < ibv.size() ; k++)
                    {
                        // Check if ib is valid if not it mean that the internal ghost does not contain information so skip it
                        if (ibv.get(k).bx.isValid() == false)
                        {continue;}
 
                        pib.bid.add();
                        pib.bid.last().box = ibv.get(k).bx;
 
                        pib.bid.last().sub_gdb_ext = convert_to_gdb_ext(i,
                                                                ibv.get(k).id,
                                                                gdb_ext,
                                                                gdb_ext_markers);
 
                        pib.bid.last().sub = dec.getLocalIGhostSub(i,j);
 
                        // It will be filled later
                        pib.bid.last().cmb = dec.getLocalIGhostPos(i,j);
                    }
                }
                else
                {
                    // Get the internal ghost boxes and transform into grid units
                    ::Box<dim,St> ib_dom = dec.getLocalIGhostBox(i,j);
                    ::Box<dim,long int> ib = cd_sm.convertDomainSpaceIntoGridUnits(ib_dom,dec.periodicity());
 
                    // Check if ib is valid if not it mean that the internal ghost does not contain information so skip it
                    if (ib.isValid() == false)
                        continue;
 
                    size_t sub_id = i;
                    size_t r_sub = dec.getLocalIGhostSub(i,j);
 
                    set_for_adjustment(sub_id,dec.getSubDomain(r_sub),
                                       dec.getLocalIGhostPos(i,j),ib,ghost_int);
 
                    // Check if ib is valid if not it mean that the internal ghost does not contain information so skip it
                    if (ib.isValid() == false)
                    continue;
 
                    pib.bid.add();
                    pib.bid.last().box = ib;
                    pib.bid.last().sub = dec.getLocalIGhostSub(i,j);
                    pib.bid.last().k.add(dec.getLocalIGhostE(i,j));
                    pib.bid.last().cmb = dec.getLocalIGhostPos(i,j);
                    pib.bid.last().sub_gdb_ext = i;
                }
            }
 
            if (use_bx_def == true)
            {
                // unfortunately boxes that define where the grid is located can generate
                // additional internal ghost boxes
 
                for (size_t j = gdb_ext_markers.get(i) ; j < gdb_ext_markers.get(i+1) ; j++)
                {
                    // intersect within box in the save sub-domain
 
                    for (size_t k = gdb_ext_markers.get(i) ; k < gdb_ext_markers.get(i+1) ; k++)
                    {
                        if (j == k) {continue;}
 
                        // extend k and calculate the internal ghost box
                        Box<dim,long int> bx_e =  gdb_ext.get(k).GDbox;
                        bx_e += gdb_ext.get(k).origin;
                        Box<dim,long int> bx = gdb_ext.get(j).Dbox;
                        bx += gdb_ext.get(j).origin;
 
                        Box<dim,long int> output;
                        if (bx.Intersect(bx_e, output) == true)
                        {
                            pib.bid.add();
 
                            pib.bid.last().box = output;
 
                            pib.bid.last().sub_gdb_ext = j;
                            pib.bid.last().sub = i;
 
                            // these ghost always in the quadrant zero
                            pib.bid.last().cmb.zero();
 
                        }
                    }
                }
            }
 
        }
 
 
        init_local_i_g_box = true;
    }
 
    void create_local_eg_box()
    {
        // Get the grid info
        auto g = cd_sm.getGrid();
 
        if (init_local_e_g_box == true) return;
 
        loc_eg_box.resize(dec.getNSubDomain());
 
        // Get the number of sub-domain
        for (size_t i = 0 ; i < dec.getNSubDomain() ; i++)
        {
            for (size_t j = 0 ; j < loc_ig_box.get(i).bid.size() ; j++)
            {
                long int volume_linked = 0;
 
                size_t le_sub = loc_ig_box.get(i).bid.get(j).sub;
                auto & pib = loc_eg_box.get(le_sub);
 
                if (use_bx_def == true)
                {
 
                    // We check if an external local ghost box intersect one
                    // or more sub-grids
                    for (size_t k = 0 ; k < gdb_ext.size() ; k++)
                    {
                        Box<dim,long int> bx = gdb_ext.get(k).Dbox;
                        bx += gdb_ext.get(k).origin;
 
                        Box<dim,long int> gbx = gdb_ext.get(k).GDbox;
                        gbx += gdb_ext.get(k).origin;
 
                        Box<dim,long int> output;
                        Box<dim,long int> flp_i = flip_box(loc_ig_box.get(i).bid.get(j).box,loc_ig_box.get(i).bid.get(j).cmb,ginfo);
 
                        bool intersect_domain = bx.Intersect(flp_i,output);
                        bool intersect_gdomain = gbx.Intersect(flp_i,output);
 
                        // it intersect one sub-grid
                        if (intersect_domain == false && intersect_gdomain == true)
                        {
                            // fill the link variable
                            loc_ig_box.get(i).bid.get(j).k.add(pib.bid.size());
                            size_t s = loc_ig_box.get(i).bid.get(j).k.last();
 
                            comb<dim> cmb = loc_ig_box.get(i).bid.get(j).cmb;
                            cmb.sign_flip();
 
                            add_loc_eg_box(le_sub,
                                           loc_ig_box.get(i).bid.get(j).sub_gdb_ext,
                                           j,
                                           k,
                                           pib.bid,
                                           output,
                                           cmb);
 
 
                            volume_linked += pib.bid.last().ebox.getVolumeKey();
                        }
                    }
                }
                else
                {
                    size_t k = loc_ig_box.get(i).bid.get(j).sub;
                    auto & pib = loc_eg_box.get(k);
 
                    size_t s = loc_ig_box.get(i).bid.get(j).k.get(0);
                    pib.bid.resize(dec.getLocalNEGhost(k));
 
                    pib.bid.get(s).ebox = flip_box(loc_ig_box.get(i).bid.get(j).box,loc_ig_box.get(i).bid.get(j).cmb,ginfo);
                    pib.bid.get(s).sub = dec.getLocalEGhostSub(k,s);
                    pib.bid.get(s).cmb = loc_ig_box.get(i).bid.get(j).cmb;
                    pib.bid.get(s).cmb.sign_flip();
                    pib.bid.get(s).k = j;
                    pib.bid.get(s).initialized = true;
                    pib.bid.get(s).sub_gdb_ext = k;
                }
            }
        }
 
        init_local_e_g_box = true;
    }
 
 
    inline void check_size(const size_t (& g_sz)[dim])
    {
        for (size_t i = 0 ; i < dim ; i++)
        {
            if (g_sz[i] < 2)
            {std::cerr << "Error: " << __FILE__ << ":" << __LINE__ << " distributed grids with size smaller than 2 are not supported\n";}
        }
    }
 
    inline void check_domain(const Box<dim,St> & dom)
    {
        for (size_t i = 0 ; i < dim ; i++)
        {
            if (dom.getLow(i) >= dom.getHigh(i))
            {std::cerr << "Error: " << __FILE__ << ":" << __LINE__ << " error the simulation domain is invalid\n";}
        }
    }
 
    void Create(openfpm::vector<Box<dim,long int>> & bx_def,
                const Ghost<dim,long int> & g,
                bool use_bx_def)
    {
        // create gdb
        create_gdb_ext<dim,Decomposition>(gdb_ext,gdb_ext_markers,dec,cd_sm,bx_def,g,use_bx_def);
 
        size_t n_grid = gdb_ext.size();
 
        // create local grids for each hyper-cube
        loc_grid.resize(n_grid);
 
        // Size of the grid on each dimension
        size_t l_res[dim];
 
        // Allocate the grids
        for (size_t i = 0 ; i < n_grid ; i++)
        {
            SpaceBox<dim,long int> sp_tg = gdb_ext.get(i).GDbox;
 
            // Get the size of the local grid
            // The boxes indicate the extension of the index the size
            // is this extension +1
            // for example a 1D box (interval) from 0 to 3 in one dimension have
            // the points 0,1,2,3 = so a total of 4 points
            for (size_t j = 0 ; j < dim ; j++)
            {l_res[j] = (sp_tg.getHigh(j) >= 0)?(sp_tg.getHigh(j)+1):0;}
 
            // Set the dimensions of the local grid
            loc_grid.get(i).resize(l_res);
        }
    }
 
 
    inline void InitializeCellDecomposer(const CellDecomposer_sm<dim,St,shift<dim,St>> & cd_old, const Box<dim,size_t> & ext)
    {
        // Initialize the cell decomposer
        cd_sm.setDimensions(cd_old,ext);
    }
 
    inline void InitializeCellDecomposer(const size_t (& g_sz)[dim], const size_t (& bc)[dim])
    {
        // check that the grid has valid size
        check_size(g_sz);
 
        // get the size of the cell decomposer
        size_t c_g[dim];
        getCellDecomposerPar<dim>(c_g,g_sz,bc);
 
