From 078e6b8a4754efec6ea4d7d676d851c837602758 Mon Sep 17 00:00:00 2001
From: Yaroslav <beorn.90@gmail.com>
Date: Tue, 14 Jun 2016 15:20:17 +0200
Subject: [PATCH] Reordering + iterator good performance
---
images/Makefile.am | 8 +-
openfpm_data | 2 +-
openfpm_devices | 2 +-
openfpm_io | 2 +-
openfpm_numerics | 2 +-
src/Makefile.am | 2 +-
src/Vector/vector_dist.hpp | 132 ++++++++++++++++---
src/Vector/vector_dist_unit_test.hpp | 181 ++++++++++++++++++++++++++-
src/main.cpp | 1 +
9 files changed, 300 insertions(+), 32 deletions(-)
diff --git a/images/Makefile.am b/images/Makefile.am
index 4634f73c3..41ca924f1 100644
--- a/images/Makefile.am
+++ b/images/Makefile.am
@@ -3,22 +3,22 @@ LINKLIBS = $(METIS_LIB) $(PARMETIS_LIB) $(PTHREAD_LIBS) $(OPT_LIBS) $(BOOST_LDF
noinst_PROGRAMS = cart_dec metis_dec dom_box vector_dist
cart_dec_SOURCES = CartDecomposition_gen_vtk.cpp ../src/lib/pdata.cpp ../openfpm_devices/src/memory/HeapMemory.cpp ../openfpm_devices/src/memory/PtrMemory.cpp ../openfpm_vcluster/src/VCluster.cpp ../openfpm_devices/src/Memleak_check.cpp
-cart_dec_CXXFLAGS = $(METIS_INCLUDE) $(PARMETIS_INCLUDE) $(CUDA_CFLAGS) $(INCLUDES_PATH) $(BOOST_CPPFLAGS) -I../src -Wno-unused-function -Wno-unused-local-typedefs
+cart_dec_CXXFLAGS = $(METIS_INCLUDE) $(PARMETIS_INCLUDE) $(CUDA_CFLAGS) $(INCLUDES_PATH) $(BOOST_CPPFLAGS) -I../src -I/usr/local/libhilbert/include -Wno-unused-function -Wno-unused-local-typedefs
cart_dec_CFLAGS = $(CUDA_CFLAGS)
cart_dec_LDADD = $(LINKLIBS) -lparmetis -lmetis
metis_dec_SOURCES = Metis_gen_vtk.cpp ../src/lib/pdata.cpp ../openfpm_devices/src/memory/HeapMemory.cpp ../openfpm_devices/src/memory/PtrMemory.cpp ../openfpm_vcluster/src/VCluster.cpp ../openfpm_devices/src/Memleak_check.cpp
-metis_dec_CXXFLAGS = $(METIS_INCLUDE) $(CUDA_CFLAGS) $(INCLUDES_PATH) $(BOOST_CPPFLAGS) -I../src -Wno-unused-function -Wno-unused-local-typedefs
+metis_dec_CXXFLAGS = $(METIS_INCLUDE) $(CUDA_CFLAGS) $(INCLUDES_PATH) $(BOOST_CPPFLAGS) -I../src -I/usr/local/libhilbert/include -Wno-unused-function -Wno-unused-local-typedefs
metis_dec_CFLAGS = $(CUDA_CFLAGS)
metis_dec_LDADD = $(LINKLIBS) -lmetis
dom_box_SOURCES = domain_gen_vtk.cpp ../src/lib/pdata.cpp ../openfpm_devices/src/memory/HeapMemory.cpp ../openfpm_devices/src/memory/PtrMemory.cpp ../openfpm_vcluster/src/VCluster.cpp ../openfpm_devices/src/Memleak_check.cpp
-dom_box_CXXFLAGS = $(METIS_INCLUDE) $(CUDA_CFLAGS) $(INCLUDES_PATH) $(BOOST_CPPFLAGS) -I../src -Wno-unused-function -Wno-unused-local-typedefs
+dom_box_CXXFLAGS = $(METIS_INCLUDE) $(CUDA_CFLAGS) $(INCLUDES_PATH) $(BOOST_CPPFLAGS) -I../src -I/usr/local/libhilbert/include -Wno-unused-function -Wno-unused-local-typedefs
dom_box_CFLAGS = $(CUDA_CFLAGS)
dom_box_LDADD = $(LINKLIBS)
vector_dist_SOURCES = vector.cpp ../openfpm_devices/src/memory/HeapMemory.cpp ../openfpm_vcluster/src/VCluster.cpp ../openfpm_devices/src/memory/PtrMemory.cpp
