PhysDeriv.cpp

Go to the documentation of this file.

///////////////////////////////////////////////////////////////////////////////
//
// File: PhysDeriv.cpp
//
// For more information, please see: http://www.nektar.info
//
// The MIT License
//
// Copyright (c) 2006 Division of Applied Mathematics, Brown University (USA),
// Department of Aeronautics, Imperial College London (UK), and Scientific
// Computing and Imaging Institute, University of Utah (USA).
//
// Permission is hereby granted, free of charge, to any person obtaining a
// copy of this software and associated documentation files (the "Software"),
// to deal in the Software without restriction, including without limitation
// the rights to use, copy, modify, merge, publish, distribute, sublicense,
// and/or sell copies of the Software, and to permit persons to whom the
// Software is furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included
// in all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
// OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
// THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
// DEALINGS IN THE SOFTWARE.
//
// Description: PhysDeriv operator implementations
//
///////////////////////////////////////////////////////////////////////////////
 
#include <Collections/Collection.h>
#include <Collections/MatrixFreeBase.h>
#include <Collections/Operator.h>
#include <MatrixFreeOps/Operator.hpp>
 
using namespace std;
 
namespace Nektar::Collections
{
 
using LibUtilities::eHexahedron;
using LibUtilities::eNodalPrism;
using LibUtilities::eNodalTet;
using LibUtilities::eNodalTri;
using LibUtilities::ePrism;
using LibUtilities::ePyramid;
using LibUtilities::eQuadrilateral;
using LibUtilities::eSegment;
using LibUtilities::eTetrahedron;
using LibUtilities::eTriangle;
 
/**
 * @brief Physical Derivative help class to calculate the size of the collection
 * that is given as an input and as an output to the PhysDeriv Operator. The
 * Operator evaluation is happenning in the physical space and the output is
 * expected to be part of the physical space too.
 */
class PhysDeriv_Helper : virtual public Operator
{
protected:
    PhysDeriv_Helper()
    {
        // expect input to be number of elements by the number of quadrature
        // points
        m_inputSize = m_numElmt * m_stdExp->GetTotPoints();
        // the derivate is using data from the physical space to evaluate the
        // derivative in the physical space
        m_outputSize = m_inputSize;
    }
};
 
/**
 * @brief Phys deriv operator using standard matrix approach
 */
class PhysDeriv_StdMat final : virtual public Operator,
                               virtual public PhysDeriv_Helper
{
public:
    OPERATOR_CREATE(PhysDeriv_StdMat)
 
    ~PhysDeriv_StdMat() final = default;
 
    void operator()(const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output0,
                    Array<OneD, NekDouble> &output1,
                    Array<OneD, NekDouble> &output2,
                    Array<OneD, NekDouble> &wsp) final
    {
        int nPhys = m_stdExp->GetTotPoints();
        int ntot  = m_numElmt * nPhys;
        Array<OneD, NekDouble> tmp0, tmp1, tmp2;
        Array<OneD, Array<OneD, NekDouble>> Diff(3);
        Array<OneD, Array<OneD, NekDouble>> out(3);
        out[0] = output0;
        out[1] = output1;
        out[2] = output2;
 
        for (int i = 0; i < m_dim; ++i)
        {
            Diff[i] = wsp + i * ntot;
        }
 
        // calculate local derivatives
        for (int i = 0; i < m_dim; ++i)
        {
            Blas::Dgemm('N', 'N', m_derivMat[i]->GetRows(), m_numElmt,
                        m_derivMat[i]->GetColumns(), 1.0,
                        m_derivMat[i]->GetRawPtr(), m_derivMat[i]->GetRows(),
                        input.data(), nPhys, 0.0, &Diff[i][0], nPhys);
        }
 
        // calculate full derivative
        if (m_isDeformed)
        {
            for (int i = 0; i < m_coordim; ++i)
            {
                Vmath::Zero(ntot, out[i], 1);
                for (int j = 0; j < m_dim; ++j)
                {
                    Vmath::Vvtvp(ntot, m_derivFac[i * m_dim + j], 1, Diff[j], 1,
                                 out[i], 1, out[i], 1);
                }
            }
        }
        else
        {
            Array<OneD, NekDouble> t;
            for (int i = 0; i < m_coordim; ++i)
            {
                Vmath::Zero(ntot, out[i], 1);
                for (int e = 0; e < m_numElmt; ++e)
                {
                    for (int j = 0; j < m_dim; ++j)
                    {
                        Vmath::Svtvp(m_nqe, m_derivFac[i * m_dim + j][e],
                                     Diff[j] + e * m_nqe, 1, out[i] + e * m_nqe,
                                     1, t = out[i] + e * m_nqe, 1);
                    }
                }
            }
        }
    }
 
    void operator()(int dir, const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output,
                    Array<OneD, NekDouble> &wsp) final
    {
        int nPhys = m_stdExp->GetTotPoints();
        int ntot  = m_numElmt * nPhys;
        Array<OneD, NekDouble> tmp0, tmp1, tmp2;
        Array<OneD, Array<OneD, NekDouble>> Diff(3);
 
        for (int i = 0; i < m_dim; ++i)
        {
            Diff[i] = wsp + i * ntot;
        }
 
        // calculate local derivatives
        for (int i = 0; i < m_dim; ++i)
        {
            Blas::Dgemm('N', 'N', m_derivMat[i]->GetRows(), m_numElmt,
                        m_derivMat[i]->GetColumns(), 1.0,
                        m_derivMat[i]->GetRawPtr(), m_derivMat[i]->GetRows(),
                        input.data(), nPhys, 0.0, &Diff[i][0], nPhys);
        }
 
        // calculate full derivative
        Vmath::Zero(ntot, output, 1);
        if (m_isDeformed)
        {
            for (int j = 0; j < m_dim; ++j)
            {
                Vmath::Vvtvp(ntot, m_derivFac[dir * m_dim + j], 1, Diff[j], 1,
                             output, 1, output, 1);
            }
        }
        else
        {
            Array<OneD, NekDouble> t;
            for (int e = 0; e < m_numElmt; ++e)
            {
                for (int j = 0; j < m_dim; ++j)
                {
                    Vmath::Svtvp(m_nqe, m_derivFac[dir * m_dim + j][e],
                                 Diff[j] + e * m_nqe, 1, output + e * m_nqe, 1,
                                 t = output + e * m_nqe, 1);
                }
            }
        }
    }
 
protected:
    Array<OneD, DNekMatSharedPtr> m_derivMat;
    Array<TwoD, const NekDouble> m_derivFac;
    int m_dim;
    int m_coordim;
 
private:
    PhysDeriv_StdMat(vector<LocalRegions::ExpansionSharedPtr> pCollExp,
                     CoalescedGeomDataSharedPtr pGeomData,
                     StdRegions::FactorMap factors)
        : Operator(pCollExp, pGeomData, factors), PhysDeriv_Helper()
    {
        int nqtot = pCollExp[0]->GetTotPoints();
        m_dim     = pCollExp[0]->GetShapeDimension();
        m_coordim = pCollExp[0]->GetCoordim();
 
        // set up a PhysDeriv StdMat.
        m_derivMat = Array<OneD, DNekMatSharedPtr>(m_dim);
        for (int i = 0; i < m_dim; ++i)
        {
            Array<OneD, NekDouble> tmp(nqtot), tmp1(nqtot);
            m_derivMat[i] =
                MemoryManager<DNekMat>::AllocateSharedPtr(nqtot, nqtot);
            for (int j = 0; j < nqtot; ++j)
            {
                Vmath::Zero(nqtot, tmp, 1);
                tmp[j] = 1.0;
                m_stdExp->PhysDeriv(i, tmp, tmp1);
                Vmath::Vcopy(nqtot, &tmp1[0], 1,
                             &(m_derivMat[i]->GetPtr())[0] + j * nqtot, 1);
            }
        }
        m_derivFac = pGeomData->GetDerivFactors(pCollExp);
        m_wspSize  = 3 * nqtot * m_numElmt;
    }
};
 
/// Factory initialisation for the PhysDeriv_StdMat operators
OperatorKey PhysDeriv_StdMat::m_typeArr[] = {
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eSegment, ePhysDeriv, eStdMat, false),
        PhysDeriv_StdMat::create, "PhysDeriv_StdMat_Seg"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eTriangle, ePhysDeriv, eStdMat, false),
        PhysDeriv_StdMat::create, "PhysDeriv_StdMat_Tri"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eNodalTri, ePhysDeriv, eStdMat, true),
        PhysDeriv_StdMat::create, "PhysDeriv_StdMat_NodalTri"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eQuadrilateral, ePhysDeriv, eStdMat, false),
        PhysDeriv_StdMat::create, "PhysDeriv_StdMat_Quad"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eTetrahedron, ePhysDeriv, eStdMat, false),
        PhysDeriv_StdMat::create, "PhysDeriv_StdMat_Tet"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eNodalTet, ePhysDeriv, eStdMat, true),
        PhysDeriv_StdMat::create, "PhysDeriv_StdMat_NodalTet"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(ePyramid, ePhysDeriv, eStdMat, false),
        PhysDeriv_StdMat::create, "PhysDeriv_StdMat_Pyr"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(ePrism, ePhysDeriv, eStdMat, false),
        PhysDeriv_StdMat::create, "PhysDeriv_StdMat_Prism"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eNodalPrism, ePhysDeriv, eStdMat, true),
        PhysDeriv_StdMat::create, "PhysDeriv_StdMat_NodalPrism"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eHexahedron, ePhysDeriv, eStdMat, false),
        PhysDeriv_StdMat::create, "PhysDeriv_StdMat_Hex"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(ePyramid, ePhysDeriv, eSumFac, false),
        PhysDeriv_StdMat::create, "PhysDeriv_SumFac_Pyr")};
 
/**
 * @brief Phys deriv operator using matrix free operators.
 */
class PhysDeriv_MatrixFree final : virtual public Operator,
                                   MatrixFreeBase,
                                   virtual public PhysDeriv_Helper
{
public:
    OPERATOR_CREATE(PhysDeriv_MatrixFree)
 
    ~PhysDeriv_MatrixFree() final = default;
 
    void operator()(const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output0,
                    Array<OneD, NekDouble> &output1,
                    Array<OneD, NekDouble> &output2,
                    [[maybe_unused]] Array<OneD, NekDouble> &wsp) final
    {
        Array<OneD, Array<OneD, NekDouble>> output(m_coordim);
        // currently using temporary local temporary space for output
        // to allow for other operator call below which is
        // directionally dependent
        switch (m_coordim)
        {
            case 1:
                output[0] = output0;
                break;
            case 2:
                output[0] = output0;
                output[1] = output1;
                break;
            case 3:
                output[0] = output0;
                output[1] = output1;
                output[2] = output2;
                break;
            default:
                NEKERROR(ErrorUtil::efatal, "Unknown coordinate dimension");
                break;
        }
        (*m_oper)(input, output);
    }
 
    void operator()(int dir, const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output,
                    [[maybe_unused]] Array<OneD, NekDouble> &wsp) final
    {
        (*m_oper)(input, m_output);
        Vmath::Vcopy(m_nOut, m_output[dir], 1, output, 1);
    }
 
private:
    std::shared_ptr<MatrixFree::PhysDeriv> m_oper;
    int m_coordim;
    Array<OneD, Array<OneD, NekDouble>> m_output;
 