        // Initialize the cell decomposer
        cd_sm.setDimensions(domain,c_g,0);
    }
 
    inline void InitializeDecomposition(const size_t (& g_sz)[dim], const size_t (& bc)[dim], const grid_sm<dim,void> & g_dist = grid_sm<dim,void>())
    {
        // fill the global size of the grid
        for (size_t i = 0 ; i < dim ; i++)  {this->g_sz[i] = g_sz[i];}
 
        // Get the number of processor and calculate the number of sub-domain
        // for decomposition
        size_t n_proc = v_cl.getProcessingUnits();
        size_t n_sub = n_proc * v_sub_unit_factor;
 
        // Calculate the maximum number (before merging) of sub-domain on
        // each dimension
        size_t div[dim];
        for (size_t i = 0 ; i < dim ; i++)
        {div[i] = openfpm::math::round_big_2(pow(n_sub,1.0/dim));}
 
        if (g_dist.size(0) != 0)
        {
            for (size_t i = 0 ; i < dim ; i++)
            {div[i] = g_dist.size(i);}
        }
 
        // Create the sub-domains
        dec.setParameters(div,domain,bc,ghost);
        dec.decompose(dec_options::DEC_SKIP_ICELL);
    }
 
    inline void InitializeStructures(const size_t (& g_sz)[dim])
    {
        // an empty
        openfpm::vector<Box<dim,long int>> empty;
 
        // Ghost zero
        Ghost<dim,long int> zero;
 
        InitializeStructures(g_sz,empty,zero,false);
    }
 
    inline void InitializeStructures(const size_t (& g_sz)[dim],
                                     openfpm::vector<Box<dim,long int>> & bx,
                                     const Ghost<dim,long int> & g,
                                     bool use_bx_def)
    {
        // fill the global size of the grid
        for (size_t i = 0 ; i < dim ; i++)  {this->g_sz[i] = g_sz[i];}
 
        // Create local grid
        Create(bx,g,use_bx_def);
    }
 
    // Ghost as integer
    Ghost<dim,long int> gint = Ghost<dim,long int>(0);
 
protected:
 
    Box<dim,size_t> getDomain(size_t i)
    {
        return gdb_ext.get(i).Dbox;
    }
 
    static inline Ghost<dim,St> convert_ghost(const Ghost<dim,long int> & gd, const CellDecomposer_sm<dim,St,shift<dim,St>> & cd_sm)
    {
        Ghost<dim,St> gc;
 
        // get the grid spacing
        Box<dim,St> sp = cd_sm.getCellBox();
 
        // enlarge 0.001 of the spacing
        sp.magnify_fix_P1(1.1);
 
        // set the ghost
        for (size_t i = 0 ; i < dim ; i++)
        {
            gc.setLow(i,gd.getLow(i)*(sp.getHigh(i)));
            gc.setHigh(i,gd.getHigh(i)*(sp.getHigh(i)));
        }
 
        return gc;
    }
 
    void setDecompositionGranularity(size_t n_sub)
    {
        this->v_sub_unit_factor = n_sub;
    }
 
    void reset_ghost_structures()
    {
        g_id_to_internal_ghost_box.clear();
        ig_box.clear();
        init_i_g_box = false;
 
        eg_box.clear();
        eb_gid_list.clear();
        init_e_g_box = false;
 
        init_local_i_g_box = false;
        loc_ig_box.clear();
 
        init_local_e_g_box = false;
        loc_eg_box.clear();
    }
 
public:
 
    typedef device_grid d_grid;
 
    typedef Decomposition decomposition;
 
    typedef T value_type;
 
    typedef St stype;
 
    typedef Memory memory_type;
 
    typedef device_grid device_grid_type;
 
    static const unsigned int dims = dim;
 
    inline const Box<dim,St> getDomain() const
    {
        return domain;
    }
 
    Point<dim,St> getOffset(size_t i)
    {
        return pmul(Point<dim,St>(gdb_ext.get(i).origin), cd_sm.getCellBox().getP2()) + getDomain().getP1();
    }
 
    inline St spacing(size_t i) const
    {
        return cd_sm.getCellBox().getHigh(i);
    }
 
    size_t size() const
    {
        return ginfo_v.size();
    }
 
    void setBackgroundValue(T & bv)
    {
        setBackground_impl<T,decltype(loc_grid)> func(bv,loc_grid);
        boost::mpl::for_each_ref< boost::mpl::range_c<int,0,T::max_prop>>(func);
    }
 
    template<unsigned int p>
    void setBackgroundValue(const typename boost::mpl::at<typename T::type,boost::mpl::int_<p>>::type & bv)
    {
        for (size_t i = 0 ; i < loc_grid.size() ; i++)
        {loc_grid.get(i).template setBackgroundValue<p>(bv);}
    }
 
    size_t size_local_inserted() const
    {
        size_t lins = 0;
 
        for (size_t i = 0 ; i < loc_grid.size() ; i++)
        {lins += loc_grid.get(i).size_inserted();}
 
        return lins;
    }
 
    size_t size(size_t i) const
    {
        return ginfo_v.size(i);
    }
 
    grid_dist_id(const grid_dist_id<dim,St,T,Decomposition,Memory,device_grid> & g)
    :grid_dist_id_comm<dim,St,T,Decomposition,Memory,device_grid>(g),
     domain(g.domain),
     ghost(g.ghost),
     ghost_int(g.ghost_int),
     loc_grid(g.loc_grid),
     loc_grid_old(g.loc_grid_old),
     dec(g.dec),
     gdb_ext_markers(g.gdb_ext_markers),
     gdb_ext(g.gdb_ext),
     gdb_ext_global(g.gdb_ext_global),
     gdb_ext_old(g.gdb_ext_old),
     cd_sm(g.cd_sm),
     v_cl(g.v_cl),
     prp_names(g.prp_names),
     g_id_to_external_ghost_box(g.g_id_to_external_ghost_box),
     g_id_to_internal_ghost_box(g.g_id_to_internal_ghost_box),
     ginfo(g.ginfo),
     ginfo_v(g.ginfo_v),
     init_local_i_g_box(g.init_local_i_g_box),
     init_local_e_g_box(g.init_local_e_g_box)
    {
#ifdef SE_CLASS2
        check_new(this,8,GRID_DIST_EVENT,4);
#endif
 
        for (size_t i = 0 ; i < dim ; i++)
        {g_sz[i] = g.g_sz[i];}
    }
 
    template<typename H>
    grid_dist_id(const grid_dist_id<dim,St,H,typename Decomposition::base_type,Memory,grid_cpu<dim,H>> & g,
                 const Ghost<dim,long int> & gh,
                 Box<dim,size_t> ext)
    :ghost_int(gh),dec(create_vcluster()),v_cl(create_vcluster())
    {
#ifdef SE_CLASS2
        check_new(this,8,GRID_DIST_EVENT,4);
#endif
 
        size_t ext_dim[dim];
        for (size_t i = 0 ; i < dim ; i++) {ext_dim[i] = g.getGridInfoVoid().size(i) + ext.getKP1().get(i) + ext.getKP2().get(i);}
 
        // Set the grid info of the extended grid
        ginfo.setDimensions(ext_dim);
        ginfo_v.setDimensions(ext_dim);
 
        InitializeCellDecomposer(g.getCellDecomposer(),ext);
 
        ghost = convert_ghost(gh,cd_sm);
 
        // Extend the grid by the extension part and calculate the domain
 
        for (size_t i = 0 ; i < dim ; i++)
        {
            g_sz[i] = g.size(i) + ext.getLow(i) + ext.getHigh(i);
 
            if (g.getDecomposition().periodicity(i) == NON_PERIODIC)
            {
                this->domain.setLow(i,g.getDomain().getLow(i) - ext.getLow(i) * g.spacing(i) - g.spacing(i) / 2.0);
                this->domain.setHigh(i,g.getDomain().getHigh(i) + ext.getHigh(i) * g.spacing(i) + g.spacing(i) / 2.0);
            }
            else
            {
                this->domain.setLow(i,g.getDomain().getLow(i) - ext.getLow(i) * g.spacing(i));
                this->domain.setHigh(i,g.getDomain().getHigh(i) + ext.getHigh(i) * g.spacing(i));
            }
        }
 
        dec.setParameters(g.getDecomposition(),ghost,this->domain);
 
        // an empty
        openfpm::vector<Box<dim,long int>> empty;
 
        InitializeStructures(g.getGridInfoVoid().getSize(),empty,gh,false);
    }
 
    template<typename Decomposition2>
    grid_dist_id(const Decomposition2 & dec,
                 const size_t (& g_sz)[dim],
                 const Ghost<dim,St> & ghost)
    :domain(dec.getDomain()),ghost(ghost),ghost_int(INVALID_GHOST),dec(create_vcluster()),v_cl(create_vcluster()),
     ginfo(g_sz),ginfo_v(g_sz)
    {
#ifdef SE_CLASS2
        check_new(this,8,GRID_DIST_EVENT,4);
#endif
 