-vector_dist_CXXFLAGS = $(PARMETIS_INCLUDE) $(METIS_INCLUDE) $(CUDA_CFLAGS) $(INCLUDES_PATH) $(HDF5_CPPFLAGS) $(BOOST_CPPFLAGS) -I../src -Wno-unused-function -Wno-unused-local-typedefs
+vector_dist_CXXFLAGS = $(PARMETIS_INCLUDE) $(METIS_INCLUDE) $(CUDA_CFLAGS) $(INCLUDES_PATH) $(HDF5_CPPFLAGS) $(BOOST_CPPFLAGS) -I../src -I/usr/local/libhilbert/include -Wno-unused-function -Wno-unused-local-typedefs
vector_dist_CFLAGS = $(CUDA_CFLAGS)
vector_dist_LDADD = $(LINKLIBS) -lparmetis -lmetis
diff --git a/openfpm_data b/openfpm_data
index 21d50e7f0..82bddfe80 160000
--- a/openfpm_data
+++ b/openfpm_data
@@ -1 +1 @@
-Subproject commit 21d50e7f0f796ceb5699dcf6939183572aba2ba0
+Subproject commit 82bddfe80a08243f00acae696b42920c439f2bb3
diff --git a/openfpm_devices b/openfpm_devices
index 68d2e016d..d46372d3d 160000
--- a/openfpm_devices
+++ b/openfpm_devices
@@ -1 +1 @@
-Subproject commit 68d2e016d10ddcdf201bbc6cd4270ff0e4623d02
+Subproject commit d46372d3db114dd2dc95b3b03d7d9b906287d253
diff --git a/openfpm_io b/openfpm_io
index d07c3c784..78f8cff03 160000
--- a/openfpm_io
+++ b/openfpm_io
@@ -1 +1 @@
-Subproject commit d07c3c7848e446437526d0bbda0843c18ab6a925
+Subproject commit 78f8cff03f684d2ddebed35b11c33bd0d9672a75
diff --git a/openfpm_numerics b/openfpm_numerics
index 759fb6002..cd85b0454 160000
--- a/openfpm_numerics
+++ b/openfpm_numerics
@@ -1 +1 @@
-Subproject commit 759fb600260b785c9d282c10a9f1d6596767ec78
+Subproject commit cd85b04542371e1ed08507f52172b10e0310a35a
diff --git a/src/Makefile.am b/src/Makefile.am
index afc0ff2e3..148b3fd52 100644
--- a/src/Makefile.am
+++ b/src/Makefile.am
@@ -2,7 +2,7 @@ LINKLIBS = $(METIS_LIB) $(PTHREAD_LIBS) $(OPT_LIBS) $(BOOST_LDFLAGS) $(BOOST_IOS
noinst_PROGRAMS = pdata
pdata_SOURCES = main.cpp lib/pdata.cpp test_multiple_o.cpp ../openfpm_devices/src/memory/HeapMemory.cpp ../openfpm_devices/src/memory/PtrMemory.cpp ../openfpm_vcluster/src/VCluster.cpp ../openfpm_devices/src/Memleak_check.cpp
-pdata_CXXFLAGS = $(HDF5_CPPFLAGS) $(CUDA_CFLAGS) $(INCLUDES_PATH) $(PARMETIS_INCLUDE) $(METIS_INCLUDE) $(BOOST_CPPFLAGS) (H5PART_INCLUDE) -DPARALLEL_IO -I/usr/local/libhilbert/include -Wno-unused-local-typedefs
+pdata_CXXFLAGS = $(HDF5_CPPFLAGS) $(CUDA_CFLAGS) $(INCLUDES_PATH) $(PARMETIS_INCLUDE) $(METIS_INCLUDE) $(BOOST_CPPFLAGS) $(H5PART_INCLUDE) -DPARALLEL_IO -I/usr/local/libhilbert/include -Wno-unused-local-typedefs
pdata_CFLAGS = $(CUDA_CFLAGS)
pdata_LDADD = $(LINKLIBS) -lparmetis -lmetis -L/usr/local/libhilbert/lib -lhilbert
nobase_include_HEADERS = Decomposition/CartDecomposition.hpp Decomposition/common.hpp Decomposition/Decomposition.hpp Decomposition/ie_ghost.hpp \
diff --git a/src/Vector/vector_dist.hpp b/src/Vector/vector_dist.hpp
index b340e276d..30b8cf298 100644
--- a/src/Vector/vector_dist.hpp
