AMReX_FabSet.H

Go to the documentation of this file.

 
#ifndef AMREX_FABSET_H_
#define AMREX_FABSET_H_
#include <AMReX_Config.H>
 
#include <AMReX_FabDataType.H>
#include <AMReX_MultiFab.H>
#include <AMReX_ParallelDescriptor.H>
#include <AMReX_BLProfiler.H>
#include <AMReX_VisMF.H>
 
#ifdef AMREX_USE_OMP
#include <omp.h>
#endif
 
#include <limits>
 
namespace amrex {
 
template <typename MF>
class FabSetT
{
    friend class FabSetIter;
    friend class FluxRegister;
public:
    using value_type = typename FabDataType<MF>::value_type;
    using FAB        = typename FabDataType<MF>::fab_type;
 
    //
    FabSetT () noexcept = default;
    //
    FabSetT (const BoxArray& grids, const DistributionMapping& dmap, int ncomp);
 
    ~FabSetT () = default;
 
    FabSetT (FabSetT<MF>&& rhs) noexcept = default;
 
    FabSetT (const FabSetT<MF>& rhs) = delete;
    FabSetT<MF>& operator= (const FabSetT<MF>& rhs) = delete;
    FabSetT<MF>& operator= (FabSetT<MF>&& rhs) = delete;
 
    //
    void define (const BoxArray& grids, const DistributionMapping& dmap, int ncomp);
 
    FAB const& operator[] (const MFIter& mfi) const noexcept { return m_mf[mfi]; }
    FAB      & operator[] (const MFIter& mfi)       noexcept { return m_mf[mfi]; }
    FAB const& operator[] (int i)             const noexcept { return m_mf[i]; }
    FAB      & operator[] (int i)                   noexcept { return m_mf[i]; }
 
    Array4<value_type const> array (const MFIter& mfi) const noexcept { return m_mf.const_array(mfi); }
    Array4<value_type      > array (const MFIter& mfi)       noexcept { return m_mf.array(mfi); }
    Array4<value_type const> array (int i) const noexcept { return m_mf.const_array(i);   }
    Array4<value_type      > array (int i)       noexcept { return m_mf.array(i);   }
    Array4<value_type const> const_array (const MFIter& mfi) const noexcept { return m_mf.const_array(mfi); }
    Array4<value_type const> const_array (int i)             const noexcept { return m_mf.const_array(i); }
 
    MultiArray4<value_type const> arrays () const noexcept { return m_mf.const_arrays(); }
    MultiArray4<value_type      > arrays ()       noexcept { return m_mf.arrays(); }
    MultiArray4<value_type const> const_arrays () const noexcept { return m_mf.const_arrays(); }
 
    [[nodiscard]] Box fabbox (int K) const noexcept { return m_mf.fabbox(K); }
 
    [[nodiscard]] int size () const noexcept { return m_mf.size(); }
 
    [[nodiscard]] const BoxArray& boxArray () const noexcept { return m_mf.boxArray(); }
 
    [[nodiscard]] const DistributionMapping& DistributionMap () const noexcept
        { return m_mf.DistributionMap(); }
 
    [[nodiscard]] MF      & multiFab ()       noexcept { return m_mf; }
    [[nodiscard]] MF const& multiFab () const noexcept { return m_mf; }
 
    [[nodiscard]] int nComp () const noexcept { return m_mf.nComp(); }
 
    void clear () { m_mf.clear(); }
 
    FabSetT<MF>& copyFrom (const FabSetT<MF>& src, int scomp, int dcomp, int ncomp);
 
    FabSetT<MF>& copyFrom (const MF& src, int ngrow, int scomp, int dcomp, int ncomp,
                           const Periodicity& period = Periodicity::NonPeriodic());
 
    FabSetT<MF>& plusFrom (const FabSetT<MF>& src, int scomp, int dcomp, int ncomp);
 
