diff --git a/src/Grid/grid_dist_id.hpp b/src/Grid/grid_dist_id.hpp
index 4b5954dbb05438fcddecef3d7766325306c30e70..182080a0c7984b2664c8c66c767b4fb9138f9cee 100644
--- a/src/Grid/grid_dist_id.hpp
+++ b/src/Grid/grid_dist_id.hpp
@@ -2636,6 +2636,35 @@ public:
 		}
 	}
 
+	/*! \brief apply a convolution using the stencil N
+	 *
+	 *
+	 */
+	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...);
+			}
+		}
+	}
+
 	/*! \brief apply a convolution using the stencil N
 	 *
 	 *
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 432cc219110021621867a3e9d9d69f120f3bbce4..9b79ea1b56f6f3aefa79afe68cc62402530a30b4 100644
--- a/src/Grid/tests/sgrid_dist_id_gpu_unit_tests.cu
+++ b/src/Grid/tests/sgrid_dist_id_gpu_unit_tests.cu
@@ -631,6 +631,101 @@ BOOST_AUTO_TEST_CASE( sgrid_gpu_test_conv2_test_3d )
 	BOOST_REQUIRE_EQUAL(match,true);
 }
 
+BOOST_AUTO_TEST_CASE( sgrid_gpu_test_conv_cross_block_test_3d )
+{
+	#ifdef CUDA_ON_CPU
+	size_t sz[3] = {20,20,20};
+	#else
+	size_t sz[3] = {60,60,60};
+	#endif
+	periodicity<3> bc = {PERIODIC,PERIODIC,PERIODIC};
+
+	Ghost<3,long int> g(1);
+
+	Box<3,float> domain({0.0,0.0,0.0},{1.0,1.0,1.0});
+
+	sgrid_dist_id_gpu<3,float,aggregate<float,float,float,float>> gdist(sz,domain,g,bc);
+
+	gdist.template setBackgroundValue<0>(666);
+	gdist.template setBackgroundValue<1>(666);
+	gdist.template setBackgroundValue<2>(666);
+	gdist.template setBackgroundValue<3>(666);
+
+	/////// GPU insert + flush
+
+	Box<3,size_t> box({1,1,1},{sz[0]-1,sz[1]-1,sz[2]-1});
+
+	/////// GPU Run kernel
+
+	float c = 5.0;
+
+	typedef typename GetAddBlockType<decltype(gdist)>::type InsertBlockT;
+
+	gdist.addPoints(box.getKP1(),box.getKP2(),
+			        [] __device__ (int i, int j, int k)
+			        {
+						return true;
+			        },
+			        [c] __device__ (InsertBlockT & data, int i, int j, int k)
+			        {
+			        	data.template get<0>() = c + i + j + k;
+			        	data.template get<1>() = c + 1000 + i + j + k;
+			        }
+			        );
+
+	gdist.template flush<smax_<0>,smax_<1>>(flush_type::FLUSH_ON_DEVICE);
+
+	gdist.template ghost_get<0,1>(RUN_ON_DEVICE);
+
+	for (int i = 0 ; i < 10 ; i++)
+	{
+		gdist.template ghost_get<0,1>(RUN_ON_DEVICE);
+	}
+
+	// Now run the convolution
+
+	typedef typename GetCpBlockType<decltype(gdist),0,1>::type CpBlockType;
+
+	gdist.template conv_cross_b<0,1,1>({2,2,2},{(int)sz[0]-2,(int)sz[1]-2,(int)sz[2]-2},[] __device__ (CpBlockType & u, auto & block, int offset,int i, int j, int k){
+		return u(i+1,j,k) - u(i-1,j,k) + u(i,j+1,k) - u(i,j-1,k) + u(i,j,k+1) - u(i,j,k-1) + block.template get<0>()[offset];
+	});
+
+	gdist.deviceToHost<0,1,2,3>();
+
+	// Now we check that ghost is correct
+
+	auto it3 = gdist.getSubDomainIterator({2,2,2},{(int)sz[0]-2,(int)sz[1]-2,(int)sz[2]-2});
+
+	bool match = true;
+
+	while (it3.isNext())
+	{
+		auto p = it3.get();
+
+		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 p_zp1 = p.move(2,1);
+		auto p_zm1 = p.move(2,-1);
+
+		float sub1 = gdist.template get<1>(p);
+
+		if (sub1 != 6.0 + gdist.template get<0>(p))
+		{
+			std::cout << sub1 << std::endl;
+			std::cout << gdist.template get<0>(p_xp1) << "   " << gdist.template get<0>(p_xm1) << std::endl;
+			std::cout << gdist.template get<1>(p_xp1) << "   " << gdist.template get<1>(p_xm1) << std::endl;
+			match = false;
+			break;
+		}
+
+		++it3;
+	}
+
+	BOOST_REQUIRE_EQUAL(match,true);
+}
+
 BOOST_AUTO_TEST_CASE( sgrid_gpu_test_ghost_point_remove )
 {
 	size_t sz[3] = {60,60,60};