        InitializeCellDecomposer(g_sz,dec.periodicity());
 
        this->dec = dec.duplicate(ghost);
 
        InitializeStructures(g_sz);
    }
 
    grid_dist_id(Decomposition && dec, const size_t (& g_sz)[dim],
                 const Ghost<dim,St> & ghost)
    :domain(dec.getDomain()),ghost(ghost),dec(dec),ginfo(g_sz),
     ginfo_v(g_sz),v_cl(create_vcluster()),ghost_int(INVALID_GHOST)
    {
#ifdef SE_CLASS2
        check_new(this,8,GRID_DIST_EVENT,4);
#endif
 
        InitializeCellDecomposer(g_sz,dec.periodicity());
 
        this->dec = dec.duplicate(ghost);
 
        InitializeStructures(g_sz);
    }
 
    grid_dist_id(const Decomposition & dec, const size_t (& g_sz)[dim],
                 const Ghost<dim,long int> & g)
    :domain(dec.getDomain()),ghost_int(g),dec(create_vcluster()),v_cl(create_vcluster()),
     ginfo(g_sz),ginfo_v(g_sz)
    {
        InitializeCellDecomposer(g_sz,dec.periodicity());
 
        ghost = convert_ghost(g,cd_sm);
        this->dec = dec.duplicate(ghost);
 
        // an empty
        openfpm::vector<Box<dim,long int>> empty;
 
        // Initialize structures
        InitializeStructures(g_sz,empty,g,false);
    }
 
    template<typename Decomposition2>
    grid_dist_id(const Decomposition2 & dec, const size_t (& g_sz)[dim],
                 const Ghost<dim,long int> & g)
    :domain(dec.getDomain()),ghost_int(g),dec(create_vcluster()),v_cl(create_vcluster()),
     ginfo(g_sz),ginfo_v(g_sz)
    {
        InitializeCellDecomposer(g_sz,dec.periodicity());
 
        ghost = convert_ghost(g,cd_sm);
        this->dec = dec.duplicate(ghost);
 
        // an empty
        openfpm::vector<Box<dim,long int>> empty;
 
        // Initialize structures
        InitializeStructures(g_sz,empty,g,false);
    }
 
    grid_dist_id(Decomposition && dec, const size_t (& g_sz)[dim],
                 const Ghost<dim,long int> & g)
    :domain(dec.getDomain()),dec(dec),v_cl(create_vcluster()),ginfo(g_sz),
     ginfo_v(g_sz),ghost_int(g)
    {
#ifdef SE_CLASS2
        check_new(this,8,GRID_DIST_EVENT,4);
#endif
        InitializeCellDecomposer(g_sz,dec.periodicity());
 
        ghost = convert_ghost(g,cd_sm);
        this->dec = dec.duplicate(ghost);
 
        // an empty
        openfpm::vector<Box<dim,long int>> empty;
 
        // Initialize structures
        InitializeStructures(g_sz,empty,g,false);
    }
 
    grid_dist_id(const size_t (& g_sz)[dim],const Box<dim,St> & domain, const Ghost<dim,St> & g, size_t opt = 0)
    :grid_dist_id(g_sz,domain,g,create_non_periodic<dim>(),opt)
    {
    }
 
    grid_dist_id(const size_t (& g_sz)[dim],const Box<dim,St> & domain, const Ghost<dim,long int> & g, size_t opt = 0)
    :grid_dist_id(g_sz,domain,g,create_non_periodic<dim>(),opt)
    {
    }
 
    grid_dist_id(const size_t (& g_sz)[dim],const Box<dim,St> & domain, const Ghost<dim,St> & g, const periodicity<dim> & p, size_t opt = 0)
    :domain(domain),ghost(g),ghost_int(INVALID_GHOST),dec(create_vcluster()),v_cl(create_vcluster()),ginfo(g_sz),ginfo_v(g_sz)
    {
#ifdef SE_CLASS2
        check_new(this,8,GRID_DIST_EVENT,4);
#endif
 
        if (opt >> 32 != 0)
        {this->setDecompositionGranularity(opt >> 32);}
 
        check_domain(domain);
        InitializeCellDecomposer(g_sz,p.bc);
        InitializeDecomposition(g_sz, p.bc);
        InitializeStructures(g_sz);
    }
 
 
    grid_dist_id(const size_t (& g_sz)[dim],const Box<dim,St> & domain, const Ghost<dim,long int> & g, const periodicity<dim> & p, size_t opt = 0, const grid_sm<dim,void> & g_dec = grid_sm<dim,void>())
    :domain(domain),ghost_int(g),dec(create_vcluster()),v_cl(create_vcluster()),ginfo(g_sz),ginfo_v(g_sz)
    {
#ifdef SE_CLASS2
        check_new(this,8,GRID_DIST_EVENT,4);
#endif
 
        if (opt >> 32 != 0)
        {this->setDecompositionGranularity(opt >> 32);}
 
        check_domain(domain);
        InitializeCellDecomposer(g_sz,p.bc);
 
        ghost = convert_ghost(g,cd_sm);
 
        InitializeDecomposition(g_sz,p.bc,g_dec);
 
        // an empty
        openfpm::vector<Box<dim,long int>> empty;
 
        // Initialize structures
        InitializeStructures(g_sz,empty,g,false);
    }
 
 
    grid_dist_id(const size_t (& g_sz)[dim],
                 const Box<dim,St> & domain,
                 const Ghost<dim,long int> & g,
                 const periodicity<dim> & p,
                 openfpm::vector<Box<dim,long int>> & bx_def)
    :domain(domain),dec(create_vcluster()),v_cl(create_vcluster()),ginfo(g_sz),ginfo_v(g_sz),gint(g)
    {
#ifdef SE_CLASS2
        check_new(this,8,GRID_DIST_EVENT,4);
#endif
 
        check_domain(domain);
        InitializeCellDecomposer(g_sz,p.bc);
 
        ghost = convert_ghost(g,cd_sm);
 
        InitializeDecomposition(g_sz, p.bc);
        InitializeStructures(g_sz,bx_def,g,true);
        this->bx_def = bx_def;
        this->use_bx_def = true;
    }
 
    const grid_sm<dim,T> & getGridInfo() const
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        return ginfo;
    }
 
    const grid_sm<dim,void> & getGridInfoVoid() const
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        return ginfo_v;
    }
 
    Decomposition & getDecomposition()
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        return dec;
    }
 
    const Decomposition & getDecomposition() const
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        return dec;
    }
 
    const CellDecomposer_sm<dim,St,shift<dim,St>> & getCellDecomposer() const
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        return cd_sm;
    }
 
    bool isInside(const grid_key_dx<dim> & gk) const
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        for (size_t i = 0 ; i < dim ; i++)
        {
            if (gk.get(i) < 0 || gk.get(i) >= (long int)g_sz[i])
            {return false;}
        }
 
        return true;
    }
 
    size_t getLocalDomainSize() const
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        size_t total = 0;
 
        for (size_t i = 0 ; i < gdb_ext.size() ; i++)
        {
            total += gdb_ext.get(i).Dbox.getVolumeKey();
        }
 
        return total;
    }
 
    size_t getLocalDomainWithGhostSize() const
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        size_t total = 0;
 
        for (size_t i = 0 ; i < gdb_ext.size() ; i++)
        {
            total += gdb_ext.get(i).GDbox.getVolumeKey();
        }
 
        return total;
    }
 
 
    const openfpm::vector<GBoxes<device_grid::dims>> & getLocalGridsInfo() const
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        return gdb_ext;
    }
 
    void getGlobalGridsInfo(openfpm::vector<GBoxes<device_grid::dims>> & gdb_ext_global) const
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        gdb_ext_global.clear();
 
        v_cl.SGather(gdb_ext,gdb_ext_global,0);
        v_cl.execute();
 
        size_t size_r;
        size_t size = gdb_ext_global.size();
 
        if (v_cl.getProcessUnitID()  == 0)
        {
            for (size_t i = 1; i < v_cl.getProcessingUnits(); i++)
                v_cl.send(i,0,&size,sizeof(size_t));
 
            size_r = size;
        }
        else
            v_cl.recv(0,0,&size_r,sizeof(size_t));
 
        v_cl.execute();
 
        gdb_ext_global.resize(size_r);
 
 
        if (v_cl.getProcessUnitID()  == 0)
        {
            for (size_t i = 1; i < v_cl.getProcessingUnits(); i++)
                v_cl.send(i,0,gdb_ext_global);
        }
        else
            v_cl.recv(0,0,gdb_ext_global);
 
        v_cl.execute();
    }
 
 
    grid_dist_iterator<dim,device_grid,
                       decltype(device_grid::type_of_subiterator()),
                       FREE>
    getOldDomainIterator() const
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
 
        grid_key_dx<dim> stop(ginfo_v.getSize());
        grid_key_dx<dim> one;
        one.one();
        stop = stop - one;
 
        grid_dist_iterator<dim,device_grid,
                            decltype(device_grid::type_of_subiterator()),
                            FREE> it(loc_grid_old,gdb_ext_old,stop);
 
        return it;
    }
 
    inline grid_dist_id_iterator_dec<Decomposition> getGridIterator(const grid_key_dx<dim> & start, const grid_key_dx<dim> & stop)
    {
        grid_dist_id_iterator_dec<Decomposition> it_dec(getDecomposition(), g_sz, start, stop);
        return it_dec;
    }
 
