olb::HeterogeneousCopyTaskDataForGpuTarget< T, DESCRIPTOR, SOURCE > Class Template Reference

Private implementation of heterogeneous copy task between CPU_* source and GPU_CUDA target. More...

#include <communicator.hh>

Inheritance diagram for olb::HeterogeneousCopyTaskDataForGpuTarget< T, DESCRIPTOR, SOURCE >:

Collaboration diagram for olb::HeterogeneousCopyTaskDataForGpuTarget< T, DESCRIPTOR, SOURCE >:

Public Member Functions
	HeterogeneousCopyTaskDataForGpuTarget (const std::vector< std::type_index > &fields, const std::vector< CellID > &targetCells, ConcreteBlockLattice< T, DESCRIPTOR, Platform::GPU_CUDA > &target, const std::vector< CellID > &sourceCells, ConcreteBlockLattice< T, DESCRIPTOR, SOURCE > &source)
void	copy () override
void	wait () override
Public Member Functions inherited from olb::ConcreteHeterogeneousCopyTask
virtual	~ConcreteHeterogeneousCopyTask ()

Detailed Description

template<typename T, typename DESCRIPTOR, Platform SOURCE>
class olb::HeterogeneousCopyTaskDataForGpuTarget< T, DESCRIPTOR, SOURCE >

Private implementation of heterogeneous copy task between CPU_* source and GPU_CUDA target.

Definition at line 590 of file communicator.hh.

Constructor & Destructor Documentation

HeterogeneousCopyTaskDataForGpuTarget()

template<typename T , typename DESCRIPTOR , Platform SOURCE>

olb::HeterogeneousCopyTaskDataForGpuTarget< T, DESCRIPTOR, SOURCE >::HeterogeneousCopyTaskDataForGpuTarget ( const std::vector< std::type_index > & fields, const std::vector< CellID > & targetCells, ConcreteBlockLattice< T, DESCRIPTOR, Platform::GPU_CUDA > & target, const std::vector< CellID > & sourceCells, ConcreteBlockLattice< T, DESCRIPTOR, SOURCE > & source )

inline

Definition at line 608 of file communicator.hh.

                                                                                                        :
    _targetFields(target.getDataRegistry().deviceFieldArrays(fields)),
    _onlyPopulationField(fields.size() == 1 && fields[0] == typeid(descriptors::POPULATION)),
    _targetCells(targetCells),
    _sourceCells(sourceCells),
    _target(target),
    _source(source, fields),
    _stream(std::make_unique<gpu::cuda::device::Stream>(cudaStreamNonBlocking)),
    _buffer(_source.size(_sourceCells))
  { }

Member Function Documentation

copy()

template<typename T , typename DESCRIPTOR , Platform SOURCE>

void olb::HeterogeneousCopyTaskDataForGpuTarget< T, DESCRIPTOR, SOURCE >::copy ( )

inlineoverridevirtual

Implements olb::ConcreteHeterogeneousCopyTask.

Definition at line 622 of file communicator.hh.

                       {
    _source.serialize(_sourceCells, _buffer.data());
    _buffer.setProcessingContext(ProcessingContext::Simulation, *_stream);
 
    if (_onlyPopulationField) {
      gpu::cuda::DeviceContext<T,DESCRIPTOR> lattice(_target);
      gpu::cuda::async_scatter_field<descriptors::POPULATION>(_stream->get(), lattice, _targetCells, _buffer.deviceData());
    } else {
      gpu::cuda::async_scatter_any_fields(_stream->get(), _targetFields, _targetCells, _buffer.deviceData());
    }
  }

References olb::gpu::cuda::async_scatter_any_fields(), olb::gpu::cuda::Column< T >::data(), olb::gpu::cuda::Column< T >::deviceData(), olb::MultiConcreteCommunicatable< COMMUNICATEE >::serialize(), olb::gpu::cuda::Column< T >::setProcessingContext(), and olb::Simulation.

Here is the call graph for this function:

wait()

template<typename T , typename DESCRIPTOR , Platform SOURCE>

void olb::HeterogeneousCopyTaskDataForGpuTarget< T, DESCRIPTOR, SOURCE >::wait ( )

inlineoverridevirtual

Implements olb::ConcreteHeterogeneousCopyTask.

Definition at line 634 of file communicator.hh.

                       {
    _stream->synchronize();
  }

---

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

• src/core/platform/gpu/cuda/communicator.hh