Merge branch 'JB/mesh' into 'develop' (f220426b) · Commits · TNL / tnl-dev

src/Benchmarks/CMakeLists.txt

+1 −0

Original line number	Diff line number	Diff line
		@@ -7,3 +7,4 @@ add_subdirectory( LinearSolvers )
		add_subdirectory( ODESolvers )
		add_subdirectory( Sorting )
		add_subdirectory( Traversers )
		add_subdirectory( Mesh )

src/Benchmarks/Mesh/CMakeLists.txt

0 → 100644

+35 −0

Original line number	Diff line number	Diff line
		if( BUILD_CUDA )
		CUDA_ADD_EXECUTABLE( tnl-benchmark-mesh-cuda tnl-benchmark-mesh.cu )

		find_package( tinyxml2 QUIET )
		if( tinyxml2_FOUND )
		target_compile_definitions(tnl-benchmark-mesh-cuda PUBLIC "-DHAVE_TINYXML2")
		target_link_libraries(tnl-benchmark-mesh-cuda tinyxml2::tinyxml2)
		endif()

		find_package( ZLIB )
		if( ZLIB_FOUND )
		target_compile_definitions(tnl-benchmark-mesh-cuda PUBLIC "-DHAVE_ZLIB")
		target_include_directories(tnl-benchmark-mesh-cuda PUBLIC ${ZLIB_INCLUDE_DIRS})
		target_link_libraries(tnl-benchmark-mesh-cuda ${ZLIB_LIBRARIES})
		endif()

		install( TARGETS tnl-benchmark-mesh-cuda RUNTIME DESTINATION bin )
		endif()

		ADD_EXECUTABLE( tnl-benchmark-mesh tnl-benchmark-mesh.cpp )

		find_package( tinyxml2 QUIET )
		if( tinyxml2_FOUND )
		target_compile_definitions(tnl-benchmark-mesh PUBLIC "-DHAVE_TINYXML2")
		target_link_libraries(tnl-benchmark-mesh tinyxml2::tinyxml2)
		endif()

		find_package( ZLIB )
		if( ZLIB_FOUND )
		target_compile_definitions(tnl-benchmark-mesh PUBLIC "-DHAVE_ZLIB")
		target_include_directories(tnl-benchmark-mesh PUBLIC ${ZLIB_INCLUDE_DIRS})
		target_link_libraries(tnl-benchmark-mesh ${ZLIB_LIBRARIES})
		endif()

		install( TARGETS tnl-benchmark-mesh RUNTIME DESTINATION bin )

src/Benchmarks/Mesh/MemoryInfo.h

0 → 100644

+195 −0

Original line number	Diff line number	Diff line
		#pragma once

		// References:
		// - https://stackoverflow.com/a/64166/4180822
		// - https://lemire.me/blog/2020/03/03/calling-free-or-delete/
		// - https://stackoverflow.com/questions/15529643/what-does-malloc-trim0-really-mean

		#include <stdlib.h>
		#include <stdio.h>
		#include <string.h>

		#include <sys/types.h>
		#include <sys/sysinfo.h>
		#include <malloc.h>

		inline long
		getTotalVirtualMemory()
		{
		struct sysinfo memInfo;
		sysinfo (&memInfo);

		long totalVirtualMem = memInfo.totalram;
		// Add other values in next statement to avoid int overflow on right hand side...
		totalVirtualMem += memInfo.totalswap;
		totalVirtualMem *= memInfo.mem_unit;

		return totalVirtualMem;
		}

		inline long
		getUsedVirtualMemory()
		{
		struct sysinfo memInfo;
		sysinfo (&memInfo);

		long virtualMemUsed = memInfo.totalram - memInfo.freeram;
		// Add other values in next statement to avoid int overflow on right hand side...
		virtualMemUsed += memInfo.totalswap - memInfo.freeswap;
		virtualMemUsed *= memInfo.mem_unit;

		return virtualMemUsed;
		}

		inline long
		parseLine(char* line)
		{
		// This assumes that a digit will be found and the line ends in " kB".
		int i = strlen(line);
		const char* p = line;
		while (p <'0' \|\| p > '9') p++;
		line[i-3] = '\0';
		return atol(p);
		}