#ifdef __NVCC__
 
    template<typename lambda_t2>
    void setPoints(lambda_t2 f2)
    {
        auto it = getGridIteratorGPU();
 
        it.template launch<1>(launch_set_dense<dim>(),f2);
    }
 
    template<typename lambda_t2>
    void setPoints(grid_key_dx<dim> k1, grid_key_dx<dim> k2, lambda_t2 f2)
    {
        auto it = getGridIteratorGPU(k1,k2);
 
        it.template launch<0>(launch_set_dense<dim>(),f2);
    }
 
    template<typename lambda_t1, typename lambda_t2>
    void addPoints(lambda_t1 f1, lambda_t2 f2)
    {
        auto it = getGridIteratorGPU();
        it.setGPUInsertBuffer(1);
 
        it.template launch<1>(launch_insert_sparse(),f1,f2);
    }
 
    template<typename lambda_t1, typename lambda_t2>
    void addPoints(grid_key_dx<dim> k1, grid_key_dx<dim> k2, lambda_t1 f1, lambda_t2 f2)
    {
        auto it = getGridIteratorGPU(k1,k2);
        it.setGPUInsertBuffer(1);
 
        it.template launch<1>(launch_insert_sparse(),f1,f2);
    }
 
    inline grid_dist_id_iterator_gpu<Decomposition,openfpm::vector<device_grid>>
    getGridIteratorGPU(const grid_key_dx<dim> & start, const grid_key_dx<dim> & stop)
    {
        grid_dist_id_iterator_gpu<Decomposition,openfpm::vector<device_grid>> it_dec(loc_grid,getDecomposition(), g_sz, start, stop);
        return it_dec;
    }
 
    inline grid_dist_id_iterator_gpu<Decomposition,openfpm::vector<device_grid>>
    getGridIteratorGPU()
    {
        grid_dist_id_iterator_gpu<Decomposition,openfpm::vector<device_grid>> it_dec(loc_grid,getDecomposition(), g_sz);
        return it_dec;
    }
 
#endif
 
    inline grid_dist_id_iterator_dec<Decomposition,true> getGridGhostIterator(const grid_key_dx<dim> & start, const grid_key_dx<dim> & stop)
    {
        grid_dist_id_iterator_dec<Decomposition,true> it_dec(getDecomposition(), g_sz, start, stop);
        return it_dec;
    }
 
    inline grid_dist_id_iterator_dec<Decomposition> getGridIterator()
    {
        grid_key_dx<dim> start;
        grid_key_dx<dim> stop;
        for (size_t i = 0; i < dim; i++)
        {
            start.set_d(i, 0);
            stop.set_d(i, g_sz[i] - 1);
        }
 
        grid_dist_id_iterator_dec<Decomposition> it_dec(getDecomposition(), g_sz, start, stop);
        return it_dec;
    }
 
    grid_dist_iterator<dim,device_grid,
                       decltype(device_grid::type_of_subiterator()),FREE>
    getDomainIterator() const
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
 
        grid_key_dx<dim> stop(ginfo_v.getSize());
        grid_key_dx<dim> one;
        one.one();
        stop = stop - one;
 
        grid_dist_iterator<dim,device_grid,
                            decltype(device_grid::type_of_subiterator()),
                            FREE> it(loc_grid,gdb_ext,stop);
 
        return it;
    }
 
    template<unsigned int Np>
    grid_dist_iterator<dim,device_grid,
                        decltype(device_grid::template type_of_subiterator<stencil_offset_compute<dim,Np>>()),
                        FREE,
                        stencil_offset_compute<dim,Np> >
    getDomainIteratorStencil(const grid_key_dx<dim> (& stencil_pnt)[Np]) const
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
 
        grid_key_dx<dim> stop(ginfo_v.getSize());
        grid_key_dx<dim> one;
        one.one();
        stop = stop - one;
 
        grid_dist_iterator<dim,device_grid,
                           decltype(device_grid::template type_of_subiterator<stencil_offset_compute<dim,Np>>()),
                           FREE,
                           stencil_offset_compute<dim,Np>> it(loc_grid,gdb_ext,stop,stencil_pnt);
 
        return it;
    }
 
    grid_dist_iterator<dim,device_grid,
    decltype(device_grid::type_of_iterator()),
    FIXED>
    getDomainGhostIterator() const
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        grid_key_dx<dim> stop;
        for (size_t i = 0 ; i < dim ; i++)
        {stop.set_d(i,0);}
 
        grid_dist_iterator<dim,device_grid,
                            decltype(device_grid::type_of_iterator()),
                            FIXED> it(loc_grid,gdb_ext,stop);
 
        return it;
    }
 
    grid_dist_iterator_sub<dim,device_grid>
    getSubDomainIterator(const grid_key_dx<dim> & start,
                         const grid_key_dx<dim> & stop) const
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        grid_dist_iterator_sub<dim,device_grid> it(start,stop,loc_grid,gdb_ext);
 
        return it;
    }
 
    grid_dist_iterator_sub<dim,device_grid> getSubDomainIterator(const long int (& start)[dim], const long int (& stop)[dim]) const
    {
        grid_dist_iterator_sub<dim,device_grid> it(grid_key_dx<dim>(start),grid_key_dx<dim>(stop),loc_grid,gdb_ext);
 
        return it;
    }
 
    ~grid_dist_id()
    {
#ifdef SE_CLASS2
        check_delete(this);
#endif
        dec.decRef();
    }
 
    Vcluster<> & getVC()
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        return v_cl;
    }
 
    void removeUnusedBuffers()
    {
        for (int i = 0 ; i < loc_grid.size() ; i++)
        {
            loc_grid.get(i).removeUnusedBuffers();
        }
    }
 
    bool is_staggered() const
    {
        return false;
    }
 
    template <typename bg_key> inline void remove(const grid_dist_key_dx<dim,bg_key> & v1)
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        return loc_grid.get(v1.getSub()).remove(v1.getKey());
    }
 
    template <typename bg_key> inline void remove_no_flush(const grid_dist_key_dx<dim,bg_key> & v1)
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        return loc_grid.get(v1.getSub()).remove_no_flush(v1.getKey());
    }
 
 
    template<typename ... v_reduce>
    void flush(flush_type opt = flush_type::FLUSH_ON_HOST)
    {
        for (size_t i = 0 ; i < loc_grid.size() ; i++)
        {
            loc_grid.get(i).template flush<v_reduce ...>(v_cl.getgpuContext(),opt);
        }
    }
 
    inline void flush_remove()
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        for (size_t i = 0 ; i < loc_grid.size() ; i++)
        {loc_grid.get(i).flush_remove();}
    }
 
    template <unsigned int p,typename bg_key>inline auto insert(const grid_dist_key_dx<dim,bg_key> & v1)
    -> decltype(loc_grid.get(v1.getSub()).template insert<p>(v1.getKey()))
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
 
        return loc_grid.get(v1.getSub()).template insert<p>(v1.getKey());
    }
 
 
    template <unsigned int p,typename bg_key>inline auto insertFlush(const grid_dist_key_dx<dim,bg_key> & v1)
    -> decltype(loc_grid.get(v1.getSub()).template insertFlush<p>(v1.getKey()))
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
 
        return loc_grid.get(v1.getSub()).template insertFlush<p>(v1.getKey());
    }
 
    template <typename bg_key>inline auto insertFlush(const grid_dist_key_dx<dim,bg_key> & v1)
    -> decltype(loc_grid.get(v1.getSub()).template insertFlush(v1.getKey()))
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
 
        return loc_grid.get(v1.getSub()).template insertFlush(v1.getKey());
    }
 
    template <unsigned int p, typename bg_key>
    inline auto get(const grid_dist_key_dx<dim,bg_key> & v1) const
    -> typename std::add_lvalue_reference<decltype(loc_grid.get(v1.getSub()).template get<p>(v1.getKey()))>::type
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        return loc_grid.get(v1.getSub()).template get<p>(v1.getKey());
    }
 
 
    template <unsigned int p, typename bg_key>
    inline auto get(const grid_dist_key_dx<dim,bg_key> & v1)
    -> decltype(loc_grid.get(v1.getSub()).template get<p>(v1.getKey()))
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        return loc_grid.get(v1.getSub()).template get<p>(v1.getKey());
    }
 
    template <unsigned int p = 0>
    inline auto get(const grid_dist_g_dx<device_grid> & v1) const
    -> decltype(v1.getSub()->template get<p>(v1.getKey()))
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        return v1.getSub()->template get<p>(v1.getKey());
    }
 
