diff --git a/src/Grid/grid_dist_id.hpp b/src/Grid/grid_dist_id.hpp
index 0c471f526fe5ce13edd63052fb3d67adb6e241da..0d24f5184e49cb8048a4a2cdfd727235ef30ae03 100644
--- a/src/Grid/grid_dist_id.hpp
+++ b/src/Grid/grid_dist_id.hpp
@@ -69,6 +69,11 @@ struct device_grid_copy
{
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>
@@ -98,6 +103,21 @@ struct copy_all_prop_sparse
}
};
+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>
{
@@ -112,6 +132,143 @@ struct device_grid_copy<true>
boost::mpl::for_each_ref< boost::mpl::range_c<int,0,decltype(block_data_dst)::max_prop> >(cp);
}
+
+/*
+ *
+ *
+ auto & v_cl = create_vcluster();
+ auto & dg = loc_grid.get(0);
+
+ typedef typename std::remove_reference<decltype(dg)>::type sparse_grid_type;
+
+ grid_key_dx<sparse_grid_type::dims> kp1 = gdb_ext.get(0).Dbox.getKP1();
+
+ for (int j = 0 ; j < std::ceil(gdb_ext_old.size() / 32) + 1 ; j++)
+ {
+ int jbase = j*32;
+ int sz = gdb_ext_old.size() - jbase;
+ if (sz < 0) {break;}
+ else if (sz > 32) {sz = 32;}
+ dg.copyRemoveReset();
+
+ std::cout << "JBASE: " << jbase << " SZ: " << sz << std::endl;
+
+ for (int i = 0 ; i < sz ; i++)
+ {
+ Box<sparse_grid_type::dims,long int> bx_dst;
+ Box<sparse_grid_type::dims,long int> bx_src;
+ for (int k = 0 ; k < sparse_grid_type::dims ; k++)
+ {
+ bx_dst.setLow(k,kp1.get(k) + gdb_ext_old.get(i+jbase).origin.get(k) + gdb_ext_old.get(i+jbase).Dbox.getKP1().get(k));
+ bx_dst.setHigh(k,kp1.get(k) + gdb_ext_old.get(i+jbase).origin.get(k) + gdb_ext_old.get(i+jbase).Dbox.getKP2().get(k));
+
+ bx_src.setLow(k,gdb_ext_old.get(i+jbase).Dbox.getKP1().get(k));
+ bx_src.setHigh(k,gdb_ext_old.get(i+jbase).Dbox.getKP2().get(k));
+ }
+
+ std::cout << jbase << " " << i << " SRC:" << gdb_ext_old.get(i+jbase).Dbox.toString() << " DST: " << bx_dst.toString() << " INDEX BUFFER: " << loc_grid_old.get(jbase + i).private_get_index_array().size() << std::endl;
+ loc_grid_old.get(jbase + i).template hostToDevice<0>();
+ dg.copy_to(loc_grid_old.get(jbase + i),gdb_ext_old.get(jbase +i).Dbox,bx_dst);
+ }
+
+ for (int i = 0 ; i < sz ; i++)
+ {
+ std::cout << "FINALIZE1" << std::endl;
+ loc_grid_old.get(jbase + i).template removeCopyToFinalize<0>(v_cl.getmgpuContext(), rem_copy_opt::PHASE1);
+ std::cout << "FINALIZE2" << std::endl;
+ loc_grid_old.get(jbase + i).template removeCopyToFinalize<0>(v_cl.getmgpuContext(), rem_copy_opt::PHASE2);
+ std::cout << "FINALIZE3" << std::endl;
+ loc_grid_old.get(jbase + i).template removeCopyToFinalize<0>(v_cl.getmgpuContext(), rem_copy_opt::PHASE3);
+ }
+ }
+
+ *
+ *
+ */
+
+ 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);
+ }
+ }
};
/*! \brief This is a distributed grid
@@ -3361,6 +3518,7 @@ public:
}
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);
@@ -3376,15 +3534,15 @@ public:
while (it_src.isNext())
{
- auto key = it_src.get();
- grid_key_dx<dim> key_dst;
+ 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));}
+ 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);
+ device_grid_copy<device_grid::isCompressed()>::assign(key,key_dst,dg,lg);
- ++it_src;
+ ++it_src;
}
}
}
diff --git a/src/Grid/tests/sgrid_dist_id_gpu_unit_tests.cu b/src/Grid/tests/sgrid_dist_id_gpu_unit_tests.cu
index a0582fd4315f550d4dfc033511cf810af1d61636..830ca9bdbf9eca38f80a1304c9cc32493a996b07 100644
--- a/src/Grid/tests/sgrid_dist_id_gpu_unit_tests.cu
+++ b/src/Grid/tests/sgrid_dist_id_gpu_unit_tests.cu
@@ -187,6 +187,81 @@ BOOST_AUTO_TEST_CASE( sgrid_gpu_test_output )
#endif
}
+template<typename grid, typename box_type>
+void check_sgrid(grid & gdist2, box_type & box, float c)
+{
+ bool match = true;
+ auto it2 = gdist2.getDomainIterator();
+
