From afc410326b30c0944c3ff26f1a14e043d9a68e38 Mon Sep 17 00:00:00 2001
From: Incardona Pietro <incardon@mpi-cbg.de>
Date: Wed, 6 Oct 2021 12:22:36 +0200
Subject: [PATCH] Fixing getProp on grid_dist_id
---
src/Grid/grid_dist_id.hpp | 39 +++++++-
.../tests/sgrid_dist_id_gpu_unit_tests.cu | 97 +++++++++++++++++++
2 files changed, 131 insertions(+), 5 deletions(-)
diff --git a/src/Grid/grid_dist_id.hpp b/src/Grid/grid_dist_id.hpp
index 182080a0c..141218883 100644
--- a/src/Grid/grid_dist_id.hpp
+++ b/src/Grid/grid_dist_id.hpp
@@ -2336,8 +2336,8 @@ public:
* \return the selected element
*
*/
- template <unsigned int p = 0>
- inline auto getProp(const grid_dist_key_dx<dim> & v1) const -> decltype(this->template get<p>(v1))
+ 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);
}
@@ -2350,8 +2350,8 @@ public:
* \return the selected element
*
*/
- template <unsigned int p = 0>
- inline auto getProp(const grid_dist_key_dx<dim> & v1) -> decltype(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);
}
@@ -2723,7 +2723,7 @@ public:
}
}
- /*! \brief apply a convolution using the stencil N
+ /*! \brief apply a convolution on 2 property on GPU
*
*
*/
@@ -2752,6 +2752,35 @@ public:
}
}
+ /*! \brief apply a convolution on 2 property on GPU
+ *
+ *
+ */
+ 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<typename NNtype>
void findNeighbours()
{
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 9b79ea1b5..4bc3e8841 100644
--- a/src/Grid/tests/sgrid_dist_id_gpu_unit_tests.cu
+++ b/src/Grid/tests/sgrid_dist_id_gpu_unit_tests.cu
@@ -726,6 +726,103 @@ BOOST_AUTO_TEST_CASE( sgrid_gpu_test_conv_cross_block_test_3d )
BOOST_REQUIRE_EQUAL(match,true);
}
+BOOST_AUTO_TEST_CASE( sgrid_gpu_test_conv2_b_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 conv2_b<0,1,2,3,1>({2,2,2},{(int)sz[0]-2,(int)sz[1]-2,(int)sz[2]-2},[] __device__ (float & u_out, float & v_out, CpBlockType & u, CpBlockType & v, auto & block, int offset,int i, int j, int k){
+ u_out = 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];
+ v_out = v(i+1,j,k) - v(i-1,j,k) + v(i,j+1,k) - v(i,j-1,k) + v(i,j,k+1) - v(i,j,k-1) + block.template get<1>()[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<2>(p) ;
+ float sub2 = gdist.template get<3>(p);
+
+ if (sub1 != 6.0 + gdist.template get<0>(p) || sub2 != 6.0 + gdist.template get<1>(p))
+ {
+ std::cout << sub1 << " " << sub2 << 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};
--
GitLab