#include <limits>

#include <MultiParticleContainer.H>
#include <WarpXParticleContainer.H>
#include <AMReX_AmrParGDB.H>
#include <WarpX_f.H>
#include <WarpX.H>
#include <WarpXAlgorithmSelection.H>

// Import low-level single-particle kernels
#include <GetAndSetPosition.H>
#include <UpdatePosition.H>
#include <CurrentDeposition.H>

using namespace amrex;

int WarpXParticleContainer::do_not_push = 0;

WarpXParIter::WarpXParIter (ContainerType& pc, int level)
    : ParIter(pc, level, MFItInfo().SetDynamic(WarpX::do_dynamic_scheduling))
{
}

#if (AMREX_SPACEDIM == 2)
void
WarpXParIter::GetPosition (Cuda::ManagedDeviceVector<Real>& x, Cuda::ManagedDeviceVector<Real>& y, Cuda::ManagedDeviceVector<Real>& z) const
{
    amrex::ParIter<0,0,PIdx::nattribs>::GetPosition(x, z);
#ifdef WARPX_DIM_RZ
    const auto& attribs = GetAttribs();
    const auto& theta = attribs[PIdx::theta];
    y.resize(x.size());
    for (unsigned int i=0 ; i < x.size() ; i++) {
        // The x stored in the particles is actually the radius
        y[i] = x[i]*std::sin(theta[i]);
        x[i] = x[i]*std::cos(theta[i]);
    }
#else
    y.resize(x.size(), std::numeric_limits<Real>::quiet_NaN());
#endif
}

void
WarpXParIter::SetPosition (const Cuda::ManagedDeviceVector<Real>& x, const Cuda::ManagedDeviceVector<Real>& y, const Cuda::ManagedDeviceVector<Real>& z)
{
#ifdef WARPX_DIM_RZ
    auto& attribs = GetAttribs();
    auto& theta = attribs[PIdx::theta];
    Cuda::ManagedDeviceVector<Real> r(x.size());
    for (unsigned int i=0 ; i < x.size() ; i++) {
        theta[i] = std::atan2(y[i], x[i]);
        r[i] = std::sqrt(x[i]*x[i] + y[i]*y[i]);
    }
    amrex::ParIter<0,0,PIdx::nattribs>::SetPosition(r, z);
#else
    amrex::ParIter<0,0,PIdx::nattribs>::SetPosition(x, z);
#endif
}
#endif

WarpXParticleContainer::WarpXParticleContainer (AmrCore* amr_core, int ispecies)
    : ParticleContainer<0,0,PIdx::nattribs>(amr_core->GetParGDB())
    , species_id(ispecies)
{
    for (unsigned int i = PIdx::Ex; i <= PIdx::Bz; ++i) {
        communicate_real_comp[i] = false; // Don't need to communicate E and B.
    }
    SetParticleSize();
    ReadParameters();

    // build up the map of string names to particle component numbers
    particle_comps["w"]  = PIdx::w;
    particle_comps["ux"] = PIdx::ux;
    particle_comps["uy"] = PIdx::uy;
    particle_comps["uz"] = PIdx::uz;
    particle_comps["Ex"] = PIdx::Ex;
    particle_comps["Ey"] = PIdx::Ey;
    particle_comps["Ez"] = PIdx::Ez;
    particle_comps["Bx"] = PIdx::Bx;
    particle_comps["By"] = PIdx::By;
    particle_comps["Bz"] = PIdx::Bz;
#ifdef WARPX_DIM_RZ
    particle_comps["theta"] = PIdx::theta;
#endif

    if (WarpX::do_boosted_frame_diagnostic && do_boosted_frame_diags)
    {
        particle_comps["xold"]  = PIdx::nattribs;
        particle_comps["yold"]  = PIdx::nattribs+1;
        particle_comps["zold"]  = PIdx::nattribs+2;
        particle_comps["uxold"] = PIdx::nattribs+3;
        particle_comps["uyold"] = PIdx::nattribs+4;
        particle_comps["uzold"] = PIdx::nattribs+5;

    }

    // Initialize temporary local arrays for charge/current deposition
    int num_threads = 1;
    #ifdef _OPENMP
    #pragma omp parallel
    #pragma omp single
    num_threads = omp_get_num_threads();
    #endif
    local_rho.resize(num_threads);
    local_jx.resize(num_threads);
    local_jy.resize(num_threads);
    local_jz.resize(num_threads);
    m_xp.resize(num_threads);
    m_yp.resize(num_threads);
    m_zp.resize(num_threads);
    m_giv.resize(num_threads);
}

void
WarpXParticleContainer::ReadParameters ()
{
    static bool initialized = false;
    if (!initialized)
    {
	ParmParse pp("particles");