    PhysDeriv_MatrixFree(vector<LocalRegions::ExpansionSharedPtr> pCollExp,
                         CoalescedGeomDataSharedPtr pGeomData,
                         StdRegions::FactorMap factors)
        : Operator(pCollExp, pGeomData, factors), PhysDeriv_Helper(),
          MatrixFreeBase(pCollExp[0]->GetTotPoints(),
                         pCollExp[0]->GetTotPoints(), pCollExp.size())
    {
        // Check if deformed
        bool deformed{pGeomData->IsDeformed(pCollExp)};
        const auto dim = pCollExp[0]->GetShapeDimension();
 
        // only used operator(dir, in, out)
        m_coordim   = pCollExp[0]->GetCoordim();
        int nOut    = pCollExp[0]->GetTotPoints();
        m_output    = Array<OneD, Array<OneD, NekDouble>>(m_coordim);
        m_output[0] = Array<OneD, NekDouble>{nOut * m_nElmtPad, 0.0};
        if (m_coordim == 2)
        {
            m_output[1] = Array<OneD, NekDouble>{nOut * m_nElmtPad, 0.0};
        }
        else if (m_coordim == 3)
        {
            m_output[1] = Array<OneD, NekDouble>{nOut * m_nElmtPad, 0.0};
            m_output[2] = Array<OneD, NekDouble>{nOut * m_nElmtPad, 0.0};
        }
 
        // Basis vector.
        std::vector<LibUtilities::BasisSharedPtr> basis(dim);
        for (unsigned int i = 0; i < dim; ++i)
        {
            basis[i] = pCollExp[0]->GetBasis(i);
        }
 
        // Get shape type
        auto shapeType = pCollExp[0]->DetShapeType();
 
        // Generate operator string and create operator.
        std::string op_string = "PhysDeriv";
        op_string += MatrixFree::GetOpstring(shapeType, deformed);
        auto oper = MatrixFree::GetOperatorFactory().CreateInstance(
            op_string, basis, pCollExp.size());
 
        oper->SetUpZW(basis);
        oper->SetUpD(basis);
 
        // Set derivative factors
        oper->SetDF(pGeomData->GetDerivFactorsInterLeave(pCollExp, m_nElmtPad));
 
        m_oper = std::dynamic_pointer_cast<MatrixFree::PhysDeriv>(oper);
        ASSERTL0(m_oper, "Failed to cast pointer.");
    }
};
 
/// Factory initialisation for the PhysDeriv_MatrixFree operators
OperatorKey PhysDeriv_MatrixFree::m_typeArr[] = {
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eSegment, ePhysDeriv, eMatrixFree, false),
        PhysDeriv_MatrixFree::create, "PhysDeriv_MatrixFree_Seg"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eTriangle, ePhysDeriv, eMatrixFree, false),
        PhysDeriv_MatrixFree::create, "PhysDeriv_MatrixFree_Tri"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eQuadrilateral, ePhysDeriv, eMatrixFree, false),
        PhysDeriv_MatrixFree::create, "PhysDeriv_MatrixFree_Quad"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eHexahedron, ePhysDeriv, eMatrixFree, false),
        PhysDeriv_MatrixFree::create, "PhysDeriv_MatrixFree_Hex"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(ePrism, ePhysDeriv, eMatrixFree, false),
        PhysDeriv_MatrixFree::create, "PhysDeriv_MatrixFree_Prism"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(ePyramid, ePhysDeriv, eMatrixFree, false),
        PhysDeriv_MatrixFree::create, "PhysDeriv_MatrixFree_Pyr"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eTetrahedron, ePhysDeriv, eMatrixFree, false),
        PhysDeriv_MatrixFree::create, "PhysDeriv_MatrixFree_Tet")
 
};
 
/**
 * @brief Phys deriv operator using element-wise operation
 */
class PhysDeriv_IterPerExp final : virtual public Operator,
                                   virtual public PhysDeriv_Helper
{
public:
    OPERATOR_CREATE(PhysDeriv_IterPerExp)
 
    ~PhysDeriv_IterPerExp() final = default;
 
    void operator()(const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output0,
                    Array<OneD, NekDouble> &output1,
                    Array<OneD, NekDouble> &output2,
                    Array<OneD, NekDouble> &wsp) final
    {
        int nPhys = m_stdExp->GetTotPoints();
        int ntot  = m_numElmt * nPhys;
        Array<OneD, NekDouble> tmp0, tmp1, tmp2;
        Array<OneD, Array<OneD, NekDouble>> Diff(3);
        Array<OneD, Array<OneD, NekDouble>> out(3);
        out[0] = output0;
        out[1] = output1;
        out[2] = output2;
 
        for (int i = 0; i < m_dim; ++i)
        {
            Diff[i] = wsp + i * ntot;
        }
 
        // calculate local derivatives
        for (int i = 0; i < m_numElmt; ++i)
        {
            m_stdExp->PhysDeriv(input + i * nPhys, tmp0 = Diff[0] + i * nPhys,
                                tmp1 = Diff[1] + i * nPhys,
                                tmp2 = Diff[2] + i * nPhys);
        }
 
        // calculate full derivative
        if (m_isDeformed)
        {
            for (int i = 0; i < m_coordim; ++i)
            {
                Vmath::Vmul(ntot, m_derivFac[i * m_dim], 1, Diff[0], 1, out[i],
                            1);
                for (int j = 1; j < m_dim; ++j)
                {
                    Vmath::Vvtvp(ntot, m_derivFac[i * m_dim + j], 1, Diff[j], 1,
                                 out[i], 1, out[i], 1);
                }
            }
        }
        else
        {
            Array<OneD, NekDouble> t;
            for (int e = 0; e < m_numElmt; ++e)
            {
                for (int i = 0; i < m_coordim; ++i)
                {
                    Vmath::Smul(m_nqe, m_derivFac[i * m_dim][e],
                                Diff[0] + e * m_nqe, 1, t = out[i] + e * m_nqe,
                                1);
                    for (int j = 1; j < m_dim; ++j)
                    {
                        Vmath::Svtvp(m_nqe, m_derivFac[i * m_dim + j][e],
                                     Diff[j] + e * m_nqe, 1, out[i] + e * m_nqe,
                                     1, t = out[i] + e * m_nqe, 1);
                    }
                }
            }
        }
    }
 
    void operator()(int dir, const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output,
                    Array<OneD, NekDouble> &wsp) final
    {
        int nPhys = m_stdExp->GetTotPoints();
        int ntot  = m_numElmt * nPhys;
        Array<OneD, NekDouble> tmp0, tmp1, tmp2;
        Array<OneD, Array<OneD, NekDouble>> Diff(3);
 
        for (int i = 0; i < m_dim; ++i)
        {
            Diff[i] = wsp + i * ntot;
        }
 
        // calculate local derivatives
        for (int i = 0; i < m_numElmt; ++i)
        {
            m_stdExp->PhysDeriv(input + i * nPhys, tmp0 = Diff[0] + i * nPhys,
                                tmp1 = Diff[1] + i * nPhys,
                                tmp2 = Diff[2] + i * nPhys);
        }
 
        Vmath::Zero(ntot, output, 1);
        if (m_isDeformed)
        {
            for (int j = 0; j < m_dim; ++j)
            {
                Vmath::Vvtvp(ntot, m_derivFac[dir * m_dim + j], 1, Diff[j], 1,
                             output, 1, output, 1);
            }
        }
        else
        {
            Array<OneD, NekDouble> t;
            for (int e = 0; e < m_numElmt; ++e)
            {
                for (int j = 0; j < m_dim; ++j)
                {
                    Vmath::Svtvp(m_nqe, m_derivFac[dir * m_dim + j][e],
                                 Diff[j] + e * m_nqe, 1, output + e * m_nqe, 1,
                                 t = output + e * m_nqe, 1);
                }
            }
        }
    }
 
protected:
    Array<TwoD, const NekDouble> m_derivFac;
    int m_dim;
    int m_coordim;
 
private:
    PhysDeriv_IterPerExp(vector<LocalRegions::ExpansionSharedPtr> pCollExp,
                         CoalescedGeomDataSharedPtr pGeomData,
                         StdRegions::FactorMap factors)
        : Operator(pCollExp, pGeomData, factors), PhysDeriv_Helper()
    {
        int nqtot = pCollExp[0]->GetTotPoints();
        m_dim     = pCollExp[0]->GetShapeDimension();
        m_coordim = pCollExp[0]->GetCoordim();
 
        m_derivFac = pGeomData->GetDerivFactors(pCollExp);
        m_wspSize  = 3 * nqtot * m_numElmt;
    }
};
 
/// Factory initialisation for the PhysDeriv_IterPerExp operators
OperatorKey PhysDeriv_IterPerExp::m_typeArr[] = {
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eSegment, ePhysDeriv, eIterPerExp, false),
        PhysDeriv_IterPerExp::create, "PhysDeriv_IterPerExp_Seg"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eTriangle, ePhysDeriv, eIterPerExp, false),
        PhysDeriv_IterPerExp::create, "PhysDeriv_IterPerExp_Tri"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eNodalTri, ePhysDeriv, eIterPerExp, true),
        PhysDeriv_IterPerExp::create, "PhysDeriv_IterPerExp_NodalTri"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eQuadrilateral, ePhysDeriv, eIterPerExp, false),
        PhysDeriv_IterPerExp::create, "PhysDeriv_IterPerExp_Quad"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eTetrahedron, ePhysDeriv, eIterPerExp, false),
        PhysDeriv_IterPerExp::create, "PhysDeriv_IterPerExp_Tet"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eNodalTet, ePhysDeriv, eIterPerExp, true),
        PhysDeriv_IterPerExp::create, "PhysDeriv_IterPerExp_NodalTet"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(ePyramid, ePhysDeriv, eIterPerExp, false),
        PhysDeriv_IterPerExp::create, "PhysDeriv_IterPerExp_Pyr"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(ePrism, ePhysDeriv, eIterPerExp, false),
        PhysDeriv_IterPerExp::create, "PhysDeriv_IterPerExp_Prism"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eNodalPrism, ePhysDeriv, eIterPerExp, true),
        PhysDeriv_IterPerExp::create, "PhysDeriv_IterPerExp_NodalPrism"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eHexahedron, ePhysDeriv, eIterPerExp, false),
        PhysDeriv_IterPerExp::create, "PhysDeriv_IterPerExp_Hex")};
 
/**
 * @brief Phys deriv operator using original LocalRegions implementation.
 */
class PhysDeriv_NoCollection final : virtual public Operator,
                                     virtual public PhysDeriv_Helper
{
public:
    OPERATOR_CREATE(PhysDeriv_NoCollection)
 
    ~PhysDeriv_NoCollection() final = default;
 
    void operator()(const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output0,
                    Array<OneD, NekDouble> &output1,
                    Array<OneD, NekDouble> &output2,
                    [[maybe_unused]] Array<OneD, NekDouble> &wsp) final
    {
        const int nPhys = m_expList[0]->GetTotPoints();
        Array<OneD, NekDouble> tmp0, tmp1, tmp2;
 