+++ b/src/Vector/vector_dist.hpp
@@ -17,6 +17,7 @@
#include "memory/PreAllocHeapMemory.hpp"
#include "memory/PtrMemory.hpp"
#include "NN/CellList/CellList.hpp"
+#include "NN/CellList/CellListFast_hilb.hpp"
#include "util/common.hpp"
#include "util/object_util.hpp"
#include "memory/ExtPreAlloc.hpp"
@@ -660,6 +661,35 @@ private:
}
}
+ /*! \brief Calculate parameters for the cell list
+ *
+ * \param div Division array
+ * \param r_cut interation radius or size of each cell
+ * \param enlarge In case of padding particles the cell list must be enlarged, like a ghost. This parameter says how much must be enlarged
+ *
+ * \return the processor bounding box
+ */
+ inline Box<dim, St> cl_param_calculate(size_t (&div)[dim], St r_cut, const Ghost<dim, St> & enlarge)
+
+ {
+ // calculate the parameters of the cell list
+
+ // get the processor bounding box
+ Box<dim, St> pbox = dec.getProcessorBounds();
+
+ // extend by the ghost
+ pbox.enlarge(enlarge);
+
+ // Calculate the division array and the cell box
+ for (size_t i = 0; i < dim; i++)
+ {
+ div[i] = static_cast<size_t>((pbox.getP2().get(i) - pbox.getP1().get(i)) / r_cut);
+ div[i]++;
+ pbox.setHigh(i,pbox.getLow(i) + div[i]*r_cut);
+ }
+ return pbox;
+ }
+
public:
/*! \brief Constructor
@@ -1036,6 +1066,24 @@ public:
return getCellList(r_cut, g);
}
+ /*! \brief Construct an hilbert cell list starting from the stored particles
+ *
+ * \tparam CellL CellList type to construct
+ *
+ * \param r_cut interation radius, or size of each cell
+ *
+ * \return the Cell list
+ *
+ */
+ template<typename CellL = CellList_hilb<dim, St, FAST, shift<dim, St> > > CellL getCellList_hilb(St r_cut)
+ {
+ // Get ghost and anlarge by 1%
+ Ghost<dim,St> g = dec.getGhost();
+ g.magnify(1.01);
+
+ return getCellList_hilb(r_cut, g);
+ }
+
/*! \brief Update a cell list using the stored particles
*
* \tparam CellL CellList type to construct
@@ -1062,6 +1110,37 @@ public:
}
}
+ /*template<typename CellL> void celllist_initialize(CellL & cell_list, St r_cut, const Ghost<dim, St> & enlarge)
+ {
+ // calculate the parameters of the cell list
+
+ // get the processor bounding box
+ Box<dim, St> pbox = dec.getProcessorBounds();
+ // extend by the ghost
+ pbox.enlarge(enlarge);
+
+ size_t div[dim];
+
+ // Calculate the division array and the cell box
+ for (size_t i = 0; i < dim; i++)
+ {
+ div[i] = static_cast<size_t>((pbox.getP2().get(i) - pbox.getP1().get(i)) / r_cut);
+ div[i]++;
+ pbox.setHigh(i,pbox.getLow(i) + div[i]*r_cut);
+ }
+
+ if(is_Hilbert<CellL>::type::value == true)
+ {
+ std::cout << "Case 1: " << demangle(typeid(CellL).name()) << std::endl;
+ cell_list.Initialize(pbox, div, g_m);
+ }
+ else
+ {
+ std::cout << "Case 2: " << demangle(typeid(CellL).name()) << std::endl;
+ cell_list.Initialize(pbox, div);
+ }
+ }*/
+