#ifdef AMREX_USE_GPU
        do_tiling = false; // By default, tiling is off on GPU
#else
        do_tiling = true;
#endif
        pp.query("do_tiling",  do_tiling);
        pp.query("do_not_push", do_not_push);

	initialized = true;
    }
}

void
WarpXParticleContainer::AllocData ()
{
    // have to resize here, not in the constructor because grids have not
    // been built when constructor was called.
    reserveData();
    resizeData();
}

void
WarpXParticleContainer::AddOneParticle (int lev, int grid, int tile,
                                        Real x, Real y, Real z,
                                        std::array<Real,PIdx::nattribs>& attribs)
{
    auto& particle_tile = GetParticles(lev)[std::make_pair(grid,tile)];
    AddOneParticle(particle_tile, x, y, z, attribs);
}

void
WarpXParticleContainer::AddOneParticle (ParticleTileType& particle_tile,
                                        Real x, Real y, Real z,
                                        std::array<Real,PIdx::nattribs>& attribs)
{
    ParticleType p;
    p.id()  = ParticleType::NextID();
    p.cpu() = ParallelDescriptor::MyProc();
#if (AMREX_SPACEDIM == 3)
    p.pos(0) = x;
    p.pos(1) = y;
    p.pos(2) = z;
#elif (AMREX_SPACEDIM == 2)
#ifdef WARPX_DIM_RZ
    attribs[PIdx::theta] = std::atan2(y, x);
    x = std::sqrt(x*x + y*y);
#endif
    p.pos(0) = x;
    p.pos(1) = z;
#endif

    particle_tile.push_back(p);
    particle_tile.push_back_real(attribs);
}

void
WarpXParticleContainer::AddNParticles (int lev,
                                       int n, const Real* x, const Real* y, const Real* z,
				       const Real* vx, const Real* vy, const Real* vz,
				       int nattr, const Real* attr, int uniqueparticles, int id)
{
    BL_ASSERT(nattr == 1);
    const Real* weight = attr;

    int ibegin, iend;
    if (uniqueparticles) {
	ibegin = 0;
	iend = n;
    } else {
	int myproc = ParallelDescriptor::MyProc();
	int nprocs = ParallelDescriptor::NProcs();
	int navg = n/nprocs;
	int nleft = n - navg * nprocs;
	if (myproc < nleft) {
	    ibegin = myproc*(navg+1);
	    iend = ibegin + navg+1;
	} else {
	    ibegin = myproc*navg + nleft;
	    iend = ibegin + navg;
	}
    }

    //  Add to grid 0 and tile 0
    // Redistribute() will move them to proper places.
    std::pair<int,int> key {0,0};
    auto& particle_tile = GetParticles(lev)[key];

    std::size_t np = iend-ibegin;

#ifdef WARPX_DIM_RZ
    Vector<Real> theta(np);
#endif

    for (int i = ibegin; i < iend; ++i)
    {
        ParticleType p;
	if (id==-1)
	{
	    p.id() = ParticleType::NextID();
	} else {
	    p.id() = id;
	}
        p.cpu() = ParallelDescriptor::MyProc();
#if (AMREX_SPACEDIM == 3)
        p.pos(0) = x[i];
        p.pos(1) = y[i];
        p.pos(2) = z[i];
#elif (AMREX_SPACEDIM == 2)
#ifdef WARPX_DIM_RZ
        theta[i-ibegin] = std::atan2(y[i], x[i]);
        p.pos(0) = std::sqrt(x[i]*x[i] + y[i]*y[i]);
#else
        p.pos(0) = x[i];
#endif
        p.pos(1) = z[i];
#endif

        if (WarpX::do_boosted_frame_diagnostic && do_boosted_frame_diags)
        {
            auto& ptile = DefineAndReturnParticleTile(0, 0, 0);
            ptile.push_back_real(particle_comps["xold"], x[i]);
            ptile.push_back_real(particle_comps["yold"], y[i]);
            ptile.push_back_real(particle_comps["zold"], z[i]);
        }

        particle_tile.push_back(p);
    }

    if (np > 0)
    {
        particle_tile.push_back_real(PIdx::w , weight + ibegin, weight + iend);
        particle_tile.push_back_real(PIdx::ux,     vx + ibegin,     vx + iend);
        particle_tile.push_back_real(PIdx::uy,     vy + ibegin,     vy + iend);
        particle_tile.push_back_real(PIdx::uz,     vz + ibegin,     vz + iend);

        if (WarpX::do_boosted_frame_diagnostic && do_boosted_frame_diags)
        {
            auto& ptile = DefineAndReturnParticleTile(0, 0, 0);
            ptile.push_back_real(particle_comps["uxold"], vx + ibegin, vx + iend);
            ptile.push_back_real(particle_comps["uyold"], vy + ibegin, vy + iend);
            ptile.push_back_real(particle_comps["uzold"], vz + ibegin, vz + iend);
        }

        for (int comp = PIdx::uz+1; comp < PIdx::nattribs; ++comp)
        {
#ifdef WARPX_DIM_RZ
            if (comp == PIdx::theta) {
                particle_tile.push_back_real(comp, theta.front(), theta.back());
            }
            else {
                particle_tile.push_back_real(comp, np, 0.0);
            }
#else
            particle_tile.push_back_real(comp, np, 0.0);
#endif
        }
    }

    Redistribute();
}

/* \brief Current Deposition for thread thread_num using PICSAR
 * \param pti         : Particle iterator
 * \param wp          : Array of particle weights
 * \param uxp uyp uzp : Array of particle
 * \param jx jy jz    : Full array of current density
 * \param offset      : Index of first particle for which current is deposited
 * \param np_to_depose: Number of particles for which current is deposited.
                        Particles [offset,offset+np_tp_depose] deposit their
                        current
 * \param thread_num  : Thread number (if tiling)
 * \param lev         : Level of box that contains particles
 * \param depos_lev   : Level on which particles deposit (if buffers are used)
 * \param dt          : Time step for particle level
 */
void
WarpXParticleContainer::DepositCurrentFortran(WarpXParIter& pti,
                                              RealVector& wp, RealVector& uxp,
                                              RealVector& uyp, RealVector& uzp,
                                              MultiFab* jx, MultiFab* jy, MultiFab* jz,
                                              const long offset, const long np_to_depose,
                                              int thread_num, int lev, int depos_lev,
                                              Real dt)
{
    AMREX_ALWAYS_ASSERT_WITH_MESSAGE((depos_lev==(lev-1)) ||
                                     (depos_lev==(lev  )),
                                     "Deposition buffers only work for lev-1");
    // If no particles, do not do anything
    if (np_to_depose == 0) return;

    const long ngJ = jx->nGrow();
    const std::array<Real,3>& dx = WarpX::CellSize(std::max(depos_lev,0));
    const long lvect = 8;
    int j_is_nodal = jx->is_nodal() and jy->is_nodal() and jz->is_nodal();