		// virtual memory currently used by the calling process
		inline long
		getSelfVirtualMemory()
		{
		// explicitly release unused memory
		malloc_trim(0);

		FILE* file = fopen("/proc/self/status", "r");
		long result = -1;
		char line[128];

		while (fgets(line, 128, file) != NULL){
		if (strncmp(line, "VmSize:", 7) == 0){
		// convert from kB to B
		result = parseLine(line) * 1024;
		break;
		}
		}
		fclose(file);
		return result;
		}

		inline long
		getTotalPhysicalMemory()
		{
		struct sysinfo memInfo;
		sysinfo (&memInfo);

		long totalPhysMem = memInfo.totalram;
		//Multiply in next statement to avoid int overflow on right hand side...
		totalPhysMem *= memInfo.mem_unit;

		return totalPhysMem;
		}

		inline long
		getUsedPhysicalMemory()
		{
		struct sysinfo memInfo;
		sysinfo (&memInfo);

		long physMemUsed = memInfo.totalram - memInfo.freeram;
		//Multiply in next statement to avoid int overflow on right hand side...
		physMemUsed *= memInfo.mem_unit;

		return physMemUsed;
		}

		inline long
		getSelfPhysicalMemory()
		{
		// explicitly release unused memory
		malloc_trim(0);

		FILE* file = fopen("/proc/self/status", "r");
		long result = -1;
		char line[128];

		while (fgets(line, 128, file) != NULL){
		if (strncmp(line, "VmRSS:", 6) == 0){
		// convert from kB to B
		result = parseLine(line) * 1024;
		break;
		}
		}
		fclose(file);
		return result;
		}


		#include <TNL/Benchmarks/Benchmarks.h>
		#include <TNL/Config/ParameterContainer.h>
		#include <TNL/Containers/StaticVector.h>

		struct MemoryBenchmarkResult
		: public TNL::Benchmarks::BenchmarkResult
		{
		using HeaderElements = TNL::Benchmarks::Logging::HeaderElements;
		using RowElements = TNL::Benchmarks::Logging::RowElements;

		double memory = std::numeric_limits<double>::quiet_NaN();
		double memstddev = std::numeric_limits<double>::quiet_NaN();

		virtual HeaderElements getTableHeader() const override
		{
		return HeaderElements({ "time", "stddev", "stddev/time", "bandwidth", "speedup", "memory", "memstddev", "memstddev/memory" });
		}

		virtual RowElements getRowElements() const override
		{
		RowElements elements;
		elements << time << stddev << stddev / time << bandwidth;
		if( speedup != 0 )
		elements << speedup;
		else
		elements << "N/A";
		elements << memory << memstddev << memstddev / memory;
		return elements;
		}
		};

		template< long MAX_COPIES = 10, typename Mesh >
		MemoryBenchmarkResult
		testMemoryUsage( const TNL::Config::ParameterContainer& parameters,
		const Mesh& mesh )
		{
		const size_t memoryLimit = parameters.getParameter< size_t >( "mem-limit" ) * 1024 * 1024;
		TNL::Containers::StaticVector< MAX_COPIES, Mesh > meshes;
		TNL::Containers::StaticVector< MAX_COPIES, double > data;
		data.setValue( 0 );

		long prevCheck = getSelfPhysicalMemory();
		meshes[0] = mesh;
		long check = getSelfPhysicalMemory();
		data[0] = check - prevCheck;
		prevCheck = check;
		const int copies = TNL::min( memoryLimit / data[0], MAX_COPIES - 1 ) + 1;

		for( int i = 1; i < copies; i++ ) {
		meshes[i] = mesh;
		check = getSelfPhysicalMemory();
		data[i] = check - prevCheck;
		prevCheck = check;
		}