        // calculate local derivatives
        switch (m_expList[0]->GetCoordim())
        {
            case 1:
            {
                for (int i = 0; i < m_numElmt; ++i)
                {
                    m_expList[i]->PhysDeriv(input + i * nPhys,
                                            tmp0 = output0 + i * nPhys);
                }
                break;
            }
            case 2:
            {
                for (int i = 0; i < m_numElmt; ++i)
                {
                    m_expList[i]->PhysDeriv(input + i * nPhys,
                                            tmp0 = output0 + i * nPhys,
                                            tmp1 = output1 + i * nPhys);
                }
                break;
            }
            case 3:
            {
                for (int i = 0; i < m_numElmt; ++i)
                {
                    m_expList[i]->PhysDeriv(
                        input + i * nPhys, tmp0 = output0 + i * nPhys,
                        tmp1 = output1 + i * nPhys, tmp2 = output2 + i * nPhys);
                }
                break;
            }
            default:
                ASSERTL0(false, "Unknown dimension.");
        }
    }
 
    void operator()(int dir, const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output,
                    [[maybe_unused]] Array<OneD, NekDouble> &wsp) final
    {
        const int nPhys = m_expList[0]->GetTotPoints();
        Array<OneD, NekDouble> tmp;
 
        // calculate local derivatives
        for (int i = 0; i < m_numElmt; ++i)
        {
            m_expList[i]->PhysDeriv(dir, input + i * nPhys,
                                    tmp = output + i * nPhys);
        }
    }
 
protected:
    vector<LocalRegions::ExpansionSharedPtr> m_expList;
 
private:
    PhysDeriv_NoCollection(vector<LocalRegions::ExpansionSharedPtr> pCollExp,
                           CoalescedGeomDataSharedPtr pGeomData,
                           StdRegions::FactorMap factors)
        : Operator(pCollExp, pGeomData, factors), PhysDeriv_Helper()
    {
        m_expList = pCollExp;
    }
};
 
/// Factory initialisation for the PhysDeriv_NoCollection operators
OperatorKey PhysDeriv_NoCollection::m_typeArr[] = {
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eSegment, ePhysDeriv, eNoCollection, false),
        PhysDeriv_NoCollection::create, "PhysDeriv_NoCollection_Seg"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eTriangle, ePhysDeriv, eNoCollection, false),
        PhysDeriv_NoCollection::create, "PhysDeriv_NoCollection_Tri"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eNodalTri, ePhysDeriv, eNoCollection, true),
        PhysDeriv_NoCollection::create, "PhysDeriv_NoCollection_NodalTri"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eQuadrilateral, ePhysDeriv, eNoCollection, false),
        PhysDeriv_NoCollection::create, "PhysDeriv_NoCollection_Quad"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eTetrahedron, ePhysDeriv, eNoCollection, false),
        PhysDeriv_NoCollection::create, "PhysDeriv_NoCollection_Tet"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eNodalTet, ePhysDeriv, eNoCollection, true),
        PhysDeriv_NoCollection::create, "PhysDeriv_NoCollection_NodalTet"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(ePyramid, ePhysDeriv, eNoCollection, false),
        PhysDeriv_NoCollection::create, "PhysDeriv_NoCollection_Pyr"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(ePrism, ePhysDeriv, eNoCollection, false),
        PhysDeriv_NoCollection::create, "PhysDeriv_NoCollection_Prism"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eNodalPrism, ePhysDeriv, eNoCollection, true),
        PhysDeriv_NoCollection::create, "PhysDeriv_NoCollection_NodalPrism"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eHexahedron, ePhysDeriv, eNoCollection, false),
        PhysDeriv_NoCollection::create, "PhysDeriv_NoCollection_Hex")};
 
/**
 * @brief Phys deriv operator using sum-factorisation (Segment)
 */
class PhysDeriv_SumFac_Seg final : virtual public Operator,
                                   virtual public PhysDeriv_Helper
{
public:
    OPERATOR_CREATE(PhysDeriv_SumFac_Seg)
 
    ~PhysDeriv_SumFac_Seg() final = default;
 
    void operator()(const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output0,
                    Array<OneD, NekDouble> &output1,
                    Array<OneD, NekDouble> &output2,
                    Array<OneD, NekDouble> &wsp) final
    {
        const int nqcol = m_nquad0 * m_numElmt;
 
        ASSERTL1(wsp.size() == m_wspSize, "Incorrect workspace size");
        ASSERTL1(input.size() >= nqcol, "Incorrect input size");
 
        Array<OneD, NekDouble> diff0(nqcol, wsp);
 
        Blas::Dgemm('N', 'N', m_nquad0, m_numElmt, m_nquad0, 1.0, m_Deriv0,
                    m_nquad0, input.data(), m_nquad0, 0.0, diff0.data(),
                    m_nquad0);
 
        if (m_isDeformed)
        {
            Vmath::Vmul(nqcol, m_derivFac[0], 1, diff0, 1, output0, 1);
 
            if (m_coordim == 2)
            {
                Vmath::Vmul(nqcol, m_derivFac[1], 1, diff0, 1, output1, 1);
            }
            else if (m_coordim == 3)
            {
                Vmath::Vmul(nqcol, m_derivFac[1], 1, diff0, 1, output1, 1);
                Vmath::Vmul(nqcol, m_derivFac[2], 1, diff0, 1, output2, 1);
            }
        }
        else
        {
            Array<OneD, NekDouble> t;
            for (int e = 0; e < m_numElmt; ++e)
            {
                Vmath::Smul(m_nqe, m_derivFac[0][e], diff0 + e * m_nqe, 1,
                            t = output0 + e * m_nqe, 1);
            }
 
            if (m_coordim == 2)
            {
                for (int e = 0; e < m_numElmt; ++e)
                {
                    Vmath::Smul(m_nqe, m_derivFac[1][e], diff0 + e * m_nqe, 1,
                                t = output1 + e * m_nqe, 1);
                }
            }
            else if (m_coordim == 3)
            {
                for (int e = 0; e < m_numElmt; ++e)
                {
                    Vmath::Smul(m_nqe, m_derivFac[1][e], diff0 + e * m_nqe, 1,
                                t = output1 + e * m_nqe, 1);
                    Vmath::Smul(m_nqe, m_derivFac[2][e], diff0 + e * m_nqe, 1,
                                t = output2 + e * m_nqe, 1);
                }
            }
        }
    }
 
    void operator()(int dir, const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output,
                    Array<OneD, NekDouble> &wsp) final
    {
        const int nqcol = m_nquad0 * m_numElmt;
 
        ASSERTL1(wsp.size() == m_wspSize, "Incorrect workspace size");
        ASSERTL1(input.size() >= nqcol, "Incorrect input size");
 
        Array<OneD, NekDouble> diff0(nqcol, wsp);
 
        Blas::Dgemm('N', 'N', m_nquad0, m_numElmt, m_nquad0, 1.0, m_Deriv0,
                    m_nquad0, input.data(), m_nquad0, 0.0, diff0.data(),
                    m_nquad0);
 
        if (m_isDeformed)
        {
            Vmath::Vmul(nqcol, m_derivFac[dir], 1, diff0, 1, output, 1);
        }
        else
        {
            Array<OneD, NekDouble> t;
            for (int e = 0; e < m_numElmt; ++e)
            {
                Vmath::Smul(m_nqe, m_derivFac[0][e], diff0 + e * m_nqe, 1,
                            t = output + e * m_nqe, 1);
            }
        }
    }
 
protected:
    int m_coordim;
    const int m_nquad0;
    Array<TwoD, const NekDouble> m_derivFac;
    NekDouble *m_Deriv0;
 
private:
    PhysDeriv_SumFac_Seg(vector<LocalRegions::ExpansionSharedPtr> pCollExp,
                         CoalescedGeomDataSharedPtr pGeomData,
                         StdRegions::FactorMap factors)
        : Operator(pCollExp, pGeomData, factors), PhysDeriv_Helper(),
          m_nquad0(m_stdExp->GetNumPoints(0))
    {
        m_coordim = pCollExp[0]->GetCoordim();
 
        m_derivFac = pGeomData->GetDerivFactors(pCollExp);
 
        m_Deriv0  = &((m_stdExp->GetBasis(0)->GetD())->GetPtr())[0];
        m_wspSize = m_nquad0 * m_numElmt;
    }
};
 
/// Factory initialisation for the PhysDeriv_SumFac_Seg operators
OperatorKey PhysDeriv_SumFac_Seg::m_type =
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eSegment, ePhysDeriv, eSumFac, false),
        PhysDeriv_SumFac_Seg::create, "PhysDeriv_SumFac_Seg");
 
/**
 * @brief Phys deriv operator using sum-factorisation (Quad)
 */
class PhysDeriv_SumFac_Quad final : virtual public Operator,
                                    virtual public PhysDeriv_Helper
{
public:
    OPERATOR_CREATE(PhysDeriv_SumFac_Quad)
 
    ~PhysDeriv_SumFac_Quad() final = default;
 
    void operator()(const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output0,
                    Array<OneD, NekDouble> &output1,
                    Array<OneD, NekDouble> &output2,
                    Array<OneD, NekDouble> &wsp) final
    {
        const int nqtot = m_nquad0 * m_nquad1;
        const int nqcol = nqtot * m_numElmt;
 
        ASSERTL1(wsp.size() == m_wspSize, "Incorrect workspace size");
        ASSERTL1(input.size() >= nqcol, "Incorrect input size");
 
        Array<OneD, NekDouble> diff0(nqcol, wsp);
        Array<OneD, NekDouble> diff1(nqcol, wsp + nqcol);
 
        Blas::Dgemm('N', 'N', m_nquad0, m_nquad1 * m_numElmt, m_nquad0, 1.0,
                    m_Deriv0, m_nquad0, input.data(), m_nquad0, 0.0,
                    diff0.data(), m_nquad0);
 
        int cnt = 0;
        for (int i = 0; i < m_numElmt; ++i, cnt += nqtot)
        {
            Blas::Dgemm('N', 'T', m_nquad0, m_nquad1, m_nquad1, 1.0,
                        input.data() + cnt, m_nquad0, m_Deriv1, m_nquad1, 0.0,
                        diff1.data() + cnt, m_nquad0);
        }
 
        if (m_isDeformed)
        {
            Vmath::Vmul(nqcol, m_derivFac[0], 1, diff0, 1, output0, 1);
            Vmath::Vvtvp(nqcol, m_derivFac[1], 1, diff1, 1, output0, 1, output0,
                         1);
            Vmath::Vmul(nqcol, m_derivFac[2], 1, diff0, 1, output1, 1);
            Vmath::Vvtvp(nqcol, m_derivFac[3], 1, diff1, 1, output1, 1, output1,
                         1);
 
            if (m_coordim == 3)
            {
                Vmath::Vmul(nqcol, m_derivFac[4], 1, diff0, 1, output2, 1);
                Vmath::Vvtvp(nqcol, m_derivFac[5], 1, diff1, 1, output2, 1,
                             output2, 1);
            }
        }
        else
        {
            Array<OneD, NekDouble> t;
            for (int e = 0; e < m_numElmt; ++e)
            {
                Vmath::Smul(m_nqe, m_derivFac[0][e], diff0 + e * m_nqe, 1,
                            t = output0 + e * m_nqe, 1);
                Vmath::Svtvp(m_nqe, m_derivFac[1][e], diff1 + e * m_nqe, 1,
                             output0 + e * m_nqe, 1, t = output0 + e * m_nqe,
                             1);
 
