io/vtkoutputmodule.hh

Go to the documentation of this file.

// -*- mode: C++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*-
// vi: set et ts=4 sw=4 sts=4:
//
// SPDX-FileCopyrightText: Copyright © DuMux Project contributors, see AUTHORS.md in root folder
// SPDX-License-Identifier: GPL-3.0-or-later
//
#ifndef DUMUX_IO_VTK_OUTPUT_MODULE_HH
#define DUMUX_IO_VTK_OUTPUT_MODULE_HH
 
#include <functional>
#include <memory>
#include <string>
 
#include <dune/common/timer.hh>
#include <dune/common/fvector.hh>
#include <dune/common/typetraits.hh>
 
#include <dune/geometry/type.hh>
#include <dune/geometry/multilineargeometry.hh>
 
#include <dune/grid/common/mcmgmapper.hh>
#include <dune/grid/common/partitionset.hh>
#include <dune/grid/io/file/vtk/vtkwriter.hh>
#include <dune/grid/io/file/vtk/vtksequencewriter.hh>
#include <dune/grid/common/partitionset.hh>
 
#include <dumux/common/parameters.hh>
#include <dumux/io/format.hh>
#include <dumux/discretization/method.hh>
 
#include <dumux/io/vtk/function.hh>
#include <dumux/io/vtk/fieldtype.hh>
#include "velocityoutput.hh"
 
namespace Dumux {
 
template<class GridGeometry>
class VtkOutputModuleBase
{
    using GridView = typename GridGeometry::GridView;
    static constexpr int dim = GridView::dimension;
 
public:
    using Field = Vtk::template Field<GridView>;
 
    VtkOutputModuleBase(const GridGeometry& gridGeometry,
                        const std::string& name,
                        const std::string& paramGroup = "",
                        Dune::VTK::DataMode dm = Dune::VTK::conforming,
                        bool verbose = true)
    : gridGeometry_(gridGeometry)
    , name_(name)
    , paramGroup_(paramGroup)
    , dm_(dm)
    , verbose_(gridGeometry.gridView().comm().rank() == 0 && verbose)
    {
        const auto precisionString = getParamFromGroup<std::string>(paramGroup, "Vtk.Precision", "Float32");
        precision_ = Dumux::Vtk::stringToPrecision(precisionString);
        const auto coordPrecision = Dumux::Vtk::stringToPrecision(getParamFromGroup<std::string>(paramGroup, "Vtk.CoordPrecision", precisionString));
        writer_ = std::make_shared<Dune::VTKWriter<GridView>>(gridGeometry.gridView(), dm, coordPrecision);
        sequenceWriter_ = std::make_unique<Dune::VTKSequenceWriter<GridView>>(writer_, name);
        addProcessRank_ = getParamFromGroup<bool>(this->paramGroup(), "Vtk.AddProcessRank", true);
    }
 
    virtual ~VtkOutputModuleBase() = default;
 
    const std::string& paramGroup() const
    { return paramGroup_; }
 
    template<typename Vector>
    void addField(const Vector& v,
                  const std::string& name,
                  Vtk::FieldType fieldType = Vtk::FieldType::automatic)
    { addField(v, name, this->precision(), fieldType); }
 
    template<typename Vector>
    void addField(const Vector& v,
                  const std::string& name,
                  Dumux::Vtk::Precision precision,
                  Vtk::FieldType fieldType = Vtk::FieldType::automatic)
    {
        // Deduce the number of components from the given vector type
        const auto nComp = getNumberOfComponents_(v);
 
        const auto numElemDofs = gridGeometry().elementMapper().size();
        const auto numVertexDofs = gridGeometry().vertexMapper().size();
 
