Commit c63bfb61 authored by Jakub Klinkovský's avatar Jakub Klinkovský
Browse files

update for TNL changes

parent e0de6ea6

Makefile

+4 −1
Original line number Diff line number Diff line
@@ -22,7 +22,10 @@ cuda: tnl-benchmark-mesh-cuda 
tnl-benchmark-mesh-cuda: tnl-benchmark-mesh-cuda.cu.o $(MESH_BENCHMARK_TEMPLATES_CU:%.cu=%.cu.o)

	$(CUDA_COMPILER) $(CUDA_LDFLAGS) -o $@ $^ $(CUDA_LDLIBS)



clean:

clean: clean_templates

.PHONY: clean_templates

clean_templates:

	$(RM) tnl-benchmark-mesh tnl-benchmark-mesh-cuda

	$(RM) -r MeshBenchmarks.templates/



-include $(SOURCES:%.cpp=%.d)

Makefile.base

+13 −2
Original line number Diff line number Diff line
@@ -33,24 +33,35 @@ endif 


## essential host options

CPPFLAGS = -MD -MP

CXXFLAGS = -std=c++11 -Wall -Wno-unused-local-typedefs -Wno-unused-variable

CXXFLAGS = -std=c++14 -Wall -Wno-unused-local-typedefs -Wno-unused-variable -Wno-deprecated

LDFLAGS = -pthread

LDLIBS =



ifeq ($(BUILD),Release)

CXXFLAGS += -O3 -march=native -mtune=native -DNDEBUG

#CXXFLAGS += -fopt-info-vec-optimized=explicit.log -ftree-vectorize

# enable link-time optimization

ifeq ($(CXX),clang++)

CXXFLAGS += -flto=thin

LDFLAGS += -O3 -march=native -mtune=native -flto=thin

endif

else ifeq ($(BUILD),Debug)

CXXFLAGS += -Og -g

LDFLAGS += -rdynamic

endif





## use the LLVM linker with clang

ifeq ($(CXX),clang++)

LDFLAGS += -fuse-ld=lld

endif





## options for nvcc

# automatic dependency generation for nvcc (gcc has it automated in the pre-processor phase with the -MD flag)

# note that $@ is expanded only when $(CUDA_CPPFLAGS) is used

CUDA_CPPFLAGS = --compiler-options -MD,-MP,-MT$@

CUDA_CXXFLAGS = -std=c++11 --compiler-bindir $(CUDA_HOST_COMPILER)

CUDA_CXXFLAGS = -std=c++14 --compiler-bindir $(CUDA_HOST_COMPILER)

CUDA_LDLIBS =

# disable false compiler warnings

#   reference for the -Xcudafe flag: http://stackoverflow.com/questions/14831051/how-to-disable-compiler-warnings-with-nvcc/17095910#17095910

MeshBenchmarks.h

+50 −44
Original line number Diff line number Diff line
@@ -17,9 +17,11 @@ 
#include <TNL/Meshes/Geometry/getEntityCenter.h>

#include <TNL/Meshes/Geometry/getEntityMeasure.h>

#include <TNL/Meshes/TypeResolver/TypeResolver.h>

#include <TNL/DevicePointer.h>

#include <TNL/ParallelFor.h>

#include <TNL/StaticFor.h>

#include <TNL/Pointers/DevicePointer.h>

#include <TNL/Algorithms/ParallelFor.h>

#include <TNL/Algorithms/TemplateStaticFor.h>

#include <TNL/Benchmarks/Benchmarks.h>

#include <TNL/Communicators/NoDistrCommunicator.h>



#ifdef HAVE_CUDA

#include <cuda_profiler_api.h>
@@ -28,11 +30,10 @@ 
#include "MeshOrdering.h"



#include "MeshConfigs.h"

#include "tnl_benchmarks.h"



using namespace TNL;

using namespace TNL::Meshes;

using namespace TNL::benchmarks;

using namespace TNL::Benchmarks;



template< typename Real >

__cuda_callable__
@@ -72,7 +73,7 @@ struct MeshBenchmarks 
      Benchmark::MetadataColumns metadataColumns = {

//         {"mesh-file", meshFile},

         {"config", Mesh::Config::getConfigType()},

         {"topology", Mesh::Config::CellTopology::getType().replace("Topologies::", "")},

         {"topology", getType< typename Mesh::Config::CellTopology >().replace("Topologies::", "")},

//         {"wrld dim", worldDimension},

         {"real", getType< typename Mesh::RealType >()},

         {"gid_t", getType< typename Mesh::GlobalIndexType >()},
@@ -82,7 +83,8 @@ struct MeshBenchmarks 
      };



