Added OpenMP paralelization to Merson solver. (9c7dab00) · Commits · TNL / tnl-dev

install

+3 −2

Original line number	Diff line number	Diff line
		@@ -108,8 +108,9 @@ then
		fi



		if test x`python src/Tools/python-path-test.py 2> /dev/null` != xOK;
		PYTHON_TEST="`python src/Tools/python-path-test.py 2> /dev/null`"
		echo "xxxxx ${PYTHON_TEST} xxxxx\n"
		if test PYTHON_TEST != "xOK";
		then
		source ${BUILD_PREFIX}/python-version
		echo ""

src/TNL/Devices/Host.cpp

+6 −6

Original line number	Diff line number	Diff line
		@@ -74,11 +74,11 @@ int Host::getThreadIdx()
		void Host::configSetup( Config::ConfigDescription& config, const String& prefix )
		{
		#ifdef HAVE_OPENMP
		config.addEntry< bool >( prefix + "omp-enabled", "Enable support of OpenMP.", true );
		config.addEntry< int >( prefix + "omp-max-threads", "Set maximum number of OpenMP threads.", omp_get_max_threads() );
		config.addEntry< bool >( prefix + "openmp-enabled", "Enable support of OpenMP.", true );
		config.addEntry< int >( prefix + "openmp-max-threads", "Set maximum number of OpenMP threads.", omp_get_max_threads() );
		#else
		config.addEntry< bool >( prefix + "omp-enabled", "Enable support of OpenMP (not supported on this system).", false );
		config.addEntry< int >( prefix + "omp-max-threads", "Set maximum number of OpenMP threads (not supported on this system).", 0 );
		config.addEntry< bool >( prefix + "openmp-enabled", "Enable support of OpenMP (not supported on this system).", false );
		config.addEntry< int >( prefix + "openmp-max-threads", "Set maximum number of OpenMP threads (not supported on this system).", 0 );
		#endif

		}
		@@ -86,11 +86,11 @@ void Host::configSetup( Config::ConfigDescription& config, const String& prefix
		bool Host::setup( const Config::ParameterContainer& parameters,
		const String& prefix )
		{
		if( parameters.getParameter< bool >( prefix + "omp-enabled" ) )
		if( parameters.getParameter< bool >( prefix + "openmp-enabled" ) )
		enableOMP();
		else
		disableOMP();
		setMaxThreadsCount( parameters.getParameter< int >( prefix + "omp-max-threads" ) );
		setMaxThreadsCount( parameters.getParameter< int >( prefix + "openmp-max-threads" ) );
		return true;
		}

src/TNL/Solvers/ODE/Merson.h

+2 −0

Original line number	Diff line number	Diff line
		@@ -70,6 +70,8 @@ class Merson : public ExplicitSolver< Problem >
		* This controls the accuracy of the solver
		*/
		RealType adaptivity;

		Containers::Vector< RealType, DeviceType, IndexType > openMPErrorEstimateBuffer;
		};

		} // namespace ODE

src/TNL/Solvers/ODE/Merson_impl.h

+21 −8

Original line number	Diff line number	Diff line
		@@ -14,6 +14,8 @@
		#include <TNL/Devices/Cuda.h>
		#include <TNL/Config/ParameterContainer.h>

		#include "Merson.h"

		namespace TNL {
		namespace Solvers {
		namespace ODE {
		@@ -85,6 +87,10 @@ template< typename Problem >
		Merson< Problem > :: Merson()
		: adaptivity( 0.00001 )
		{
		if( std::is_same< DeviceType, Devices::Host >::value )
		{
		this->openMPErrorEstimateBuffer.setSize( std::max( 1, Devices::Host::getMaxThreadsCount() ) );
		}
		};

		template< typename Problem >
		@@ -378,7 +384,11 @@ typename Problem :: RealType Merson< Problem > :: computeError( const RealType t
		RealType eps( 0.0 ), maxEps( 0.0 );
		if( std::is_same< DeviceType, Devices::Host >::value )
		{
		// TODO: implement OpenMP support
		this->openMPErrorEstimateBuffer.setValue( 0.0 );
		#pragma omp parallel if( Devices::Host::isOMPEnabled() )
		{
		RealType localEps( 0.0 );
		#pragma omp for
		for( IndexType i = 0; i < size; i ++ )
		{
		RealType err = ( RealType ) ( tau / 3.0 *
		@@ -386,8 +396,11 @@ typename Problem :: RealType Merson< Problem > :: computeError( const RealType t
		-0.9 * _k3[ i ] +
		0.8 * _k4[ i ] +
		-0.1 * _k5[ i ] ) );
		eps = max( eps, err );
		localEps = max( localEps, err );
		}
		this->openMPErrorEstimateBuffer[ Devices::Host::getThreadIdx() ] = localEps;
		}
		eps = this->openMPErrorEstimateBuffer.max();
		}
		if( std::is_same< DeviceType, Devices::Cuda >::value )
		{