    template <unsigned int p = 0>
    inline auto get(const grid_dist_g_dx<device_grid> & v1) -> decltype(v1.getSub()->template get<p>(v1.getKey()))
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        return v1.getSub()->template get<p>(v1.getKey());
    }
 
    template <unsigned int p = 0>
    inline auto get(const grid_dist_lin_dx & v1) const -> decltype(loc_grid.get(v1.getSub()).template get<p>(v1.getKey()))
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        return loc_grid.get(v1.getSub()).template get<p>(v1.getKey());
    }
 
    template <unsigned int p = 0>
    inline auto get(const grid_dist_lin_dx & v1) -> decltype(loc_grid.get(v1.getSub()).template get<p>(v1.getKey()))
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        return loc_grid.get(v1.getSub()).template get<p>(v1.getKey());
    }
 
    template <typename bg_key>
    inline Point<dim,St> getPos(const grid_dist_key_dx<dim,bg_key> & v1)
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        Point<dim,St> p;
 
        for (int i = 0 ; i < dim ; i++)
        {
            p.get(i) = (gdb_ext.get(v1.getSub()).origin.get(i) + v1.getKeyRef().get(i)) * this->spacing(i) + domain.getLow(i);
        }
 
        return p;
    }
 
    template<typename bg_key>
    inline bool existPoint(const grid_dist_key_dx<dim,bg_key> & v1) const
    {
        return loc_grid.get(v1.getSub()).existPoint(v1.getKey());
    }
 
    template <unsigned int p = 0, typename bgkey>
    inline auto getProp(const grid_dist_key_dx<dim,bgkey> & v1) const -> decltype(this->template get<p>(v1))
    {
        return this->template get<p>(v1);
    }
 
    template <unsigned int p = 0, typename bgkey>
    inline auto getProp(const grid_dist_key_dx<dim,bgkey> & v1) -> decltype(this->template get<p>(v1))
    {
        return this->template get<p>(v1);
    }
 
    template<int... prp> void ghost_get(size_t opt = 0)
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
 
        // Convert the ghost  internal boxes into grid unit boxes
        create_ig_box();
 
        // Convert the ghost external boxes into grid unit boxes
        create_eg_box();
 
        // Convert the local ghost internal boxes into grid unit boxes
        create_local_ig_box();
 
        // Convert the local external ghost boxes into grid unit boxes
        create_local_eg_box();
 
        grid_dist_id_comm<dim,St,T,Decomposition,Memory,device_grid>::template ghost_get_<prp...>(ig_box,
                                                                                                  eg_box,
                                                                                                  loc_ig_box,
                                                                                                  loc_eg_box,
                                                                                                  gdb_ext,
                                                                                                  eb_gid_list,
                                                                                                  use_bx_def,
                                                                                                  loc_grid,
                                                                                                  ginfo_v,
                                                                                                  g_id_to_external_ghost_box,
                                                                                                  opt);
    }
 
    template<template<typename,typename> class op,int... prp> void ghost_put()
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
 
        // Convert the ghost  internal boxes into grid unit boxes
        create_ig_box();
 
        // Convert the ghost external boxes into grid unit boxes
        create_eg_box();
 
        // Convert the local ghost internal boxes into grid unit boxes
        create_local_ig_box();
 
        // Convert the local external ghost boxes into grid unit boxes
        create_local_eg_box();
 
        grid_dist_id_comm<dim,St,T,Decomposition,Memory,device_grid>::template ghost_put_<op,prp...>(dec,
                                                                                                     ig_box,
                                                                                                     eg_box,
                                                                                                     loc_ig_box,
                                                                                                     loc_eg_box,
                                                                                                     gdb_ext,
                                                                                                     loc_grid,
                                                                                                     g_id_to_internal_ghost_box);
    }
 
 
    grid_dist_id<dim,St,T,Decomposition,Memory,device_grid> & copy(grid_dist_id<dim,St,T,Decomposition,Memory,device_grid> & g, bool use_memcpy = true)
    {
        if (T::noPointers() == true && use_memcpy)
        {
            for (size_t i = 0 ; i < this->getN_loc_grid() ; i++)
            {
                auto & gs_src = this->get_loc_grid(i).getGrid();
 
                long int start = gs_src.LinId(gdb_ext.get(i).Dbox.getKP1());
                long int stop = gs_src.LinId(gdb_ext.get(i).Dbox.getKP2());
 
                if (stop < start) {continue;}
 
                void * dst = static_cast<void *>(static_cast<char *>(this->get_loc_grid(i).getPointer()) + start*sizeof(T));
                void * src = static_cast<void *>(static_cast<char *>(g.get_loc_grid(i).getPointer()) + start*sizeof(T));
 
                memcpy(dst,src,sizeof(T) * (stop + 1 - start));
            }
        }
        else
        {
            grid_key_dx<dim> cnt[1];
            cnt[0].zero();
 
            for (size_t i = 0 ; i < this->getN_loc_grid() ; i++)
            {
                auto & dst = this->get_loc_grid(i);
                auto & src = g.get_loc_grid(i);
 
                auto it = this->get_loc_grid_iterator_stencil(i,cnt);
 
                while (it.isNext())
                {
                    // center point
                    auto Cp = it.template getStencil<0>();
 
                    dst.insert_o(Cp) = src.get_o(Cp);
 
                    ++it;
                }
            }
        }
 
        return *this;
    }
 
    grid_dist_id<dim,St,T,Decomposition,Memory,device_grid> & copy_sparse(grid_dist_id<dim,St,T,Decomposition,Memory,device_grid> & g, bool use_memcpy = true)
    {
        grid_key_dx<dim> cnt[1];
        cnt[0].zero();
 
        for (size_t i = 0 ; i < this->getN_loc_grid() ; i++)
        {
            auto & dst = this->get_loc_grid(i);
            auto & src = g.get_loc_grid(i);
 
            dst = src;
        }
        return *this;
    }
 
    Point<dim,St> getSpacing()
    {
        return cd_sm.getCellBox().getP2();
    }
 
    inline grid_key_dx<dim> getGKey(const grid_dist_key_dx<dim> & k) const
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        // Get the sub-domain id
        size_t sub_id = k.getSub();
 
        grid_key_dx<dim> k_glob = k.getKey();
 
        // shift
        k_glob = k_glob + gdb_ext.get(sub_id).origin;
 
        return k_glob;
    }
 
    template <typename Model>inline void addComputationCosts(Model md=Model(), size_t ts = 1)
    {
        CellDecomposer_sm<dim, St, shift<dim,St>> cdsm;
 
        Decomposition & dec = getDecomposition();
        auto & dist = getDecomposition().getDistribution();
 
        cdsm.setDimensions(dec.getDomain(), dec.getDistGrid().getSize(), 0);
 
        // Invert the id to positional
 
        Point<dim,St> p;
        for (size_t i = 0; i < dist.getNOwnerSubSubDomains() ; i++)
        {
            dist.getSubSubDomainPos(i,p);
            dec.setSubSubDomainComputationCost(dist.getOwnerSubSubDomain(i) , 1 + md.resolution(p));
        }
 
        dec.computeCommunicationAndMigrationCosts(ts);
 
        dist.setDistTol(md.distributionTol());
    }
 
    template<unsigned int prop_src, unsigned int prop_dst, unsigned int stencil_size, unsigned int N, typename lambda_f, typename ... ArgsT >
    void conv(int (& stencil)[N][dim], grid_key_dx<3> start, grid_key_dx<3> stop , lambda_f func, ArgsT ... args)
    {
        for (int i = 0 ; i < loc_grid.size() ; i++)
        {
            Box<dim,long int> inte;
 
            Box<dim,long int> base;
            for (int j = 0 ; j < dim ; j++)
            {
                base.setLow(j,(long int)start.get(j) - (long int)gdb_ext.get(i).origin.get(j));
                base.setHigh(j,(long int)stop.get(j) - (long int)gdb_ext.get(i).origin.get(j));
            }
 
            Box<dim,long int> dom = gdb_ext.get(i).Dbox;
 
            bool overlap = dom.Intersect(base,inte);
 
            if (overlap == true)
            {
                loc_grid.get(i).template conv<prop_src,prop_dst,stencil_size>(stencil,inte.getKP1(),inte.getKP2(),func,args...);
            }
        }
    }
 
    template<unsigned int prop_src, unsigned int prop_dst, unsigned int stencil_size, typename lambda_f, typename ... ArgsT >
    void conv_cross(grid_key_dx<3> start, grid_key_dx<3> stop , lambda_f func, ArgsT ... args)
    {
        for (int i = 0 ; i < loc_grid.size() ; i++)
        {
            Box<dim,long int> inte;
 
            Box<dim,long int> base;
            for (int j = 0 ; j < dim ; j++)
            {
                base.setLow(j,(long int)start.get(j) - (long int)gdb_ext.get(i).origin.get(j));
                base.setHigh(j,(long int)stop.get(j) - (long int)gdb_ext.get(i).origin.get(j));
            }
 