    FabSetT<MF>& plusFrom (const MF& src, int ngrow, int scomp, int dcomp, int ncomp,
                           const Periodicity& period = Periodicity::NonPeriodic(),
                           bool deterministic = false);
 
    void copyTo (MF& dest, int ngrow, int scomp, int dcomp, int ncomp,
                 const Periodicity& period = Periodicity::NonPeriodic()) const;
 
    void plusTo (MF& dest, int ngrow, int scomp, int dcomp, int ncomp,
                 const Periodicity& period = Periodicity::NonPeriodic()) const;
 
    void setVal (value_type val);
 
    void setVal (value_type val, int comp, int num_comp);
 
    FabSetT<MF>& linComb (value_type a, value_type b, const FabSetT<MF>& src,
                          int scomp, int dcomp, int ncomp);
 
    FabSetT<MF>& linComb (value_type a, const MF& mfa, int a_comp,
                          value_type b, const MF& mfb, int b_comp,
                          int dcomp, int ncomp, int ngrow);
 
    //
    void write (const std::string& name) const;
    //
    void read (const std::string& name);
 
    static void Copy (FabSetT<MF>& dst, const FabSetT<MF>& src);
 
private:
    MF m_mf;
};
 
class FabSetIter
    : public MFIter
{
public:
    template <typename MF>
    explicit FabSetIter (const FabSetT<MF>& fs)
        : MFIter(fs.m_mf) { }
};
 
template <typename MF>
FabSetT<MF>::FabSetT (const BoxArray& grids, const DistributionMapping& dmap, int ncomp)
    : m_mf(grids,dmap,ncomp,0)
{}
 
template <typename MF>
void
FabSetT<MF>::define (const BoxArray& grids, const DistributionMapping& dm, int ncomp)
{
    m_mf.define(grids, dm, ncomp, 0);
}
 
template <typename MF>
FabSetT<MF>&
FabSetT<MF>::copyFrom (const FabSetT<MF>& src, int scomp, int dcomp, int ncomp)
{
    if (boxArray() == src.boxArray() && DistributionMap() == src.DistributionMap()) {
#ifdef AMREX_USE_OMP
#pragma omp parallel if (Gpu::notInLaunchRegion())
#endif
        for (FabSetIter fsi(*this); fsi.isValid(); ++fsi) {
            const Box& bx = fsi.validbox();
            auto const srcfab =   src.array(fsi);
            auto       dstfab = this->array(fsi);
            AMREX_HOST_DEVICE_PARALLEL_FOR_4D ( bx, ncomp, i, j, k, n,
            {
                dstfab(i,j,k,n+dcomp) = srcfab(i,j,k,n+scomp);
            });
        }
    } else {
        m_mf.ParallelCopy(src.m_mf,scomp,dcomp,ncomp);
    }
    return *this;
}
 
template <typename MF>
FabSetT<MF>&
FabSetT<MF>::copyFrom (const MF& src, int ngrow, int scomp, int dcomp, int ncomp,
                       const Periodicity& period)
{
    BL_ASSERT(boxArray() != src.boxArray());
    m_mf.ParallelCopy(src,scomp,dcomp,ncomp,ngrow,0,period);
    return *this;
}
 
template <typename MF>
FabSetT<MF>&
FabSetT<MF>::plusFrom (const FabSetT<MF>& src, int scomp, int dcomp, int ncomp)
{
    if (boxArray() == src.boxArray() && DistributionMap() == src.DistributionMap()) {
#ifdef AMREX_USE_OMP
#pragma omp parallel if (Gpu::notInLaunchRegion())
#endif
        for (FabSetIter fsi(*this); fsi.isValid(); ++fsi) {
            const Box& bx = fsi.validbox();
            auto const srcfab =   src.array(fsi);
            auto       dstfab = this->array(fsi);
            AMREX_HOST_DEVICE_PARALLEL_FOR_4D ( bx, ncomp, i, j, k, n,
            {
                dstfab(i,j,k,n+dcomp) += srcfab(i,j,k,n+scomp);
            });
        }
    } else {
        amrex::Abort("FabSetT<MF>::plusFrom: parallel plusFrom not supported");
    }
    return *this;
}
 