        // Automatically deduce the field type ...
        if(fieldType == Vtk::FieldType::automatic)
        {
            if(numElemDofs == numVertexDofs)
                DUNE_THROW(Dune::InvalidStateException, "Automatic deduction of FieldType failed. Please explicitly specify FieldType::element or FieldType::vertex.");
 
            if(v.size() == numElemDofs)
                fieldType = Vtk::FieldType::element;
            else if(v.size() == numVertexDofs)
                fieldType = Vtk::FieldType::vertex;
            else
                DUNE_THROW(Dune::RangeError, "Size mismatch of added field!");
        }
        // ... or check if the user-specified type matches the size of v
        else
        {
            if(fieldType == Vtk::FieldType::element)
                if(v.size() != numElemDofs)
                    DUNE_THROW(Dune::RangeError, "Size mismatch of added field!");
 
            if(fieldType == Vtk::FieldType::vertex)
                if(v.size() != numVertexDofs)
                    DUNE_THROW(Dune::RangeError, "Size mismatch of added field!");
        }
 
        // add the appropriate field
        if (fieldType == Vtk::FieldType::element)
            addField(Field(gridGeometry_.gridView(), gridGeometry_.elementMapper(), v, name, nComp, 0, dm_, precision));
        else
            addField(Field(gridGeometry_.gridView(), gridGeometry_.vertexMapper(), v, name, nComp, dim, dm_, precision));
    }
 
    void addField(Field&& field)
    {
        // data arrays in the vtk output have to be unique within cell and point data
        // look if we have a field by the same name with the same codim, if so, replace it
        for (auto i = 0UL; i < fields_.size(); ++i)
        {
            if (fields_[i].name() == field.name() && fields_[i].codim() == field.codim())
            {
                fields_[i] = std::move(field);
                std::cout << Fmt::format(
                    "VtkOutputModule: Replaced field \"{}\" (codim {}). "
                    "A field by the same name & codim had already been registered previously.\n",
                    field.name(), field.codim()
                );
                return;
            }
        }
 
        // otherwise add it to the end of the fields
        fields_.push_back(std::move(field));
    }
 
    void write(double time, Dune::VTK::OutputType type = Dune::VTK::ascii)
    {
        Dune::Timer timer;
 
        // write to file depending on data mode
        if (dm_ == Dune::VTK::conforming)
            writeConforming_(time, type);
        else if (dm_ == Dune::VTK::nonconforming)
            writeNonConforming_(time, type);
        else
            DUNE_THROW(Dune::NotImplemented, "Output for provided VTK data mode");
 
        timer.stop();
        if (verbose_)
            std::cout << Fmt::format("Writing output for problem \"{}\". Took {:.2g} seconds.\n", name_, timer.elapsed());
    }
 
protected:
    const GridGeometry& gridGeometry() const { return gridGeometry_; }
 
    bool verbose() const { return verbose_; }
    const std::string& name() const { return name_; }
    Dune::VTK::DataMode dataMode() const { return dm_; }
    Dumux::Vtk::Precision precision() const { return precision_; }
 
    Dune::VTKWriter<GridView>& writer() { return *writer_; }
    Dune::VTKSequenceWriter<GridView>& sequenceWriter() { return *sequenceWriter_; }
 
    const std::vector<Field>& fields() const { return fields_; }
 
    void addCellData(const Field& field)
    {
        if (std::ranges::find(addedCellData_, field.name()) == addedCellData_.end())
        {
            this->sequenceWriter().addCellData(field.get());
            addedCellData_.push_back(field.name());
        }
    }
 
    void addVertexData(const Field& field)
    {
        if (std::ranges::find(addedVertexData_, field.name()) == addedVertexData_.end())
        {
            this->sequenceWriter().addVertexData(field.get());
            addedVertexData_.push_back(field.name());
        }
    }
 
    // keep track of what has been already added because Dune::VTK::Writer
    // potentially adds the same field twice which is not allowed in VTK/Paraview
    std::vector<std::string> addedCellData_;
    std::vector<std::string> addedVertexData_;
 
private:
    virtual void writeConforming_(double time, Dune::VTK::OutputType type)
    {
 
        // process rank
        std::vector<int> rank;
 
        if (!fields_.empty() || addProcessRank_)
        {
            const auto numCells = gridGeometry_.gridView().size(0);
 
            // maybe allocate space for the process rank
            if (addProcessRank_)
            {
                rank.resize(numCells);
 
                for (const auto& element : elements(gridGeometry_.gridView(), Dune::Partitions::interior))
                {
                    const auto eIdxGlobal = gridGeometry_.elementMapper().index(element);
                    rank[eIdxGlobal] = gridGeometry_.gridView().comm().rank();
                }
            }
 
