Opm::GpuBridge< BridgeMatrix, BridgeVector, block_size > Class Template Reference

GpuBridge acts as interface between opm-simulators with the GpuSolvers. More...

#include <GpuBridge.hpp>

Public Member Functions
	GpuBridge (std::string accelerator_mode, int linear_solver_verbosity, int maxit, Scalar tolerance, unsigned int platformID, unsigned int deviceID, bool opencl_ilu_parallel, std::string linsolver)
	Construct a GpuBridge.
void	solve_system (BridgeMatrix *bridgeMat, BridgeMatrix *jacMat, int numJacobiBlocks, BridgeVector &b, WellContributions< Scalar > &wellContribs, InverseOperatorResult &result)
	Solve linear system, A*x = b.
void	get_result (BridgeVector &x)
	Get the resulting x vector.
bool	getUseGpu ()
	Return whether the GpuBridge will use the GPU or not return whether the GpuBridge will use the GPU or not.
void	initWellContributions (WellContributions< Scalar > &wellContribs, unsigned N)
	Initialize the WellContributions object with opencl context and queue those must be set before calling BlackOilWellModel::getWellContributions() in ISTL.
std::string	getAccleratorName ()
	Return the selected accelerator mode, this is input via the command-line.

Static Public Member Functions
static void	copySparsityPatternFromISTL (const BridgeMatrix &mat, std::vector< int > &h_rows, std::vector< int > &h_cols)
	Store sparsity pattern into vectors.

Detailed Description

template<class BridgeMatrix, class BridgeVector, int block_size>
class Opm::GpuBridge< BridgeMatrix, BridgeVector, block_size >

GpuBridge acts as interface between opm-simulators with the GpuSolvers.

Constructor & Destructor Documentation

GpuBridge()

template<class BridgeMatrix, class BridgeVector, int block_size>

Opm::GpuBridge< BridgeMatrix, BridgeVector, block_size >::GpuBridge	(	std::string	accelerator_mode,
		int	linear_solver_verbosity,
		int	maxit,
		Scalar	tolerance,
		unsigned int	platformID,
		unsigned int	deviceID,
		bool	opencl_ilu_parallel,
		std::string	linsolver )

Construct a GpuBridge.

Parameters

[in]	accelerator_mode	to select if an accelerated solver is used, is passed via command-line: '–accelerator-mode=[none|cusparse|opencl|amgcl|rocalution|rocsparse]'
[in]	linear_solver_verbosity	verbosity of GpuSolver
[in]	maxit	maximum number of iterations for GpuSolver
[in]	tolerance	required relative tolerance for GpuSolver
[in]	platformID	the OpenCL platform ID to be used
[in]	deviceID	the device ID to be used by the cusparse- and openclSolvers, too high values could cause runtime errors
[in]	opencl_ilu_parallel	whether to parallelize the ILU decomposition and application in OpenCL with level_scheduling
[in]	linsolver	indicating the preconditioner, equal to the –linear-solver cmdline argument

Member Function Documentation

copySparsityPatternFromISTL()

template<class BridgeMatrix, class BridgeVector, int block_size>

void Opm::GpuBridge< BridgeMatrix, BridgeVector, block_size >::copySparsityPatternFromISTL ( const BridgeMatrix & mat, std::vector< int > & h_rows, std::vector< int > & h_cols )

static

Store sparsity pattern into vectors.

Parameters

[in]	mat	input matrix, probably BCRSMatrix
[out]	h_rows	rowpointers
[out]	h_cols	columnindices

get_result()

template<class BridgeMatrix, class BridgeVector, int block_size>

void Opm::GpuBridge< BridgeMatrix, BridgeVector, block_size >::get_result

(

BridgeVector &

x

)

Get the resulting x vector.

Parameters

[in,out]

x

vector x, should be of type Dune::BlockVector

initWellContributions()

template<class BridgeMatrix, class BridgeVector, int block_size>

void Opm::GpuBridge< BridgeMatrix, BridgeVector, block_size >::initWellContributions	(	WellContributions< Scalar > &	wellContribs,
		unsigned	N )

Initialize the WellContributions object with opencl context and queue those must be set before calling BlackOilWellModel::getWellContributions() in ISTL.

Parameters

[in]	wellContribs	container to hold all WellContributions
[in]	N	number of rows in scalar vector that wellContribs will be applied on

solve_system()

template<class BridgeMatrix, class BridgeVector, int block_size>

void Opm::GpuBridge< BridgeMatrix, BridgeVector, block_size >::solve_system	(	BridgeMatrix *	bridgeMat,
		BridgeMatrix *	jacMat,
		int	numJacobiBlocks,
		BridgeVector &	b,
		WellContributions< Scalar > &	wellContribs,
		InverseOperatorResult &	result )

Solve linear system, A*x = b.

Warning

Values of A might get overwritten!

Parameters

[in]	bridgeMat	matrix A, should be of type Dune::BCRSMatrix
[in]	jacMat	matrix A, but modified for the preconditioner, should be of type Dune::BCRSMatrix
[in]	numJacobiBlocks	number of jacobi blocks in jacMat
[in]	b	vector b, should be of type Dune::BlockVector
[in]	wellContribs	contains all WellContributions, to apply them separately, instead of adding them to matrix A
[in,out]	result	summary of solver result

---

The documentation for this class was generated from the following files:

• opm/simulators/linalg/gpubridge/GpuBridge.hpp
• opm/simulators/linalg/gpubridge/GpuBridge.cpp