template <typename MF>
FabSetT<MF>&
FabSetT<MF>::plusFrom (const MF& src, int ngrow, int scomp, int dcomp, int ncomp,
                       const Periodicity& period, bool deterministic)
{
    BL_ASSERT(boxArray() != src.boxArray());
    m_mf.ParallelCopy(src,scomp,dcomp,ncomp,IntVect(ngrow),IntVect(0),period,
                      FabArrayBase::ADD,nullptr,deterministic);
    return *this;
}
 
template <typename MF>
void
FabSetT<MF>::copyTo (MF& dest, int ngrow, int scomp, int dcomp, int ncomp,
                     const Periodicity& period) const
{
    BL_ASSERT(boxArray() != dest.boxArray());
    dest.ParallelCopy(m_mf,scomp,dcomp,ncomp,0,ngrow,period);
}
 
template <typename MF>
void
FabSetT<MF>::plusTo (MF& dest, int ngrow, int scomp, int dcomp, int ncomp,
                     const Periodicity& period) const
{
    BL_ASSERT(boxArray() != dest.boxArray());
    dest.ParallelCopy(m_mf,scomp,dcomp,ncomp,0,ngrow,period,FabArrayBase::ADD);
}
 
template <typename MF>
void
FabSetT<MF>::setVal (value_type val)
{
    const int ncomp = nComp();
#ifdef AMREX_USE_OMP
#pragma omp parallel if (Gpu::notInLaunchRegion())
#endif
    for (FabSetIter fsi(*this); fsi.isValid(); ++fsi) {
        const Box& bx = fsi.validbox();
        auto fab = this->array(fsi);
        AMREX_HOST_DEVICE_PARALLEL_FOR_4D ( bx, ncomp, i, j, k, n,
        {
            fab(i,j,k,n) = val;
        });
    }
}
 
template <typename MF>
void
FabSetT<MF>::setVal (value_type val, int comp, int num_comp)
{
#ifdef AMREX_USE_OMP
#pragma omp parallel if (Gpu::notInLaunchRegion())
#endif
    for (FabSetIter fsi(*this); fsi.isValid(); ++fsi) {
        const Box& bx = fsi.validbox();
        auto fab = this->array(fsi);
        AMREX_HOST_DEVICE_PARALLEL_FOR_4D ( bx, num_comp, i, j, k, n,
        {
            fab(i,j,k,n+comp) = val;
        });
    }
}
 
// Linear combination this := a*this + b*src
// Note: corresponding fabsets must be commensurate.
template <typename MF>
FabSetT<MF>&
FabSetT<MF>::linComb (value_type a, value_type b, const FabSetT<MF>& src,
                      int scomp, int dcomp, int ncomp)
{
    BL_ASSERT(size() == src.size());
 
#ifdef AMREX_USE_OMP
#pragma omp parallel if (Gpu::notInLaunchRegion())
#endif
    for (FabSetIter fsi(*this); fsi.isValid(); ++fsi)
    {
        const Box& bx = fsi.validbox();
        auto const srcfab =   src.array(fsi);
        auto       dstfab = this->array(fsi);
        AMREX_HOST_DEVICE_PARALLEL_FOR_4D ( bx, ncomp, i, j, k, n,
        {
            dstfab(i,j,k,n+dcomp) = a*dstfab(i,j,k,n+dcomp) + b*srcfab(i,j,k,n+scomp);
        });
    }
    return *this;
}
 