            Box<dim,long int> dom = gdb_ext.get(i).Dbox;
 
            bool overlap = dom.Intersect(base,inte);
 
            if (overlap == true)
            {
                loc_grid.get(i).template conv_cross<prop_src,prop_dst,stencil_size>(inte.getKP1(),inte.getKP2(),func,args...);
            }
        }
    }
 
    template<unsigned int prop_src, unsigned int prop_dst, unsigned int stencil_size, typename lambda_f, typename ... ArgsT >
    void conv_cross_b(grid_key_dx<3> start, grid_key_dx<3> stop , lambda_f func, ArgsT ... args)
    {
        for (int i = 0 ; i < loc_grid.size() ; i++)
        {
            Box<dim,long int> inte;
 
            Box<dim,long int> base;
            for (int j = 0 ; j < dim ; j++)
            {
                base.setLow(j,(long int)start.get(j) - (long int)gdb_ext.get(i).origin.get(j));
                base.setHigh(j,(long int)stop.get(j) - (long int)gdb_ext.get(i).origin.get(j));
            }
 
            Box<dim,long int> dom = gdb_ext.get(i).Dbox;
 
            bool overlap = dom.Intersect(base,inte);
 
            if (overlap == true)
            {
                loc_grid.get(i).template conv_cross_b<prop_src,prop_dst,stencil_size>(inte.getKP1(),inte.getKP2(),func,args...);
            }
        }
    }
 
    template<unsigned int stencil_size, typename v_type, typename lambda_f, typename ... ArgsT >
    void conv_cross_ids(grid_key_dx<3> start, grid_key_dx<3> stop , lambda_f func, ArgsT ... args)
    {
        for (int i = 0 ; i < loc_grid.size() ; i++)
        {
            Box<dim,long int> inte;
 
            Box<dim,long int> base;
            for (int j = 0 ; j < dim ; j++)
            {
                base.setLow(j,(long int)start.get(j) - (long int)gdb_ext.get(i).origin.get(j));
                base.setHigh(j,(long int)stop.get(j) - (long int)gdb_ext.get(i).origin.get(j));
            }
 
            Box<dim,long int> dom = gdb_ext.get(i).Dbox;
 
            bool overlap = dom.Intersect(base,inte);
 
            if (overlap == true)
            {
                loc_grid.get(i).template conv_cross_ids<stencil_size,v_type>(inte.getKP1(),inte.getKP2(),func,args...);
            }
        }
    }
 
    template<unsigned int prop_src1, unsigned int prop_src2, unsigned int prop_dst1, unsigned int prop_dst2, unsigned int stencil_size, unsigned int N, typename lambda_f, typename ... ArgsT >
    void conv2(int (& stencil)[N][dim], grid_key_dx<dim> start, grid_key_dx<dim> stop , lambda_f func, ArgsT ... args)
    {
        for (int i = 0 ; i < loc_grid.size() ; i++)
        {
            Box<dim,long int> inte;
 
            Box<dim,long int> base;
            for (int j = 0 ; j < dim ; j++)
            {
                base.setLow(j,(long int)start.get(j) - (long int)gdb_ext.get(i).origin.get(j));
                base.setHigh(j,(long int)stop.get(j) - (long int)gdb_ext.get(i).origin.get(j));
            }
 
            Box<dim,long int> dom = gdb_ext.get(i).Dbox;
 
            bool overlap = dom.Intersect(base,inte);
 
            if (overlap == true)
            {
                loc_grid.get(i).template conv2<prop_src1,prop_src2,prop_dst1,prop_dst2,stencil_size>(stencil,inte.getKP1(),inte.getKP2(),func,create_vcluster().rank(),args...);
            }
        }
    }
 
    template<unsigned int prop_src1, unsigned int prop_src2, unsigned int prop_dst1, unsigned int prop_dst2, unsigned int stencil_size, typename lambda_f, typename ... ArgsT >
    void conv2(grid_key_dx<dim> start, grid_key_dx<dim> stop , lambda_f func, ArgsT ... args)
    {
        for (int i = 0 ; i < loc_grid.size() ; i++)
        {
            Box<dim,long int> inte;
 
            Box<dim,long int> base;
            for (int j = 0 ; j < dim ; j++)
            {
                base.setLow(j,(long int)start.get(j) - (long int)gdb_ext.get(i).origin.get(j));
                base.setHigh(j,(long int)stop.get(j) - (long int)gdb_ext.get(i).origin.get(j));
            }
 
            Box<dim,long int> dom = gdb_ext.get(i).Dbox;
 
            bool overlap = dom.Intersect(base,inte);
 
            if (overlap == true)
            {
                loc_grid.get(i).template conv2<prop_src1,prop_src2,prop_dst1,prop_dst2,stencil_size>(inte.getKP1(),inte.getKP2(),func,args...);
            }
        }
    }
 
    template<unsigned int prop_src1, unsigned int prop_dst1, unsigned int stencil_size, typename lambda_f, typename ... ArgsT >
    void conv(grid_key_dx<dim> start, grid_key_dx<dim> stop , lambda_f func, ArgsT ... args)
    {
        for (int i = 0 ; i < loc_grid.size() ; i++)
        {
            Box<dim,long int> inte;
 
            Box<dim,long int> base;
            for (int j = 0 ; j < dim ; j++)
            {
                base.setLow(j,(long int)start.get(j) - (long int)gdb_ext.get(i).origin.get(j));
                base.setHigh(j,(long int)stop.get(j) - (long int)gdb_ext.get(i).origin.get(j));
            }
 
            Box<dim,long int> dom = gdb_ext.get(i).Dbox;
 
            bool overlap = dom.Intersect(base,inte);
 
            if (overlap == true)
            {
                loc_grid.get(i).template conv<prop_src1,prop_dst1,stencil_size>(inte.getKP1(),inte.getKP2(),func,args...);
            }
        }
    }
 
    template<unsigned int prop_src1, unsigned int prop_src2, unsigned int prop_dst1, unsigned int prop_dst2, unsigned int stencil_size, typename lambda_f, typename ... ArgsT >
    void conv2_b(grid_key_dx<dim> start, grid_key_dx<dim> stop , lambda_f func, ArgsT ... args)
    {
        for (int i = 0 ; i < loc_grid.size() ; i++)
        {
            Box<dim,long int> inte;
 
            Box<dim,long int> base;
            for (int j = 0 ; j < dim ; j++)
            {
                base.setLow(j,(long int)start.get(j) - (long int)gdb_ext.get(i).origin.get(j));
                base.setHigh(j,(long int)stop.get(j) - (long int)gdb_ext.get(i).origin.get(j));
            }
 
            Box<dim,long int> dom = gdb_ext.get(i).Dbox;
 
            bool overlap = dom.Intersect(base,inte);
 
            if (overlap == true)
            {
                loc_grid.get(i).template conv2_b<prop_src1,prop_src2,prop_dst1,prop_dst2,stencil_size>(inte.getKP1(),inte.getKP2(),func,args...);
            }
        }
    }
 
    template<unsigned int prop_src1, unsigned int prop_src2, unsigned int prop_src3, 
             unsigned int prop_dst1, unsigned int prop_dst2, unsigned int prop_dst3, 
             unsigned int stencil_size, typename lambda_f, typename ... ArgsT >
    void conv3_b(grid_key_dx<dim> start, grid_key_dx<dim> stop , lambda_f func, ArgsT ... args)
    {
        for (int i = 0 ; i < loc_grid.size() ; i++)
        {
            Box<dim,long int> inte;
 
            Box<dim,long int> base;
            for (int j = 0 ; j < dim ; j++)
            {
                base.setLow(j,(long int)start.get(j) - (long int)gdb_ext.get(i).origin.get(j));
                base.setHigh(j,(long int)stop.get(j) - (long int)gdb_ext.get(i).origin.get(j));
            }
 
            Box<dim,long int> dom = gdb_ext.get(i).Dbox;
 
            bool overlap = dom.Intersect(base,inte);
 
            if (overlap == true)
            {
                loc_grid.get(i).template conv3_b<prop_src1,prop_src2,prop_src3,prop_dst1,prop_dst2,prop_dst3,stencil_size>(inte.getKP1(),inte.getKP2(),func,args...);
            }
        }
    }
    
    template<typename NNtype>
    void findNeighbours()
    {
        for (int i = 0 ; i < loc_grid.size() ; i++)
        {
            loc_grid.get(i).findNeighbours();
        }
    }
 
    template<unsigned int prop_src1, unsigned int prop_src2, unsigned int prop_dst1, unsigned int prop_dst2, unsigned int stencil_size, typename lambda_f, typename ... ArgsT >
    void conv_cross2(grid_key_dx<3> start, grid_key_dx<3> stop , lambda_f func, ArgsT ... args)
    {
        for (int i = 0 ; i < loc_grid.size() ; i++)
        {
            Box<dim,long int> inte;
 
