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

Flip rrtmgp to default to kokkos #2987

Merged
merged 5 commits into from
Sep 24, 2024
Merged

Flip rrtmgp to default to kokkos #2987

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
86b6965
Update rrtmgp branch
jgfouca Aug 29, 2024
5afe73d
Set kokkos to be the default rrtmgp backend
jgfouca Aug 29, 2024
4c3e575
Allow option to force rrtmgp inputs to be layoutleft
jgfouca Aug 29, 2024
06ab547
Merge remote-tracking branch 'origin/master' into jgfouca/flip_rrtmgp…
jgfouca Sep 6, 2024
c346b79
Test fix
jgfouca Sep 6, 2024
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
2 changes: 1 addition & 1 deletion components/eam/src/physics/rrtmgp/external
Open in desktop
Submodule external updated 24 files
+12 −12 cpp/examples/all-sky/mo_garand_atmos_io.cpp
+12 −12 cpp/examples/all-sky/mo_garand_atmos_io.h
+20 −20 cpp/extensions/cloud_optics/mo_cloud_optics.h
+4 −4 cpp/extensions/fluxes_byband/mo_fluxes_byband_kernels.cpp
+4 −4 cpp/extensions/fluxes_byband/mo_fluxes_byband_kernels.h
+34 −38 cpp/rrtmgp/kernels/mo_gas_optics_kernels.cpp
+125 −80 cpp/rrtmgp/kernels/mo_gas_optics_kernels.h
+12 −16 cpp/rrtmgp/kernels/mo_rrtmgp_util_reorder_kernels.h
+22 −22 cpp/rrtmgp/mo_gas_concentrations.h
+42 −42 cpp/rrtmgp/mo_gas_optics_rrtmgp.h
+0 −19 cpp/rrtmgp_const.h
+91 −8 cpp/rrtmgp_conversion.h
+2 −2 cpp/rte/expand_and_transpose.cpp
+2 −2 cpp/rte/expand_and_transpose.h
+8 −8 cpp/rte/kernels/mo_fluxes_broadband_kernels.cpp
+8 −8 cpp/rte/kernels/mo_fluxes_broadband_kernels.h
+46 −46 cpp/rte/kernels/mo_optical_props_kernels.cpp
+12 −12 cpp/rte/kernels/mo_optical_props_kernels.h
+46 −46 cpp/rte/kernels/mo_rte_solver_kernels.cpp
+65 −56 cpp/rte/kernels/mo_rte_solver_kernels.h
+24 −24 cpp/rte/mo_optical_props.h
+4 −4 cpp/rte/mo_rte_lw.h
+15 −0 cpp/test/build/reanalyze.sh
+204 −0 cpp/test/build/rrtmgp-perf-analysis
4 changes: 2 additions & 2 deletions components/eamxx/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -212,8 +212,8 @@ endif()
# #cmakedefine RRTMGP_EXPENSIVE_CHECKS
option (SCREAM_RRTMGP_DEBUG "Turn on extra debug checks in RRTMGP" ${SCREAM_DEBUG})

option(SCREAM_RRTMGP_ENABLE_YAKL "Use YAKL under rrtmgp" TRUE)
option(SCREAM_RRTMGP_ENABLE_KOKKOS "Use Kokkos under rrtmgp" FALSE)
option(SCREAM_RRTMGP_ENABLE_YAKL "Use YAKL under rrtmgp" FALSE)
option(SCREAM_RRTMGP_ENABLE_KOKKOS "Use Kokkos under rrtmgp" TRUE)
if (SCREAM_RRTMGP_ENABLE_YAKL)
add_definitions("-DRRTMGP_ENABLE_YAKL")
endif()
Expand Down
260 changes: 146 additions & 114 deletions components/eamxx/src/physics/rrtmgp/eamxx_rrtmgp_process_interface.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,48 @@ using KT = KokkosTypes<DefaultDevice>;
using ExeSpace = KT::ExeSpace;
using MemberType = KT::MemberType;

namespace {

struct ConvertToRrtmgpSubview
{
int beg;
int ncol;

template <typename View>
View subview1d(const View& v) const {
return View(v, std::make_pair(beg, beg+ncol));
}

template <typename View>
View subview2d_impl(const View& v, const int inner_dim) const {
return View(v, std::make_pair(beg, beg+ncol), std::make_pair(0, inner_dim));
}