 
            // the process rank
            if (addProcessRank_)
                this->addCellData(Field(gridGeometry_.gridView(), gridGeometry_.elementMapper(), rank, "process rank", 1, 0));
 
            // also register additional (non-standardized) user fields if any
            for (auto&& field : fields_)
            {
                if (field.codim() == 0)
                    this->addCellData(field);
                else if (field.codim() == dim)
                    this->addVertexData(field);
                else
                    DUNE_THROW(Dune::RangeError, "Cannot add wrongly sized vtk scalar field!");
            }
        }
 
        this->sequenceWriter().write(time, type);
 
        this->writer().clear();
 
        this->addedCellData_.clear();
        this->addedVertexData_.clear();
    }
 
    virtual void writeNonConforming_(double time, Dune::VTK::OutputType type)
    {
        DUNE_THROW(Dune::NotImplemented, "Non-conforming VTK output");
    }
 
    template<class Vector>
    std::size_t getNumberOfComponents_(const Vector& v)
    {
        if constexpr (Dune::IsIndexable<decltype(std::declval<Vector>()[0])>::value)
            return v[0].size();
        else
            return 1;
    }
 
    const GridGeometry& gridGeometry_;
    std::string name_;
    const std::string paramGroup_;
    Dune::VTK::DataMode dm_;
    bool verbose_;
    Dumux::Vtk::Precision precision_;
 
    std::shared_ptr<Dune::VTKWriter<GridView>> writer_;
    std::unique_ptr<Dune::VTKSequenceWriter<GridView>> sequenceWriter_;
 
    std::vector<Field> fields_; 
 
    bool addProcessRank_ = true;
};
 
template<class GridVariables, class SolutionVector>
class VtkOutputModule : public VtkOutputModuleBase<typename GridVariables::GridGeometry>
{
    using ParentType = VtkOutputModuleBase<typename GridVariables::GridGeometry>;
    using GridGeometry = typename GridVariables::GridGeometry;
 
    using VV = typename GridVariables::VolumeVariables;
    using Scalar = typename GridVariables::Scalar;
 
    using GridView = typename GridGeometry::GridView;
 
    enum {
        dim = GridView::dimension,
        dimWorld = GridView::dimensionworld
    };
 
    using Element = typename GridView::template Codim<0>::Entity;
    using VolVarsVector = Dune::FieldVector<Scalar, dimWorld>;
 
    static constexpr bool isBox = GridGeometry::discMethod == DiscretizationMethods::box;
    static constexpr bool isDiamond = GridGeometry::discMethod == DiscretizationMethods::fcdiamond;
    static constexpr bool isPQ1Bubble = GridGeometry::discMethod == DiscretizationMethods::pq1bubble;
    static constexpr bool isPQ2 = GridGeometry::discMethod == DiscretizationMethods::pq2;
 
    struct VolVarScalarDataInfo { std::function<Scalar(const VV&)> get; std::string name; Dumux::Vtk::Precision precision_; };
    struct VolVarVectorDataInfo { std::function<VolVarsVector(const VV&)> get; std::string name; Dumux::Vtk::Precision precision_; };
 
    using VelocityOutputType = Dumux::VelocityOutput<GridVariables>;
 
public:
    using Field = Vtk::template Field<GridView>;
    using VolumeVariables = VV;
 
