Opm::gpuistl::PreconditionerConvertFieldTypeAdapter< CudaPreconditionerType, M, X, Y, l > Class Template Reference

Converts the field type (eg. More...

#include <PreconditionerConvertFieldTypeAdapter.hpp>

Inheritance diagram for Opm::gpuistl::PreconditionerConvertFieldTypeAdapter< CudaPreconditionerType, M, X, Y, l >:

Public Types
using	matrix_type = typename std::remove_const<M>::type
	The matrix type the preconditioner is for.
using	domain_type = X
	The domain type of the preconditioner.
using	range_type = Y
	The range type of the preconditioner.
using	field_type = typename X::field_type
	The field type of the preconditioner.
using	domain_type_to = typename CudaPreconditionerType::domain_type
using	range_type_to = typename CudaPreconditionerType::range_type
	The range type of the preconditioner.
using	field_type_to = typename domain_type_to::field_type
	The field type of the preconditioner.
using	block_type = typename domain_type::block_type
using	XTo = Dune::BlockVector<Dune::FieldVector<field_type_to, block_type::dimension>>
using	YTo = Dune::BlockVector<Dune::FieldVector<field_type_to, block_type::dimension>>
using	matrix_type_to

Public Member Functions
	PreconditionerConvertFieldTypeAdapter (const M &matrix)
	Constructor.
virtual void	pre (X &x, Y &b) override
	Not used at the moment.
virtual void	apply (X &v, const Y &d) override
	Apply the preconditoner.
virtual void	post (X &x) override
	Not used at the moment.
virtual Dune::SolverCategory::Category	category () const override
	Category of the preconditioner (see SolverCategory::Category).
virtual void	update () override
const matrix_type_to &	getConvertedMatrix () const
void	setUnderlyingPreconditioner (const std::shared_ptr< CudaPreconditionerType > &conditioner)
virtual bool	hasPerfectUpdate () const override

Detailed Description

template<class CudaPreconditionerType, class M, class X, class Y, int l = 1>
class Opm::gpuistl::PreconditionerConvertFieldTypeAdapter< CudaPreconditionerType, M, X, Y, l >

Converts the field type (eg.

double to float) to benchmark single precision preconditioners

Note

This is not a fast conversion, it is simply meant to benchmark the potential of some preconditioners on consumer grade GPUs where the double precision performance is often artificially limited.

In theory this can do any field_type conversion that is meaningful, but it is only tested on double to float conversion.

Remember to set the underlying preconditioner with setUnderlyingPreconditioner (should use the matrix from getConvertedMatrix())

One could simply change the constructor design by accepting a creator function for the underlying preconditioner. For the current use cases this is however not needed.

To use this, use something like the following code:

Member Typedef Documentation

matrix_type_to

template<class CudaPreconditionerType, class M, class X, class Y, int l = 1>

using Opm::gpuistl::PreconditionerConvertFieldTypeAdapter< CudaPreconditionerType, M, X, Y, l >::matrix_type_to

Initial value:

Constructor & Destructor Documentation

PreconditionerConvertFieldTypeAdapter()

template<class CudaPreconditionerType, class M, class X, class Y, int l = 1>

Opm::gpuistl::PreconditionerConvertFieldTypeAdapter< CudaPreconditionerType, M, X, Y, l >::PreconditionerConvertFieldTypeAdapter ( const M & matrix )

inlineexplicit

Constructor.

Parameters

matrix

The matrix to operate on.

Note

After the PreconditionerConvertFieldTypeAdapter you can get the converted matrix by calling getConvertedMatrix(), which in turn can be used to create the underlying preconditioner. Once the underlying preconditioner has been called, this must be supplied to setUnderlyingPreconditioner.

Member Function Documentation

hasPerfectUpdate()

template<class CudaPreconditionerType, class M, class X, class Y, int l = 1>

virtual bool Opm::gpuistl::PreconditionerConvertFieldTypeAdapter< CudaPreconditionerType, M, X, Y, l >::hasPerfectUpdate ( ) const

inlineoverridevirtual

Implements Dune::PreconditionerWithUpdate< X, Y >.

update()

template<class CudaPreconditionerType, class M, class X, class Y, int l = 1>

virtual void Opm::gpuistl::PreconditionerConvertFieldTypeAdapter< CudaPreconditionerType, M, X, Y, l >::update ( )

inlineoverridevirtual

Implements Dune::PreconditionerWithUpdate< X, Y >.

---

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

• opm/simulators/linalg/gpuistl/PreconditionerConvertFieldTypeAdapter.hpp