                Vmath::Smul(m_nqe, m_derivFac[2][e], diff0 + e * m_nqe, 1,
                            t = output1 + e * m_nqe, 1);
                Vmath::Svtvp(m_nqe, m_derivFac[3][e], diff1 + e * m_nqe, 1,
                             output1 + e * m_nqe, 1, t = output1 + e * m_nqe,
                             1);
            }
 
            if (m_coordim == 3)
            {
                for (int e = 0; e < m_numElmt; ++e)
                {
                    Vmath::Smul(m_nqe, m_derivFac[4][e], diff0 + e * m_nqe, 1,
                                t = output2 + e * m_nqe, 1);
                    Vmath::Svtvp(m_nqe, m_derivFac[5][e], diff1 + e * m_nqe, 1,
                                 output2 + e * m_nqe, 1,
                                 t = output2 + e * m_nqe, 1);
                }
            }
        }
    }
 
    void operator()(int dir, const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output,
                    Array<OneD, NekDouble> &wsp) final
    {
        const int nqtot = m_nquad0 * m_nquad1;
        const int nqcol = nqtot * m_numElmt;
 
        ASSERTL1(wsp.size() == m_wspSize, "Incorrect workspace size");
        ASSERTL1(input.size() >= nqcol, "Incorrect input size");
 
        Array<OneD, NekDouble> diff0(nqcol, wsp);
        Array<OneD, NekDouble> diff1(nqcol, wsp + nqcol);
 
        Blas::Dgemm('N', 'N', m_nquad0, m_nquad1 * m_numElmt, m_nquad0, 1.0,
                    m_Deriv0, m_nquad0, input.data(), m_nquad0, 0.0,
                    diff0.data(), m_nquad0);
 
        int cnt = 0;
        for (int i = 0; i < m_numElmt; ++i, cnt += nqtot)
        {
            Blas::Dgemm('N', 'T', m_nquad0, m_nquad1, m_nquad1, 1.0,
                        input.data() + cnt, m_nquad0, m_Deriv1, m_nquad1, 0.0,
                        diff1.data() + cnt, m_nquad0);
        }
 
        if (m_isDeformed)
        {
            Vmath::Vmul(nqcol, m_derivFac[2 * dir], 1, diff0, 1, output, 1);
            Vmath::Vvtvp(nqcol, m_derivFac[2 * dir + 1], 1, diff1, 1, output, 1,
                         output, 1);
        }
        else
        {
            Array<OneD, NekDouble> t;
            for (int e = 0; e < m_numElmt; ++e)
            {
                Vmath::Smul(m_nqe, m_derivFac[2 * dir][e], diff0 + e * m_nqe, 1,
                            t = output + e * m_nqe, 1);
                Vmath::Svtvp(m_nqe, m_derivFac[2 * dir + 1][e],
                             diff1 + e * m_nqe, 1, output + e * m_nqe, 1,
                             t = output + e * m_nqe, 1);
            }
        }
    }
 
protected:
    int m_coordim;
    const int m_nquad0;
    const int m_nquad1;
    Array<TwoD, const NekDouble> m_derivFac;
    NekDouble *m_Deriv0;
    NekDouble *m_Deriv1;
 
private:
    PhysDeriv_SumFac_Quad(vector<LocalRegions::ExpansionSharedPtr> pCollExp,
                          CoalescedGeomDataSharedPtr pGeomData,
                          StdRegions::FactorMap factors)
        : Operator(pCollExp, pGeomData, factors), PhysDeriv_Helper(),
          m_nquad0(m_stdExp->GetNumPoints(0)),
          m_nquad1(m_stdExp->GetNumPoints(1))
    {
        m_coordim = pCollExp[0]->GetCoordim();
 
        m_derivFac = pGeomData->GetDerivFactors(pCollExp);
 
        m_Deriv0  = &((m_stdExp->GetBasis(0)->GetD())->GetPtr())[0];
        m_Deriv1  = &((m_stdExp->GetBasis(1)->GetD())->GetPtr())[0];
        m_wspSize = 2 * m_nquad0 * m_nquad1 * m_numElmt;
    }
};
 
/// Factory initialisation for the PhysDeriv_SumFac_Quad operators
OperatorKey PhysDeriv_SumFac_Quad::m_type =
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eQuadrilateral, ePhysDeriv, eSumFac, false),
        PhysDeriv_SumFac_Quad::create, "PhysDeriv_SumFac_Quad");
 
/**
 * @brief Phys deriv operator using sum-factorisation (Tri)
 */
class PhysDeriv_SumFac_Tri final : virtual public Operator,
                                   virtual public PhysDeriv_Helper
{
public:
    OPERATOR_CREATE(PhysDeriv_SumFac_Tri)
 
    ~PhysDeriv_SumFac_Tri() final = default;
 
    void operator()(const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output0,
                    Array<OneD, NekDouble> &output1,
                    Array<OneD, NekDouble> &output2,
                    Array<OneD, NekDouble> &wsp) final
    {
        const int nqtot = m_nquad0 * m_nquad1;
        const int nqcol = nqtot * m_numElmt;
 
        ASSERTL1(wsp.size() == m_wspSize, "Incorrect workspace size");
        ASSERTL1(input.size() >= nqcol, "Incorrect input size");
 
        Array<OneD, NekDouble> diff0(nqcol, wsp);
        Array<OneD, NekDouble> diff1(nqcol, wsp + nqcol);
 
        // Tensor Product Derivative
        Blas::Dgemm('N', 'N', m_nquad0, m_nquad1 * m_numElmt, m_nquad0, 1.0,
                    m_Deriv0, m_nquad0, input.data(), m_nquad0, 0.0,
                    diff0.data(), m_nquad0);
 
        int cnt = 0;
        for (int i = 0; i < m_numElmt; ++i, cnt += nqtot)
        {
            // scale diff0 by geometric factor: 2/(1-z1)
            Vmath::Vmul(nqtot, &m_fac1[0], 1, diff0.data() + cnt, 1,
                        diff0.data() + cnt, 1);
 
            Blas::Dgemm('N', 'T', m_nquad0, m_nquad1, m_nquad1, 1.0,
                        input.data() + cnt, m_nquad0, m_Deriv1, m_nquad1, 0.0,
                        diff1.data() + cnt, m_nquad0);
 
            // add to diff1 by diff0 scaled by: (1_z0)/(1-z1)
            Vmath::Vvtvp(nqtot, m_fac0.data(), 1, diff0.data() + cnt, 1,
                         diff1.data() + cnt, 1, diff1.data() + cnt, 1);
        }
 
        if (m_isDeformed)
        {
            Vmath::Vmul(nqcol, m_derivFac[0], 1, diff0, 1, output0, 1);
            Vmath::Vvtvp(nqcol, m_derivFac[1], 1, diff1, 1, output0, 1, output0,
                         1);
            Vmath::Vmul(nqcol, m_derivFac[2], 1, diff0, 1, output1, 1);
            Vmath::Vvtvp(nqcol, m_derivFac[3], 1, diff1, 1, output1, 1, output1,
                         1);
 
            if (m_coordim == 3)
            {
                Vmath::Vmul(nqcol, m_derivFac[4], 1, diff0, 1, output2, 1);
                Vmath::Vvtvp(nqcol, m_derivFac[5], 1, diff1, 1, output2, 1,
                             output2, 1);
            }
        }
        else
        {
            Array<OneD, NekDouble> t;
            for (int e = 0; e < m_numElmt; ++e)
            {
                Vmath::Smul(m_nqe, m_derivFac[0][e], diff0 + e * m_nqe, 1,
                            t = output0 + e * m_nqe, 1);
                Vmath::Svtvp(m_nqe, m_derivFac[1][e], diff1 + e * m_nqe, 1,
                             output0 + e * m_nqe, 1, t = output0 + e * m_nqe,
                             1);
 
                Vmath::Smul(m_nqe, m_derivFac[2][e], diff0 + e * m_nqe, 1,
                            t = output1 + e * m_nqe, 1);
                Vmath::Svtvp(m_nqe, m_derivFac[3][e], diff1 + e * m_nqe, 1,
                             output1 + e * m_nqe, 1, t = output1 + e * m_nqe,
                             1);
            }
 
            if (m_coordim == 3)
            {
                for (int e = 0; e < m_numElmt; ++e)
                {
                    Vmath::Smul(m_nqe, m_derivFac[4][e], diff0 + e * m_nqe, 1,
                                t = output2 + e * m_nqe, 1);
                    Vmath::Svtvp(m_nqe, m_derivFac[5][e], diff1 + e * m_nqe, 1,
                                 output2 + e * m_nqe, 1,
                                 t = output2 + e * m_nqe, 1);
                }
            }
        }
    }
 
    void operator()(int dir, const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output,
                    Array<OneD, NekDouble> &wsp) final
    {
        const int nqtot = m_nquad0 * m_nquad1;
        const int nqcol = nqtot * m_numElmt;
 
        ASSERTL1(wsp.size() == m_wspSize, "Incorrect workspace size");
        ASSERTL1(input.size() >= nqcol, "Incorrect input size");
 
        Array<OneD, NekDouble> diff0(nqcol, wsp);
        Array<OneD, NekDouble> diff1(nqcol, wsp + nqcol);
 
        // Tensor Product Derivative
        Blas::Dgemm('N', 'N', m_nquad0, m_nquad1 * m_numElmt, m_nquad0, 1.0,
                    m_Deriv0, m_nquad0, input.data(), m_nquad0, 0.0,
                    diff0.data(), m_nquad0);
 
        int cnt = 0;
        for (int i = 0; i < m_numElmt; ++i, cnt += nqtot)
        {
            // scale diff0 by geometric factor: 2/(1-z1)
            Vmath::Vmul(nqtot, &m_fac1[0], 1, diff0.data() + cnt, 1,
                        diff0.data() + cnt, 1);
 
            Blas::Dgemm('N', 'T', m_nquad0, m_nquad1, m_nquad1, 1.0,
                        input.data() + cnt, m_nquad0, m_Deriv1, m_nquad1, 0.0,
                        diff1.data() + cnt, m_nquad0);
 