    BL_PROFILE_VAR_NS("PICSAR::CurrentDeposition", blp_pxr_cd);
    BL_PROFILE_VAR_NS("PPC::Evolve::Accumulate", blp_accumulate);

    // Get tile box where current is deposited.
    // The tile box is different when depositing in the buffers (depos_lev<lev)
    // or when depositing inside the level (depos_lev=lev)
    Box tilebox;
    if (lev == depos_lev) {
        tilebox = pti.tilebox();
    } else {
        const IntVect& ref_ratio = WarpX::RefRatio(depos_lev);
        tilebox = amrex::coarsen(pti.tilebox(),ref_ratio);
    }
    
    // Staggered tile boxes (different in each direction)
    Box tbx = convert(tilebox, WarpX::jx_nodal_flag);
    Box tby = convert(tilebox, WarpX::jy_nodal_flag);
    Box tbz = convert(tilebox, WarpX::jz_nodal_flag);

    // Lower corner of tile box physical domain
    const std::array<Real,3>& xyzmin_tile = WarpX::LowerCorner(tilebox, depos_lev);
    const std::array<Real, 3>& xyzmin = xyzmin_tile;

#ifdef AMREX_USE_GPU
    // No tiling on GPU: jx_ptr points to the full
    // jx array (same for jy_ptr and jz_ptr).
    Real* jx_ptr = (*jx)[pti].dataPtr();
    Real* jy_ptr = (*jy)[pti].dataPtr();
    Real* jz_ptr = (*jz)[pti].dataPtr();
    
    auto jxntot = (*jx)[pti].length();
    auto jyntot = (*jy)[pti].length();
    auto jzntot = (*jz)[pti].length();
#else
    // Tiling is on: jx_ptr points to local_jx[thread_num]
    // (same for jy_ptr and jz_ptr)
    tbx.grow(ngJ);
    tby.grow(ngJ);
    tbz.grow(ngJ);

    local_jx[thread_num].resize(tbx, jx->nComp());
    local_jy[thread_num].resize(tby, jy->nComp());
    local_jz[thread_num].resize(tbz, jz->nComp());

    Real* jx_ptr = local_jx[thread_num].dataPtr();
    Real* jy_ptr = local_jy[thread_num].dataPtr();
    Real* jz_ptr = local_jz[thread_num].dataPtr();

    // local_jx[thread_num] is set to zero
    local_jx[thread_num].setVal(0.0);
    local_jy[thread_num].setVal(0.0);
    local_jz[thread_num].setVal(0.0);

    auto jxntot = local_jx[thread_num].length();
    auto jyntot = local_jy[thread_num].length();
    auto jzntot = local_jz[thread_num].length();
#endif
    // GPU, no tiling: deposit directly in jx
    // CPU, tiling: deposit into local_jx
    // (same for jx and jz)
    BL_PROFILE_VAR_START(blp_pxr_cd);
    warpx_current_deposition(
        jx_ptr, &ngJ, jxntot.getVect(),
        jy_ptr, &ngJ, jyntot.getVect(),
        jz_ptr, &ngJ, jzntot.getVect(),
        &WarpX::n_rz_azimuthal_modes,
        &np_to_depose,
        m_xp[thread_num].dataPtr() + offset,
        m_yp[thread_num].dataPtr() + offset,
        m_zp[thread_num].dataPtr() + offset,
        uxp.dataPtr() + offset,
        uyp.dataPtr() + offset,
        uzp.dataPtr() + offset,
        m_giv[thread_num].dataPtr() + offset,
        wp.dataPtr() + offset, &this->charge,
        &xyzmin[0], &xyzmin[1], &xyzmin[2],
        &dt, &dx[0], &dx[1], &dx[2],
        &WarpX::nox,&WarpX::noy,&WarpX::noz, &j_is_nodal,
        &lvect,&WarpX::current_deposition_algo);

    BL_PROFILE_VAR_STOP(blp_pxr_cd);

#ifndef AMREX_USE_GPU
    BL_PROFILE_VAR_START(blp_accumulate);
    // CPU, tiling: atomicAdd local_jx into jx
    // (same for jx and jz)
    (*jx)[pti].atomicAdd(local_jx[thread_num], tbx, tbx, 0, 0, local_jx[thread_num].nComp());
    (*jy)[pti].atomicAdd(local_jy[thread_num], tby, tby, 0, 0, local_jy[thread_num].nComp());
    (*jz)[pti].atomicAdd(local_jz[thread_num], tbz, tbz, 0, 0, local_jz[thread_num].nComp());
    BL_PROFILE_VAR_STOP(blp_accumulate);
#endif
}

