average_dispatch.h

Go to the documentation of this file.

#pragma once
 
#include "../blas2.h"
#include "average_cpu.h"
#if THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_CUDA
#include "average_gpu.cuh"
#elif THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_OMP
#include "average_omp.h"
#endif
 
namespace dg{
 
template<class ContainerType>
void transpose( unsigned nx, unsigned ny, const ContainerType& in, ContainerType& out)
{
    assert(&in != &out);
    using value_type = get_value_type<ContainerType>;
    dg::blas2::parallel_for( [nx,ny]DG_DEVICE( unsigned k, const value_type* ii, value_type* oo)
        {
            unsigned i = k/nx, j =  k%nx;
            oo[j*ny+i] = ii[i*nx+j];
        }, nx*ny, in, out);
}
 
template<class ContainerType>
void extend_line( unsigned nx, unsigned ny, const ContainerType& in, ContainerType& out)
{
    assert(&in != &out);
    using value_type = get_value_type<ContainerType>;
    dg::blas2::parallel_for( [nx]DG_DEVICE( unsigned k, const value_type* ii, value_type* oo)
        {
            unsigned i = k/nx, j =  k%nx;
            oo[i*nx+j] = ii[j];
        }, nx*ny, in, out);
}
template<class ContainerType>
void extend_column( unsigned nx, unsigned ny, const ContainerType& in, ContainerType& out)
{
    assert(&in != &out);
    using value_type = get_value_type<ContainerType>;
    dg::blas2::parallel_for( [nx]DG_DEVICE( unsigned k, const value_type* ii, value_type* oo)
        {
            unsigned i = k/nx, j =  k%nx;
            oo[i*nx+j] = ii[i];
        }, nx*ny, in, out);
}
 
template<class ContainerType>
void average( unsigned nx, unsigned ny, const ContainerType& in0, const ContainerType& in1, ContainerType& out)
{
    static_assert( std::is_same<get_value_type<ContainerType>, double>::value, "We only support double precision dot products at the moment!");
    const double* in0_ptr = thrust::raw_pointer_cast( in0.data());
    const double* in1_ptr = thrust::raw_pointer_cast( in1.data());
          double* out_ptr = thrust::raw_pointer_cast( out.data());
    average( get_execution_policy<ContainerType>(), nx, ny, in0_ptr, in1_ptr, out_ptr);
}
 
#ifdef MPI_VERSION
template<class ContainerType>
void mpi_average( unsigned nx, unsigned ny, const ContainerType& in0, const ContainerType& in1, ContainerType& out, MPI_Comm comm, MPI_Comm comm_mod, MPI_Comm comm_mod_reduce)
{
    static_assert( std::is_same<get_value_type<ContainerType>, double>::value, "We only support double precision dot products at the moment!");
    const double* in0_ptr = thrust::raw_pointer_cast( in0.data());
    const double* in1_ptr = thrust::raw_pointer_cast( in1.data());
          double* out_ptr = thrust::raw_pointer_cast( out.data());
    average_mpi( get_execution_policy<ContainerType>(), nx, ny, in0_ptr, in1_ptr, out_ptr, comm, comm_mod, comm_mod_reduce);
}
#endif //MPI_VERSION
 
}//namespace dg