            // add to diff1 by diff0 scaled by: (1_z0)/(1-z1)
            Vmath::Vvtvp(nqtot, m_fac0.data(), 1, diff0.data() + cnt, 1,
                         diff1.data() + cnt, 1, diff1.data() + cnt, 1);
        }
 
        if (m_isDeformed)
        {
            Vmath::Vmul(nqcol, m_derivFac[2 * dir], 1, diff0, 1, output, 1);
            Vmath::Vvtvp(nqcol, m_derivFac[2 * dir + 1], 1, diff1, 1, output, 1,
                         output, 1);
        }
        else
        {
            Array<OneD, NekDouble> t;
            for (int e = 0; e < m_numElmt; ++e)
            {
                Vmath::Smul(m_nqe, m_derivFac[2 * dir][e], diff0 + e * m_nqe, 1,
                            t = output + e * m_nqe, 1);
                Vmath::Svtvp(m_nqe, m_derivFac[2 * dir + 1][e],
                             diff1 + e * m_nqe, 1, output + e * m_nqe, 1,
                             t = output + e * m_nqe, 1);
            }
        }
    }
 
protected:
    int m_coordim;
    const int m_nquad0;
    const int m_nquad1;
    Array<TwoD, const NekDouble> m_derivFac;
    NekDouble *m_Deriv0;
    NekDouble *m_Deriv1;
    Array<OneD, NekDouble> m_fac0;
    Array<OneD, NekDouble> m_fac1;
 
private:
    PhysDeriv_SumFac_Tri(vector<LocalRegions::ExpansionSharedPtr> pCollExp,
                         CoalescedGeomDataSharedPtr pGeomData,
                         StdRegions::FactorMap factors)
        : Operator(pCollExp, pGeomData, factors), PhysDeriv_Helper(),
          m_nquad0(m_stdExp->GetNumPoints(0)),
          m_nquad1(m_stdExp->GetNumPoints(1))
    {
        m_coordim = pCollExp[0]->GetCoordim();
 
        m_derivFac = pGeomData->GetDerivFactors(pCollExp);
 
        const Array<OneD, const NekDouble> &z0 = m_stdExp->GetBasis(0)->GetZ();
        const Array<OneD, const NekDouble> &z1 = m_stdExp->GetBasis(1)->GetZ();
        m_fac0 = Array<OneD, NekDouble>(m_nquad0 * m_nquad1);
        // set up geometric factor: 0.5*(1+z0)
        for (int i = 0; i < m_nquad0; ++i)
        {
            for (int j = 0; j < m_nquad1; ++j)
            {
                m_fac0[i + j * m_nquad0] = 0.5 * (1 + z0[i]);
            }
        }
 
        m_fac1 = Array<OneD, NekDouble>(m_nquad0 * m_nquad1);
        // set up geometric factor: 2/(1-z1)
        for (int i = 0; i < m_nquad0; ++i)
        {
            for (int j = 0; j < m_nquad1; ++j)
            {
                m_fac1[i + j * m_nquad0] = 2.0 / (1 - z1[j]);
            }
        }
 
        m_Deriv0  = &((m_stdExp->GetBasis(0)->GetD())->GetPtr())[0];
        m_Deriv1  = &((m_stdExp->GetBasis(1)->GetD())->GetPtr())[0];
        m_wspSize = 2 * m_nquad0 * m_nquad1 * m_numElmt;
    }
};
 
/// Factory initialisation for the PhysDeriv_SumFac_Tri operators
OperatorKey PhysDeriv_SumFac_Tri::m_typeArr[] = {
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eTriangle, ePhysDeriv, eSumFac, false),
        PhysDeriv_SumFac_Tri::create, "PhysDeriv_SumFac_Tri"),
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eNodalTri, ePhysDeriv, eSumFac, true),
        PhysDeriv_SumFac_Tri::create, "PhysDeriv_SumFac_NodalTri")};
 
/**
 * @brief Phys deriv operator using sum-factorisation (Hex)
 */
class PhysDeriv_SumFac_Hex final : virtual public Operator,
                                   virtual public PhysDeriv_Helper
{
public:
    OPERATOR_CREATE(PhysDeriv_SumFac_Hex)
 
    ~PhysDeriv_SumFac_Hex() final = default;
 
    void operator()(const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output0,
                    Array<OneD, NekDouble> &output1,
                    Array<OneD, NekDouble> &output2,
                    Array<OneD, NekDouble> &wsp) final
    {
        int nPhys = m_stdExp->GetTotPoints();
        int ntot  = m_numElmt * nPhys;
        Array<OneD, NekDouble> tmp0, tmp1, tmp2;
        Array<OneD, Array<OneD, NekDouble>> Diff(3);
        Array<OneD, Array<OneD, NekDouble>> out(3);
        out[0] = output0;
        out[1] = output1;
        out[2] = output2;
 
        for (int i = 0; i < 3; ++i)
        {
            Diff[i] = wsp + i * ntot;
        }
 
        Blas::Dgemm('N', 'N', m_nquad0, m_nquad1 * m_nquad2 * m_numElmt,
                    m_nquad0, 1.0, m_Deriv0, m_nquad0, &input[0], m_nquad0, 0.0,
                    &Diff[0][0], m_nquad0);
 
        for (int i = 0; i < m_numElmt; ++i)
        {
            for (int j = 0; j < m_nquad2; ++j)
            {
                Blas::Dgemm('N', 'T', m_nquad0, m_nquad1, m_nquad1, 1.0,
                            &input[i * nPhys + j * m_nquad0 * m_nquad1],
                            m_nquad0, m_Deriv1, m_nquad1, 0.0,
                            &Diff[1][i * nPhys + j * m_nquad0 * m_nquad1],
                            m_nquad0);
            }
 
            Blas::Dgemm('N', 'T', m_nquad0 * m_nquad1, m_nquad2, m_nquad2, 1.0,
                        &input[i * nPhys], m_nquad0 * m_nquad1, m_Deriv2,
                        m_nquad2, 0.0, &Diff[2][i * nPhys],
                        m_nquad0 * m_nquad1);
        }
 
        // calculate full derivative
        if (m_isDeformed)
        {
            for (int i = 0; i < m_coordim; ++i)
            {
                Vmath::Vmul(ntot, m_derivFac[i * 3], 1, Diff[0], 1, out[i], 1);
                for (int j = 1; j < 3; ++j)
                {
                    Vmath::Vvtvp(ntot, m_derivFac[i * 3 + j], 1, Diff[j], 1,
                                 out[i], 1, out[i], 1);
                }
            }
        }
        else
        {
            Array<OneD, NekDouble> t;
            for (int e = 0; e < m_numElmt; ++e)
            {
                for (int i = 0; i < m_coordim; ++i)
                {
                    Vmath::Smul(m_nqe, m_derivFac[i * 3][e],
                                Diff[0] + e * m_nqe, 1, t = out[i] + e * m_nqe,
                                1);
 
                    for (int j = 1; j < 3; ++j)
                    {
                        Vmath::Svtvp(m_nqe, m_derivFac[i * 3 + j][e],
                                     Diff[j] + e * m_nqe, 1, out[i] + e * m_nqe,
                                     1, t = out[i] + e * m_nqe, 1);
                    }
                }
            }
        }
    }
 
    void operator()(int dir, const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output,
                    Array<OneD, NekDouble> &wsp) final
    {
        int nPhys = m_stdExp->GetTotPoints();
        int ntot  = m_numElmt * nPhys;
        Array<OneD, NekDouble> tmp0, tmp1, tmp2;
        Array<OneD, Array<OneD, NekDouble>> Diff(3);
 
        for (int i = 0; i < 3; ++i)
        {
            Diff[i] = wsp + i * ntot;
        }
 
        Blas::Dgemm('N', 'N', m_nquad0, m_nquad1 * m_nquad2 * m_numElmt,
                    m_nquad0, 1.0, m_Deriv0, m_nquad0, &input[0], m_nquad0, 0.0,
                    &Diff[0][0], m_nquad0);
 
        for (int i = 0; i < m_numElmt; ++i)
        {
            for (int j = 0; j < m_nquad2; ++j)
            {
                Blas::Dgemm('N', 'T', m_nquad0, m_nquad1, m_nquad1, 1.0,
                            &input[i * nPhys + j * m_nquad0 * m_nquad1],
                            m_nquad0, m_Deriv1, m_nquad1, 0.0,
                            &Diff[1][i * nPhys + j * m_nquad0 * m_nquad1],
                            m_nquad0);
            }
 
            Blas::Dgemm('N', 'T', m_nquad0 * m_nquad1, m_nquad2, m_nquad2, 1.0,
                        &input[i * nPhys], m_nquad0 * m_nquad1, m_Deriv2,
                        m_nquad2, 0.0, &Diff[2][i * nPhys],
                        m_nquad0 * m_nquad1);
        }
 
        // calculate full derivative
        if (m_isDeformed)
        {
            // calculate full derivative
            Vmath::Vmul(ntot, m_derivFac[dir * 3], 1, Diff[0], 1, output, 1);
            for (int j = 1; j < 3; ++j)
            {
                Vmath::Vvtvp(ntot, m_derivFac[dir * 3 + j], 1, Diff[j], 1,
                             output, 1, output, 1);
            }
        }
        else
        {
            Array<OneD, NekDouble> t;
            for (int e = 0; e < m_numElmt; ++e)
            {
                Vmath::Smul(m_nqe, m_derivFac[dir * 3][e], Diff[0] + e * m_nqe,
                            1, t = output + e * m_nqe, 1);
 
                for (int j = 1; j < 3; ++j)
                {
                    Vmath::Svtvp(m_nqe, m_derivFac[dir * 3 + j][e],
                                 Diff[j] + e * m_nqe, 1, output + e * m_nqe, 1,
                                 t = output + e * m_nqe, 1);
                }
            }
        }
    }
 
protected:
    Array<TwoD, const NekDouble> m_derivFac;
    int m_coordim;
    const int m_nquad0;
    const int m_nquad1;
    const int m_nquad2;
    NekDouble *m_Deriv0;
    NekDouble *m_Deriv1;
    NekDouble *m_Deriv2;
 
private:
    PhysDeriv_SumFac_Hex(vector<LocalRegions::ExpansionSharedPtr> pCollExp,
                         CoalescedGeomDataSharedPtr pGeomData,
                         StdRegions::FactorMap factors)
        : Operator(pCollExp, pGeomData, factors), PhysDeriv_Helper(),
          m_nquad0(m_stdExp->GetNumPoints(0)),
          m_nquad1(m_stdExp->GetNumPoints(1)),
          m_nquad2(m_stdExp->GetNumPoints(2))
    {
        m_coordim = pCollExp[0]->GetCoordim();
 
        m_derivFac = pGeomData->GetDerivFactors(pCollExp);
 
        m_Deriv0 = &((m_stdExp->GetBasis(0)->GetD())->GetPtr())[0];
        m_Deriv1 = &((m_stdExp->GetBasis(1)->GetD())->GetPtr())[0];
        m_Deriv2 = &((m_stdExp->GetBasis(2)->GetD())->GetPtr())[0];
 
        m_wspSize = 3 * m_nquad0 * m_nquad1 * m_nquad2 * m_numElmt;
    }
};
 
/// Factory initialisation for the PhysDeriv_SumFac_Hex operators
OperatorKey PhysDeriv_SumFac_Hex::m_typeArr[] = {
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eHexahedron, ePhysDeriv, eSumFac, false),
        PhysDeriv_SumFac_Hex::create, "PhysDeriv_SumFac_Hex")};
 
/**
 * @brief Phys deriv operator using sum-factorisation (Tet)
 */
class PhysDeriv_SumFac_Tet final : virtual public Operator,
                                   virtual public PhysDeriv_Helper
{
public:
    OPERATOR_CREATE(PhysDeriv_SumFac_Tet)
 
    ~PhysDeriv_SumFac_Tet() final = default;
 
    void operator()(const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output0,
                    Array<OneD, NekDouble> &output1,
                    Array<OneD, NekDouble> &output2,
                    Array<OneD, NekDouble> &wsp) final
    {
        int nPhys = m_stdExp->GetTotPoints();
        int ntot  = m_numElmt * nPhys;
        Array<OneD, NekDouble> tmp0, tmp1, tmp2;
        Array<OneD, Array<OneD, NekDouble>> Diff(3);
        Array<OneD, Array<OneD, NekDouble>> out(3);
        out[0] = output0;
        out[1] = output1;
        out[2] = output2;
 
        for (int i = 0; i < 3; ++i)
        {
            Diff[i] = wsp + i * ntot;
        }
 