#ifdef RRTMGP_LAYOUT_LEFT
template <typename FieldView, typename BufferView>
BufferView subview2d(const FieldView&, const BufferView& buffer_view, const int inner_dim) const {
return BufferView(buffer_view, std::make_pair(0, ncol), Kokkos::ALL);
}
#else
// Be sure to trim the excess
// items from the field manager views due to simd packs. If we don't trim, then
// check_range_k will fail due to looking at unused values. Once rrtmgp can handle
// packs, this won't be necessary.
template <typename FieldView, typename BufferView>
FieldView subview2d(const FieldView& field_view, const BufferView&, const int inner_dim) const {
return subview2d_impl(field_view, inner_dim);
}
#endif

template <typename View>
View subview3d(const View& v) const {
// The range assumes these are buffer views, not fields
return View(v, std::make_pair(0, ncol), Kokkos::ALL, Kokkos::ALL);
}
};

}

RRTMGPRadiation::
RRTMGPRadiation (const ekat::Comm& comm, const ekat::ParameterList& params)
: AtmosphereProcess(comm, params)
Expand Down Expand Up @@ -255,7 +297,12 @@ size_t RRTMGPRadiation::requested_buffer_size_in_bytes() const
Buffer::num_3d_nlay_nswbands*m_col_chunk_size*(m_nlay)*m_nswbands +
Buffer::num_3d_nlay_nlwbands*m_col_chunk_size*(m_nlay)*m_nlwbands +
Buffer::num_3d_nlay_nswgpts*m_col_chunk_size*(m_nlay)*m_nswgpts +
Buffer::num_3d_nlay_nlwgpts*m_col_chunk_size*(m_nlay)*m_nlwgpts;
Buffer::num_3d_nlay_nlwgpts*m_col_chunk_size*(m_nlay)*m_nlwgpts *
#if defined(RRTMGP_ENABLE_YAKL) && defined(RRTMGP_ENABLE_KOKKOS)
2;
#else
1;
#endif

return interface_request * sizeof(Real);
} // RRTMGPRadiation::requested_buffer_size
Expand Down Expand Up @@ -409,13 +456,7 @@ void RRTMGPRadiation::init_buffers(const ATMBufferManager &buffer_manager)
mem += m_buffer.cld_tau_lw_bnd.totElems();
#endif

// During the transition to kokkos, the buffer views/arrays will point to the same memory,
// so care is needed to avoid repeating calculations in such a way that answers change.
// Example: buff_view(x) += foo;
// Stuff like this cannot be done twice when both kokkos and yakl are enabled
#ifdef RRTMGP_ENABLE_KOKKOS
mem = reinterpret_cast<Real*>(buffer_manager.get_memory());

// 1d arrays
m_buffer.mu0_k = decltype(m_buffer.mu0_k)(mem, m_col_chunk_size);
mem += m_buffer.mu0_k.size();
Expand Down Expand Up @@ -847,7 +888,11 @@ void RRTMGPRadiation::run_impl (const double dt) {
this->log(LogLevel::debug,
"[RRTMGP::run_impl] Col chunk beg,end: " + std::to_string(beg) + ", " + std::to_string(beg+ncol) + "\n");