		MemoryBenchmarkResult result;

		const double mean = TNL::sum( data ) / (double) copies;
		result.memory = mean / 1024.0 / 1024.0; // MiB

		if( copies > 1 ) {
		for( int i = copies; i < MAX_COPIES; i++ ) {
		data[i] = mean;
		}

		const double stddev = 1.0 / std::sqrt( copies ) * TNL::l2Norm( data - mean );
		result.memstddev = stddev / 1024.0 / 1024.0; // MiB
		}

		return result;
		}

src/Benchmarks/Mesh/MeshBenchmarks.h

0 → 100644

+396 −0

Original line number	Diff line number	Diff line
		/***************************************************************************
		MeshBenchmarks.h - description
		-------------------
		begin : Nov 21, 2017
		copyright : (C) 2017 by Tomas Oberhuber et al.
		email : tomas.oberhuber@fjfi.cvut.cz
		***************************************************************************/

		/* See Copyright Notice in tnl/Copyright */

		// Implemented by: Ján Bobot, Jakub Klinkovský

		#pragma once

		#include <TNL/Meshes/Mesh.h>
		#include <TNL/Meshes/Geometry/getEntityMeasure.h>
		#include <TNL/Meshes/Geometry/getDecomposedMesh.h>
		#include <TNL/Meshes/Geometry/getPlanarMesh.h>
		#include <TNL/Pointers/DevicePointer.h>
		#include <TNL/Algorithms/ParallelFor.h>
		#include <TNL/Algorithms/staticFor.h>
		#include <TNL/Meshes/TypeResolver/resolveMeshType.h>
		#include <TNL/Meshes/Readers/MeshReader.h>
		#include <TNL/Meshes/Topologies/IsDynamicTopology.h>
		#include <TNL/Benchmarks/Benchmarks.h>

		#include "MeshConfigs.h"
		#include "MemoryInfo.h"

		using namespace TNL;
		using namespace TNL::Meshes;
		using namespace TNL::Meshes::Readers;
		using namespace TNL::Benchmarks;

		template< typename Device >
		bool checkDevice( const Config::ParameterContainer& parameters )
		{
		const String device = parameters.getParameter< String >( "devices" );
		if( device == "all" )
		return true;
		if( std::is_same< Device, Devices::Host >::value && device == "host" )
		return true;
		if( std::is_same< Device, Devices::Cuda >::value && device == "cuda" )
		return true;
		return false;
		}

		std::string removeNamespaces( const String & topology )
		{
		std::size_t found = topology.find_last_of("::");
		return topology.substr( found + 1 );
		}

		template< typename Mesh >
		struct MeshBenchmarks
		{
		static_assert( std::is_same< typename Mesh::DeviceType, Devices::Host >::value, "The mesh should be loaded on the host." );

		static bool run( Benchmark<> & benchmark, const Config::ParameterContainer & parameters )
		{
		Logging::MetadataColumns metadataColumns = {
		// {"mesh-file", meshFile},
		{"config", Mesh::Config::getConfigType()},
		//{"topology", removeNamespaces( getType< typename Mesh::Config::CellTopology >() ) },
		//{"space dim", std::to_string( Mesh::Config::spaceDimension )},
		//{"real", getType< typename Mesh::RealType >()},
		//{"gid_t", getType< typename Mesh::GlobalIndexType >()},
		//{"lid_t", getType< typename Mesh::LocalIndexType >()}
		};
		benchmark.setMetadataColumns( metadataColumns );

		const String & meshFile = parameters.getParameter< String >( "mesh-file" );
		auto reader = getMeshReader( meshFile, "auto" );
		Mesh mesh;

		try {
		reader->loadMesh( mesh );
		}
		catch( const Meshes::Readers::MeshReaderError& e ) {
		std::cerr << "Failed to load mesh from file '" << meshFile << "'." << std::endl;
		return false;
		}

		dispatchTests( benchmark, parameters, mesh, reader );

		return true;
		}

		static void dispatchTests( Benchmark<> & benchmark, const Config::ParameterContainer & parameters, const Mesh & mesh, std::shared_ptr< MeshReader > reader )
		{
		ReaderDispatch::exec( benchmark, parameters, reader );
		InitDispatch::exec( benchmark, parameters, reader );
		DecompositionDispatch::exec( benchmark, parameters, mesh );
		PlanarDispatch::exec( benchmark, parameters, mesh );
		MeasuresDispatch::exec( benchmark, parameters, mesh );
		MemoryDispatch::exec( benchmark, parameters, mesh );
		CopyDispatch::exec( benchmark, parameters, mesh );
		}

