olb::StatisticsPostProcessor::type< ConcreteBlockLattice< T, DESCRIPTOR, Platform::GPU_CUDA > > Struct Template Reference

#include <statistics.h>

Collaboration diagram for olb::StatisticsPostProcessor::type< ConcreteBlockLattice< T, DESCRIPTOR, Platform::GPU_CUDA > >:

Public Member Functions
void	setup (ConcreteBlockLattice< T, DESCRIPTOR, Platform::GPU_CUDA > &blockLattice)
void	apply (ConcreteBlockLattice< T, DESCRIPTOR, Platform::GPU_CUDA > &blockLattice)

Detailed Description

template<typename T, typename DESCRIPTOR>
struct olb::StatisticsPostProcessor::type< ConcreteBlockLattice< T, DESCRIPTOR, Platform::GPU_CUDA > >

Definition at line 32 of file statistics.h.

Member Function Documentation

apply()

template<typename T , typename DESCRIPTOR >

void olb::StatisticsPostProcessor::type< ConcreteBlockLattice< T, DESCRIPTOR, Platform::GPU_CUDA > >::apply

(

ConcreteBlockLattice< T, DESCRIPTOR, Platform::GPU_CUDA > &

blockLattice

)

Todo: Reimplement these four separate reductions as single kernel

Definition at line 80 of file statistics.hh.

{
  if (!blockLattice.statisticsEnabled()) {
    return;
  }
 
  auto& statisticGenerated = blockLattice.template getField<descriptors::STATISTIC_GENERATED>();
  auto& statistic = blockLattice.template getField<descriptors::STATISTIC>();
  std::size_t nCells = thrust::reduce(thrust::device,
                                      statisticGenerated[0].deviceData(),
                                      statisticGenerated[0].deviceData() + blockLattice.getNcells(),
                                      T{0},
                                      thrust::plus<T>());
  T rhoSum = thrust::reduce(thrust::device,
                            statistic[0].deviceData(),
                            statistic[0].deviceData() + blockLattice.getNcells(),
                            T{0},
                            thrust::plus<T>());
  auto [maxU, uSqrSum]  = thrust::reduce(thrust::device,
                                         statistic[1].deviceData(),
                                         statistic[1].deviceData() + blockLattice.getNcells(),
                                         gpu::cuda::pair{std::numeric_limits<T>::min(), T{0}},
                                         gpu::cuda::maximum_and_plus<T>());
 
  typename LatticeStatistics<T>::Aggregatable statistics{
    .nCells = nCells,
    .avRho = rhoSum,
    .avEnergy = uSqrSum,
    .maxU = maxU
  };
  blockLattice.getStatistics().incrementStats(statistics);
 
  blockLattice.getStatistics().reset();
}

References olb::BlockStructureD< D >::getNcells(), and olb::BlockLattice< T, DESCRIPTOR >::statisticsEnabled().

Here is the call graph for this function:

setup()

template<typename T , typename DESCRIPTOR >

void olb::StatisticsPostProcessor::type< ConcreteBlockLattice< T, DESCRIPTOR, Platform::GPU_CUDA > >::setup ( ConcreteBlockLattice< T, DESCRIPTOR, Platform::GPU_CUDA > & blockLattice )

inline

Definition at line 34 of file statistics.h.

{
  // Allocate on-device statistic buffer
  blockLattice.template getField<descriptors::STATISTIC_GENERATED>();
  blockLattice.template getField<descriptors::STATISTIC>();
}

---

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

• src/core/platform/gpu/cuda/statistics.h
• src/core/platform/gpu/cuda/statistics.hh