// d_tint and d_dz are used in eamxx calls and therefore
// must be layout right
ulrreal2dk d_tint = ulrreal2dk(m_buffer.d_tint.data(), m_col_chunk_size, m_nlay+1);
ulrreal2dk d_dz = ulrreal2dk(m_buffer.d_dz.data(), m_col_chunk_size, m_nlay);
auto d_mu0 = m_buffer.cosine_zenith;
#ifdef RRTMGP_ENABLE_YAKL
// Create YAKL arrays. RRTMGP expects YAKL arrays with styleFortran, i.e., data has ncol
// as the fastest index. For this reason we must copy the data.
Expand Down Expand Up @@ -920,123 +965,68 @@ void RRTMGPRadiation::run_impl (const double dt) {
auto cld_tau_lw_gpt = subview_3d(m_buffer.cld_tau_lw_gpt);
#endif
#ifdef RRTMGP_ENABLE_KOKKOS
// If YAKL is on, we don't want aliased memory in both the yakl and kokos
// subviews, so make new views and deep_copy. Also, be sure to trim the excess
// items from the field manager views due to simd packs. If we don't trim, then
// check_range_k will fail due to looking at unused values. Once rrtmgp can handle
// packs, this won't be necessary.
auto subview_1dk = [&](const ureal1dk& v) -> ureal1dk {
ureal1dk subv(v, std::make_pair(beg, beg+ncol));
#ifdef RRTMGP_ENABLE_YAKL
real1dk rv(v.label(), ncol);
Kokkos::deep_copy(rv, subv);
return rv;
#else
return subv;
#endif
};
auto subview_1dkc = [&](const cureal1dk& v) -> cureal1dk {
cureal1dk subv(v, std::make_pair(beg, beg+ncol));
#ifdef RRTMGP_ENABLE_YAKL
creal1dk rv(v.label(), ncol);
Kokkos::deep_copy(rv, subv);
return rv;
#else
return subv;
#endif
};
auto subview_2dk = [&](const ureal2dk& v, const int inner_dim) -> ureal2dk {
ureal2dk subv(v, std::make_pair(beg, beg+ncol), std::make_pair(0, inner_dim));
#ifdef RRTMGP_ENABLE_YAKL
real2dk rv(v.label(), ncol, v.extent(1));
Kokkos::deep_copy(rv, subv);
return rv;
#else
return subv;
#endif
};
auto subview_2dkc = [&](const cureal2dk& v, const int inner_dim) -> cureal2dk {
cureal2dk subv(v, std::make_pair(beg, beg+ncol), std::make_pair(0, inner_dim));
#ifdef RRTMGP_ENABLE_YAKL
creal2dk rv(v.label(), ncol, v.extent(1));
Kokkos::deep_copy(rv, subv);
return rv;
#else
return subv;
#endif
};
auto subview_3dk = [&](const ureal3dk& v) -> ureal3dk {
ureal3dk subv(v, std::make_pair(0, ncol), Kokkos::ALL, Kokkos::ALL); // The range assumes these are buffer views, not fields
#ifdef RRTMGP_ENABLE_YAKL
real3dk rv(v.label(), ncol, v.extent(1), v.extent(2));
Kokkos::deep_copy(rv, subv);
return rv;
#else
return subv;
#endif
};
ConvertToRrtmgpSubview conv = {beg, ncol};