    VtkOutputModule(const GridVariables& gridVariables,
                    const SolutionVector& sol,
                    const std::string& name,
                    const std::string& paramGroup = "",
                    Dune::VTK::DataMode dm = Dune::VTK::conforming,
                    bool verbose = true)
    : ParentType(gridVariables.gridGeometry(), name, paramGroup, dm, verbose)
    , gridVariables_(gridVariables)
    , sol_(sol)
    , velocityOutput_(std::make_shared<VelocityOutputType>())
    {
        enableVelocityOutput_ = getParamFromGroup<bool>(this->paramGroup(), "Vtk.AddVelocity", false);
        addProcessRank_ = getParamFromGroup<bool>(this->paramGroup(), "Vtk.AddProcessRank", true);
    }
 
 
    void addVelocityOutput(std::shared_ptr<VelocityOutputType> velocityOutput)
    { velocityOutput_ = velocityOutput; }
 
    void addVolumeVariable(std::function<Scalar(const VolumeVariables&)>&& f,
                                                const std::string& name)
    {
        // data arrays in the vtk output have to be unique within cell and point data
        // look if we have a field by the same name with the same codim, if so, replace it
        for (auto i = 0UL; i < volVarScalarDataInfo_.size(); ++i)
        {
            if (volVarScalarDataInfo_[i].name == name)
            {
                volVarScalarDataInfo_[i] = VolVarScalarDataInfo{f, name, this->precision()};
                std::cout << Fmt::format(
                    "VtkOutputModule: Replaced volume variable output \"{}\". "
                    "A field by the same name had already been registered previously.\n",
                    name
                );
                return;
            }
        }
 
        // otherwise add it to the end of the fields
        volVarScalarDataInfo_.push_back(VolVarScalarDataInfo{f, name, this->precision()});
    }
 
    template<class VVV = VolVarsVector, typename std::enable_if_t<(VVV::dimension > 1), int> = 0>
    void addVolumeVariable(std::function<VolVarsVector(const VolumeVariables&)>&& f,
                                                       const std::string& name)
    {
        // data arrays in the vtk output have to be unique within cell and point data
        // look if we have a field by the same name with the same codim, if so, replace it
        for (auto i = 0UL; i < volVarVectorDataInfo_.size(); ++i)
        {
            if (volVarVectorDataInfo_[i].name == name)
            {
                volVarVectorDataInfo_[i] = VolVarVectorDataInfo{f, name, this->precision()};
                std::cout << Fmt::format(
                    "VtkOutputModule: Replaced volume variable output \"{}\". "
                    "A field by the same name had already been registered previously.\n",
                    name
                );
                return;
            }
        }
 
        // otherwise add it to the end of the fields
        volVarVectorDataInfo_.push_back(VolVarVectorDataInfo{f, name, this->precision()});
    }
 
protected:
    // some return functions for differing implementations to use
    const auto& problem() const { return gridVariables_.curGridVolVars().problem(); }
    const GridVariables& gridVariables() const { return gridVariables_; }
    const GridGeometry& gridGeometry() const { return gridVariables_.gridGeometry(); }
    const SolutionVector& sol() const { return sol_; }
 
    const std::vector<VolVarScalarDataInfo>& volVarScalarDataInfo() const { return volVarScalarDataInfo_; }
    const std::vector<VolVarVectorDataInfo>& volVarVectorDataInfo() const { return volVarVectorDataInfo_; }
 
    using VelocityOutput = VelocityOutputType;
    const VelocityOutput& velocityOutput() const { return *velocityOutput_; }
 
private:
 
    void writeConforming_(double time, Dune::VTK::OutputType type) override
    {
        const Dune::VTK::DataMode dm = Dune::VTK::conforming;
 
        // instantiate the velocity output
        using VelocityVector = typename VelocityOutput::VelocityVector;
        std::vector<VelocityVector> velocity(velocityOutput_->numFluidPhases());
 
        // process rank
        std::vector<double> rank;
 
        // volume variable data
        std::vector<std::vector<Scalar>> volVarScalarData;
        std::vector<std::vector<VolVarsVector>> volVarVectorData;
 
        if (!volVarScalarDataInfo_.empty()
            || !volVarVectorDataInfo_.empty()
            || !this->fields().empty()
            || velocityOutput_->enableOutput()
            || addProcessRank_)
        {
            const auto numCells = gridGeometry().gridView().size(0);
            const auto numDofs = numDofs_();
 