+ while (it2.isNext())
+ {
+ auto p = it2.get();
+
+ auto key = it2.getGKey(p);
+
+ auto p_xp1 = p.move(0,1);
+ auto p_xm1 = p.move(0,-1);
+ auto p_yp1 = p.move(1,1);
+ auto p_ym1 = p.move(1,-1);
+
+ auto key_xp1 = key.move(0,1);
+ auto key_xm1 = key.move(0,-1);
+ auto key_yp1 = key.move(1,1);
+ auto key_ym1 = key.move(1,-1);
+
+ if (box.isInside(key_xp1.toPoint()))
+ {
+ match &= gdist2.template get<0>(p_xp1) == c + key_xp1.get(0) + key_xp1.get(1);
+
+ if (match == false)
+ {
+ std::cout << gdist2.template get<0>(p_xp1) << " " << c + key_xp1.get(0) + key_xp1.get(1) << std::endl;
+ std::cout << "1 " << p_xp1.getKey().toPoint().to_string() << " " << &gdist2.template get<0>(p_xp1) << std::endl;
+ break;
+ }
+ }
+
+ if (box.isInside(key_xm1.toPoint()))
+ {
+ match &= gdist2.template get<0>(p_xm1) == c + key_xm1.get(0) + key_xm1.get(1);
+
+ if (match == false)
+ {
+ std::cout << gdist2.template get<0>(p_xm1) << " " << c + key_xm1.get(0) + key_xm1.get(1) << std::endl;
+ std::cout << "2 " << key_xm1.to_string() << std::endl;
+ break;
+ }
+ }
+
+ if (box.isInside(key_yp1.toPoint()))
+ {
+ match &= gdist2.template get<0>(p_yp1) == c + key_yp1.get(0) + key_yp1.get(1);
+
+ if (match == false)
+ {
+ std::cout << gdist2.template get<0>(p_yp1) << " " << c + key_yp1.get(0) + key_yp1.get(1) << std::endl;
+ std::cout << "3 " << key_yp1.to_string() << std::endl;
+ break;
+ }
+ }
+
+ if (box.isInside(key_ym1.toPoint()))
+ {
+ match &= gdist2.template get<0>(p_ym1) == c + key_ym1.get(0) + key_ym1.get(1);
+
+ if (match == false)
+ {
+ std::cout << gdist2.template get<0>(p_ym1) << " " << c + key_ym1.get(0) + key_ym1.get(1) << std::endl;
+ std::cout << "4 " << key_ym1.to_string() << std::endl;
+ break;
+ }
+ }
+
+ ++it2;
+ }
+
+ BOOST_REQUIRE_EQUAL(match,true);
+}
BOOST_AUTO_TEST_CASE( sgrid_gpu_test_save_and_load )
{
@@ -232,19 +307,23 @@ BOOST_AUTO_TEST_CASE( sgrid_gpu_test_save_and_load )
gdist.template deviceToHost<0>();
gdist.save("sgrid_gpu_output_hdf5");
+ check_sgrid(gdist,box,c);
// Now load
sgrid_dist_id_gpu<2,float,aggregate<float,float>> gdist2(sz,domain,g,bc);
gdist2.load("sgrid_gpu_output_hdf5");
+ check_sgrid(gdist2,box,c);
+ gdist2.template hostToDevice<0>();
gdist2.template ghost_get<0,1>(RUN_ON_DEVICE);
gdist2.deviceToHost<0,1>();
gdist.deviceToHost<0,1>();
+ check_sgrid(gdist2,box,c);
- bool match = true;
+/* bool match = true;
auto it2 = gdist2.getDomainIterator();
@@ -272,6 +351,7 @@ BOOST_AUTO_TEST_CASE( sgrid_gpu_test_save_and_load )
if (match == false)
{
std::cout << gdist.template get<0>(p_xp1) << " " << c + key_xp1.get(0) + key_xp1.get(1) << std::endl;
+ std::cout << key_xp1.to_string() << std::endl;
break;
}
}
@@ -283,6 +363,7 @@ BOOST_AUTO_TEST_CASE( sgrid_gpu_test_save_and_load )
if (match == false)
{
std::cout << gdist.template get<0>(p_xm1) << " " << c + key_xm1.get(0) + key_xm1.get(1) << std::endl;
+ std::cout << key_xm1.to_string() << std::endl;
break;
}
}
@@ -294,6 +375,7 @@ BOOST_AUTO_TEST_CASE( sgrid_gpu_test_save_and_load )
if (match == false)
{
std::cout << gdist.template get<0>(p_yp1) << " " << c + key_yp1.get(0) + key_yp1.get(1) << std::endl;
+ std::cout << key_yp1.to_string() << std::endl;
break;
}
}
@@ -305,6 +387,7 @@ BOOST_AUTO_TEST_CASE( sgrid_gpu_test_save_and_load )
if (match == false)
{
std::cout << gdist.template get<0>(p_ym1) << " " << c + key_ym1.get(0) + key_ym1.get(1) << std::endl;
+ std::cout << key_ym1.to_string() << std::endl;
break;
}
}
@@ -313,7 +396,7 @@ BOOST_AUTO_TEST_CASE( sgrid_gpu_test_save_and_load )
}
- BOOST_REQUIRE_EQUAL(match,true);
+ BOOST_REQUIRE_EQUAL(match,true);*/
}
void sgrid_ghost_get(size_t (& sz)[2],size_t (& sz2)[2])