            Box<dim,long int> base;
            for (int j = 0 ; j < dim ; j++)
            {
                base.setLow(j,(long int)start.get(j) - (long int)gdb_ext.get(i).origin.get(j));
                base.setHigh(j,(long int)stop.get(j) - (long int)gdb_ext.get(i).origin.get(j));
            }
 
            Box<dim,long int> dom = gdb_ext.get(i).Dbox;
 
            bool overlap = dom.Intersect(base,inte);
 
            if (overlap == true)
            {
                loc_grid.get(i).template conv_cross2<prop_src1,prop_src2,prop_dst1,prop_dst2,stencil_size>(inte.getKP1(),inte.getKP2(),func,args...);
            }
        }
    }
 
    bool write(std::string output, size_t opt = VTK_WRITER | FORMAT_BINARY )
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
 
        file_type ft = file_type::ASCII;
 
        if (opt & FORMAT_BINARY)
        {ft = file_type::BINARY;}
 
        // Create a writer and write
        VTKWriter<boost::mpl::pair<device_grid,float>,VECTOR_GRIDS> vtk_g;
        for (size_t i = 0 ; i < loc_grid.size() ; i++)
        {
            Point<dim,St> offset = getOffset(i);
 
            if (opt & PRINT_GHOST)
            {vtk_g.add(loc_grid.get(i),offset,cd_sm.getCellBox().getP2(),gdb_ext.get(i).GDbox);}
            else
            {vtk_g.add(loc_grid.get(i),offset,cd_sm.getCellBox().getP2(),gdb_ext.get(i).Dbox);}
        }
        vtk_g.write(output + "_" + std::to_string(v_cl.getProcessUnitID()) + ".vtk", prp_names, "grids", ft);
 
        return true;
    }
 
    bool write_debug(std::string output)
    {
        for (int i = 0 ; i < getN_loc_grid() ; i++)
        {
            Point<dim,St> sp;
            Point<dim,St> offset;
 
            for (int j = 0 ; j < dim ; j++)
            {sp.get(j) = this->spacing(j);}
 
            offset = gdb_ext.get(i).origin;
 
            get_loc_grid(i).write_debug(output + "_" + std::to_string(i) + "_" + std::to_string(v_cl.getProcessUnitID()) + ".vtk",sp,offset);
        }
 
        return true;
    }
 
    bool write_frame(std::string output, size_t i, size_t opt = VTK_WRITER | FORMAT_ASCII)
    {
#ifdef SE_CLASS2
        check_valid(this,8);
#endif
        file_type ft = file_type::ASCII;
 
        if (opt & FORMAT_BINARY)
            ft = file_type::BINARY;
 
        // Create a writer and write
        VTKWriter<boost::mpl::pair<device_grid,float>,VECTOR_GRIDS> vtk_g;
        for (size_t i = 0 ; i < loc_grid.size() ; i++)
        {
            Point<dim,St> offset = getOffset(i);
            vtk_g.add(loc_grid.get(i),offset,cd_sm.getCellBox().getP2(),gdb_ext.get(i).Dbox);
        }
        vtk_g.write(output + "_" + std::to_string(v_cl.getProcessUnitID()) + "_" + std::to_string(i) + ".vtk",prp_names,"grids",ft);
 
        return true;
    }
 
 
 
    device_grid & get_loc_grid(size_t i)
    {
        return loc_grid.get(i);
    }
 
    const device_grid & get_loc_grid(size_t i) const
    {
        return loc_grid.get(i);
    }
 
    grid_key_dx_iterator_sub<dim,no_stencil> get_loc_grid_iterator(size_t i)
    {
        return grid_key_dx_iterator_sub<dim,no_stencil>(loc_grid.get(i).getGrid(),
                 gdb_ext.get(i).Dbox.getKP1(),
                 gdb_ext.get(i).Dbox.getKP2());
    }
 
    template<unsigned int Np>
    grid_key_dx_iterator_sub<dim,stencil_offset_compute<dim,Np>,typename device_grid::linearizer_type>
    get_loc_grid_iterator_stencil(size_t i,const grid_key_dx<dim> (& stencil_pnt)[Np])
    {
        return grid_key_dx_iterator_sub<dim,stencil_offset_compute<dim,Np>,typename device_grid::linearizer_type>(loc_grid.get(i).getGrid(),
                                                     gdb_ext.get(i).Dbox.getKP1(),
                                                     gdb_ext.get(i).Dbox.getKP2(),
                                                     stencil_pnt);
    }
 
    size_t getN_loc_grid() const
    {
        return loc_grid.size();
    }
 
 
    long int who()
    {
#ifdef SE_CLASS2
        return check_whoami(this,8);
#else
            return -1;
#endif
    }
 
    void debugPrint()
    {
        size_t tot_volume = 0;
 
        std::cout << "-------- External Ghost boxes ---------- " << std::endl;
 
        for (size_t i = 0 ; i < eg_box.size() ; i++)
        {
            std::cout << "Processor: " << eg_box.get(i).prc << " Boxes:" << std::endl;
 
            for (size_t j = 0; j < eg_box.get(i).bid.size() ; j++)
            {
                std::cout << " Box: " << eg_box.get(i).bid.get(j).g_e_box.toString() << "   Id: " << eg_box.get(i).bid.get(j).g_id << std::endl;
                tot_volume += eg_box.get(i).bid.get(j).g_e_box.getVolumeKey();
            }
        }
 
        std::cout << "TOT volume external ghost " << tot_volume << std::endl;
 
        std::cout << "-------- Internal Ghost boxes ---------- " << std::endl;
 
        tot_volume = 0;
        for (size_t i = 0 ; i < ig_box.size() ; i++)
        {
            std::cout << "Processor: " << ig_box.get(i).prc << " Boxes:" << std::endl;
 
            for (size_t j = 0 ; j < ig_box.get(i).bid.size() ; j++)
            {
                std::cout << " Box: " << ig_box.get(i).bid.get(j).box.toString() << "   Id: " << ig_box.get(i).bid.get(j).g_id << std::endl;
                tot_volume += ig_box.get(i).bid.get(j).box.getVolumeKey();
            }
        }
 
        std::cout << "TOT volume internal ghost " << tot_volume << std::endl;
    }
 
    void setPropNames(const openfpm::vector<std::string> & names)
    {
        prp_names = names;
    }
 
    const openfpm::vector<std::string> & getPropNames()
    {
        return prp_names;
    }
 
    void clear()
    {
        for (size_t i = 0 ; i < loc_grid.size() ; i++)
        {loc_grid.get(i).clear();}
    }
 
    void construct_link(self & grid_up, self & grid_dw)
    {
        for (int i = 0 ; i < loc_grid.size() ; i++)
        {
            loc_grid.get(i).construct_link(grid_up.get_loc_grid(i),grid_dw.get_loc_grid(i),v_cl.getgpuContext());
        }
    }
 
    void construct_link_dw(self & grid_dw, openfpm::vector<offset_mv<dim>> & mvof)
    {
        for (int i = 0 ; i < loc_grid.size() ; i++)
        {
            Point<dim,int> p_dw;
            for(int j = 0 ; j < dim ; j++)
            {p_dw.get(j) = mvof.get(i).dw.get(j);}
 
            loc_grid.get(i).construct_link_dw(grid_dw.get_loc_grid(i),gdb_ext.get(i).Dbox,p_dw,v_cl.getgpuContext());
        }
    }
 
    void construct_link_up(self & grid_up, openfpm::vector<offset_mv<dim>> & mvof)
    {
        for (int i = 0 ; i < loc_grid.size() ; i++)
        {
            Point<dim,int> p_up;
            for(int j = 0 ; j < dim ; j++)
            {p_up.get(j) = mvof.get(i).up.get(j);}
 
            loc_grid.get(i).construct_link_up(grid_up.get_loc_grid(i),gdb_ext.get(i).Dbox,p_up,v_cl.getgpuContext());
        }
    }
 
    template<typename stencil_type>
    void tagBoundaries()
    {
        for (int i = 0 ; i < loc_grid.size() ; i++)
        {
            // we limit to the domain subset for tagging
 
            Box_check<dim,unsigned int> chk(gdb_ext.get(i).Dbox);
 
 
            loc_grid.get(i).template tagBoundaries<stencil_type>(v_cl.getgpuContext(),chk);
        }
    }
 
    void map(size_t opt = 0)
    {
        // Save the background values
        T bv;
 
        copy_aggregate_dual<decltype(loc_grid.get(0).getBackgroundValue()),
                            T> ca(loc_grid.get(0).getBackgroundValue(),bv);
 
        boost::mpl::for_each_ref<boost::mpl::range_c<int,0,T::max_prop>>(ca);
 
        if (!(opt & NO_GDB_EXT_SWITCH))
        {
            gdb_ext_old = gdb_ext;
            loc_grid_old = loc_grid;
 