        // dEta0
        Blas::Dgemm('N', 'N', m_nquad0, m_nquad1 * m_nquad2 * m_numElmt,
                    m_nquad0, 1.0, m_Deriv0, m_nquad0, &input[0], m_nquad0, 0.0,
                    &Diff[0][0], m_nquad0);
 
        // dEta2
        for (int i = 0; i < m_numElmt; ++i)
        {
            Blas::Dgemm('N', 'T', m_nquad0 * m_nquad1, m_nquad2, m_nquad2, 1.0,
                        &input[i * nPhys], m_nquad0 * m_nquad1, m_Deriv2,
                        m_nquad2, 0.0, &Diff[2][i * nPhys],
                        m_nquad0 * m_nquad1);
        }
 
        for (int i = 0; i < m_numElmt; ++i)
        {
            // dEta1
            for (int j = 0; j < m_nquad2; ++j)
            {
                Blas::Dgemm('N', 'T', m_nquad0, m_nquad1, m_nquad1, 1.0,
                            &input[i * nPhys + j * m_nquad0 * m_nquad1],
                            m_nquad0, m_Deriv1, m_nquad1, 0.0,
                            &Diff[1][i * nPhys + j * m_nquad0 * m_nquad1],
                            m_nquad0);
            }
 
            // dxi2 = (1 + eta_1)/(1 -eta_2)*dEta1 + dEta2
            Vmath::Vvtvp(nPhys, m_fac3.data(), 1, Diff[1].data() + i * nPhys, 1,
                         Diff[2].data() + i * nPhys, 1,
                         Diff[2].data() + i * nPhys, 1);
 
            // dxi1 =  2/(1 - eta_2) dEta1
            Vmath::Vmul(nPhys, m_fac2.data(), 1, Diff[1].data() + i * nPhys, 1,
                        Diff[1].data() + i * nPhys, 1);
 
            // dxi1 = 2.0(1+eta_0)/((1-eta_1)(1-eta_2)) dEta0 + dxi1
            Vmath::Vvtvp(nPhys, m_fac1.data(), 1, Diff[0].data() + i * nPhys, 1,
                         Diff[1].data() + i * nPhys, 1,
                         Diff[1].data() + i * nPhys, 1);
 
            // dxi2 = 2.0(1+eta_0)/((1-eta_1)(1-eta_2)) dEta0 + dxi2
            Vmath::Vvtvp(nPhys, m_fac1.data(), 1, Diff[0].data() + i * nPhys, 1,
                         Diff[2].data() + i * nPhys, 1,
                         Diff[2].data() + i * nPhys, 1);
 
            // dxi0 = 4.0/((1-eta_1)(1-eta_2)) dEta0
            Vmath::Vmul(nPhys, m_fac0.data(), 1, Diff[0].data() + i * nPhys, 1,
                        Diff[0].data() + i * nPhys, 1);
        }
 
        // calculate full derivative
        if (m_isDeformed)
        {
            for (int i = 0; i < m_coordim; ++i)
            {
                Vmath::Vmul(ntot, m_derivFac[i * 3], 1, Diff[0], 1, out[i], 1);
                for (int j = 1; j < 3; ++j)
                {
                    Vmath::Vvtvp(ntot, m_derivFac[i * 3 + j], 1, Diff[j], 1,
                                 out[i], 1, out[i], 1);
                }
            }
        }
        else
        {
            Array<OneD, NekDouble> t;
            for (int e = 0; e < m_numElmt; ++e)
            {
                for (int i = 0; i < m_coordim; ++i)
                {
                    Vmath::Smul(m_nqe, m_derivFac[i * 3][e],
                                Diff[0] + e * m_nqe, 1, t = out[i] + e * m_nqe,
                                1);
                    for (int j = 1; j < 3; ++j)
                    {
                        Vmath::Svtvp(m_nqe, m_derivFac[i * 3 + j][e],
                                     Diff[j] + e * m_nqe, 1, out[i] + e * m_nqe,
                                     1, t = out[i] + e * m_nqe, 1);
                    }
                }
            }
        }
    }
 
    void operator()(int dir, const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output,
                    Array<OneD, NekDouble> &wsp) final
    {
        int nPhys = m_stdExp->GetTotPoints();
        int ntot  = m_numElmt * nPhys;
        Array<OneD, NekDouble> tmp0, tmp1, tmp2;
        Array<OneD, Array<OneD, NekDouble>> Diff(3);
 
        for (int i = 0; i < 3; ++i)
        {
            Diff[i] = wsp + i * ntot;
        }
 
        // dEta0
        Blas::Dgemm('N', 'N', m_nquad0, m_nquad1 * m_nquad2 * m_numElmt,
                    m_nquad0, 1.0, m_Deriv0, m_nquad0, &input[0], m_nquad0, 0.0,
                    &Diff[0][0], m_nquad0);
 
        // dEta2
        for (int i = 0; i < m_numElmt; ++i)
        {
            Blas::Dgemm('N', 'T', m_nquad0 * m_nquad1, m_nquad2, m_nquad2, 1.0,
                        &input[i * nPhys], m_nquad0 * m_nquad1, m_Deriv2,
                        m_nquad2, 0.0, &Diff[2][i * nPhys],
                        m_nquad0 * m_nquad1);
        }
 
        for (int i = 0; i < m_numElmt; ++i)
        {
            // dEta1
            for (int j = 0; j < m_nquad2; ++j)
            {
                Blas::Dgemm('N', 'T', m_nquad0, m_nquad1, m_nquad1, 1.0,
                            &input[i * nPhys + j * m_nquad0 * m_nquad1],
                            m_nquad0, m_Deriv1, m_nquad1, 0.0,
                            &Diff[1][i * nPhys + j * m_nquad0 * m_nquad1],
                            m_nquad0);
            }
 
            // dxi2 = (1 + eta_1)/(1 -eta_2)*dEta1 + dEta2
            Vmath::Vvtvp(nPhys, m_fac3.data(), 1, Diff[1].data() + i * nPhys, 1,
                         Diff[2].data() + i * nPhys, 1,
                         Diff[2].data() + i * nPhys, 1);
 
            // dxi1 =  2/(1 - eta_2) dEta1
            Vmath::Vmul(nPhys, m_fac2.data(), 1, Diff[1].data() + i * nPhys, 1,
                        Diff[1].data() + i * nPhys, 1);
 
            // dxi1 = 2.0(1+eta_0)/((1-eta_1)(1-eta_2)) dEta0 + dxi1
            Vmath::Vvtvp(nPhys, m_fac1.data(), 1, Diff[0].data() + i * nPhys, 1,
                         Diff[1].data() + i * nPhys, 1,
                         Diff[1].data() + i * nPhys, 1);
 
            // dxi2 = 2.0(1+eta_0)/((1-eta_1)(1-eta_2)) dEta0 + dxi2
            Vmath::Vvtvp(nPhys, m_fac1.data(), 1, Diff[0].data() + i * nPhys, 1,
                         Diff[2].data() + i * nPhys, 1,
                         Diff[2].data() + i * nPhys, 1);
 
            // dxi0 = 4.0/((1-eta_1)(1-eta_2)) dEta0
            Vmath::Vmul(nPhys, m_fac0.data(), 1, Diff[0].data() + i * nPhys, 1,
                        Diff[0].data() + i * nPhys, 1);
        }
 
        // calculate full derivative
        if (m_isDeformed)
        {
            // calculate full derivative
            Vmath::Vmul(ntot, m_derivFac[dir * 3], 1, Diff[0], 1, output, 1);
            for (int j = 1; j < 3; ++j)
            {
                Vmath::Vvtvp(ntot, m_derivFac[dir * 3 + j], 1, Diff[j], 1,
                             output, 1, output, 1);
            }
        }
        else
        {
            Array<OneD, NekDouble> t;
            for (int e = 0; e < m_numElmt; ++e)
            {
                Vmath::Smul(m_nqe, m_derivFac[dir * 3][e], Diff[0] + e * m_nqe,
                            1, t = output + e * m_nqe, 1);
                for (int j = 1; j < 3; ++j)
                {
                    Vmath::Svtvp(m_nqe, m_derivFac[dir * 3 + j][e],
                                 Diff[j] + e * m_nqe, 1, output + e * m_nqe, 1,
                                 t = output + e * m_nqe, 1);
                }
            }
        }
    }
 
protected:
    Array<TwoD, const NekDouble> m_derivFac;
    int m_coordim;
    const int m_nquad0;
    const int m_nquad1;
    const int m_nquad2;
    NekDouble *m_Deriv0;
    NekDouble *m_Deriv1;
    NekDouble *m_Deriv2;
    Array<OneD, NekDouble> m_fac0;
    Array<OneD, NekDouble> m_fac1;
    Array<OneD, NekDouble> m_fac2;
    Array<OneD, NekDouble> m_fac3;
 
private:
    PhysDeriv_SumFac_Tet(vector<LocalRegions::ExpansionSharedPtr> pCollExp,
                         CoalescedGeomDataSharedPtr pGeomData,
                         StdRegions::FactorMap factors)
        : Operator(pCollExp, pGeomData, factors), PhysDeriv_Helper(),
          m_nquad0(m_stdExp->GetNumPoints(0)),
          m_nquad1(m_stdExp->GetNumPoints(1)),
          m_nquad2(m_stdExp->GetNumPoints(2))
    {
        m_coordim = pCollExp[0]->GetCoordim();
 
        m_derivFac = pGeomData->GetDerivFactors(pCollExp);
 
        m_Deriv0 = &((m_stdExp->GetBasis(0)->GetD())->GetPtr())[0];
        m_Deriv1 = &((m_stdExp->GetBasis(1)->GetD())->GetPtr())[0];
        m_Deriv2 = &((m_stdExp->GetBasis(2)->GetD())->GetPtr())[0];
 
        m_wspSize = 3 * m_nquad0 * m_nquad1 * m_nquad2 * m_numElmt;
 
        const Array<OneD, const NekDouble> &z0 = m_stdExp->GetBasis(0)->GetZ();
        const Array<OneD, const NekDouble> &z1 = m_stdExp->GetBasis(1)->GetZ();
        const Array<OneD, const NekDouble> &z2 = m_stdExp->GetBasis(2)->GetZ();
 
        m_fac0 = Array<OneD, NekDouble>(m_nquad0 * m_nquad1 * m_nquad2);
        m_fac1 = Array<OneD, NekDouble>(m_nquad0 * m_nquad1 * m_nquad2);
        m_fac2 = Array<OneD, NekDouble>(m_nquad0 * m_nquad1 * m_nquad2);
        m_fac3 = Array<OneD, NekDouble>(m_nquad0 * m_nquad1 * m_nquad2);
 
        // calculate 2.0/((1-eta_1)(1-eta_2))
        for (int i = 0; i < m_nquad0; ++i)
        {
            for (int j = 0; j < m_nquad1; ++j)
            {
                for (int k = 0; k < m_nquad2; ++k)
                {
                    m_fac0[i + j * m_nquad0 + k * m_nquad0 * m_nquad1] =
                        4.0 / ((1 - z1[j]) * (1 - z2[k]));
                    m_fac1[i + j * m_nquad0 + k * m_nquad0 * m_nquad1] =
                        2.0 * (1 + z0[i]) / ((1 - z1[j]) * (1 - z2[k]));
                    m_fac2[i + j * m_nquad0 + k * m_nquad0 * m_nquad1] =
                        2.0 / (1 - z2[k]);
                    m_fac3[i + j * m_nquad0 + k * m_nquad0 * m_nquad1] =
                        (1 + z1[j]) / (1 - z2[k]);
                }
            }
        }
    }
};
 