            // get fields for all volume variables
            if (!volVarScalarDataInfo_.empty())
                volVarScalarData.resize(volVarScalarDataInfo_.size(), std::vector<Scalar>(numDofs));
            if (!volVarVectorDataInfo_.empty())
                volVarVectorData.resize(volVarVectorDataInfo_.size(), std::vector<VolVarsVector>(numDofs));
 
            if (velocityOutput_->enableOutput())
            {
                for (int phaseIdx = 0; phaseIdx < velocityOutput_->numFluidPhases(); ++phaseIdx)
                {
                    if (velocityOutput_->fieldType() == VelocityOutput::FieldType::element)
                        velocity[phaseIdx].resize(numCells);
                    else if (velocityOutput_->fieldType() == VelocityOutput::FieldType::vertex)
                        velocity[phaseIdx].resize(numDofs);
                    else
                    {
                        if(isBox && dim == 1)
                            velocity[phaseIdx].resize(numCells);
                        else
                            velocity[phaseIdx].resize(numDofs);
                    }
                }
            }
 
            // maybe allocate space for the process rank
            if (addProcessRank_) rank.resize(numCells);
 
            auto fvGeometry = localView(gridGeometry());
            auto elemVolVars = localView(gridVariables_.curGridVolVars());
            for (const auto& element : elements(gridGeometry().gridView()))
            {
                if (!velocityOutput_->enableOutput() &&
                    element.partitionType() != Dune::PartitionType::InteriorEntity)
                {
                    continue;
                }
                const auto eIdxGlobal = gridGeometry().elementMapper().index(element);
                // If velocity output is enabled we need to bind to the whole stencil
                // otherwise element-local data is sufficient
                if (velocityOutput_->enableOutput())
                {
                    fvGeometry.bind(element);
                    elemVolVars.bind(element, fvGeometry, sol_);
                }
                else
                {
                    fvGeometry.bindElement(element);
                    elemVolVars.bindElement(element, fvGeometry, sol_);
                }
 
                // velocity output
                if (velocityOutput_->enableOutput())
                {
                    const auto elemFluxVarsCache = localView(gridVariables_.gridFluxVarsCache()).bind(element, fvGeometry, elemVolVars);
 
                    for (int phaseIdx = 0; phaseIdx < velocityOutput_->numFluidPhases(); ++phaseIdx)
                        velocityOutput_->calculateVelocity(velocity[phaseIdx], element, fvGeometry, elemVolVars, elemFluxVarsCache, phaseIdx);
                }
                else if (element.partitionType() != Dune::PartitionType::InteriorEntity)
                {
                    continue;
                }
 
                if (!volVarScalarDataInfo_.empty() || !volVarVectorDataInfo_.empty())
                {
                    for (const auto& scv : scvs(fvGeometry))
                    {
                        const auto dofIdxGlobal = scv.dofIndex();
                        const auto& volVars = elemVolVars[scv];
 
                        // get the scalar-valued data
                        for (std::size_t i = 0; i < volVarScalarDataInfo_.size(); ++i)
                            volVarScalarData[i][dofIdxGlobal] = volVarScalarDataInfo_[i].get(volVars);
 
                        // get the vector-valued data
                        for (std::size_t i = 0; i < volVarVectorDataInfo_.size(); ++i)
                            volVarVectorData[i][dofIdxGlobal] = volVarVectorDataInfo_[i].get(volVars);
                    }
                }
 
                if (addProcessRank_)
                    rank[eIdxGlobal] = static_cast<double>(gridGeometry().gridView().comm().rank());
            }
 
 
            // volume variables if any
            if constexpr (isBox || isPQ1Bubble || isPQ2)
            {
                for (std::size_t i = 0; i < volVarScalarDataInfo_.size(); ++i)
                    this->addVertexData( Field(gridGeometry().gridView(), gridGeometry().dofMapper(), volVarScalarData[i],
                                          volVarScalarDataInfo_[i].name, /*numComp*/1, /*codim*/dim, dm, this->precision()) );
                for (std::size_t i = 0; i < volVarVectorDataInfo_.size(); ++i)
                    this->addVertexData( Field(gridGeometry().gridView(), gridGeometry().dofMapper(), volVarVectorData[i],
                                          volVarVectorDataInfo_[i].name, /*numComp*/dimWorld, /*codim*/dim, dm, this->precision()) );
 