      Mesh mesh;

      if( ! loadMesh( meshFile, mesh ) ) {

      DistributedMeshes::DistributedMesh<Mesh> distributedMesh;

      if( ! loadMesh< Communicators::NoDistrCommunicator >( meshFile, mesh, distributedMesh ) ) {

         std::cerr << "Failed to load mesh from file '" << meshFile << "'." << std::endl;

         return false;

      }
@@ -121,8 +123,8 @@ struct MeshBenchmarks 


   static void dispatchAlgorithms( Benchmark & benchmark, const Mesh & mesh )

   {

      StaticFor< int, 1, Mesh::getMeshDimension() + 1, CentersDispatch >::execHost( benchmark, mesh );

      StaticFor< int, 1, Mesh::getMeshDimension() + 1, MeasuresDispatch >::execHost( benchmark, mesh );

      Algorithms::TemplateStaticFor< int, 1, Mesh::getMeshDimension() + 1, CentersDispatch >::execHost( benchmark, mesh );

      Algorithms::TemplateStaticFor< int, 1, Mesh::getMeshDimension() + 1, MeasuresDispatch >::execHost( benchmark, mesh );

      DualMeasuresDispatch::exec( benchmark, mesh );

      SpheresDispatch::exec( benchmark, mesh );

   }
@@ -134,7 +136,7 @@ struct MeshBenchmarks 
                typename = typename std::enable_if< M::template entitiesAvailable< EntityDimension >() >::type >

      static void exec( Benchmark & benchmark, const M & mesh )

      {

         benchmark.setOperation( String("Centers (d = ") + String(EntityDimension) + ")" );

         benchmark.setOperation( String("Centers (d = ") + convertToString(EntityDimension) + ")" );

         benchmark_centers< EntityDimension, Devices::Host >( benchmark, mesh );

#ifdef HAVE_CUDA

         benchmark_centers< EntityDimension, Devices::Cuda >( benchmark, mesh );
@@ -156,7 +158,7 @@ struct MeshBenchmarks 
                typename = typename std::enable_if< M::template entitiesAvailable< EntityDimension >() >::type >

      static void exec( Benchmark & benchmark, const M & mesh )

      {

         benchmark.setOperation( String("Measures (d = ") + String(EntityDimension) + ")" );

         benchmark.setOperation( String("Measures (d = ") + convertToString(EntityDimension) + ")" );

         benchmark_measures< EntityDimension, Devices::Host >( benchmark, mesh );

#ifdef HAVE_CUDA

         benchmark_measures< EntityDimension, Devices::Cuda >( benchmark, mesh );
@@ -238,9 +240,9 @@ struct MeshBenchmarks 
      const Index entitiesCount = mesh_src.template getEntitiesCount< EntityDimension >();



      const DeviceMesh mesh = mesh_src;

      DevicePointer< const DeviceMesh > meshPointer( mesh );

      Pointers::DevicePointer< const DeviceMesh > meshPointer( mesh );

      Containers::Array< PointType, Device, Index > centers;

      centers.setSize( PointType::size * entitiesCount );

      centers.setSize( PointType::getSize() * entitiesCount );



      auto kernel_measures = [] __cuda_callable__

         ( Index i,
@@ -256,13 +258,14 @@ struct MeshBenchmarks 
      };



      auto benchmark_func = [&] () {

         ParallelFor< Device >::exec( (Index) 0, entitiesCount,

         Algorithms::ParallelFor< Device >::exec(

               (Index) 0, entitiesCount,

               kernel_measures,

               &meshPointer.template getData< Device >(),

               centers.getData() );

      };



      benchmark.time( reset,

      benchmark.time< Device >( reset,

                                (std::is_same< Device, Devices::Host >::value) ? "CPU" : "GPU",

                                benchmark_func );

   }
@@ -277,7 +280,7 @@ struct MeshBenchmarks 
      const Index entitiesCount = mesh_src.template getEntitiesCount< EntityDimension >();



      const DeviceMesh mesh = mesh_src;

      DevicePointer< const DeviceMesh > meshPointer( mesh );

      Pointers::DevicePointer< const DeviceMesh > meshPointer( mesh );

      Containers::Array< Real, Device, Index > measures;

      measures.setSize( entitiesCount );
@@ -295,13 +298,14 @@ struct MeshBenchmarks 
      };



      auto benchmark_func = [&] () {

         ParallelFor< Device >::exec( (Index) 0, entitiesCount,

         Algorithms::ParallelFor< Device >::exec(

               (Index) 0, entitiesCount,

               kernel_measures,

               &meshPointer.template getData< Device >(),

               measures.getData() );

      };



      benchmark.time( reset,

      benchmark.time< Device >( reset,

                                (std::is_same< Device, Devices::Host >::value) ? "CPU" : "GPU",

                                benchmark_func );