/*! \brief Construct a cell list starting from the stored particles
*
* It differ from the get getCellList for an additional parameter, in case the
@@ -1079,23 +1158,43 @@ public:
{
CellL cell_list;
- // calculate the parameters of the cell list
+ // Division array
+ size_t div[dim];
- // get the processor bounding box
- Box<dim, St> pbox = dec.getProcessorBounds();
- // extend by the ghost
- pbox.enlarge(enlarge);
+ // Processor bounding box
+ auto pbox = cl_param_calculate(div, r_cut, enlarge);
+
+ cell_list.Initialize(pbox, div);
+
+ updateCellList(cell_list);
+
+ return cell_list;
+ }
+ /*! \brief Construct an hilbert cell list starting from the stored particles
+ *
+ * It differ from the get getCellList for an additional parameter, in case the
+ * domain + ghost is not big enough to contain additional padding particles, a Cell list
+ * with bigger space can be created
+ * (padding particles in general are particles added by the user out of the domains)
+ *
+ * \tparam CellL CellList type to construct
+ *
+ * \param r_cut interation radius, or size of each cell
+ * \param enlarge In case of padding particles the cell list must be enlarged, like a ghost this parameter say how much must be enlarged
+ *
+ */
+ template<typename CellL = CellList_hilb<dim, St, FAST, shift<dim, St> > > CellL getCellList_hilb(St r_cut, const Ghost<dim, St> & enlarge)
+ {
+ CellL cell_list;
+
+ // Division array
size_t div[dim];
- // Calculate the division array and the cell box
- for (size_t i = 0; i < dim; i++)
- {
- div[i] = static_cast<size_t>((pbox.getP2().get(i) - pbox.getP1().get(i)) / r_cut);
- div[i]++;
- }
+ // Processor bounding box
+ auto pbox = cl_param_calculate(div, r_cut, enlarge);
- cell_list.Initialize(pbox, div);
+ cell_list.Initialize(pbox, div, g_m);
updateCellList(cell_list);
@@ -1202,20 +1301,15 @@ public:
// extend by the ghost
pbox.enlarge(enlarge);
- Box<dim,St> cell_box;
-
size_t div[dim];
// Calculate the division array and the cell box
for (size_t i = 0 ; i < dim ; i++)
{
div[i] = 1 << m;
-
- cell_box.setLow(i,0.0);
- cell_box.setHigh(i,pbox.getP2().get(i) - pbox.getP1().get(i));
}
- cell_list.Initialize(cell_box,div,pbox.getP1());
+ cell_list.Initialize(pbox,div);
// for each particle add the particle to the cell list
@@ -1225,7 +1319,7 @@ public:
{
auto key = it.get();
- cell_list.add(this->template getPos<0>(key),key.getKey());
+ cell_list.add(this->getPos(key),key.getKey());
++it;
}
diff --git a/src/Vector/vector_dist_unit_test.hpp b/src/Vector/vector_dist_unit_test.hpp
index 43e23886e..3f1007dcc 100644
--- a/src/Vector/vector_dist_unit_test.hpp
+++ b/src/Vector/vector_dist_unit_test.hpp
@@ -11,6 +11,7 @@
#include <random>
#include "Vector/vector_dist.hpp"
#include "data_type/aggregate.hpp"
+#include "Vector/vector_dist_cl_hilb_performance_tests.hpp"
/*! \brief Count the total number of particles
*
