Add latent_heat views to temporary workspace instead of allocating ev…

…ery timestep

components/eamxx/src/physics/p3/CMakeLists.txt

@@ -45,7 +45,6 @@ if (NOT EAMXX_ENABLE_GPU OR Kokkos_ENABLE_CUDA_RELOCATABLE_DEVICE_CODE OR Kokkos
     eti/p3_ice_relaxation_timescale.cpp
     eti/p3_ice_nucleation.cpp
     eti/p3_ice_cldliq_wet_growth.cpp
-    eti/p3_get_latent_heat.cpp
     eti/p3_check_values.cpp
     eti/p3_incloud_mixingratios.cpp
     eti/p3_subgrid_variance_scaling.cpp
@@ -63,13 +62,13 @@ endif()
 
 # List of dispatch source files if monolithic kernels are off
 set(P3_SK_SRCS
-    disp/p3_check_values_impl_disp.cpp  
-    disp/p3_ice_sed_impl_disp.cpp  
-    disp/p3_main_impl_part1_disp.cpp  
+    disp/p3_check_values_impl_disp.cpp
+    disp/p3_ice_sed_impl_disp.cpp
+    disp/p3_main_impl_part1_disp.cpp
     disp/p3_main_impl_part3_disp.cpp
-    disp/p3_cloud_sed_impl_disp.cpp     
-    disp/p3_main_impl_disp.cpp     
-    disp/p3_main_impl_part2_disp.cpp  
+    disp/p3_cloud_sed_impl_disp.cpp
+    disp/p3_main_impl_disp.cpp
+    disp/p3_main_impl_part2_disp.cpp
     disp/p3_rain_sed_impl_disp.cpp
     )
 

components/eamxx/src/physics/p3/disp/p3_main_impl_disp.cpp

@@ -22,6 +22,7 @@ ::p3_main_init_disp(
   const uview_2d<Spack>& diag_eff_radius_qr,
   const uview_2d<Spack>& inv_cld_frac_i, const uview_2d<Spack>& inv_cld_frac_l, const uview_2d<Spack>& inv_cld_frac_r,
   const uview_2d<Spack>& exner, const uview_2d<Spack>& T_atm, const uview_2d<Spack>& qv, const uview_2d<Spack>& inv_dz,
+  const uview_2d<Spack>& latent_heat_vapor, const uview_2d<Spack>& latent_heat_sublim, const uview_2d<Spack>& latent_heat_fusion,
   const uview_1d<Scalar>& precip_liq_surf, const uview_1d<Scalar>& precip_ice_surf,
   const uview_2d<Spack>& mu_r, const uview_2d<Spack>& lamr, const uview_2d<Spack>& logn0r, const uview_2d<Spack>& nu,
   const uview_2d<Spack>& cdist, const uview_2d<Spack>& cdist1, const uview_2d<Spack>& cdistr,
@@ -43,6 +44,9 @@ ::p3_main_init_disp(
     precip_liq_surf(i) = 0;
     precip_ice_surf(i) = 0;
 
+    constexpr Scalar latvap = C::LatVap;
+    constexpr Scalar latice = C::LatIce;
+
     Kokkos::parallel_for(
       Kokkos::TeamVectorRange(team, nk_pack), [&] (Int k) {
         diag_equiv_reflectivity(i,k) = -99;
@@ -58,6 +62,9 @@ ::p3_main_init_disp(
         T_atm(i,k)                 = th_atm(i,k) * exner(i,k);
         qv(i,k)                = max(qv(i,k), 0);
         inv_dz(i,k)            = 1 / dz(i,k);
+        latent_heat_vapor(i,k)  = latvap;
+        latent_heat_sublim(i,k) = latvap+latice;
+        latent_heat_fusion(i,k) = latice;
         mu_r(i,k)               = 0.;
         lamr(i,k)               = 0.;
         logn0r(i,k)             = 0.;
@@ -115,10 +122,6 @@ ::p3_main_internal_disp(
 {
   using ExeSpace = typename KT::ExeSpace;
 