   }
@@ -323,7 +327,7 @@ struct MeshBenchmarks 
      const Index entitiesCount = mesh_src.template getEntitiesCount< Mesh::getMeshDimension() >();



      const DeviceMesh mesh = mesh_src;

      DevicePointer< const DeviceMesh > meshPointer( mesh );

      Pointers::DevicePointer< const DeviceMesh > meshPointer( mesh );

      Containers::Array< Real, Device, Index > measures;

      measures.setSize( entitiesCount );
@@ -360,13 +364,14 @@ struct MeshBenchmarks 
      };



      auto benchmark_func = [&] () {

         ParallelFor< Device >::exec( (Index) 0, entitiesCount,

         Algorithms::ParallelFor< Device >::exec(

               (Index) 0, entitiesCount,

               kernel_measures,

               &meshPointer.template getData< Device >(),

               measures.getData() );

      };



      benchmark.time( reset,

      benchmark.time< Device >( reset,

                                (std::is_same< Device, Devices::Host >::value) ? "CPU" : "GPU",

                                benchmark_func );

   }
@@ -386,7 +391,7 @@ struct MeshBenchmarks 
      const Index entitiesCount = mesh_src.template getEntitiesCount< 0 >();



      const DeviceMesh mesh = mesh_src;

      DevicePointer< const DeviceMesh > meshPointer( mesh );

      Pointers::DevicePointer< const DeviceMesh > meshPointer( mesh );

      Containers::Array< Real, Device, Index > spheres;

      spheres.setSize( entitiesCount );
@@ -452,13 +457,14 @@ struct MeshBenchmarks 
      };



      auto benchmark_func = [&] () {

         ParallelFor< Device >::exec( (Index) 0, entitiesCount,

         Algorithms::ParallelFor< Device >::exec(

               (Index) 0, entitiesCount,

               kernel_spheres,

               &meshPointer.template getData< Device >(),

               spheres.getData() );

      };



      benchmark.time( reset,

      benchmark.time< Device >( reset,

                                (std::is_same< Device, Devices::Host >::value) ? "CPU" : "GPU",

                                benchmark_func );

   }

MeshConfigs.h

+3 −26
Original line number Diff line number Diff line
@@ -13,7 +13,6 @@ 
#pragma once



#include <TNL/String.h>

#include <TNL/param-types.h>

#include <TNL/Meshes/Topologies/SubentityVertexMap.h>



template< typename Cell,
@@ -33,17 +32,6 @@ struct FullConfig 
   static constexpr int worldDimension = WorldDimension;

   static constexpr int meshDimension = Cell::dimension;



   static TNL::String getType()

   {

      return TNL::String( "Meshes::FullConfig< " ) +

             Cell::getType() + ", " +

             TNL::String( WorldDimension ) + ", " +

             TNL::getType< Real >() + ", " +

             TNL::getType< GlobalIndex >() + ", " +

             TNL::getType< LocalIndex >() + ", " +

             TNL::getType< Id >() + " >";

   }



   static TNL::String getConfigType()

   {

      return "Full";
@@ -118,17 +106,6 @@ struct MinimalConfig 
   static constexpr int worldDimension = WorldDimension;

   static constexpr int meshDimension = Cell::dimension;



   static TNL::String getType()

   {

      return TNL::String( "Meshes::MinimalConfig< " ) +

             Cell::getType() + ", " +

             TNL::String( WorldDimension ) + ", " +

             TNL::getType< Real >() + ", " +

             TNL::getType< GlobalIndex >() + ", " +

             TNL::getType< LocalIndex >() + ", " +

             TNL::getType< Id >() + " >";

   }



   static TNL::String getConfigType()

   {

      return "Minimal";

MeshOrdering.h

+6 −6
Original line number Diff line number Diff line
@@ -27,7 +27,7 @@ struct KdTreeOrdering 
        perm.setSize( numberOfEntities );

        iperm.setSize( numberOfEntities );



        spatial::idle_point_multimap< PointType::size, PointType, IndexType > container;

        spatial::idle_point_multimap< PointType::getSize(), PointType, IndexType > container;



        for( IndexType i = 0; i < numberOfEntities; i++ ) {

            const auto& entity = mesh.template getEntity< MeshEntity >( i );
@@ -312,7 +312,7 @@ public: 
        Vector tmp;

        tmp.setLike( vector );



        ParallelFor< Device >::exec( (IndexType) 0, vector.getSize(),

        Algorithms::ParallelFor< Device >::exec( (IndexType) 0, vector.getSize(),

                                                 kernel,

                                                 vector.getData(),

                                                 tmp.getData(),