/* \brief Current Deposition for thread thread_num
 * \param pti         : Particle iterator
 * \param wp          : Array of particle weights
 * \param uxp uyp uzp : Array of particle
 * \param jx jy jz    : Full array of current density
 * \param offset      : Index of first particle for which current is deposited
 * \param np_to_depose: Number of particles for which current is deposited.
                        Particles [offset,offset+np_tp_depose] deposit current
 * \param thread_num  : Thread number (if tiling)
 * \param lev         : Level of box that contains particles
 * \param depos_lev   : Level on which particles deposit (if buffers are used)
 * \param dt          : Time step for particle level
 */
void
WarpXParticleContainer::DepositCurrent(WarpXParIter& pti,
                                       RealVector& wp, RealVector& uxp,
                                       RealVector& uyp, RealVector& uzp,
                                       MultiFab* jx, MultiFab* jy, MultiFab* jz,
                                       const long offset, const long np_to_depose,
                                       int thread_num, int lev, int depos_lev,
                                       Real dt)
{    
    AMREX_ALWAYS_ASSERT_WITH_MESSAGE((depos_lev==(lev-1)) ||
                                     (depos_lev==(lev  )),
                                     "Deposition buffers only work for lev-1");
    // If no particles, do not do anything
    if (np_to_depose == 0) return;

    const long ngJ = jx->nGrow();
    const std::array<Real,3>& dx = WarpX::CellSize(std::max(depos_lev,0));
    int j_is_nodal = jx->is_nodal() and jy->is_nodal() and jz->is_nodal();
    Real q = this->charge;
    const Real stagger_shift = j_is_nodal ? 0.0 : 0.5;

    BL_PROFILE_VAR_NS("PPC::Evolve::Accumulate", blp_accumulate);

    // Get tile box where current is deposited.
    // The tile box is different when depositing in the buffers (depos_lev<lev)
    // or when depositing inside the level (depos_lev=lev)
    Box tilebox;
    if (lev == depos_lev) {
        tilebox = pti.tilebox();
    } else {
        const IntVect& ref_ratio = WarpX::RefRatio(depos_lev);
        tilebox = amrex::coarsen(pti.tilebox(),ref_ratio);
    }
    
    // Staggered tile boxes (different in each direction)
    Box tbx = convert(tilebox, WarpX::jx_nodal_flag);
    Box tby = convert(tilebox, WarpX::jy_nodal_flag);
    Box tbz = convert(tilebox, WarpX::jz_nodal_flag);
    tilebox.grow(ngJ);

#ifdef AMREX_USE_GPU
    // No tiling on GPU: jx_ptr points to the full
    // jx array (same for jy_ptr and jz_ptr).
    Array4<Real> const& jx_arr = jx->array(pti);
    Array4<Real> const& jy_arr = jy->array(pti);
    Array4<Real> const& jz_arr = jz->array(pti);
#else
    // Tiling is on: jx_ptr points to local_jx[thread_num]
    // (same for jy_ptr and jz_ptr)
    tbx.grow(ngJ);
    tby.grow(ngJ);
    tbz.grow(ngJ);

    local_jx[thread_num].resize(tbx);
    local_jy[thread_num].resize(tby);
    local_jz[thread_num].resize(tbz);

    // local_jx[thread_num] is set to zero
    local_jx[thread_num].setVal(0.0);
    local_jy[thread_num].setVal(0.0);
    local_jz[thread_num].setVal(0.0);

    Array4<Real> const& jx_arr = local_jx[thread_num].array();
    Array4<Real> const& jy_arr = local_jy[thread_num].array();
    Array4<Real> const& jz_arr = local_jz[thread_num].array();
#endif
    // GPU, no tiling: deposit directly in jx
    // CPU, tiling: deposit into local_jx
    // (same for jx and jz)

    Real* AMREX_RESTRICT xp = m_xp[thread_num].dataPtr() + offset;
    Real* AMREX_RESTRICT zp = m_zp[thread_num].dataPtr() + offset;
    Real* AMREX_RESTRICT yp = m_yp[thread_num].dataPtr() + offset;

    // Lower corner of tile box physical domain
    // Note that this includes guard cells since it is after tilebox.ngrow
    const std::array<Real, 3>& xyzmin = WarpX::LowerCorner(tilebox, depos_lev);
    // xyzmin is built on pti.tilebox(), so it does 
    // not include staggering, so the stagger_shift has to be done by hand.
    // Alternatively, we could define xyzminx from tbx (and the same for 3 
    // directions and for jx, jy, jz). This way, sx0 would not be needed.
    // Better for memory? worth trying?    
    const Dim3 lo = lbound(tilebox);