                if constexpr (isPQ1Bubble)
                {
                    for (std::size_t i = 0; i < volVarScalarDataInfo_.size(); ++i)
                        this->addCellData( Field(gridGeometry().gridView(), gridGeometry().dofMapper(), volVarScalarData[i],
                                            volVarScalarDataInfo_[i].name, /*numComp*/1, /*codim*/0,dm, this->precision()) );
                    for (std::size_t i = 0; i < volVarVectorDataInfo_.size(); ++i)
                        this->addCellData( Field(gridGeometry().gridView(), gridGeometry().dofMapper(), volVarVectorData[i],
                                            volVarVectorDataInfo_[i].name, /*numComp*/dimWorld, /*codim*/0,dm, this->precision()) );
                }
 
            }
            else
            {
                for (std::size_t i = 0; i < volVarScalarDataInfo_.size(); ++i)
                    this->addCellData( Field(gridGeometry().gridView(), gridGeometry().elementMapper(), volVarScalarData[i],
                                        volVarScalarDataInfo_[i].name, /*numComp*/1, /*codim*/0,dm, this->precision()) );
                for (std::size_t i = 0; i < volVarVectorDataInfo_.size(); ++i)
                    this->addCellData( Field(gridGeometry().gridView(), gridGeometry().elementMapper(), volVarVectorData[i],
                                        volVarVectorDataInfo_[i].name, /*numComp*/dimWorld, /*codim*/0,dm, this->precision()) );
            }
 
            // the velocity field
            if (velocityOutput_->enableOutput())
            {
                if (velocityOutput_->fieldType() == VelocityOutput::FieldType::vertex
                    || ( (velocityOutput_->fieldType() == VelocityOutput::FieldType::automatic) && dim > 1 && isBox ))
                {
                    for (int phaseIdx = 0; phaseIdx < velocityOutput_->numFluidPhases(); ++phaseIdx)
                        this->addVertexData( Field(gridGeometry().gridView(), gridGeometry().vertexMapper(), velocity[phaseIdx],
                                              "velocity_" + velocityOutput_->phaseName(phaseIdx) + " (m/s)",
                                              /*numComp*/dimWorld, /*codim*/dim, dm, this->precision()) );
                }
                // cell-centered models
                else
                {
                    for (int phaseIdx = 0; phaseIdx < velocityOutput_->numFluidPhases(); ++phaseIdx)
                        this->addCellData( Field(gridGeometry().gridView(), gridGeometry().elementMapper(), velocity[phaseIdx],
                                            "velocity_" + velocityOutput_->phaseName(phaseIdx) + " (m/s)",
                                            /*numComp*/dimWorld, /*codim*/0, dm, this->precision()) );
                }
            }
 
            // the process rank
            if (addProcessRank_)
                this->addCellData(Field(gridGeometry().gridView(), gridGeometry().elementMapper(), rank, "process rank", 1, 0));
 
            // also register additional (non-standardized) user fields if any
            for (auto&& field : this->fields())
            {
                if (field.codim() == 0)
                    this->addCellData(field);
                else if (field.codim() == dim)
                    this->addVertexData(field);
                else
                    DUNE_THROW(Dune::RangeError, "Cannot add wrongly sized vtk scalar field!");
            }
        }
 
        this->sequenceWriter().write(time, type);
 
        this->writer().clear();
 
        this->addedCellData_.clear();
        this->addedVertexData_.clear();
    }
 
    void writeNonConforming_(double time, Dune::VTK::OutputType type) override
    {
        const Dune::VTK::DataMode dm = Dune::VTK::nonconforming;
 
        // only supports finite-element-like discretization schemes
        if(!isBox && !isDiamond)
            DUNE_THROW(Dune::NotImplemented,
                "Non-conforming output for discretization scheme " << GridGeometry::discMethod
            );
 
 
        // check the velocity output
        if (enableVelocityOutput_ && !velocityOutput_->enableOutput())
            std::cerr << "Warning! Velocity output was enabled in the input file"
                    << " but no velocity output policy was set for the VTK output module:"
                    << " There will be no velocity output."
                    << " Use the addVelocityOutput member function of the VTK output module." << std::endl;
        using VelocityVector = typename VelocityOutput::VelocityVector;
        std::vector<VelocityVector> velocity(velocityOutput_->numFluidPhases());
 