// Note, ncol will not necessary be m_col_chunk_size because the number of cols
// will not always be evenly divided by m_col_chunk_size. In most cases, the
// extra space will not cause any problems, but it does sometimes.
auto p_lay_k = subview_2dkc(d_pmid, m_nlay);
auto t_lay_k = subview_2dkc(d_tmid, m_nlay);
auto p_lev_k = subview_2dkc(d_pint, m_nlay+1);
auto p_del_k = subview_2dkc(d_pdel, m_nlay);
auto p_lay_k = conv.subview2d(d_pmid, m_buffer.p_lay_k, m_nlay);
auto t_lay_k = conv.subview2d(d_tmid, m_buffer.t_lay_k, m_nlay);
auto p_lev_k = conv.subview2d(d_pint, m_buffer.p_lev_k, m_nlay+1);
auto z_del_k = conv.subview2d(d_dz, m_buffer.z_del_k, m_nlay);
auto p_del_k = conv.subview2d(d_pdel, m_buffer.p_del_k, m_nlay);
auto t_lev_k = conv.subview2d(d_tint, m_buffer.t_lev_k, m_nlay+1);
auto sfc_alb_dir_k = m_buffer.sfc_alb_dir_k;
auto sfc_alb_dif_k = m_buffer.sfc_alb_dif_k;
auto sfc_alb_dir_vis_k = subview_1dkc(d_sfc_alb_dir_vis);
auto sfc_alb_dir_nir_k = subview_1dkc(d_sfc_alb_dir_nir);
auto sfc_alb_dif_vis_k = subview_1dkc(d_sfc_alb_dif_vis);
auto sfc_alb_dif_nir_k = subview_1dkc(d_sfc_alb_dif_nir);
auto qc_k = subview_2dkc(d_qc, m_nlay);
auto nc_k = subview_2dkc(d_nc, m_nlay);
auto qi_k = subview_2dkc(d_qi, m_nlay);
auto sfc_alb_dir_vis_k = conv.subview1d(d_sfc_alb_dir_vis);
auto sfc_alb_dir_nir_k = conv.subview1d(d_sfc_alb_dir_nir);
auto sfc_alb_dif_vis_k = conv.subview1d(d_sfc_alb_dif_vis);
auto sfc_alb_dif_nir_k = conv.subview1d(d_sfc_alb_dif_nir);
auto qc_k = conv.subview2d(d_qc, m_buffer.qc_k, m_nlay);
auto nc_k = conv.subview2d(d_nc, m_buffer.nc_k, m_nlay);
auto qi_k = conv.subview2d(d_qi, m_buffer.qi_k, m_nlay);
auto cldfrac_tot_k = m_buffer.cldfrac_tot_k;
auto rel_k = subview_2dkc(d_rel, m_nlay);
auto rei_k = subview_2dkc(d_rei, m_nlay);
auto sw_flux_up_k = subview_2dk(d_sw_flux_up, m_nlay+1);
auto sw_flux_dn_k = subview_2dk(d_sw_flux_dn, m_nlay+1);
auto sw_flux_dn_dir_k = subview_2dk(d_sw_flux_dn_dir, m_nlay+1);
auto lw_flux_up_k = subview_2dk(d_lw_flux_up, m_nlay+1);
auto lw_flux_dn_k = subview_2dk(d_lw_flux_dn, m_nlay+1);
auto sw_clnclrsky_flux_up_k = subview_2dk(d_sw_clnclrsky_flux_up, m_nlay+1);
auto sw_clnclrsky_flux_dn_k = subview_2dk(d_sw_clnclrsky_flux_dn, m_nlay+1);
auto sw_clnclrsky_flux_dn_dir_k = subview_2dk(d_sw_clnclrsky_flux_dn_dir, m_nlay+1);
auto sw_clrsky_flux_up_k = subview_2dk(d_sw_clrsky_flux_up, m_nlay+1);
auto sw_clrsky_flux_dn_k = subview_2dk(d_sw_clrsky_flux_dn, m_nlay+1);
auto sw_clrsky_flux_dn_dir_k = subview_2dk(d_sw_clrsky_flux_dn_dir, m_nlay+1);
auto sw_clnsky_flux_up_k = subview_2dk(d_sw_clnsky_flux_up, m_nlay+1);
auto sw_clnsky_flux_dn_k = subview_2dk(d_sw_clnsky_flux_dn, m_nlay+1);
auto sw_clnsky_flux_dn_dir_k = subview_2dk(d_sw_clnsky_flux_dn_dir, m_nlay+1);
auto lw_clnclrsky_flux_up_k = subview_2dk(d_lw_clnclrsky_flux_up, m_nlay+1);
auto lw_clnclrsky_flux_dn_k = subview_2dk(d_lw_clnclrsky_flux_dn, m_nlay+1);
auto lw_clrsky_flux_up_k = subview_2dk(d_lw_clrsky_flux_up, m_nlay+1);
auto lw_clrsky_flux_dn_k = subview_2dk(d_lw_clrsky_flux_dn, m_nlay+1);
auto lw_clnsky_flux_up_k = subview_2dk(d_lw_clnsky_flux_up, m_nlay+1);
auto lw_clnsky_flux_dn_k = subview_2dk(d_lw_clnsky_flux_dn, m_nlay+1);
auto rel_k = conv.subview2d(d_rel, m_buffer.eff_radius_qc_k, m_nlay);
auto rei_k = conv.subview2d(d_rei, m_buffer.eff_radius_qi_k, m_nlay);
auto sw_flux_up_k = conv.subview2d(d_sw_flux_up, m_buffer.sw_flux_up_k, m_nlay+1);
auto sw_flux_dn_k = conv.subview2d(d_sw_flux_dn, m_buffer.sw_flux_dn_k, m_nlay+1);
