Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Pre- and Post-interpolation hook interface #2991

Merged
merged 2 commits into from
Oct 26, 2022
Merged

Pre- and Post-interpolation hook interface #2991

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d2455f6
Pre- and Post-interpolation hook interface
WeiqunZhang Oct 15, 2022
91f8a63
Because the pre-interp hook might modify the data, we need to make a …
WeiqunZhang Oct 26, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
69 changes: 30 additions & 39 deletions Src/AmrCore/AMReX_FillPatchUtil_I.H
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,31 @@

namespace amrex {

namespace detail {

template <typename F, typename MF>
auto call_interp_hook (F const& f, MF& mf, int icomp, int ncomp)
-> decltype(f(mf[0],Box(),icomp,ncomp))
{
#ifdef AMREX_USE_OMP
#pragma omp parallel if (Gpu::notInLaunchRegion())
#endif
for (MFIter mfi(mf); mfi.isValid(); ++mfi) {
auto& dfab = mf[mfi];
const Box& dbx = dfab.box();
f(dfab, dbx, icomp, ncomp);
}
}

template <typename F, typename MF>
auto call_interp_hook (F const& f, MF& mf, int icomp, int ncomp)
-> decltype(f(mf,icomp,ncomp))
{
f(mf, icomp, ncomp);
}

}

template <typename Interp>
bool ProperlyNested (const IntVect& ratio, const IntVect& blocking_factor, int ngrow,
const IndexType& boxType, Interp* mapper)
Expand Down Expand Up @@ -459,9 +484,6 @@ namespace {

if ( ! fpc.ba_crse_patch.empty())
{

using FAB = typename MF::FABType::value_type;

MF mf_crse_patch = make_mf_crse_patch<MF> (fpc, ncomp, mf.boxArray().ixType());
// Must make sure fine exists under needed coarse faces.
// It stores values for the final (interior) interpolation,
Expand Down Expand Up @@ -491,20 +513,12 @@ namespace {
solve_mask.setVal(1); // Values to solve.
solve_mask.setVal(0, mask_cpc, 0, 1); // Known values.

for (MFIter mfi(mf_refined_patch); mfi.isValid(); ++mfi)
{
FAB& sfab = mf_crse_patch[mfi];
pre_interp(sfab, sfab.box(), 0, ncomp);
}
detail::call_interp_hook(pre_interp, mf_crse_patch, 0, ncomp);

InterpFace(mapper, mf_crse_patch, 0, mf_refined_patch, 0, ncomp,
ratio, solve_mask, cgeom, fgeom, bcscomp, RunOn::Gpu, bcs);

for (MFIter mfi(mf_refined_patch); mfi.isValid(); ++mfi)
{
FAB& dfab = mf_refined_patch[mfi];
post_interp(dfab, dfab.box(), 0, ncomp);
}
detail::call_interp_hook(post_interp, mf_refined_patch, 0, ncomp);

bool aliasing = false;
for (auto const& fmf_a : fmf) {
Expand Down Expand Up @@ -538,30 +552,14 @@ namespace {

MF mf_fine_patch = make_mf_fine_patch<MF>(fpc, ncomp);

#ifdef AMREX_USE_OMP
#pragma omp parallel if (Gpu::notInLaunchRegion())
#endif
for (MFIter mfi(mf_crse_patch); mfi.isValid(); ++mfi)
{
auto& sfab = mf_crse_patch[mfi];
const Box& sbx = sfab.box();
pre_interp(sfab, sbx, 0, ncomp);
}
detail::call_interp_hook(pre_interp, mf_crse_patch, 0, ncomp);

FillPatchInterp(mf_fine_patch, 0, mf_crse_patch, 0,
ncomp, IntVect(0), cgeom, fgeom,
amrex::grow(amrex::convert(fgeom.Domain(),mf.ixType()),nghost),
ratio, mapper, bcs, bcscomp);

#ifdef AMREX_USE_OMP
#pragma omp parallel if (Gpu::notInLaunchRegion())
#endif
for (MFIter mfi(mf_fine_patch); mfi.isValid(); ++mfi)
{
auto& dfab = mf_fine_patch[mfi];
const Box& dbx = dfab.box();
post_interp(dfab, dbx, 0, ncomp);
}
detail::call_interp_hook(post_interp, mf_fine_patch, 0, ncomp);

mf.ParallelCopy(mf_fine_patch, 0, dcomp, ncomp, IntVect{0}, nghost);
}
Expand Down Expand Up @@ -1024,14 +1022,7 @@ InterpFromCoarseLevel (MF& mf, IntVect const& nghost, Real time,

cbc(mf_crse_patch, 0, ncomp, mf_crse_patch.nGrowVect(), time, cbccomp);