@@ -1161,11 +1162,11 @@ BOOST_AUTO_TEST_CASE( vector_dist_cell_verlet_test )
}
}
-BOOST_AUTO_TEST_CASE( vector_dist_hilbert_timer_test )
+BOOST_AUTO_TEST_CASE( vector_dist_hilbert_2d_benchmark_test )
{
typedef Point<2,float> s;
- Vcluster & v_cl = *global_v_cluster;
+ Vcluster & v_cl = create_vcluster();
// set the seed
// create the random generator engine
@@ -1200,8 +1201,8 @@ BOOST_AUTO_TEST_CASE( vector_dist_hilbert_timer_test )
{
auto key = it.get();
- vd.template getPos<s::x>(key)[0] = ud(eg);
- vd.template getPos<s::x>(key)[1] = ud(eg);
+ vd.getPos(key)[0] = ud(eg);
+ vd.getPos(key)[1] = ud(eg);
++it;
}
@@ -1236,6 +1237,178 @@ BOOST_AUTO_TEST_CASE( vector_dist_hilbert_timer_test )
}
}
+BOOST_AUTO_TEST_CASE( vector_dist_cl_hilb_forces_test )
+{
+ ///////// INPUT DATA //////////
+
+ // Dimensionality of the space
+ const size_t dim = 2;
+ // Cut-off radiuses. Can be put different number of values
+ openfpm::vector<float> cl_r_cutoff {0.05};
+ // The starting amount of particles (remember that this number is multiplied by number of processors you use for testing)
+ size_t cl_k_start = 100000;
+ // The lower threshold for number of particles
+ size_t cl_k_min = 10000;
+ // Ghost part of distributed vector
+ double ghost_part = 0.05;
+
+ ///////////////////////////////
+
+ //For different r_cut
+ for (size_t r = 0; r < cl_r_cutoff.size(); r++ )
+ {
+ Vcluster & v_cl = create_vcluster();
+
+ //Cut-off radius
+ float r_cut = cl_r_cutoff.get(r);
+
+ //Number of particles
+ size_t k = cl_k_start * v_cl.getProcessingUnits();
+
+ std::string str("Testing " + std::to_string(dim) + "D vector's forces k<=");
+
+ vector_dist_test::print_test_v(str,k);
+
+ std::cout << std::endl << "r_cut is " << r_cut << std::endl << std::endl;
+
+ //For different number of particles
+ for (size_t k_int = k ; k_int >= cl_k_min ; k_int/=2 )
+ {
+ BOOST_TEST_CHECKPOINT( "Testing " << dim << "D vector's forces k<=" << k_int );
+
+ std::cout << "k_int: " << k_int << std::endl;
+
+ Box<dim,float> box;
+
+ for (size_t i = 0; i < dim; i++)
+ {
+ box.setLow(i,0.0);
+ box.setHigh(i,1.0);
+ }
+
+ // Boundary conditions
+ size_t bc[dim];
+
+ for (size_t i = 0; i < dim; i++)
+ bc[i] = PERIODIC;
+
+ vector_dist<dim,float, aggregate<float[dim]>, CartDecomposition<dim,float> > vd(k_int,box,bc,Ghost<dim,float>(ghost_part));
+
+ vector_dist<dim,float, aggregate<float[dim]>, CartDecomposition<dim,float> > vd2(k_int,box,bc,Ghost<dim,float>(ghost_part));
+
+ // Initialize dist vectors
+ vd_initialize_double<dim>(vd, vd2, v_cl, k_int);
+
+ vd.template ghost_get<0>();
+ vd2.template ghost_get<0>();
+
+ //Get a cell list
+
+ auto NN = vd.getCellList(r_cut);
+
+ //Calculate forces
+
+ auto t_forces = vector_dist_cl_perf_test::calculate_forces<dim>(NN,vd,r_cut);
+
+ //Get a cell list hilb
+
+ auto NN_hilb = vd2.getCellList_hilb(r_cut);