    if (WarpX::current_deposition_algo == CurrentDepositionAlgo::Esirkepov) {
        if        (WarpX::nox == 1){
            doEsirkepovDepositionShapeN<1>(xp, yp, zp, wp.dataPtr() + offset, uxp.dataPtr() + offset, 
                                           uyp.dataPtr() + offset, uzp.dataPtr() + offset, jx_arr, jy_arr, 
                                           jz_arr, np_to_depose, dt, dx,
                                           xyzmin, lo, q);
        } else if (WarpX::nox == 2){
            doEsirkepovDepositionShapeN<2>(xp, yp, zp, wp.dataPtr() + offset, uxp.dataPtr() + offset, 
                                           uyp.dataPtr() + offset, uzp.dataPtr() + offset, jx_arr, jy_arr, 
                                           jz_arr, np_to_depose, dt, dx,
                                           xyzmin, lo, q);
        } else if (WarpX::nox == 3){
            doEsirkepovDepositionShapeN<3>(xp, yp, zp, wp.dataPtr() + offset, uxp.dataPtr() + offset, 
                                           uyp.dataPtr() + offset, uzp.dataPtr() + offset, jx_arr, jy_arr, 
                                           jz_arr, np_to_depose, dt, dx,
                                           xyzmin, lo, q);
        }
    } else {
        if        (WarpX::nox == 1){
            doDepositionShapeN<1>(xp, yp, zp, wp.dataPtr() + offset, uxp.dataPtr() + offset, 
                                  uyp.dataPtr() + offset, uzp.dataPtr() + offset, jx_arr, jy_arr, 
                                  jz_arr, np_to_depose, dt, dx,
                                  xyzmin, lo, stagger_shift, q);
        } else if (WarpX::nox == 2){
            doDepositionShapeN<2>(xp, yp, zp, wp.dataPtr() + offset, uxp.dataPtr() + offset, 
                                  uyp.dataPtr() + offset, uzp.dataPtr() + offset, jx_arr, jy_arr, 
                                  jz_arr, np_to_depose, dt, dx,
                                  xyzmin, lo, stagger_shift, q);
        } else if (WarpX::nox == 3){
            doDepositionShapeN<3>(xp, yp, zp, wp.dataPtr() + offset, uxp.dataPtr() + offset, 
                                  uyp.dataPtr() + offset, uzp.dataPtr() + offset, jx_arr, jy_arr, 
                                  jz_arr, np_to_depose, dt, dx,
                                  xyzmin, lo, stagger_shift, q);
        }
    }

#ifndef AMREX_USE_GPU
    BL_PROFILE_VAR_START(blp_accumulate);
    // CPU, tiling: atomicAdd local_jx into jx
    // (same for jx and jz)
    (*jx)[pti].atomicAdd(local_jx[thread_num], tbx, tbx, 0, 0, 1);
    (*jy)[pti].atomicAdd(local_jy[thread_num], tby, tby, 0, 0, 1);
    (*jz)[pti].atomicAdd(local_jz[thread_num], tbz, tbz, 0, 0, 1);
    BL_PROFILE_VAR_STOP(blp_accumulate);
#endif
}

void
WarpXParticleContainer::DepositCharge ( WarpXParIter& pti, RealVector& wp,
                                        MultiFab* rhomf, MultiFab* crhomf, int icomp,
                                        const long np_current,
                                        const long np, int thread_num, int lev )
{

  BL_PROFILE_VAR_NS("PICSAR::ChargeDeposition", blp_pxr_chd);
  BL_PROFILE_VAR_NS("PPC::Evolve::Accumulate", blp_accumulate);

  const std::array<Real,3>& xyzmin_tile = WarpX::LowerCorner(pti.tilebox(), lev);
  const long lvect = 8;

  long ngRho = rhomf->nGrow();
  Real* data_ptr;
  Box tile_box = convert(pti.tilebox(), IntVect::TheUnitVector());

  const std::array<Real,3>& dx = WarpX::CellSize(lev);
  const std::array<Real,3>& cdx = WarpX::CellSize(std::max(lev-1,0));

  // Deposit charge for particles that are not in the current buffers
  if (np_current > 0)
  {
      const std::array<Real, 3>& xyzmin = xyzmin_tile;

#ifdef AMREX_USE_GPU
      data_ptr = (*rhomf)[pti].dataPtr(icomp*(rhomf->nComp()/2));
      auto rholen = (*rhomf)[pti].length();
#else
      tile_box.grow(ngRho);
      local_rho[thread_num].resize(tile_box, rhomf->nComp());

      data_ptr = local_rho[thread_num].dataPtr();
      auto rholen = local_rho[thread_num].length();

      local_rho[thread_num].setVal(0.0);
#endif

#if (AMREX_SPACEDIM == 3)
      const long nx = rholen[0]-1-2*ngRho;
      const long ny = rholen[1]-1-2*ngRho;
      const long nz = rholen[2]-1-2*ngRho;
#else
      const long nx = rholen[0]-1-2*ngRho;
      const long ny = 0;
      const long nz = rholen[1]-1-2*ngRho;
#endif
      BL_PROFILE_VAR_START(blp_pxr_chd);
      warpx_charge_deposition(data_ptr, &np_current,
                              m_xp[thread_num].dataPtr(),
                              m_yp[thread_num].dataPtr(),
                              m_zp[thread_num].dataPtr(),
                              wp.dataPtr(),
                              &this->charge,
                              &xyzmin[0], &xyzmin[1], &xyzmin[2],
                              &dx[0], &dx[1], &dx[2], &nx, &ny, &nz,
                              &ngRho, &ngRho, &ngRho,
                              &WarpX::nox,&WarpX::noy,&WarpX::noz,
                              &lvect, &WarpX::charge_deposition_algo);
#ifdef WARPX_DIM_RZ
      warpx_charge_deposition_rz_volume_scaling(
                               data_ptr, &ngRho, rholen.getVect(),
                               &xyzmin[0], &dx[0]);
#endif
      BL_PROFILE_VAR_STOP(blp_pxr_chd);

#ifndef AMREX_USE_GPU
      BL_PROFILE_VAR_START(blp_accumulate);

      (*rhomf)[pti].atomicAdd(local_rho[thread_num], tile_box, tile_box, 0, icomp*(rhomf->nComp()/2), (rhomf->nComp()/2));

      BL_PROFILE_VAR_STOP(blp_accumulate);
#endif
  }

  // Deposit charge for particles that are in the current buffers
  if (np_current < np)
  {
      const IntVect& ref_ratio = WarpX::RefRatio(lev-1);
      const Box& ctilebox = amrex::coarsen(pti.tilebox(), ref_ratio);
      const std::array<Real,3>& cxyzmin_tile = WarpX::LowerCorner(ctilebox, lev-1);

