Benchmarks: moved plain functions into Utils.h (c31b2268) · Commits · TNL / tnl-dev

src/Benchmarks/Benchmarks.h

+0 −59

Original line number	Diff line number	Diff line
		@@ -20,12 +20,6 @@
		#include <TNL/String.h>
		#include <TNL/Solvers/IterativeSolverMonitor.h>

		#include <TNL/Devices/Host.h>
		#include <TNL/SystemInfo.h>
		#include <TNL/Cuda/DeviceInfo.h>
		#include <TNL/Config/ConfigDescription.h>
		#include <TNL/MPI/Wrappers.h>

		namespace TNL {
		namespace Benchmarks {

		@@ -176,59 +170,6 @@ class Benchmark
		SolverMonitorType monitor;
		};


		inline typename Logging::MetadataMap getHardwareMetadata()
		{
		const int cpu_id = 0;
		const CacheSizes cacheSizes = SystemInfo::getCPUCacheSizes( cpu_id );
		String cacheInfo = convertToString( cacheSizes.L1data ) + ", "
		+ convertToString( cacheSizes.L1instruction ) + ", "
		+ convertToString( cacheSizes.L2 ) + ", "
		+ convertToString( cacheSizes.L3 );
		#ifdef HAVE_CUDA
		const int activeGPU = Cuda::DeviceInfo::getActiveDevice();
		const String deviceArch = convertToString( Cuda::DeviceInfo::getArchitectureMajor( activeGPU ) ) + "." +
		convertToString( Cuda::DeviceInfo::getArchitectureMinor( activeGPU ) );
		#endif

		#ifdef HAVE_MPI
		int nproc = 1;
		// check if MPI was initialized (some benchmarks do not initialize MPI even when
		// they are built with HAVE_MPI and thus MPI::GetSize() cannot be used blindly)
		if( TNL::MPI::Initialized() )
		nproc = TNL::MPI::GetSize();
		#endif

		typename Logging::MetadataMap metadata {
		{ "host name", SystemInfo::getHostname() },
		{ "architecture", SystemInfo::getArchitecture() },
		{ "system", SystemInfo::getSystemName() },
		{ "system release", SystemInfo::getSystemRelease() },
		{ "start time", SystemInfo::getCurrentTime() },
		#ifdef HAVE_MPI
		{ "number of MPI processes", convertToString( nproc ) },
		#endif
		{ "OpenMP enabled", convertToString( Devices::Host::isOMPEnabled() ) },
		{ "OpenMP threads", convertToString( Devices::Host::getMaxThreadsCount() ) },
		{ "CPU model name", SystemInfo::getCPUModelName( cpu_id ) },
		{ "CPU cores", convertToString( SystemInfo::getNumberOfCores( cpu_id ) ) },
		{ "CPU threads per core", convertToString( SystemInfo::getNumberOfThreads( cpu_id ) / SystemInfo::getNumberOfCores( cpu_id ) ) },
		{ "CPU max frequency (MHz)", convertToString( SystemInfo::getCPUMaxFrequency( cpu_id ) / 1e3 ) },
		{ "CPU cache sizes (L1d, L1i, L2, L3) (kiB)", cacheInfo },
		#ifdef HAVE_CUDA
		{ "GPU name", Cuda::DeviceInfo::getDeviceName( activeGPU ) },
		{ "GPU architecture", deviceArch },
		{ "GPU CUDA cores", convertToString( Cuda::DeviceInfo::getCudaCores( activeGPU ) ) },
		{ "GPU clock rate (MHz)", convertToString( (double) Cuda::DeviceInfo::getClockRate( activeGPU ) / 1e3 ) },
		{ "GPU global memory (GB)", convertToString( (double) Cuda::DeviceInfo::getGlobalMemory( activeGPU ) / 1e9 ) },
		{ "GPU memory clock rate (MHz)", convertToString( (double) Cuda::DeviceInfo::getMemoryClockRate( activeGPU ) / 1e3 ) },
		{ "GPU memory ECC enabled", convertToString( Cuda::DeviceInfo::getECCEnabled( activeGPU ) ) },
		#endif
		};

		return metadata;
		}

		} // namespace Benchmarks
		} // namespace TNL

src/Benchmarks/Benchmarks.hpp

+1 −1

Original line number	Diff line number	Diff line
		@@ -14,7 +14,7 @@
		#pragma once

		#include "Benchmarks.h"
		#include "FunctionTimer.h"
		#include "Utils.h"