		struct ReaderDispatch
		{
		static void exec( Benchmark<> & benchmark, const Config::ParameterContainer & parameters, std::shared_ptr< MeshReader > reader )
		{
		benchmark.setOperation( String( "Reader" ) );
		benchmark_reader( benchmark, parameters, reader );
		}
		};

		struct InitDispatch
		{
		static void exec( Benchmark<> & benchmark, const Config::ParameterContainer & parameters, std::shared_ptr< MeshReader > reader )
		{
		benchmark.setOperation( String( "Init" ) );
		benchmark_init( benchmark, parameters, reader );
		}
		};

		struct DecompositionDispatch
		{
		// Polygonal Mesh
		template< typename M,
		std::enable_if_t< std::is_same< typename M::Config::CellTopology, Topologies::Polygon >::value, bool > = true >
		static void exec( Benchmark<> & benchmark, const Config::ParameterContainer & parameters, const M & mesh_src )
		{
		benchmark.setOperation( String( "Decomposition (c)" ) );
		benchmark_decomposition< EntityDecomposerVersion::ConnectEdgesToCentroid >( benchmark, parameters, mesh_src );

		benchmark.setOperation( String( "Decomposition (p)" ) );
		benchmark_decomposition< EntityDecomposerVersion::ConnectEdgesToPoint >( benchmark, parameters, mesh_src );
		}

		// Polyhedral Mesh
		template< typename M,
		std::enable_if_t< std::is_same< typename M::Config::CellTopology, Topologies::Polyhedron >::value, bool > = true >
		static void exec( Benchmark<> & benchmark, const Config::ParameterContainer & parameters, const M & mesh_src )
		{
		benchmark.setOperation( String( "Decomposition (cc)" ) );
		benchmark_decomposition< EntityDecomposerVersion::ConnectEdgesToCentroid,
		EntityDecomposerVersion::ConnectEdgesToCentroid >( benchmark, parameters, mesh_src );

		benchmark.setOperation( String( "Decomposition (cp)" ) );
		benchmark_decomposition< EntityDecomposerVersion::ConnectEdgesToCentroid,
		EntityDecomposerVersion::ConnectEdgesToPoint >( benchmark, parameters, mesh_src );

		benchmark.setOperation( String( "Decomposition (pc)" ) );
		benchmark_decomposition< EntityDecomposerVersion::ConnectEdgesToPoint,
		EntityDecomposerVersion::ConnectEdgesToCentroid >( benchmark, parameters, mesh_src );

		benchmark.setOperation( String( "Decomposition (pp)" ) );
		benchmark_decomposition< EntityDecomposerVersion::ConnectEdgesToPoint,
		EntityDecomposerVersion::ConnectEdgesToPoint >( benchmark, parameters, mesh_src );
		}

		// Other than Polygonal and Polyhedral Mesh
		template< typename M,
		std::enable_if_t< ! std::is_same< typename M::Config::CellTopology, Topologies::Polygon >::value &&
		! std::is_same< typename M::Config::CellTopology, Topologies::Polyhedron >::value, bool > = true >
		static void exec( Benchmark<> & benchmark, const Config::ParameterContainer & parameters, const M & mesh_src )
		{
		}
		};

		struct PlanarDispatch
		{
		template< typename M,
		std::enable_if_t< M::Config::spaceDimension == 3 &&
		(std::is_same< typename M::Config::CellTopology, Topologies::Polygon >::value \|\|
		std::is_same< typename M::Config::CellTopology, Topologies::Polyhedron >::value ), bool > = true >
		static void exec( Benchmark<> & benchmark, const Config::ParameterContainer & parameters, const M & mesh_src )
		{
		benchmark.setOperation( String( "Planar Correction (c)" ) );
		benchmark_planar< EntityDecomposerVersion::ConnectEdgesToCentroid >( benchmark, parameters, mesh_src );