#ifdef AMREX_USE_OMP
#pragma omp parallel if (Gpu::notInLaunchRegion())
#endif
for (MFIter mfi(mf_crse_patch); mfi.isValid(); ++mfi)
{
FAB& sfab = mf_crse_patch[mfi];
pre_interp(sfab, sfab.box(), 0, ncomp);
}
detail::call_interp_hook(pre_interp, mf_crse_patch, 0, ncomp);

FillPatchInterp(mf, dcomp, mf_crse_patch, 0, ncomp, nghost, cgeom, fgeom, fdomain_g,
ratio, mapper, bcs, bcscomp);
Expand Down
31 changes: 15 additions & 16 deletions Src/AmrCore/AMReX_FillPatcher.H
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -330,26 +330,25 @@ FillPatcher<MF>::fillCoarseFineBoundary (MF& mf, IntVect const& nghost, Real tim
}
}

MF mf_crse_patch;
if (m_cf_crse_data_tmp == nullptr) {
m_cf_crse_data_tmp = std::make_unique<MF>
(make_mf_crse_patch<MF>(fpc, m_ncomp));
}

if (m_cf_crse_data.size() > 0 &&
amrex::almostEqual(time, m_cf_crse_data[0].first,5))
{
mf_crse_patch = MF(*m_cf_crse_data[0].second, amrex::make_alias,
scomp, ncomp);
amrex::Copy(*m_cf_crse_data_tmp, *m_cf_crse_data[0].second,
scomp, 0, ncomp, 0);
}
else if (m_cf_crse_data.size() > 1 &&
amrex::almostEqual(time, m_cf_crse_data[1].first,5))
{
mf_crse_patch = MF(*m_cf_crse_data[1].second, amrex::make_alias,
scomp, ncomp);
amrex::Copy(*m_cf_crse_data_tmp, *m_cf_crse_data[1].second,
scomp, 0, ncomp, 0);
}
else if (m_cf_crse_data.size() == 2)
{
if (m_cf_crse_data_tmp == nullptr) {
m_cf_crse_data_tmp = std::make_unique<MF>
(make_mf_crse_patch<MF>(fpc, m_ncomp));
}
mf_crse_patch = MF(*m_cf_crse_data_tmp, amrex::make_alias, scomp, ncomp);
int const ng_space_interp = 8; // Need to be big enough
Box domain = m_cgeom.growPeriodicDomain(ng_space_interp);
domain.convert(mf.ixType());
Expand All @@ -358,10 +357,10 @@ FillPatcher<MF>::fillCoarseFineBoundary (MF& mf, IntVect const& nghost, Real tim
Real alpha = (t1-time)/(t1-t0);
Real beta = (time-t0)/(t1-t0);
AMREX_ASSERT(alpha >= 0._rt && beta >= 0._rt);
auto const& a = mf_crse_patch.arrays();
auto const& a = m_cf_crse_data_tmp->arrays();
auto const& a0 = m_cf_crse_data[0].second->const_arrays();
auto const& a1 = m_cf_crse_data[1].second->const_arrays();
amrex::ParallelFor(mf_crse_patch, IntVect(0), ncomp,
amrex::ParallelFor(*m_cf_crse_data_tmp, IntVect(0), ncomp,
[=] AMREX_GPU_DEVICE (int bi, int i, int j, int k, int n) noexcept
{
if (domain.contains(i,j,k)) {
Expand All @@ -377,17 +376,17 @@ FillPatcher<MF>::fillCoarseFineBoundary (MF& mf, IntVect const& nghost, Real tim
amrex::Abort("FillPatcher: High order interpolation in time not supported. Or FillPatcher was not properly deleted.");
}

cbc(mf_crse_patch, 0, ncomp, nghost, time, cbccomp);
cbc(*m_cf_crse_data_tmp, 0, ncomp, nghost, time, cbccomp);

pre_interp(mf_crse_patch, 0, ncomp);
detail::call_interp_hook(pre_interp, *m_cf_crse_data_tmp, 0, ncomp);

FillPatchInterp(*m_cf_fine_data, scomp, mf_crse_patch, 0,
FillPatchInterp(*m_cf_fine_data, scomp, *m_cf_crse_data_tmp, 0,
ncomp, IntVect(0), m_cgeom, m_fgeom,
amrex::grow(amrex::convert(m_fgeom.Domain(),
mf.ixType()),nghost),
m_ratio, m_interp, bcs, bcscomp);

post_interp(*m_cf_fine_data, scomp, ncomp);
detail::call_interp_hook(post_interp, *m_cf_fine_data, scomp, ncomp);

mf.ParallelCopy(*m_cf_fine_data, scomp, dcomp, ncomp, IntVect{0}, nghost);
}
Expand Down