		#include <iostream>
		#include <exception>

src/Benchmarks/CMakeLists.txt

+1 −1

Original line number	Diff line number	Diff line
		@@ -11,10 +11,10 @@ add_subdirectory( Traversers )
		set( headers
		Benchmarks.h
		Benchmarks.hpp
		FunctionTimer.h
		Logging.h
		CustomLogging.h
		JsonLogging.h
		Utils.h
		)

		install( FILES ${headers} DESTINATION ${TNL_TARGET_INCLUDE_DIRECTORY}/Benchmarks )

src/Benchmarks/FunctionTimer.h→src/Benchmarks/Utils.h

+150 −0

Original line number	Diff line number	Diff line
		/***************************************************************************
		FunctionTimer.h - description
		Utils.h - description
		-------------------
		begin : Dec 25, 2018
		copyright : (C) 2018 by Tomas Oberhuber et al.
		@@ -14,12 +14,19 @@
		#pragma once

		#include <tuple>
		#include <map>

		#include <TNL/Timer.h>
		#include <TNL/Devices/Cuda.h>
		#include <TNL/Containers/Vector.h>
		#include <TNL/Solvers/IterativeSolverMonitor.h>

		#include <TNL/Devices/Host.h>
		#include <TNL/SystemInfo.h>
		#include <TNL/Cuda/DeviceInfo.h>
		#include <TNL/Config/ConfigDescription.h>
		#include <TNL/MPI/Wrappers.h>

		namespace TNL {
		namespace Benchmarks {

		@@ -87,5 +94,57 @@ timeFunction( ComputeFunction compute,
		return std::make_tuple( loops, mean, stddev );
		}

		inline std::map< std::string, std::string > getHardwareMetadata()
		{
		const int cpu_id = 0;
		const CacheSizes cacheSizes = SystemInfo::getCPUCacheSizes( cpu_id );
		const std::string cacheInfo = std::to_string( cacheSizes.L1data ) + ", "
		+ std::to_string( cacheSizes.L1instruction ) + ", "
		+ std::to_string( cacheSizes.L2 ) + ", "
		+ std::to_string( cacheSizes.L3 );
		#ifdef HAVE_CUDA
		const int activeGPU = Cuda::DeviceInfo::getActiveDevice();
		const std::string deviceArch = std::to_string( Cuda::DeviceInfo::getArchitectureMajor( activeGPU ) ) + "."
		+ std::to_string( Cuda::DeviceInfo::getArchitectureMinor( activeGPU ) );
		#endif

		#ifdef HAVE_MPI
		int nproc = 1;
		// check if MPI was initialized (some benchmarks do not initialize MPI even when
		// they are built with HAVE_MPI and thus MPI::GetSize() cannot be used blindly)
		if( TNL::MPI::Initialized() )
		nproc = TNL::MPI::GetSize();
		#endif

		std::map< std::string, std::string > metadata {
		{ "host name", SystemInfo::getHostname() },
		{ "architecture", SystemInfo::getArchitecture() },
		{ "system", SystemInfo::getSystemName() },
		{ "system release", SystemInfo::getSystemRelease() },
		{ "start time", SystemInfo::getCurrentTime() },
		#ifdef HAVE_MPI
		{ "number of MPI processes", std::to_string( nproc ) },
		#endif
		{ "OpenMP enabled", std::to_string( Devices::Host::isOMPEnabled() ) },
		{ "OpenMP threads", std::to_string( Devices::Host::getMaxThreadsCount() ) },
		{ "CPU model name", SystemInfo::getCPUModelName( cpu_id ) },
		{ "CPU cores", std::to_string( SystemInfo::getNumberOfCores( cpu_id ) ) },
		{ "CPU threads per core", std::to_string( SystemInfo::getNumberOfThreads( cpu_id ) / SystemInfo::getNumberOfCores( cpu_id ) ) },
		{ "CPU max frequency (MHz)", std::to_string( SystemInfo::getCPUMaxFrequency( cpu_id ) / 1e3 ) },
		{ "CPU cache sizes (L1d, L1i, L2, L3) (kiB)", cacheInfo },
		#ifdef HAVE_CUDA
		{ "GPU name", Cuda::DeviceInfo::getDeviceName( activeGPU ) },
		{ "GPU architecture", deviceArch },
		{ "GPU CUDA cores", std::to_string( Cuda::DeviceInfo::getCudaCores( activeGPU ) ) },
		{ "GPU clock rate (MHz)", std::to_string( (double) Cuda::DeviceInfo::getClockRate( activeGPU ) / 1e3 ) },
		{ "GPU global memory (GB)", std::to_string( (double) Cuda::DeviceInfo::getGlobalMemory( activeGPU ) / 1e9 ) },
		{ "GPU memory clock rate (MHz)", std::to_string( (double) Cuda::DeviceInfo::getMemoryClockRate( activeGPU ) / 1e3 ) },
		{ "GPU memory ECC enabled", std::to_string( Cuda::DeviceInfo::getECCEnabled( activeGPU ) ) },
		#endif
		};

		return metadata;
		}

		} // namespace Benchmarks
		} // namespace TNL