Added functinality for attaching GDB to MPI processes. (52f8128a) · Commits · TNL / tnl-dev

src/CMakeLists.txt

+1 −1

Original line number	Diff line number	Diff line
		#ADD_SUBDIRECTORY( Python )
		ADD_SUBDIRECTORY( Python )
		ADD_SUBDIRECTORY( TNL )
		ADD_SUBDIRECTORY( Tools )
		ADD_SUBDIRECTORY( UnitTests )

src/TNL/Communicators/MpiCommunicator.h

+67 −26

Original line number	Diff line number	Diff line
		@@ -16,6 +16,7 @@

		#ifdef HAVE_MPI
		#include <mpi.h>
		#include <mpi-ext.h>

		#ifdef HAVE_CUDA
		#include <TNL/Devices/Cuda.h>
		@@ -72,6 +73,8 @@ class MpiCommunicator
		{
		#ifdef HAVE_MPI
		config.addEntry< bool >( "redirect-mpi-output", "Only process with rank 0 prints to console. Other processes are redirected to files.", true );
		config.addEntry< bool >( "mpi-gdb-debug", "Wait for GDB to attach the master MPI process.", false );
		config.addEntry< int >( "mpi-process-to-attach", "Number of the MPI process to be attached by GDB.", 0 );
		#endif
		}

		@@ -83,8 +86,47 @@ class MpiCommunicator
		{
		redirect = parameters.getParameter< bool >( "redirect-mpi-output" );
		setupRedirection();
		}
		#ifdef HAVE_CUDA
		#if defined(MPIX_CUDA_AWARE_SUPPORT) && MPIX_CUDA_AWARE_SUPPORT
		std::cout << "CUDA-aware MPI detected on this system ... " << std::endl;
		#elif defined(MPIX_CUDA_AWARE_SUPPORT) && !MPIX_CUDA_AWARE_SUPPORT
		std::cerr << "MPI is not CUDA-aware. Please install correct version of MPI." << std::endl;
		return false;
		#else
		std::cerr << "WARNING: TNL cannot detect if you have CUDA-aware MPI. Some problems may occur." << std::endl;
		#endif
		#endif // HAVE_CUDA
		bool gdbDebug = parameters.getParameter< bool >( "mpi-gdb-debug" );
		int processToAttach = parameters.getParameter< int >( "mpi-process-to-attach" );

		if( gdbDebug )
		{
		int rank = GetRank( MPI_COMM_WORLD );
		int pid = getpid();

		volatile int tnlMPIDebugAttached = 0;
		MPI_Send( &pid, 1, MPI_INT, 0, 0, MPI_COMM_WORLD );
		if( rank == 0 )
		std::cerr << "Attach GDB to MPI process(es) by entering:" << std::endl;
		for( int i = 0; i < GetSize( MPI_COMM_WORLD ); i++ )
		{
		MPI_Status status;
		int recvPid;
		MPI_Recv( &recvPid, 1, MPI_INT, i, 0, MPI_COMM_WORLD, &status );

		if( i == processToAttach \|\| processToAttach == -1 )
		{
		std::cerr << " For MPI process " << i << ": gdb -q -ex \"attach " << recvPid << "\""
		<< " -ex \"set variable tnlMPIDebugAttached=1\""
		<< " -ex \"finish\"" << std::endl;
		}
		}
		if( rank == processToAttach \|\| processToAttach == -1 )
		while( ! tnlMPIDebugAttached );
		MPI_Barrier( MPI_COMM_WORLD );
		}
		}
		#endif // HAVE_MPI
		return true;
		}

		@@ -188,9 +230,8 @@ class MpiCommunicator
		int sum=0;
		for(int i=0;i<dim;i++)
		sum+=distr[i];
		if(sum==0) //uživatel neovlivňuje distribuci
		if(sum==0)
		{
		std::cout << "vynucuji distribuci" <<std::endl;
		for(int i=0;i<dim-1;i++)
		{
		distr[i]=1;

src/TNL/Config/ParameterContainer.cpp

+11 −11

Original line number	Diff line number	Diff line
		@@ -203,7 +203,7 @@ parseCommandLine( int argc, char* argv[],
		Containers::List< String > parsedEntryType;
		if( ! parseObjectType( entryType, parsedEntryType ) )
		{
		std::cerr << "Internal error: Uknown config entry type " << entryType << "." << std::endl;
		std::cerr << "Internal error: Unknown config entry type " << entryType << "." << std::endl;
		return false;
		}
		if( parsedEntryType[ 0 ] == "List" )
		@@ -308,7 +308,7 @@ parseCommandLine( int argc, char* argv[],
		}
		if( parsedEntryType[ 0 ] == "int" )
		{
		/*if( ! isdigit( value ) )
		/*if( ! std::isdigit( value ) ) //TODO: Check for real number
		{
		std::cerr << "Integer constant is required for the parameter " << option << "." << std::endl;
		parse_error = true;
		@@ -320,7 +320,7 @@ parseCommandLine( int argc, char* argv[],
		}
		if( parsedEntryType[ 0 ] == "double" )
		{
		/*if( ! isdigit( value ) )
		/*if( ! std::isdigit( value ) ) //TODO: Check for real number
		{
		std::cerr << "Real constant is required for the parameter " << option << "." << std::endl;
		parse_error = true;

src/TNL/Solvers/IterativeSolverMonitor_impl.h

+0 −2

Original line number	Diff line number	Diff line
		@@ -171,8 +171,6 @@ void IterativeSolverMonitor< Real, Index > :: refresh()
		if( nodesPerIteration ) // otherwise MLUPS: 0 is printed
		{
		const RealType mlups = nodesPerIteration * (iterations - iterations_before_refresh) / (getElapsedTime() - elapsed_time_before_refresh) * 1e-6;
		//std::cerr << std::endl << " iterations - iterations_before_refresh = " << iterations - iterations_before_refresh
		// << " getElapsedTime() - elapsed_time_before_refresh = " << getElapsedTime() - elapsed_time_before_refresh << std::endl;
		print_item( " MLUPS:", 0 );
		if( mlups > 0 )
		{

src/TNL/Solvers/IterativeSolver_impl.h

+6 −1

Original line number	Diff line number	Diff line
		@@ -107,7 +107,12 @@ bool IterativeSolver< Real, Index> :: checkNextIteration()
		{
		this->refreshSolverMonitor();

		if( std::isnan( this->getResidue() ) \|\|
		if(
		#ifndef HAVE_CUDA
		std::isnan( this->getResidue() ) \|\|
		// TODO: Fix this !!!!
		// this does not work (at least) with nvcc 8.0 and g++ 5.4
		#endif
		this->getIterations() > this->getMaxIterations() \|\|
		( this->getResidue() > this->getDivergenceResidue() && this->getIterations() >= this->getMinIterations() ) \|\|
		( this->getResidue() < this->getConvergenceResidue() && this->getIterations() >= this->getMinIterations() ) )

src/TNL/Config/ParameterContainer.h

+1 −1

File changed.

Contains only whitespace changes.