+
+ //Calculate forces
+ auto t_forces_hilb = calculate_forces_hilb<dim>(NN_hilb,vd2,r_cut);
+
+ auto it_v = vd.getIterator();
+
+ while (it_v.isNext())
+ {
+ //key
+ vect_dist_key_dx key = it_v.get();
+
+ for (size_t i = 0; i < dim; i++)
+ {
+ auto a1 = vd.template getProp<0>(key)[i];
+ auto a2 = vd2.template getProp<0>(key)[i];
+ //Check that the forces are equal
+ BOOST_REQUIRE_EQUAL(a1,a2);
+ }
+
+ ++it_v;
+ }
+ }
+ }
+}
+
+BOOST_AUTO_TEST_CASE( stupid_test )
+{
+ ///////// INPUT DATA //////////
+
+ // Dimensionality of the space
+ const size_t dim = 2;
+ // Cut-off radiuses. Can be put different number of values
+ openfpm::vector<float> cl_r_cutoff {0.05};
+ // The starting amount of particles (remember that this number is multiplied by number of processors you use for testing)
+ size_t cl_k_start = 100000;
+ // The lower threshold for number of particles
+ size_t cl_k_min = 10000;
+ // Ghost part of distributed vector
+ double ghost_part = 0.05;
+
+ ///////////////////////////////
+
+ //For different r_cut
+ for (size_t r = 0; r < cl_r_cutoff.size(); r++ )
+ {
+ Vcluster & v_cl = create_vcluster();
+
+ //Cut-off radius
+ float r_cut = cl_r_cutoff.get(r);
+
+ //Number of particles
+ size_t k = cl_k_start * v_cl.getProcessingUnits();
+
+ std::string str("Testing " + std::to_string(dim) + "D vector's forces k<=");
+
+ vector_dist_test::print_test_v(str,k);
+
+ std::cout << std::endl << "r_cut is " << r_cut << std::endl << std::endl;
+
+ //For different number of particles
+ for (size_t k_int = k ; k_int >= cl_k_min ; k_int/=2 )
+ {
+ BOOST_TEST_CHECKPOINT( "Testing " << dim << "D vector's forces k<=" << k_int );
+
+ std::cout << "k_int: " << k_int << std::endl;
+
+ Box<dim,float> box;
+
+ for (size_t i = 0; i < dim; i++)
+ {
+ box.setLow(i,0.0);
+ box.setHigh(i,1.0);
+ }
+
+ // Boundary conditions
+ size_t bc[dim];
+
+ for (size_t i = 0; i < dim; i++)
+ bc[i] = PERIODIC;
+
+ vector_dist<dim,float, aggregate<float[dim]>, CartDecomposition<dim,float> > vd(k_int,box,bc,Ghost<dim,float>(ghost_part));
+
+ vector_dist<dim,float, aggregate<float[dim]>, CartDecomposition<dim,float> > vd2(k_int,box,bc,Ghost<dim,float>(ghost_part));
+
+ // Initialize dist vectors
+ vd_initialize_double<dim>(vd, vd2, v_cl, k_int);
+
+ vd.template ghost_get<0>();
+ vd2.template ghost_get<0>();
+
+ //Get a cell list
+
+ auto NN = vd.getCellList(r_cut);
+ }
+ }
+}
BOOST_AUTO_TEST_SUITE_END()
#endif /* VECTOR_DIST_UNIT_TEST_HPP_ */
diff --git a/src/main.cpp b/src/main.cpp
index 4a178acf8..01ef322b5 100644
--- a/src/main.cpp
+++ b/src/main.cpp
@@ -26,6 +26,7 @@
#include "dec_optimizer_unit_test.hpp"
#include "Grid/grid_dist_id_unit_test.hpp"
#include "Vector/vector_dist_unit_test.hpp"
+#include "Vector/vector_dist_HDF5_save_test.hpp"
//#ifdef PERFORMANCE_TEST
#include "pdata_performance.hpp"
//#endif
--
GitLab