auto sw_flux_dn_dir_k = conv.subview2d(d_sw_flux_dn_dir, m_buffer.sw_flux_dn_dir_k, m_nlay+1);
auto lw_flux_up_k = conv.subview2d(d_lw_flux_up, m_buffer.lw_flux_up_k, m_nlay+1);
auto lw_flux_dn_k = conv.subview2d(d_lw_flux_dn, m_buffer.lw_flux_dn_k, m_nlay+1);
auto sw_clnclrsky_flux_up_k = conv.subview2d(d_sw_clnclrsky_flux_up, m_buffer.sw_clnclrsky_flux_up_k, m_nlay+1);
auto sw_clnclrsky_flux_dn_k = conv.subview2d(d_sw_clnclrsky_flux_dn, m_buffer.sw_clnclrsky_flux_dn_k, m_nlay+1);
auto sw_clnclrsky_flux_dn_dir_k = conv.subview2d(d_sw_clnclrsky_flux_dn_dir, m_buffer.sw_clnclrsky_flux_dn_dir_k, m_nlay+1);
auto sw_clrsky_flux_up_k = conv.subview2d(d_sw_clrsky_flux_up, m_buffer.sw_clrsky_flux_up_k, m_nlay+1);
auto sw_clrsky_flux_dn_k = conv.subview2d(d_sw_clrsky_flux_dn, m_buffer.sw_clrsky_flux_dn_k, m_nlay+1);
auto sw_clrsky_flux_dn_dir_k = conv.subview2d(d_sw_clrsky_flux_dn_dir, m_buffer.sw_clrsky_flux_dn_dir_k, m_nlay+1);
auto sw_clnsky_flux_up_k = conv.subview2d(d_sw_clnsky_flux_up, m_buffer.sw_clnsky_flux_up_k, m_nlay+1);
auto sw_clnsky_flux_dn_k = conv.subview2d(d_sw_clnsky_flux_dn, m_buffer.sw_clnsky_flux_dn_k, m_nlay+1);
auto sw_clnsky_flux_dn_dir_k = conv.subview2d(d_sw_clnsky_flux_dn_dir, m_buffer.sw_clnsky_flux_dn_dir_k, m_nlay+1);
auto lw_clnclrsky_flux_up_k = conv.subview2d(d_lw_clnclrsky_flux_up, m_buffer.lw_clnclrsky_flux_up_k, m_nlay+1);
auto lw_clnclrsky_flux_dn_k = conv.subview2d(d_lw_clnclrsky_flux_dn, m_buffer.lw_clnclrsky_flux_dn_k, m_nlay+1);
auto lw_clrsky_flux_up_k = conv.subview2d(d_lw_clrsky_flux_up, m_buffer.lw_clrsky_flux_up_k, m_nlay+1);
auto lw_clrsky_flux_dn_k = conv.subview2d(d_lw_clrsky_flux_dn, m_buffer.lw_clrsky_flux_dn_k, m_nlay+1);
auto lw_clnsky_flux_up_k = conv.subview2d(d_lw_clnsky_flux_up, m_buffer.lw_clnsky_flux_up_k, m_nlay+1);
auto lw_clnsky_flux_dn_k = conv.subview2d(d_lw_clnsky_flux_dn, m_buffer.lw_clnsky_flux_dn_k, m_nlay+1);
auto sw_bnd_flux_up_k = m_buffer.sw_bnd_flux_up_k;
auto sw_bnd_flux_dn_k = m_buffer.sw_bnd_flux_dn_k;
auto sw_bnd_flux_dir_k = m_buffer.sw_bnd_flux_dir_k;
auto sw_bnd_flux_dif_k = m_buffer.sw_bnd_flux_dif_k;
auto lw_bnd_flux_up_k = m_buffer.lw_bnd_flux_up_k;
auto lw_bnd_flux_dn_k = m_buffer.lw_bnd_flux_dn_k;
auto sfc_flux_dir_vis_k = subview_1dk(d_sfc_flux_dir_vis);
auto sfc_flux_dir_nir_k = subview_1dk(d_sfc_flux_dir_nir);
auto sfc_flux_dif_vis_k = subview_1dk(d_sfc_flux_dif_vis);
auto sfc_flux_dif_nir_k = subview_1dk(d_sfc_flux_dif_nir);
auto sfc_flux_dir_vis_k = conv.subview1d(d_sfc_flux_dir_vis);
auto sfc_flux_dir_nir_k = conv.subview1d(d_sfc_flux_dir_nir);
auto sfc_flux_dif_vis_k = conv.subview1d(d_sfc_flux_dif_vis);
auto sfc_flux_dif_nir_k = conv.subview1d(d_sfc_flux_dif_nir);
auto aero_tau_sw_k = m_buffer.aero_tau_sw_k;
auto aero_ssa_sw_k = m_buffer.aero_ssa_sw_k;
auto aero_g_sw_k = m_buffer.aero_g_sw_k;
auto aero_tau_lw_k = m_buffer.aero_tau_lw_k;
auto cld_tau_sw_bnd_k = subview_3dk(m_buffer.cld_tau_sw_bnd_k);
auto cld_tau_lw_bnd_k = subview_3dk(m_buffer.cld_tau_lw_bnd_k);
auto cld_tau_sw_gpt_k = subview_3dk(m_buffer.cld_tau_sw_gpt_k);
auto cld_tau_lw_gpt_k = subview_3dk(m_buffer.cld_tau_lw_gpt_k);
auto cld_tau_sw_bnd_k = conv.subview3d(m_buffer.cld_tau_sw_bnd_k);
auto cld_tau_lw_bnd_k = conv.subview3d(m_buffer.cld_tau_lw_bnd_k);
auto cld_tau_sw_gpt_k = conv.subview3d(m_buffer.cld_tau_sw_gpt_k);
auto cld_tau_lw_gpt_k = conv.subview3d(m_buffer.cld_tau_lw_gpt_k);
#endif
auto d_tint = m_buffer.d_tint;
auto d_dz = m_buffer.d_dz;
auto d_mu0 = m_buffer.cosine_zenith;