            InitializeStructures(g_sz,bx_def,gint,bx_def.size() != 0);
        }
 
        getGlobalGridsInfo(gdb_ext_global);
 
 
        typedef typename to_int_sequence<0,T::max_prop-1>::type result;
 
        variadic_caller<result>::call(this,dec,cd_sm,loc_grid,loc_grid_old,gdb_ext,gdb_ext_old,gdb_ext_global,opt);
 
        loc_grid_old.clear();
        loc_grid_old.shrink_to_fit();
        gdb_ext_old.clear();
 
        // reset ghost structure to recalculate
        reset_ghost_structures();
 
        // Reset the background values
        setBackgroundValue(bv);
    }
 
    inline void save(const std::string & filename) const
    {
        HDF5_writer<GRID_DIST> h5s;
 
        h5s.save(filename,loc_grid,gdb_ext);
    }
 
    inline void load(const std::string & filename)
    {
        HDF5_reader<GRID_DIST> h5l;
 
        h5l.load<device_grid>(filename,loc_grid_old,gdb_ext_old);
 
        if (v_cl.size() != 1)
        {
            // move information from host to device
            for (int i = 0 ; i < loc_grid_old.size() ; i++)
            {loc_grid_old.get(i).template hostToDevice<0>();}
            // Map the distributed grid
            size_t opt_ = NO_GDB_EXT_SWITCH;
            if (std::is_same<Memory,CudaMemory>::value == true)
            {opt_ |= RUN_ON_DEVICE;}
            map(opt_);
        }
        else
        {
            device_grid_copy<device_grid::isCompressed()>::pre_load(gdb_ext_old,loc_grid_old,gdb_ext,loc_grid);
            for (int i = 0 ; i < gdb_ext_old.size() ; i++)
            {
                    auto & lg = loc_grid_old.get(i);
                    auto it_src = lg.getIterator(gdb_ext_old.get(i).Dbox.getKP1(),gdb_ext_old.get(i).Dbox.getKP2());
                    auto & dg = loc_grid.get(0);
                    grid_key_dx<dim> kp1 = gdb_ext.get(0).Dbox.getKP1();
 
                    grid_key_dx<dim> orig;
                    for (int j = 0 ; j < dim ; j++)
                    {
                            orig.set_d(j,gdb_ext_old.get(i).origin.get(j));
                    }
 
                    while (it_src.isNext())
                    {
                        auto key = it_src.get();
                        grid_key_dx<dim> key_dst;
 
                        for (int j = 0 ; j < dim ; j++)
                        {key_dst.set_d(j,key.get(j) + orig.get(j) + kp1.get(j));}
 
                        device_grid_copy<device_grid::isCompressed()>::assign(key,key_dst,dg,lg);
 
                        ++it_src;
                    }
 
                    dg.template hostToDevice<0>();
            }
        }
    }
 
    template <typename stencil = no_stencil>
    static grid_dist_iterator_sub<dim,device_grid> type_of_subiterator()
    {
        return grid_key_dx_iterator_sub<dim, stencil>();
    }
 
    const openfpm::vector<i_lbox_grid<dim>> & get_loc_ig_box()
    {
        return this->loc_ig_box;
    }
 
    const openfpm::vector<i_lbox_grid<dim>> & get_ig_box()
    {
        return this->ig_box;
    }
 
    void print_stats()
    {
        std::cout << "-- REPORT --" << std::endl;
#ifdef ENABLE_GRID_DIST_ID_PERF_STATS
        std::cout << "Processor: " << v_cl.rank() << " Time spent in packing data: " << tot_pack << std::endl;
        std::cout << "Processor: " << v_cl.rank() << " Time spent in sending and receving data: " << tot_sendrecv << std::endl;
        std::cout << "Processor: " << v_cl.rank() << " Time spent in merging: " << tot_merge << std::endl;
        std::cout << "Processor: " << v_cl.rank() << " Time spent in local merging: " << tot_loc_merge << std::endl;
#else
 
        std::cout << "Enable ENABLE_GRID_DIST_ID_PERF_STATS if you want to activate this feature" << std::endl;
 
#endif
    }
 
    void clear_stats()
    {
#ifdef ENABLE_GRID_DIST_ID_PERF_STATS
        tot_pack = 0;
        tot_sendrecv = 0;
        tot_merge = 0;
#else
 
        std::cout << "Enable ENABLE_GRID_DIST_ID_PERF_STATS if you want to activate this feature" << std::endl;
 
#endif
    }
 
#ifdef __NVCC__
 
    void setNumberOfInsertPerThread(size_t n_ins)
    {
        gpu_n_insert_thread = n_ins;
    }
 
    template<typename func_t,typename it_t, typename ... args_t>
    void iterateGridGPU(it_t & it, args_t ... args)
    {
        // setGPUInsertBuffer must be called in anycase even with 0 points to insert
        // the loop "it.isNextGrid()" does not guarantee to call it for all local grids
        for (size_t i = 0 ; i < loc_grid.size() ; i++)
        {loc_grid.get(i).setGPUInsertBuffer(0ul,1ul);}
 
        while(it.isNextGrid())
        {
            Box<dim,size_t> b = it.getGridBox();
 
            size_t i = it.getGridId();
 
            auto ite = loc_grid.get(i).getGridGPUIterator(b.getKP1int(),b.getKP2int());
 
            loc_grid.get(i).setGPUInsertBuffer(ite.nblocks(),ite.nthrs());
            loc_grid.get(i).initializeGPUInsertBuffer();
 
            ite_gpu_dist<dim> itd = ite;
 
            for (int j = 0 ; j < dim ; j++)
            {
                itd.origin.set_d(j,gdb_ext.get(i).origin.get(j));
                itd.start_base.set_d(j,0);
            }
 
            CUDA_LAUNCH((grid_apply_functor),ite,loc_grid.get(i).toKernel(),itd,func_t(),args...);
 
            it.nextGrid();
        }
    }
 
    void removePoints(Box<dim,size_t> & box)
    {
        Box<dim,long int> box_i = box;
 
        for (size_t i = 0 ; i < loc_grid.size() ; i++)
        {
            Box<dim,long int> bx = gdb_ext.get(i).Dbox + gdb_ext.get(i).origin;
 
            Box<dim,long int> bout;
            bool inte = bx.Intersect(box,bout);
            bout -= gdb_ext.get(i).origin;
 
            if (inte == true)
            {
                loc_grid.get(i).copyRemoveReset();
                loc_grid.get(i).remove(bout);
                loc_grid.get(i).removePoints(v_cl.getgpuContext());
            }
        }
    }
 
    template<unsigned int ... prp> void deviceToHost()
    {
        for (size_t i = 0 ; i < loc_grid.size() ; i++)
        {
            loc_grid.get(i).template deviceToHost<prp ...>();
        }
    }
 
    template<unsigned int ... prp> void hostToDevice()
    {
        for (size_t i = 0 ; i < loc_grid.size() ; i++)
        {
            loc_grid.get(i).template hostToDevice<prp ...>();
        }
    }
 
#endif
 
 
    //\cond
    friend grid_dist_id<dim,St,T,typename Decomposition::extended_type,Memory,device_grid>;
    //\endcond
};
 
 
template<unsigned int dim, typename St, typename T, typename Memory = HeapMemory, typename Decomposition = CartDecomposition<dim,St> >
using sgrid_dist_id = grid_dist_id<dim,St,T,Decomposition,Memory,sgrid_cpu<dim,T,Memory>>;
 
template<unsigned int dim, typename St, typename T, typename Memory = HeapMemory, typename Decomposition = CartDecomposition<dim,St>>
using sgrid_dist_soa = grid_dist_id<dim,St,T,Decomposition,Memory,sgrid_soa<dim,T,Memory>>;
 
template<unsigned int dim, typename St, typename T, typename devg, typename Memory = HeapMemory, typename Decomposition = CartDecomposition<dim,St>>
using grid_dist_id_devg = grid_dist_id<dim,St,T,Decomposition,Memory,devg>;
 
#ifdef __NVCC__
template<unsigned int dim, typename St, typename T, typename Memory = CudaMemory, typename Decomposition = CartDecomposition<dim,St,CudaMemory,memory_traits_inte> >
using sgrid_dist_id_gpu = grid_dist_id<dim,St,T,Decomposition,Memory,SparseGridGpu<dim,T>>;
 
template<unsigned int dim, typename St, typename T, typename Memory = CudaMemory, typename Decomposition = CartDecomposition<dim,St,CudaMemory,memory_traits_inte> >
using grid_dist_id_gpu = grid_dist_id<dim,St,T,Decomposition,Memory,grid_gpu<dim,T>>;
 
template<unsigned int dim, typename St, typename T, typename Memory = CudaMemory, typename Decomposition = CartDecomposition<dim,St,CudaMemory,memory_traits_inte> >
using sgrid_dist_sid_gpu = grid_dist_id<dim,St,T,Decomposition,Memory,SparseGridGpu<dim,T,default_edge<dim>::type::value,default_edge<dim>::tb::value,int>>;
#endif
 
#endif