		benchmark.setOperation( String( "Planar Correction (p)" ) );
		benchmark_planar< EntityDecomposerVersion::ConnectEdgesToPoint >( benchmark, parameters, mesh_src );
		}

		template< typename M,
		std::enable_if_t< M::Config::spaceDimension < 3 \|\|
		(! std::is_same< typename M::Config::CellTopology, Topologies::Polygon >::value &&
		! std::is_same< typename M::Config::CellTopology, Topologies::Polyhedron >::value ), bool > = true >
		static void exec( Benchmark<> & benchmark, const Config::ParameterContainer & parameters, const M & mesh_src )
		{
		}
		};

		struct MeasuresDispatch
		{
		template< typename M >
		static void exec( Benchmark<> & benchmark, const Config::ParameterContainer & parameters, const M & mesh )
		{
		benchmark.setOperation( String("Measures") );
		benchmark_measures< Devices::Host >( benchmark, parameters, mesh );
		#ifdef HAVE_CUDA
		benchmark_measures< Devices::Cuda >( benchmark, parameters, mesh );
		#endif
		}
		};

		struct MemoryDispatch
		{
		template< typename M >
		static void exec( Benchmark<> & benchmark, const Config::ParameterContainer& parameters, const M& mesh_src )
		{
		benchmark.setOperation( String("Memory") );
		benchmark_memory( benchmark, parameters, mesh_src );
		}
		};

		struct CopyDispatch
		{
		template< typename M >
		static void exec( Benchmark<> & benchmark, const Config::ParameterContainer & parameters, const M & mesh_src )
		{
		#ifdef HAVE_CUDA
		benchmark.setOperation( String("Copy CPU->GPU") );
		benchmark_copy< Devices::Host, Devices::Cuda >( benchmark, parameters, mesh_src );
		benchmark.setOperation( String("Copy GPU->CPU") );
		benchmark_copy< Devices::Cuda, Devices::Host >( benchmark, parameters, mesh_src );
		#endif
		}
		};

		static void benchmark_reader( Benchmark<> & benchmark, const Config::ParameterContainer & parameters, std::shared_ptr< MeshReader > reader )
		{
		if( ! checkDevice< Devices::Host >( parameters ) )
		return;

		auto reset = [&]() {
		reader->reset();
		};

		auto benchmark_func = [&] () {
		reader->detectMesh();
		};

		benchmark.time< Devices::Host >( reset,
		"CPU",
		benchmark_func );
		}

		static void benchmark_init( Benchmark<> & benchmark, const Config::ParameterContainer & parameters, std::shared_ptr< MeshReader > reader )
		{
		if( ! checkDevice< Devices::Host >( parameters ) )
		return;

		auto reset = [&]() {
		reader->detectMesh();
		};

		auto benchmark_func = [&] () {
		Mesh mesh;
		reader->loadMesh( mesh );
		};

		benchmark.time< Devices::Host >( reset,
		"CPU",
		benchmark_func );
		}

		template< EntityDecomposerVersion DecomposerVersion,
		EntityDecomposerVersion SubDecomposerVersion = EntityDecomposerVersion::ConnectEdgesToPoint,
		typename M >
		static void benchmark_decomposition( Benchmark<> & benchmark, const Config::ParameterContainer & parameters, const M & mesh_src )
		{
		// skip benchmarks on devices which the user did not select
		if( ! checkDevice< Devices::Host >( parameters ) )
		return;

		auto benchmark_func = [&] () {
		auto meshBuilder = decomposeMesh< DecomposerVersion, SubDecomposerVersion >( mesh_src );
		};

		benchmark.time< Devices::Host >( "CPU",
		benchmark_func );
		}

		template< EntityDecomposerVersion DecomposerVersion,
		typename M,
		std::enable_if_t< M::Config::spaceDimension == 3 &&
		(std::is_same< typename M::Config::CellTopology, Topologies::Polygon >::value \|\|
		std::is_same< typename M::Config::CellTopology, Topologies::Polyhedron >::value ), bool > = true >
		static void benchmark_planar( Benchmark<> & benchmark, const Config::ParameterContainer & parameters, const M & mesh_src )
		{
		if( ! checkDevice< Devices::Host >( parameters ) )
		return;