// Set gas concs to "view" only the first ncol columns
#ifdef RRTMGP_ENABLE_YAKL
Expand All @@ -1045,7 +1035,7 @@ void RRTMGPRadiation::run_impl (const double dt) {
#endif
#ifdef RRTMGP_ENABLE_KOKKOS
m_gas_concs_k.ncol = ncol;
m_gas_concs_k.concs = subview_3dk(gas_concs_k);
m_gas_concs_k.concs = conv.subview3d(gas_concs_k);
#endif

// Copy data from the FieldManager to the YAKL arrays
Expand Down Expand Up @@ -1156,6 +1146,26 @@ void RRTMGPRadiation::run_impl (const double dt) {
}
#endif
#ifdef RRTMGP_ENABLE_KOKKOS
#ifdef RRTMGP_LAYOUT_LEFT
// Copy to layout left buffer views
Kokkos::parallel_for(Kokkos::TeamVectorRange(team, nlay), [&] (const int& k) {
p_lay_k(i,k) = d_pmid(icol,k);
t_lay_k(i,k) = d_tmid(icol,k);
z_del_k(i,k) = d_dz(i,k);
p_del_k(i,k) = d_pdel(icol,k);
qc_k(i,k) = d_qc(icol,k);
nc_k(i,k) = d_nc(icol,k);
qi_k(i,k) = d_qi(icol,k);
rel_k(i,k) = d_rel(icol,k);
rei_k(i,k) = d_rei(icol,k);
p_lev_k(i,k) = d_pint(icol,k);
t_lev_k(i,k) = d_tint(i,k);
});

p_lev_k(i,nlay) = d_pint(icol,nlay);
t_lev_k(i,nlay) = d_tint(i,nlay);
#endif

// Note that RRTMGP expects ordering (col,lay,bnd) but the FM keeps things in (col,bnd,lay) order
if (do_aerosol_rad) {
Kokkos::parallel_for(Kokkos::TeamVectorRange(team, nswbands*nlay), [&] (const int&idx) {
Expand Down Expand Up @@ -1215,7 +1225,7 @@ void RRTMGPRadiation::run_impl (const double dt) {
auto f = name=="o3" ? get_field_in(full_name) : get_field_out(full_name);
auto d_vmr = f.get_view<const Real**>();
#ifdef RRTMGP_ENABLE_KOKKOS
auto tmp2d_k = subview_2dkc(d_vmr, m_nlay);
auto tmp2d_k = conv.subview2d_impl(d_vmr, m_nlay);
#endif

#ifdef RRTMGP_ENABLE_YAKL
Expand Down Expand Up @@ -1328,11 +1338,8 @@ void RRTMGPRadiation::run_impl (const double dt) {
iwp(i+1,k+1) *= 1e3;
#endif
#ifdef RRTMGP_ENABLE_KOKKOS
#ifndef RRTMGP_ENABLE_YAKL
// lwp and lwp_k point to the same memory
lwp_k(i,k) *= 1e3;
iwp_k(i,k) *= 1e3;
#endif
#endif
});
});
Expand Down Expand Up @@ -1385,7 +1392,7 @@ void RRTMGPRadiation::run_impl (const double dt) {
#ifdef RRTMGP_ENABLE_KOKKOS
interface_t::rrtmgp_main(
ncol, m_nlay,
p_lay_k, t_lay_k, p_lev_k, d_tint,
p_lay_k, t_lay_k, p_lev_k, t_lev_k,
m_gas_concs_k,
sfc_alb_dir_k, sfc_alb_dif_k, d_mu0,
lwp_k, iwp_k, rel_k, rei_k, cldfrac_tot_k,
Expand Down Expand Up @@ -1592,7 +1599,7 @@ void RRTMGPRadiation::run_impl (const double dt) {
real1dk eff_radius_qi_at_cldtop_k (d_eff_radius_qi_at_cldtop.data() + m_col_chunk_beg[ic], ncol);