/// Factory initialisation for the PhysDeriv_SumFac_Tet operators
OperatorKey PhysDeriv_SumFac_Tet::m_typeArr[] = {
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(eTetrahedron, ePhysDeriv, eSumFac, false),
        PhysDeriv_SumFac_Tet::create, "PhysDeriv_SumFac_Tet")};
 
/**
 * @brief Phys deriv operator using sum-factorisation (Prism)
 */
class PhysDeriv_SumFac_Prism final : virtual public Operator,
                                     virtual public PhysDeriv_Helper
{
public:
    OPERATOR_CREATE(PhysDeriv_SumFac_Prism)
 
    ~PhysDeriv_SumFac_Prism() final = default;
 
    void operator()(const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output0,
                    Array<OneD, NekDouble> &output1,
                    Array<OneD, NekDouble> &output2,
                    Array<OneD, NekDouble> &wsp) final
    {
        int nPhys = m_stdExp->GetTotPoints();
        int ntot  = m_numElmt * nPhys;
        Array<OneD, NekDouble> tmp0, tmp1, tmp2;
        Array<OneD, Array<OneD, NekDouble>> Diff(3);
        Array<OneD, Array<OneD, NekDouble>> out(3);
        out[0] = output0;
        out[1] = output1;
        out[2] = output2;
 
        for (int i = 0; i < 3; ++i)
        {
            Diff[i] = wsp + i * ntot;
        }
 
        // dEta0
        Blas::Dgemm('N', 'N', m_nquad0, m_nquad1 * m_nquad2 * m_numElmt,
                    m_nquad0, 1.0, m_Deriv0, m_nquad0, &input[0], m_nquad0, 0.0,
                    &Diff[0][0], m_nquad0);
 
        int cnt = 0;
        for (int i = 0; i < m_numElmt; ++i)
        {
            // dEta 1
            for (int j = 0; j < m_nquad2; ++j)
            {
                Blas::Dgemm('N', 'T', m_nquad0, m_nquad1, m_nquad1, 1.0,
                            &input[i * nPhys + j * m_nquad0 * m_nquad1],
                            m_nquad0, m_Deriv1, m_nquad1, 0.0,
                            &Diff[1][i * nPhys + j * m_nquad0 * m_nquad1],
                            m_nquad0);
            }
 
            // dEta 2
            Blas::Dgemm('N', 'T', m_nquad0 * m_nquad1, m_nquad2, m_nquad2, 1.0,
                        &input[i * nPhys], m_nquad0 * m_nquad1, m_Deriv2,
                        m_nquad2, 0.0, &Diff[2][i * nPhys],
                        m_nquad0 * m_nquad1);
 
            // dxi0 = 2/(1-eta_2) d Eta_0
            Vmath::Vmul(nPhys, &m_fac0[0], 1, Diff[0].data() + cnt, 1,
                        Diff[0].data() + cnt, 1);
 
            // dxi2 = (1+eta0)/(1-eta_2) d Eta_0 + d/dEta2;
            Vmath::Vvtvp(nPhys, &m_fac1[0], 1, Diff[0].data() + cnt, 1,
                         Diff[2].data() + cnt, 1, Diff[2].data() + cnt, 1);
            cnt += nPhys;
        }
 
        // calculate full derivative
        if (m_isDeformed)
        {
            for (int i = 0; i < m_coordim; ++i)
            {
                Vmath::Vmul(ntot, m_derivFac[i * 3], 1, Diff[0], 1, out[i], 1);
                for (int j = 1; j < 3; ++j)
                {
                    Vmath::Vvtvp(ntot, m_derivFac[i * 3 + j], 1, Diff[j], 1,
                                 out[i], 1, out[i], 1);
                }
            }
        }
        else
        {
            Array<OneD, NekDouble> t;
            for (int e = 0; e < m_numElmt; ++e)
            {
                for (int i = 0; i < m_coordim; ++i)
                {
                    Vmath::Smul(m_nqe, m_derivFac[i * 3][e],
                                Diff[0] + e * m_nqe, 1, t = out[i] + e * m_nqe,
                                1);
 
                    for (int j = 1; j < 3; ++j)
                    {
                        Vmath::Svtvp(m_nqe, m_derivFac[i * 3 + j][e],
                                     Diff[j] + e * m_nqe, 1, out[i] + e * m_nqe,
                                     1, t = out[i] + e * m_nqe, 1);
                    }
                }
            }
        }
    }
 
    void operator()(int dir, const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output,
                    Array<OneD, NekDouble> &wsp) final
    {
        int nPhys = m_stdExp->GetTotPoints();
        int ntot  = m_numElmt * nPhys;
        Array<OneD, NekDouble> tmp0, tmp1, tmp2;
        Array<OneD, Array<OneD, NekDouble>> Diff(3);
 
        for (int i = 0; i < 3; ++i)
        {
            Diff[i] = wsp + i * ntot;
        }
 
        // dEta0
        Blas::Dgemm('N', 'N', m_nquad0, m_nquad1 * m_nquad2 * m_numElmt,
                    m_nquad0, 1.0, m_Deriv0, m_nquad0, &input[0], m_nquad0, 0.0,
                    &Diff[0][0], m_nquad0);
 
        int cnt = 0;
        for (int i = 0; i < m_numElmt; ++i)
        {
            // dEta 1
            for (int j = 0; j < m_nquad2; ++j)
            {
                Blas::Dgemm('N', 'T', m_nquad0, m_nquad1, m_nquad1, 1.0,
                            &input[i * nPhys + j * m_nquad0 * m_nquad1],
                            m_nquad0, m_Deriv1, m_nquad1, 0.0,
                            &Diff[1][i * nPhys + j * m_nquad0 * m_nquad1],
                            m_nquad0);
            }
 
            // dEta 2
            Blas::Dgemm('N', 'T', m_nquad0 * m_nquad1, m_nquad2, m_nquad2, 1.0,
                        &input[i * nPhys], m_nquad0 * m_nquad1, m_Deriv2,
                        m_nquad2, 0.0, &Diff[2][i * nPhys],
                        m_nquad0 * m_nquad1);
 
            // dxi0 = 2/(1-eta_2) d Eta_0
            Vmath::Vmul(nPhys, &m_fac0[0], 1, Diff[0].data() + cnt, 1,
                        Diff[0].data() + cnt, 1);
 
            // dxi2 = (1+eta0)/(1-eta_2) d Eta_0 + d/dEta2;
            Vmath::Vvtvp(nPhys, &m_fac1[0], 1, Diff[0].data() + cnt, 1,
                         Diff[2].data() + cnt, 1, Diff[2].data() + cnt, 1);
            cnt += nPhys;
        }
 
        // calculate full derivative
        if (m_isDeformed)
        {
            // calculate full derivative
            Vmath::Vmul(ntot, m_derivFac[dir * 3], 1, Diff[0], 1, output, 1);
            for (int j = 1; j < 3; ++j)
            {
                Vmath::Vvtvp(ntot, m_derivFac[dir * 3 + j], 1, Diff[j], 1,
                             output, 1, output, 1);
            }
        }
        else
        {
            Array<OneD, NekDouble> t;
            for (int e = 0; e < m_numElmt; ++e)
            {
                Vmath::Smul(m_nqe, m_derivFac[dir * 3][e], Diff[0] + e * m_nqe,
                            1, t = output + e * m_nqe, 1);
 
                for (int j = 1; j < 3; ++j)
                {
                    Vmath::Svtvp(m_nqe, m_derivFac[dir * 3 + j][e],
                                 Diff[j] + e * m_nqe, 1, output + e * m_nqe, 1,
                                 t = output + e * m_nqe, 1);
                }
            }
        }
    }
 
protected:
    Array<TwoD, const NekDouble> m_derivFac;
    int m_coordim;
    const int m_nquad0;
    const int m_nquad1;
    const int m_nquad2;
    NekDouble *m_Deriv0;
    NekDouble *m_Deriv1;
    NekDouble *m_Deriv2;
    Array<OneD, NekDouble> m_fac0;
    Array<OneD, NekDouble> m_fac1;
 
private:
    PhysDeriv_SumFac_Prism(vector<LocalRegions::ExpansionSharedPtr> pCollExp,
                           CoalescedGeomDataSharedPtr pGeomData,
                           StdRegions::FactorMap factors)
        : Operator(pCollExp, pGeomData, factors), PhysDeriv_Helper(),
          m_nquad0(m_stdExp->GetNumPoints(0)),
          m_nquad1(m_stdExp->GetNumPoints(1)),
          m_nquad2(m_stdExp->GetNumPoints(2))
    {
        m_coordim = pCollExp[0]->GetCoordim();
 
        m_derivFac = pGeomData->GetDerivFactors(pCollExp);
 
        const Array<OneD, const NekDouble> &z0 = m_stdExp->GetBasis(0)->GetZ();
        const Array<OneD, const NekDouble> &z2 = m_stdExp->GetBasis(2)->GetZ();
        m_fac0 = Array<OneD, NekDouble>(m_nquad0 * m_nquad1 * m_nquad2);
        m_fac1 = Array<OneD, NekDouble>(m_nquad0 * m_nquad1 * m_nquad2);
        for (int i = 0; i < m_nquad0; ++i)
        {
            for (int j = 0; j < m_nquad1; ++j)
            {
                for (int k = 0; k < m_nquad2; ++k)
                {
                    m_fac0[i + j * m_nquad0 + k * m_nquad0 * m_nquad1] =
                        2.0 / (1 - z2[k]);
                    m_fac1[i + j * m_nquad0 + k * m_nquad0 * m_nquad1] =
                        0.5 * (1 + z0[i]);
                }
            }
        }
 
        m_Deriv0 = &((m_stdExp->GetBasis(0)->GetD())->GetPtr())[0];
        m_Deriv1 = &((m_stdExp->GetBasis(1)->GetD())->GetPtr())[0];
        m_Deriv2 = &((m_stdExp->GetBasis(2)->GetD())->GetPtr())[0];
 
        m_wspSize = 3 * m_nquad0 * m_nquad1 * m_nquad2 * m_numElmt;
    }
};
 
/// Factory initialisation for the PhysDeriv_SumFac_Prism operators
OperatorKey PhysDeriv_SumFac_Prism::m_typeArr[] = {
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(ePrism, ePhysDeriv, eSumFac, false),
        PhysDeriv_SumFac_Prism::create, "PhysDeriv_SumFac_Prism")};
 
/**
 * @brief Phys deriv operator using sum-factorisation (Pyramid)
 */
class PhysDeriv_SumFac_Pyr final : virtual public Operator,
                                   virtual public PhysDeriv_Helper
{
public:
    OPERATOR_CREATE(PhysDeriv_SumFac_Pyr)
 
