Function ParallelCopyFaceToFace

Tests/MultiBlock/IndexType/main.cpp:89–131

}

bool ParallelCopyFaceToFace(amrex::iMultiFab& dest, const amrex::Box& domain_dest, amrex::Orientation::Side dest_side,
                            const amrex::iMultiFab& src, const amrex::Box& domain_src, amrex::Orientation::Side src_side)
{
    int sdir = GetFaceDir(domain_src.ixType());
    int ddir = GetFaceDir(domain_dest.ixType());
    const amrex::Box src_box = GetFaceBoundary(domain_src, src_side);
    const amrex::Box dest_box = GetFaceBoundary(domain_dest, dest_side);

    // Default construction is identity
    amrex::NonLocalBC::MultiBlockIndexMapping dtos{};
    // Change permutation to get a correct mapping between index types
    std::swap(dtos.permutation[sdir], dtos.permutation[ddir]);
    // Map smallest destination box index as an index in the source space
    const amrex::IntVect dest_smallEnd_in_src = amrex::NonLocalBC::Apply(dtos, dest_box.smallEnd());
    // Compute the offset to get the proper shift in the source space
    dtos.offset =  dest_smallEnd_in_src - src_box.smallEnd();
    // Sanity-Check that the correct box is being mapped
    // Note, this checks index types, too!
    AMREX_ASSERT(amrex::NonLocalBC::Image(dtos, dest_box) == src_box);

    amrex::NonLocalBC::ParallelCopy(dest, dest_box, src, 0, 0, 1, amrex::IntVect{0}, dtos);
    int fails = 0;
    const int nx = domain_src.length(0);
    const int ny = domain_src.length(1);
    amrex::ReduceOps<amrex::ReduceOpSum> reduce_op;
    amrex::ReduceData<int> reduce_data(reduce_op);
    using ReduceTuple = amrex::GpuTuple<int>;
    for (amrex::MFIter mfi(dest); mfi.isValid(); ++mfi) {
        const amrex::Box section = dest_box & mfi.tilebox();
        if (section.isEmpty()) { continue; }
        auto darray = dest.const_array(mfi);
        reduce_op.eval(section, reduce_data,
                       [=] AMREX_GPU_DEVICE (int i, int j, int k) -> ReduceTuple
        {
            amrex::Dim3 si = dtos(amrex::Dim3{.x = i, .y = j, .z = k});
            int value = si.x + si.y*nx + si.z*nx*ny;
            auto r = int(darray(i,j,k) != value);
            return { r };
        });
    }
    fails += amrex::get<0>(reduce_data.value());
    return fails == 0;
}

int MyMain()
{