interface_t::compute_aerocom_cloudtop(
ncol, nlay, t_lay_k, p_lay_k, p_del_k, d_dz, qc_k, qi_k, rel_k, rei_k, cldfrac_tot_k,
ncol, nlay, t_lay_k, p_lay_k, p_del_k, z_del_k, qc_k, qi_k, rel_k, rei_k, cldfrac_tot_k,
nc_k, T_mid_at_cldtop_k, p_mid_at_cldtop_k, cldfrac_ice_at_cldtop_k,
cldfrac_liq_at_cldtop_k, cldfrac_tot_at_cldtop_k, cdnc_at_cldtop_k,
eff_radius_qc_at_cldtop_k, eff_radius_qi_at_cldtop_k);
Expand Down Expand Up @@ -1658,6 +1665,31 @@ void RRTMGPRadiation::run_impl (const double dt) {
const int icol = i + beg;
d_sfc_flux_sw_net(icol) = sw_flux_dn_k(i,kbot_k) - sw_flux_up_k(i,kbot_k);
d_sfc_flux_lw_dn(icol) = lw_flux_dn_k(i,kbot_k);
#ifdef RRTMGP_LAYOUT_LEFT
// Copy from layout left buffer views back to layout right fields
Kokkos::parallel_for(Kokkos::TeamVectorRange(team, nlay+1), [&] (const int& k) {
d_sw_flux_up(icol,k) = sw_flux_up_k(i,k);
d_sw_flux_dn(icol,k) = sw_flux_dn_k(i,k);
d_sw_flux_dn_dir(icol,k) = sw_flux_dn_dir_k(i,k);
d_lw_flux_up(icol,k) = lw_flux_up_k(i,k);
d_lw_flux_dn(icol,k) = lw_flux_dn_k(i,k);
d_sw_clnclrsky_flux_up(icol,k) = sw_clnclrsky_flux_up_k(i,k);
d_sw_clnclrsky_flux_dn(icol,k) = sw_clnclrsky_flux_dn_k(i,k);
d_sw_clnclrsky_flux_dn_dir(icol,k) = sw_clnclrsky_flux_dn_dir_k(i,k);
d_sw_clrsky_flux_up(icol,k) = sw_clrsky_flux_up_k(i,k);
d_sw_clrsky_flux_dn(icol,k) = sw_clrsky_flux_dn_k(i,k);
d_sw_clrsky_flux_dn_dir(icol,k) = sw_clrsky_flux_dn_dir_k(i,k);
d_sw_clnsky_flux_up(icol,k) = sw_clnsky_flux_up_k(i,k);
d_sw_clnsky_flux_dn(icol,k) = sw_clnsky_flux_dn_k(i,k);
d_sw_clnsky_flux_dn_dir(icol,k) = sw_clnsky_flux_dn_dir_k(i,k);
d_lw_clnclrsky_flux_up(icol,k) = lw_clnclrsky_flux_up_k(i,k);
d_lw_clnclrsky_flux_dn(icol,k) = lw_clnclrsky_flux_dn_k(i,k);
d_lw_clrsky_flux_up(icol,k) = lw_clrsky_flux_up_k(i,k);
d_lw_clrsky_flux_dn(icol,k) = lw_clrsky_flux_dn_k(i,k);
d_lw_clnsky_flux_up(icol,k) = lw_clnsky_flux_up_k(i,k);
d_lw_clnsky_flux_dn(icol,k) = lw_clnsky_flux_dn_k(i,k);
});
#endif
// Extract optical properties for COSP
Kokkos::parallel_for(Kokkos::TeamVectorRange(team, nlay), [&] (const int& k) {
d_dtau067(icol,k) = cld_tau_sw_bnd_k(i,k,idx_067_k);
Expand Down
Loading