// Linear combination: this := a*mfa + b*mfb
// CastroRadiation is the only code that uses this function.
template <typename MF>
FabSetT<MF>&
FabSetT<MF>::linComb (value_type a, const MF& mfa, int a_comp,
                      value_type b, const MF& mfb, int b_comp,
                      int dcomp, int ncomp, int ngrow)
{
    BL_PROFILE("FabSetT<MF>::linComb()");
    BL_ASSERT(mfa.nGrowVect().allGE(ngrow) && mfb.nGrowVect().allGE(ngrow));
    BL_ASSERT(mfa.boxArray() == mfb.boxArray());
    BL_ASSERT(boxArray() != mfa.boxArray());
 
    MF bdrya(boxArray(),DistributionMap(),ncomp,0,MFInfo());
    MF bdryb(boxArray(),DistributionMap(),ncomp,0,MFInfo());
 
    const auto huge = static_cast<value_type>(sizeof(value_type) == 8 ? 1.e200 : 1.e30);
 
#ifdef AMREX_USE_OMP
#pragma omp parallel if (Gpu::notInLaunchRegion())
#endif
    for (MFIter mfi(bdrya); mfi.isValid(); ++mfi) // tiling is not safe for this BoxArray
    {
        const Box& bx = mfi.validbox();
        auto afab = bdrya.array(mfi);
        auto bfab = bdryb.array(mfi);
        AMREX_HOST_DEVICE_PARALLEL_FOR_4D ( bx, ncomp, i, j, k, n,
        {
            afab(i,j,k,n) = huge;
            bfab(i,j,k,n) = huge;
        });
    }
 
    bdrya.ParallelCopy(mfa,a_comp,0,ncomp,ngrow,0);
    bdryb.ParallelCopy(mfb,b_comp,0,ncomp,ngrow,0);
 
#ifdef AMREX_USE_OMP
#pragma omp parallel if (Gpu::notInLaunchRegion())
#endif
    for (FabSetIter fsi(*this); fsi.isValid(); ++fsi)
    {
        const Box& bx = fsi.validbox();
        auto const afab = bdrya.array(fsi);
        auto const bfab = bdryb.array(fsi);
        auto       dfab = this->array(fsi);
        AMREX_HOST_DEVICE_PARALLEL_FOR_4D ( bx, ncomp, i, j, k, n,
        {
            dfab(i,j,k,n+dcomp) = a*afab(i,j,k,n) + b*bfab(i,j,k,n);
        });
    }
 
    return *this;
}
 
template <typename MF>
void
FabSetT<MF>::write (const std::string& name) const
{
    if (AsyncOut::UseAsyncOut()) {
        VisMF::AsyncWrite(m_mf,name);
    } else {
        VisMF::Write(m_mf,name);
    }
}
 
template <typename MF>
void
FabSetT<MF>::read (const std::string& name)
{
    if (m_mf.empty()) {
        amrex::Abort("FabSetT<MF>::read: not predefined");
    }
    VisMF::Read(m_mf,name);
}
 
template <typename MF>
void
FabSetT<MF>::Copy (FabSetT<MF>& dst, const FabSetT<MF>& src)
{
    BL_ASSERT(amrex::match(dst.boxArray(), src.boxArray()));
    BL_ASSERT(dst.DistributionMap() == src.DistributionMap());
    int ncomp = dst.nComp();
#ifdef AMREX_USE_OMP
#pragma omp parallel if (Gpu::notInLaunchRegion())
#endif
    for (FabSetIter fsi(dst); fsi.isValid(); ++fsi) {
        const Box& bx = fsi.validbox();
        auto const srcfab = src.array(fsi);
        auto       dstfab = dst.array(fsi);
        AMREX_HOST_DEVICE_PARALLEL_FOR_4D ( bx, ncomp, i, j, k, n,
        {
            dstfab(i,j,k,n) = srcfab(i,j,k,n);
        });
    }
}
 
using FabSet = FabSetT<MultiFab>;
using fFabSet = FabSetT<fMultiFab>;
 
}
 
#endif /*_FABSET_H_*/