    ~PhysDeriv_SumFac_Pyr() final = default;
 
    void operator()(const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output0,
                    Array<OneD, NekDouble> &output1,
                    Array<OneD, NekDouble> &output2,
                    Array<OneD, NekDouble> &wsp) final
    {
        int nPhys = m_stdExp->GetTotPoints();
        int ntot  = m_numElmt * nPhys;
        Array<OneD, NekDouble> tmp0, tmp1, tmp2;
        Array<OneD, Array<OneD, NekDouble>> Diff(3);
        Array<OneD, Array<OneD, NekDouble>> out(3);
        out[0] = output0;
        out[1] = output1;
        out[2] = output2;
 
        for (int i = 0; i < 3; ++i)
        {
            Diff[i] = wsp + i * ntot;
        }
 
        // dEta0
        Blas::Dgemm('N', 'N', m_nquad0, m_nquad1 * m_nquad2 * m_numElmt,
                    m_nquad0, 1.0, m_Deriv0, m_nquad0, &input[0], m_nquad0, 0.0,
                    &Diff[0][0], m_nquad0);
 
        int cnt = 0;
        for (int i = 0; i < m_numElmt; ++i)
        {
            // dEta 1
            for (int j = 0; j < m_nquad2; ++j)
            {
                Blas::Dgemm('N', 'T', m_nquad0, m_nquad1, m_nquad1, 1.0,
                            &input[i * nPhys + j * m_nquad0 * m_nquad1],
                            m_nquad0, m_Deriv1, m_nquad1, 0.0,
                            &Diff[1][i * nPhys + j * m_nquad0 * m_nquad1],
                            m_nquad0);
            }
 
            // dEta 2
            Blas::Dgemm('N', 'T', m_nquad0 * m_nquad1, m_nquad2, m_nquad2, 1.0,
                        &input[i * nPhys], m_nquad0 * m_nquad1, m_Deriv2,
                        m_nquad2, 0.0, &Diff[2][i * nPhys],
                        m_nquad0 * m_nquad1);
 
            // dxi0 = 2/(1-eta_2) d Eta_0
            Vmath::Vmul(nPhys, &m_fac0[0], 1, Diff[0].data() + cnt, 1,
                        Diff[0].data() + cnt, 1);
 
            // dxi1 = 2/(1-eta_2) d Eta_1
            Vmath::Vmul(nPhys, &m_fac0[0], 1, Diff[1].data() + cnt, 1,
                        Diff[1].data() + cnt, 1);
 
            // dxi2 = (1+eta0)/(1-eta_2) d Eta_0 + d/dEta2;
            Vmath::Vvtvp(nPhys, &m_fac1[0], 1, Diff[0].data() + cnt, 1,
                         Diff[2].data() + cnt, 1, Diff[2].data() + cnt, 1);
 
            // dxi2 += (1+eta1)/(1-eta_2) d Eta_1
            Vmath::Vvtvp(nPhys, &m_fac2[0], 1, Diff[1].data() + cnt, 1,
                         Diff[2].data() + cnt, 1, Diff[2].data() + cnt, 1);
            cnt += nPhys;
        }
 
        // calculate full derivative
        if (m_isDeformed)
        {
            for (int i = 0; i < m_coordim; ++i)
            {
                Vmath::Vmul(ntot, m_derivFac[i * 3], 1, Diff[0], 1, out[i], 1);
                for (int j = 1; j < 3; ++j)
                {
                    Vmath::Vvtvp(ntot, m_derivFac[i * 3 + j], 1, Diff[j], 1,
                                 out[i], 1, out[i], 1);
                }
            }
        }
        else
        {
            Array<OneD, NekDouble> t;
            for (int e = 0; e < m_numElmt; ++e)
            {
                for (int i = 0; i < m_coordim; ++i)
                {
                    Vmath::Smul(m_nqe, m_derivFac[i * 3][e],
                                Diff[0] + e * m_nqe, 1, t = out[i] + e * m_nqe,
                                1);
 
                    for (int j = 1; j < 3; ++j)
                    {
                        Vmath::Svtvp(m_nqe, m_derivFac[i * 3 + j][e],
                                     Diff[j] + e * m_nqe, 1, out[i] + e * m_nqe,
                                     1, t = out[i] + e * m_nqe, 1);
                    }
                }
            }
        }
    }
 
    void operator()(int dir, const Array<OneD, const NekDouble> &input,
                    Array<OneD, NekDouble> &output,
                    Array<OneD, NekDouble> &wsp) final
    {
        int nPhys = m_stdExp->GetTotPoints();
        int ntot  = m_numElmt * nPhys;
        Array<OneD, NekDouble> tmp0, tmp1, tmp2;
        Array<OneD, Array<OneD, NekDouble>> Diff(3);
 
        for (int i = 0; i < 3; ++i)
        {
            Diff[i] = wsp + i * ntot;
        }
 
        // dEta0
        Blas::Dgemm('N', 'N', m_nquad0, m_nquad1 * m_nquad2 * m_numElmt,
                    m_nquad0, 1.0, m_Deriv0, m_nquad0, &input[0], m_nquad0, 0.0,
                    &Diff[0][0], m_nquad0);
 
        int cnt = 0;
        for (int i = 0; i < m_numElmt; ++i)
        {
            // dEta 1
            for (int j = 0; j < m_nquad2; ++j)
            {
                Blas::Dgemm('N', 'T', m_nquad0, m_nquad1, m_nquad1, 1.0,
                            &input[i * nPhys + j * m_nquad0 * m_nquad1],
                            m_nquad0, m_Deriv1, m_nquad1, 0.0,
                            &Diff[1][i * nPhys + j * m_nquad0 * m_nquad1],
                            m_nquad0);
            }
 
            // dEta 2
            Blas::Dgemm('N', 'T', m_nquad0 * m_nquad1, m_nquad2, m_nquad2, 1.0,
                        &input[i * nPhys], m_nquad0 * m_nquad1, m_Deriv2,
                        m_nquad2, 0.0, &Diff[2][i * nPhys],
                        m_nquad0 * m_nquad1);
 
            // dxi0 = 2/(1-eta_2) d Eta_0
            Vmath::Vmul(nPhys, &m_fac0[0], 1, Diff[0].data() + cnt, 1,
                        Diff[0].data() + cnt, 1);
 
            // dxi1 = 2/(1-eta_2) d Eta_1
            Vmath::Vmul(nPhys, &m_fac0[0], 1, Diff[1].data() + cnt, 1,
                        Diff[1].data() + cnt, 1);
 
            // dxi2 = (1+eta0)/(1-eta_2) d Eta_0 + d/dEta2;
            Vmath::Vvtvp(nPhys, &m_fac1[0], 1, Diff[0].data() + cnt, 1,
                         Diff[2].data() + cnt, 1, Diff[2].data() + cnt, 1);
 
            // dxi2 = (1+eta1)/(1-eta_2) d Eta_1 + d/dEta2;
            Vmath::Vvtvp(nPhys, &m_fac2[0], 1, Diff[1].data() + cnt, 1,
                         Diff[2].data() + cnt, 1, Diff[2].data() + cnt, 1);
            cnt += nPhys;
        }
 
        // calculate full derivative
        if (m_isDeformed)
        {
            // calculate full derivative
            Vmath::Vmul(ntot, m_derivFac[dir * 3], 1, Diff[0], 1, output, 1);
            for (int j = 1; j < 3; ++j)
            {
                Vmath::Vvtvp(ntot, m_derivFac[dir * 3 + j], 1, Diff[j], 1,
                             output, 1, output, 1);
            }
        }
        else
        {
            Array<OneD, NekDouble> t;
            for (int e = 0; e < m_numElmt; ++e)
            {
                Vmath::Smul(m_nqe, m_derivFac[dir * 3][e], Diff[0] + e * m_nqe,
                            1, t = output + e * m_nqe, 1);
 
                for (int j = 1; j < 3; ++j)
                {
                    Vmath::Svtvp(m_nqe, m_derivFac[dir * 3 + j][e],
                                 Diff[j] + e * m_nqe, 1, output + e * m_nqe, 1,
                                 t = output + e * m_nqe, 1);
                }
            }
        }
    }
 
protected:
    Array<TwoD, const NekDouble> m_derivFac;
    int m_coordim;
    const int m_nquad0;
    const int m_nquad1;
    const int m_nquad2;
    NekDouble *m_Deriv0;
    NekDouble *m_Deriv1;
    NekDouble *m_Deriv2;
    Array<OneD, NekDouble> m_fac0;
    Array<OneD, NekDouble> m_fac1;
    Array<OneD, NekDouble> m_fac2;
 
private:
    PhysDeriv_SumFac_Pyr(vector<LocalRegions::ExpansionSharedPtr> pCollExp,
                         CoalescedGeomDataSharedPtr pGeomData,
                         StdRegions::FactorMap factors)
        : Operator(pCollExp, pGeomData, factors), PhysDeriv_Helper(),
          m_nquad0(m_stdExp->GetNumPoints(0)),
          m_nquad1(m_stdExp->GetNumPoints(1)),
          m_nquad2(m_stdExp->GetNumPoints(2))
    {
        m_coordim = pCollExp[0]->GetCoordim();
 
        m_derivFac = pGeomData->GetDerivFactors(pCollExp);
 
        const Array<OneD, const NekDouble> &z0 = m_stdExp->GetBasis(0)->GetZ();
        const Array<OneD, const NekDouble> &z1 = m_stdExp->GetBasis(1)->GetZ();
        const Array<OneD, const NekDouble> &z2 = m_stdExp->GetBasis(2)->GetZ();
        m_fac0 = Array<OneD, NekDouble>(m_nquad0 * m_nquad1 * m_nquad2);
        m_fac1 = Array<OneD, NekDouble>(m_nquad0 * m_nquad1 * m_nquad2);
        m_fac2 = Array<OneD, NekDouble>(m_nquad0 * m_nquad1 * m_nquad2);
 
        int nq0_nq1 = m_nquad0 * m_nquad1;
        for (int i = 0; i < m_nquad0; ++i)
        {
            for (int j = 0; j < m_nquad1; ++j)
            {
                int ifac = i + j * m_nquad0;
                for (int k = 0; k < m_nquad2; ++k)
                {
                    m_fac0[ifac + k * nq0_nq1] = 2.0 / (1 - z2[k]);
                    m_fac1[ifac + k * nq0_nq1] = 0.5 * (1 + z0[i]);
                    m_fac2[ifac + k * nq0_nq1] = 0.5 * (1 + z1[j]);
                }
            }
        }
 
        m_Deriv0 = &((m_stdExp->GetBasis(0)->GetD())->GetPtr())[0];
        m_Deriv1 = &((m_stdExp->GetBasis(1)->GetD())->GetPtr())[0];
        m_Deriv2 = &((m_stdExp->GetBasis(2)->GetD())->GetPtr())[0];
 
        m_wspSize = 3 * m_nquad0 * m_nquad1 * m_nquad2 * m_numElmt;
    }
};
 
/// Factory initialisation for the PhysDeriv_SumFac_Pyr operators
OperatorKey PhysDeriv_SumFac_Pyr::m_typeArr[] = {
    GetOperatorFactory().RegisterCreatorFunction(
        OperatorKey(ePyramid, ePhysDeriv, eSumFac, false),
        PhysDeriv_SumFac_Pyr::create, "PhysDeriv_SumFac_Pyr")};
 
} // namespace Nektar::Collections