		auto benchmark_func = [&] () {
		auto meshBuilder = planarCorrection< DecomposerVersion >( mesh_src );
		};

		benchmark.time< Devices::Host >( "CPU",
		benchmark_func );
		}

		template< typename Device,
		typename M >
		static void benchmark_measures( Benchmark<> & benchmark, const Config::ParameterContainer & parameters, const M & mesh_src )
		{
		using Real = typename M::RealType;
		using Index = typename M::GlobalIndexType;
		using DeviceMesh = Meshes::Mesh< typename M::Config, Device >;

		// skip benchmarks on devices which the user did not select
		if( ! checkDevice< Device >( parameters ) )
		return;

		const Index entitiesCount = mesh_src.template getEntitiesCount< M::getMeshDimension() >();

		const DeviceMesh mesh = mesh_src;
		Pointers::DevicePointer< const DeviceMesh > meshPointer( mesh );
		Containers::Array< Real, Device, Index > measures;
		measures.setSize( entitiesCount );

		auto kernel_measures = [] __cuda_callable__
		( Index i,
		const DeviceMesh* mesh,
		Real* array )
		{
		const auto& entity = mesh->template getEntity< M::getMeshDimension() >( i );
		array[ i ] = getEntityMeasure( *mesh, entity );
		};

		auto reset = [&]() {
		measures.setValue( 0.0 );
		};

		auto benchmark_func = [&] () {
		Algorithms::ParallelFor< Device >::exec(
		(Index) 0, entitiesCount,
		kernel_measures,
		&meshPointer.template getData< Device >(),
		measures.getData() );
		};

		benchmark.time< Device >( reset,
		(std::is_same< Device, Devices::Host >::value) ? "CPU" : "GPU",
		benchmark_func );
		}

		template< typename M >
		static void benchmark_memory( Benchmark<> & benchmark, const Config::ParameterContainer & parameters, const M & mesh_src )
		{
		if( ! checkDevice< Devices::Host >( parameters ) )
		return;

		MemoryBenchmarkResult memResult = testMemoryUsage( parameters, mesh_src );
		auto noop = [](){};
		benchmark.time< TNL::Devices::Host >( "CPU", noop, memResult );
		}

		template< typename DeviceFrom,
		typename DeviceTo,
		typename M >
		static void benchmark_copy( Benchmark<> & benchmark, const Config::ParameterContainer & parameters, const M & mesh_src )
		{
		using MeshFrom = Meshes::Mesh< typename M::Config, DeviceFrom >;
		using MeshTo = Meshes::Mesh< typename M::Config, DeviceTo >;
		using Device = typename std::conditional_t< std::is_same< DeviceFrom, Devices::Host >::value &&
		std::is_same< DeviceTo, Devices::Host >::value,
		Devices::Host,
		Devices::Cuda >;

		// skip benchmarks on devices which the user did not select
		if( ! checkDevice< Device >( parameters ) )
		return;

		const MeshFrom meshFrom = mesh_src;

		auto benchmark_func = [&] () {
		MeshTo meshTo = meshFrom;
		};

		benchmark.time< Device >( [] () {},
		(std::is_same< Device, Devices::Host >::value) ? "CPU" : "GPU",
		benchmark_func );
		}
		};

		template< template< typename, int, typename, typename, typename > class ConfigTemplate,
		typename CellTopology,
		int SpaceDimension,
		typename Real,
		typename GlobalIndex,
		typename LocalIndex >
		struct MeshBenchmarksRunner
		{
		static bool
		run( Benchmark<> & benchmark,
		const Config::ParameterContainer & parameters )
		{
		using Config = ConfigTemplate< CellTopology, SpaceDimension, Real, GlobalIndex, LocalIndex >;
		using MeshType = Mesh< Config, Devices::Host >;
		return MeshBenchmarks< MeshType >::run( benchmark, parameters );
		}
		};

src/Benchmarks/Mesh/MeshConfigs.h

0 → 100644

+138 −0

File added.

Preview size limit exceeded, changes collapsed.