#ifdef AMREX_USE_GPU
      data_ptr = (*crhomf)[pti].dataPtr(icomp);
      auto rholen = (*crhomf)[pti].length();
#else
      tile_box = amrex::convert(ctilebox, IntVect::TheUnitVector());
      tile_box.grow(ngRho);
      local_rho[thread_num].resize(tile_box, crhomf->nComp());

      data_ptr = local_rho[thread_num].dataPtr();
      auto rholen = local_rho[thread_num].length();

      local_rho[thread_num].setVal(0.0);
#endif

#if (AMREX_SPACEDIM == 3)
      const long nx = rholen[0]-1-2*ngRho;
      const long ny = rholen[1]-1-2*ngRho;
      const long nz = rholen[2]-1-2*ngRho;
#else
      const long nx = rholen[0]-1-2*ngRho;
      const long ny = 0;
      const long nz = rholen[1]-1-2*ngRho;
#endif

      long ncrse = np - np_current;
      BL_PROFILE_VAR_START(blp_pxr_chd);
      warpx_charge_deposition(data_ptr, &ncrse,
                              m_xp[thread_num].dataPtr() + np_current,
                              m_yp[thread_num].dataPtr() + np_current,
                              m_zp[thread_num].dataPtr() + np_current,
                              wp.dataPtr() + np_current,
                              &this->charge,
                              &cxyzmin_tile[0], &cxyzmin_tile[1], &cxyzmin_tile[2],
                              &cdx[0], &cdx[1], &cdx[2], &nx, &ny, &nz,
                              &ngRho, &ngRho, &ngRho,
                              &WarpX::nox,&WarpX::noy,&WarpX::noz,
                              &lvect, &WarpX::charge_deposition_algo);
#ifdef WARPX_DIM_RZ
      warpx_charge_deposition_rz_volume_scaling(
                               data_ptr, &ngRho, rholen.getVect(),
                               &cxyzmin_tile[0], &cdx[0]);
#endif
      BL_PROFILE_VAR_STOP(blp_pxr_chd);

#ifndef AMREX_USE_GPU
      BL_PROFILE_VAR_START(blp_accumulate);

      (*crhomf)[pti].atomicAdd(local_rho[thread_num], tile_box, tile_box, 0, icomp*(crhomf->nComp()/2), (crhomf->nComp()/2));

      BL_PROFILE_VAR_STOP(blp_accumulate);
#endif
    }
};

void
WarpXParticleContainer::DepositCharge (Vector<std::unique_ptr<MultiFab> >& rho, bool local)
{

    int num_levels = rho.size();
    int finest_level = num_levels - 1;

    // each level deposits it's own particles
    const int ng = rho[0]->nGrow();
    for (int lev = 0; lev < num_levels; ++lev) {

        rho[lev]->setVal(0.0, ng);

        const auto& gm = m_gdb->Geom(lev);
        const auto& ba = m_gdb->ParticleBoxArray(lev);

        const Real* dx  = gm.CellSize();
        const Real* plo = gm.ProbLo();
        BoxArray nba = ba;
        nba.surroundingNodes();

        for (WarpXParIter pti(*this, lev); pti.isValid(); ++pti) {
            const Box& box = nba[pti];

            auto& wp = pti.GetAttribs(PIdx::w);
            const auto& particles = pti.GetArrayOfStructs();
            int nstride = particles.dataShape().first;
            const long np  = pti.numParticles();

            FArrayBox& rhofab = (*rho[lev])[pti];

            WRPX_DEPOSIT_CIC(particles.dataPtr(), nstride, np,
                             wp.dataPtr(), &this->charge,
                             rhofab.dataPtr(), box.loVect(), box.hiVect(),
                             plo, dx, &ng);
        }

        if (!local) rho[lev]->SumBoundary(gm.periodicity());
    }

    // now we average down fine to crse
    std::unique_ptr<MultiFab> crse;
    for (int lev = finest_level - 1; lev >= 0; --lev) {
        const BoxArray& fine_BA = rho[lev+1]->boxArray();
        const DistributionMapping& fine_dm = rho[lev+1]->DistributionMap();
        BoxArray coarsened_fine_BA = fine_BA;
        coarsened_fine_BA.coarsen(m_gdb->refRatio(lev));

        MultiFab coarsened_fine_data(coarsened_fine_BA, fine_dm, rho[lev+1]->nComp(), 0);
        coarsened_fine_data.setVal(0.0);

        IntVect ratio(AMREX_D_DECL(2, 2, 2));  // FIXME

        for (MFIter mfi(coarsened_fine_data); mfi.isValid(); ++mfi) {
            const Box& bx = mfi.validbox();
            const Box& crse_box = coarsened_fine_data[mfi].box();
            const Box& fine_box = (*rho[lev+1])[mfi].box();
            WRPX_SUM_FINE_TO_CRSE_NODAL(bx.loVect(), bx.hiVect(), ratio.getVect(),
                                        coarsened_fine_data[mfi].dataPtr(), crse_box.loVect(), crse_box.hiVect(),
                                        (*rho[lev+1])[mfi].dataPtr(), fine_box.loVect(), fine_box.hiVect());
        }

        rho[lev]->copy(coarsened_fine_data, m_gdb->Geom(lev).periodicity(), FabArrayBase::ADD);
    }
}

std::unique_ptr<MultiFab>
WarpXParticleContainer::GetChargeDensity (int lev, bool local)
{
    const auto& gm = m_gdb->Geom(lev);
    const auto& ba = m_gdb->ParticleBoxArray(lev);
    const auto& dm = m_gdb->DistributionMap(lev);
    BoxArray nba = ba;
    nba.surroundingNodes();

    const std::array<Real,3>& dx = WarpX::CellSize(lev);

    const int ng = WarpX::nox;

    auto rho = std::unique_ptr<MultiFab>(new MultiFab(nba,dm,WarpX::ncomps,ng));
    rho->setVal(0.0);