        // process rank
        std::vector<double> rank;
 
        // volume variable data (indexing: volvardata/element/localindex)
        using ScalarDataContainer = std::vector< std::vector<Scalar> >;
        using VectorDataContainer = std::vector< std::vector<VolVarsVector> >;
        std::vector< ScalarDataContainer > volVarScalarData;
        std::vector< VectorDataContainer > volVarVectorData;
 
        if (!volVarScalarDataInfo_.empty()
            || !volVarVectorDataInfo_.empty()
            || !this->fields().empty()
            || velocityOutput_->enableOutput()
            || addProcessRank_)
        {
            const auto numCells = gridGeometry().gridView().size(0);
            const auto outputSize = numDofs_();
 
            // get fields for all volume variables
            if (!volVarScalarDataInfo_.empty())
                volVarScalarData.resize(volVarScalarDataInfo_.size(), ScalarDataContainer(numCells));
            if (!volVarVectorDataInfo_.empty())
                volVarVectorData.resize(volVarVectorDataInfo_.size(), VectorDataContainer(numCells));
 
            if (velocityOutput_->enableOutput())
            {
                for (int phaseIdx = 0; phaseIdx < velocityOutput_->numFluidPhases(); ++phaseIdx)
                {
                    if((isBox && dim == 1) || isDiamond)
                        velocity[phaseIdx].resize(numCells);
                    else
                        velocity[phaseIdx].resize(outputSize);
                }
            }
 
            // maybe allocate space for the process rank
            if (addProcessRank_) rank.resize(numCells);
 
            // now we go element-local to extract values at local dof locations
            auto fvGeometry = localView(gridGeometry());
            auto elemVolVars = localView(gridVariables_.curGridVolVars());
            for (const auto& element : elements(gridGeometry().gridView()))
            {
                if (!velocityOutput_->enableOutput() &&
                    element.partitionType() != Dune::PartitionType::InteriorEntity)
                {
                    continue;
                }
                const auto eIdxGlobal = gridGeometry().elementMapper().index(element);
                // If velocity output is enabled we need to bind to the whole stencil
                // otherwise element-local data is sufficient
                if (velocityOutput_->enableOutput())
                {
                    fvGeometry.bind(element);
                    elemVolVars.bind(element, fvGeometry, sol_);
                }
                else
                {
                    fvGeometry.bindElement(element);
                    elemVolVars.bindElement(element, fvGeometry, sol_);
                }
 
                const auto numLocalDofs = fvGeometry.numScv();
                // resize element-local data containers
                for (std::size_t i = 0; i < volVarScalarDataInfo_.size(); ++i)
                    volVarScalarData[i][eIdxGlobal].resize(numLocalDofs);
                for (std::size_t i = 0; i < volVarVectorDataInfo_.size(); ++i)
                    volVarVectorData[i][eIdxGlobal].resize(numLocalDofs);
 
                // velocity output
                if (velocityOutput_->enableOutput())
                {
                    const auto elemFluxVarsCache = localView(gridVariables_.gridFluxVarsCache()).bind(element, fvGeometry, elemVolVars);
                    for (int phaseIdx = 0; phaseIdx < velocityOutput_->numFluidPhases(); ++phaseIdx)
                        velocityOutput_->calculateVelocity(velocity[phaseIdx], element, fvGeometry, elemVolVars, elemFluxVarsCache, phaseIdx);
                }
                else if (element.partitionType() != Dune::PartitionType::InteriorEntity)
                {
                    continue;
                }
 
                if (!volVarScalarDataInfo_.empty() || !volVarVectorDataInfo_.empty())
                {
                    for (const auto& scv : scvs(fvGeometry))
                    {
                        const auto& volVars = elemVolVars[scv];
 
                        // get the scalar-valued data
                        for (std::size_t i = 0; i < volVarScalarDataInfo_.size(); ++i)
                            volVarScalarData[i][eIdxGlobal][scv.localDofIndex()] = volVarScalarDataInfo_[i].get(volVars);
 