#ifdef _OPENMP
#pragma omp parallel
    {
#endif
        Cuda::ManagedDeviceVector<Real> xp, yp, zp;
#ifdef _OPENMP
        FArrayBox rho_loc;
#endif

        for (WarpXParIter pti(*this, lev); pti.isValid(); ++pti)
        {
            auto& wp = pti.GetAttribs(PIdx::w);

            const long np  = pti.numParticles();

            pti.GetPosition(xp, yp, zp);

            // Data on the grid
            Real* data_ptr;
            FArrayBox& rhofab = (*rho)[pti];
#ifdef _OPENMP
            const std::array<Real,3>& xyzmin_tile = WarpX::LowerCorner(pti.tilebox(), lev);
            Box tile_box = convert(pti.tilebox(), IntVect::TheUnitVector());
            const std::array<Real, 3>& xyzmin = xyzmin_tile;
            tile_box.grow(ng);
            rho_loc.resize(tile_box, rho->nComp());
            rho_loc = 0.0;
            data_ptr = rho_loc.dataPtr();
            auto rholen = rho_loc.length();
#else
            const Box& box = pti.validbox();
            const std::array<Real,3>& xyzmin_grid = WarpX::LowerCorner(box, lev);
            const std::array<Real, 3>& xyzmin = xyzmin_grid;
            data_ptr = rhofab.dataPtr();
            auto rholen = rhofab.length();
#endif

#if (AMREX_SPACEDIM == 3)
            const long nx = rholen[0]-1-2*ng;
            const long ny = rholen[1]-1-2*ng;
            const long nz = rholen[2]-1-2*ng;
#else
            const long nx = rholen[0]-1-2*ng;
            const long ny = 0;
            const long nz = rholen[1]-1-2*ng;
#endif

            long nxg = ng;
            long nyg = ng;
            long nzg = ng;
            long lvect = 8;

            warpx_charge_deposition(data_ptr,
                                    &np,
                                    xp.dataPtr(),
                                    yp.dataPtr(),
                                    zp.dataPtr(), wp.dataPtr(),
                                    &this->charge, &xyzmin[0], &xyzmin[1], &xyzmin[2],
                                    &dx[0], &dx[1], &dx[2], &nx, &ny, &nz,
                                    &nxg, &nyg, &nzg, &WarpX::nox,&WarpX::noy,&WarpX::noz,
                                    &lvect, &WarpX::charge_deposition_algo);
#ifdef WARPX_DIM_RZ
            long ngRho = WarpX::nox;
            warpx_charge_deposition_rz_volume_scaling(
                                     data_ptr, &ngRho, rholen.getVect(),
                                     &xyzmin[0], &dx[0]);
#endif

#ifdef _OPENMP
            rhofab.atomicAdd(rho_loc);
        }
#endif
    }

    if (!local) rho->SumBoundary(gm.periodicity());

    return rho;
}

Real WarpXParticleContainer::sumParticleCharge(bool local) {

    amrex::Real total_charge = 0.0;

    for (int lev = 0; lev < finestLevel(); ++lev)
    {

#ifdef _OPENMP
#pragma omp parallel reduction(+:total_charge)
#endif
        for (WarpXParIter pti(*this, lev); pti.isValid(); ++pti)
        {
            auto& wp = pti.GetAttribs(PIdx::w);
            for (unsigned long i = 0; i < wp.size(); i++) {
                total_charge += wp[i];
            }
        }
    }

    if (!local) ParallelDescriptor::ReduceRealSum(total_charge);
    total_charge *= this->charge;
    return total_charge;
}

std::array<Real, 3> WarpXParticleContainer::meanParticleVelocity(bool local) {

    amrex::Real vx_total = 0.0;
    amrex::Real vy_total = 0.0;
    amrex::Real vz_total = 0.0;

    long np_total = 0;

    amrex::Real inv_clight_sq = 1.0/PhysConst::c/PhysConst::c;

    for (int lev = 0; lev <= finestLevel(); ++lev) {

#ifdef _OPENMP
#pragma omp parallel reduction(+:vx_total, vy_total, vz_total, np_total)
#endif
        for (WarpXParIter pti(*this, lev); pti.isValid(); ++pti)
        {
            auto& ux = pti.GetAttribs(PIdx::ux);
            auto& uy = pti.GetAttribs(PIdx::uy);
            auto& uz = pti.GetAttribs(PIdx::uz);

            np_total += pti.numParticles();

            for (unsigned long i = 0; i < ux.size(); i++) {
                Real usq = (ux[i]*ux[i] + uy[i]*uy[i] + uz[i]*uz[i])*inv_clight_sq;
                Real gaminv = 1.0/std::sqrt(1.0 + usq);
                vx_total += ux[i]*gaminv;
                vy_total += uy[i]*gaminv;
                vz_total += uz[i]*gaminv;
            }
        }
    }

    if (!local) {
        ParallelDescriptor::ReduceRealSum(vx_total);
        ParallelDescriptor::ReduceRealSum(vy_total);
        ParallelDescriptor::ReduceRealSum(vz_total);
        ParallelDescriptor::ReduceLongSum(np_total);
    }