                        // get the vector-valued data
                        for (std::size_t i = 0; i < volVarVectorDataInfo_.size(); ++i)
                            volVarVectorData[i][eIdxGlobal][scv.localDofIndex()] = volVarVectorDataInfo_[i].get(volVars);
                    }
                }
 
                if (addProcessRank_)
                    rank[eIdxGlobal] = static_cast<double>(gridGeometry().gridView().comm().rank());
            }
 
 
            // volume variables if any
            static constexpr int dofLocCodim = isDiamond ? 1 : dim;
            for (std::size_t i = 0; i < volVarScalarDataInfo_.size(); ++i)
                this->addVertexData(Field(
                    gridGeometry().gridView(), gridGeometry().elementMapper(),
                    volVarScalarData[i], volVarScalarDataInfo_[i].name,
                    /*numComp*/1, /*codim*/dofLocCodim, /*nonconforming*/dm, this->precision()
                ));
 
            for (std::size_t i = 0; i < volVarVectorDataInfo_.size(); ++i)
                this->addVertexData(Field(
                    gridGeometry().gridView(), gridGeometry().elementMapper(),
                    volVarVectorData[i], volVarVectorDataInfo_[i].name,
                    /*numComp*/dimWorld, /*codim*/dofLocCodim, /*nonconforming*/dm, this->precision()
                ));
 
            // the velocity field
            if (velocityOutput_->enableOutput())
            {
                // node-wise velocities
                if (dim > 1 && !isDiamond)
                    for (int phaseIdx = 0; phaseIdx < velocityOutput_->numFluidPhases(); ++phaseIdx)
                        this->addVertexData(Field(
                            gridGeometry().gridView(), gridGeometry().vertexMapper(), velocity[phaseIdx],
                            "velocity_" + velocityOutput_->phaseName(phaseIdx) + " (m/s)",
                            /*numComp*/dimWorld, /*codim*/dofLocCodim, dm, this->precision()
                        ));
 
                // cell-wise velocities
                else
                    for (int phaseIdx = 0; phaseIdx < velocityOutput_->numFluidPhases(); ++phaseIdx)
                        this->addCellData(Field(
                            gridGeometry().gridView(), gridGeometry().elementMapper(), velocity[phaseIdx],
                            "velocity_" + velocityOutput_->phaseName(phaseIdx) + " (m/s)",
                            /*numComp*/dimWorld, /*codim*/0, dm, this->precision()
                        ));
            }
        }
 
        // the process rank
        if (addProcessRank_)
            this->addCellData(Field(
                gridGeometry().gridView(), gridGeometry().elementMapper(),
                rank, "process rank", /*numComp*/1, /*codim*/0
            ));
 
        // also register additional (non-standardized) user fields if any
        for (const auto& field : this->fields())
        {
            if (field.codim() == 0)
                this->addCellData(field);
            else if (field.codim() == dim || field.codim() == 1)
                this->addVertexData(field);
            else
                DUNE_THROW(Dune::RangeError, "Cannot add wrongly sized vtk scalar field!");
        }
 
        this->sequenceWriter().write(time, type);
 
        this->writer().clear();
 
        this->addedCellData_.clear();
        this->addedVertexData_.clear();
    }
 
    std::size_t numDofs_() const
    {
        // TODO this should actually always be dofMapper.size()
        // maybe some discretizations needs special treatment (?)
        if constexpr (isBox || isDiamond || isPQ1Bubble || isPQ2)
            return gridGeometry().dofMapper().size();
        else
            return gridGeometry().elementMapper().size();
    }
 
    const GridVariables& gridVariables_;
    const SolutionVector& sol_;
 
    std::vector<VolVarScalarDataInfo> volVarScalarDataInfo_; 
    std::vector<VolVarVectorDataInfo> volVarVectorDataInfo_; 
 
    std::shared_ptr<VelocityOutput> velocityOutput_; 
    bool enableVelocityOutput_ = false;
    bool addProcessRank_ = true;
};
 
} // end namespace Dumux
 
#endif