    std::array<Real, 3> mean_v;
    if (np_total > 0) {
        mean_v[0] = vx_total / np_total;
        mean_v[1] = vy_total / np_total;
        mean_v[2] = vz_total / np_total;
    }

    return mean_v;
}

Real WarpXParticleContainer::maxParticleVelocity(bool local) {

    amrex::Real max_v = 0.0;

    for (int lev = 0; lev <= finestLevel(); ++lev)
    {

#ifdef _OPENMP
#pragma omp parallel reduction(max:max_v)
#endif
        for (WarpXParIter pti(*this, lev); pti.isValid(); ++pti)
        {
            auto& ux = pti.GetAttribs(PIdx::ux);
            auto& uy = pti.GetAttribs(PIdx::uy);
            auto& uz = pti.GetAttribs(PIdx::uz);
            for (unsigned long i = 0; i < ux.size(); i++) {
                max_v = std::max(max_v, sqrt(ux[i]*ux[i] + uy[i]*uy[i] + uz[i]*uz[i]));
            }
        }
    }

    if (!local) ParallelDescriptor::ReduceRealMax(max_v);
    return max_v;
}

void
WarpXParticleContainer::PushXES (Real dt)
{
    BL_PROFILE("WPC::PushXES()");

    int num_levels = finestLevel() + 1;

    for (int lev = 0; lev < num_levels; ++lev) {
        const auto& gm = m_gdb->Geom(lev);
        const RealBox& prob_domain = gm.ProbDomain();
        for (WarpXParIter pti(*this, lev); pti.isValid(); ++pti) {
            auto& particles = pti.GetArrayOfStructs();
            int nstride = particles.dataShape().first;
            const long np  = pti.numParticles();

            auto& attribs = pti.GetAttribs();
            auto& uxp = attribs[PIdx::ux];
            auto& uyp = attribs[PIdx::uy];
            auto& uzp = attribs[PIdx::uz];

            WRPX_PUSH_LEAPFROG_POSITIONS(particles.dataPtr(), nstride, np,
                                         uxp.dataPtr(), uyp.dataPtr(),
#if AMREX_SPACEDIM == 3
                                         uzp.dataPtr(),
#endif
                                         &dt,
                                         prob_domain.lo(), prob_domain.hi());
        }
    }
}

void
WarpXParticleContainer::PushX (Real dt)
{
    for (int lev = 0; lev <= finestLevel(); ++lev) {
        PushX(lev, dt);
    }
}

void
WarpXParticleContainer::PushX (int lev, Real dt)
{
    BL_PROFILE("WPC::PushX()");

    if (do_not_push) return;

    MultiFab* cost = WarpX::getCosts(lev);

#ifdef _OPENMP
#pragma omp parallel
#endif
    {

        for (WarpXParIter pti(*this, lev); pti.isValid(); ++pti)
        {
            Real wt = amrex::second();

            //
            // Particle Push
            //
            // Extract pointers to particle position and momenta, for this particle tile
            // - positions are stored as an array of struct, in `ParticleType`
            ParticleType * AMREX_RESTRICT pstructs = &(pti.GetArrayOfStructs()[0]);
            // - momenta are stored as a struct of array, in `attribs`
            auto& attribs = pti.GetAttribs();
            Real* AMREX_RESTRICT ux = attribs[PIdx::ux].dataPtr();
            Real* AMREX_RESTRICT uy = attribs[PIdx::uy].dataPtr();
            Real* AMREX_RESTRICT uz = attribs[PIdx::uz].dataPtr();
#ifdef WARPX_DIM_RZ
            Real* AMREX_RESTRICT theta = attribs[PIdx::theta].dataPtr();
#endif
            // Loop over the particles and update their position
            amrex::ParallelFor( pti.numParticles(),
                [=] AMREX_GPU_DEVICE (long i) {
                    ParticleType& p = pstructs[i]; // Particle object that gets updated
                    Real x, y, z; // Temporary variables
#ifndef WARPX_DIM_RZ
                    GetPosition( x, y, z, p ); // Initialize x, y, z
                    UpdatePosition( x, y, z, ux[i], uy[i], uz[i], dt);
                    SetPosition( p, x, y, z ); // Update the object p
#else
                    // For WARPX_DIM_RZ, the particles are still pushed in 3D Cartesian
                    GetCartesianPositionFromCylindrical( x, y, z, p, theta[i] );
                    UpdatePosition( x, y, z, ux[i], uy[i], uz[i], dt);
                    SetCylindricalPositionFromCartesian( p, theta[i], x, y, z );
#endif
                }
            );

            if (cost) {
                const Box& tbx = pti.tilebox();
                wt = (amrex::second() - wt) / tbx.d_numPts();
                Array4<Real> const& costarr = cost->array(pti);
                amrex::ParallelFor(tbx,
                [=] AMREX_GPU_DEVICE (int i, int j, int k) noexcept
                {
                    costarr(i,j,k) += wt;
                });
            }
        }
    }
}

// This function is called in Redistribute, just after locate
void
WarpXParticleContainer::particlePostLocate(ParticleType& p,
                                           const ParticleLocData& pld,
                                           const int lev)
{
    // Tag particle if goes to higher level.
    // It will be split later in the loop
    if (pld.m_lev == lev+1
        and p.m_idata.id != NoSplitParticleID
        and p.m_idata.id >= 0)
    {
        p.m_idata.id = DoSplitParticleID;
    }
    // For the moment, do not do anything if particles goes
    // to lower level.
    if (pld.m_lev == lev-1){
    }
}