-  view_2d<Spack> latent_heat_sublim("latent_heat_sublim", nj, nk), latent_heat_vapor("latent_heat_vapor", nj, nk), latent_heat_fusion("latent_heat_fusion", nj, nk);
-
-  get_latent_heat(nj, nk, latent_heat_vapor, latent_heat_sublim, latent_heat_fusion);
-
   const Int nk_pack = ekat::npack<Spack>(nk);
 
   // load constants into local vars
@@ -225,6 +228,9 @@ ::p3_main_internal_disp(
   auto flux_qit                = temporaries.flux_qit;
   auto v_qr                    = temporaries.v_qr;
   auto v_nr                    = temporaries.v_nr;
+  auto latent_heat_vapor       = temporaries.latent_heat_vapor;
+  auto latent_heat_sublim      = temporaries.latent_heat_sublim;
+  auto latent_heat_fusion      = temporaries.latent_heat_fusion;
 
   // we do not want to measure init stuff
   auto start = std::chrono::steady_clock::now();
@@ -234,6 +240,7 @@ ::p3_main_internal_disp(
       nj, nk_pack, cld_frac_i, cld_frac_l, cld_frac_r, inv_exner, th, dz, diag_equiv_reflectivity,
       ze_ice, ze_rain, diag_eff_radius_qc, diag_eff_radius_qi, diag_eff_radius_qr,
       inv_cld_frac_i, inv_cld_frac_l, inv_cld_frac_r, exner, T_atm, qv, inv_dz,
+      latent_heat_vapor, latent_heat_sublim, latent_heat_fusion,
       diagnostic_outputs.precip_liq_surf, diagnostic_outputs.precip_ice_surf,
       mu_r, lamr, logn0r, nu, cdist, cdist1, cdistr,
       qc_incld, qr_incld, qi_incld, qm_incld, nc_incld, nr_incld, ni_incld, bm_incld,

components/eamxx/src/physics/p3/eamxx_p3_process_interface.cpp

@@ -139,7 +139,7 @@ size_t P3Microphysics::requested_buffer_size_in_bytes() const
 
   // Number of Reals needed by the WorkspaceManager passed to p3_main
   const auto policy       = ekat::ExeSpaceUtils<KT::ExeSpace>::get_default_team_policy(m_num_cols, nk_pack);
-  const size_t wsm_request   = WSM::get_total_bytes_needed(nk_pack_p1, 52, policy);
+  const size_t wsm_request   = WSM::get_total_bytes_needed(nk_pack_p1, 55, policy);
 
   return interface_request + wsm_request;
 }
@@ -188,7 +188,7 @@ void P3Microphysics::init_buffers(const ATMBufferManager &buffer_manager)
     &m_buffer.ntend_ignore, &m_buffer.mu_c, &m_buffer.lamc, &m_buffer.qr_evap_tend, &m_buffer.v_qc,
     &m_buffer.v_nc, &m_buffer.flux_qx, &m_buffer.flux_nx, &m_buffer.v_qit, &m_buffer.v_nit,
     &m_buffer.flux_nit, &m_buffer.flux_bir, &m_buffer.flux_qir, &m_buffer.flux_qit, &m_buffer.v_qr,
-    &m_buffer.v_nr
+    &m_buffer.v_nr, &m_buffer.latent_heat_vapor, &m_buffer.latent_heat_sublim, &m_buffer.latent_heat_fusion
 #endif
   };
   for (int i=0; i<Buffer::num_2d_vector; ++i) {
@@ -210,7 +210,7 @@ void P3Microphysics::init_buffers(const ATMBufferManager &buffer_manager)
   // Compute workspace manager size to check used memory
   // vs. requested memory
   const auto policy  = ekat::ExeSpaceUtils<KT::ExeSpace>::get_default_team_policy(m_num_cols, nk_pack);
-  const int wsm_size = WSM::get_total_bytes_needed(nk_pack_p1, 52, policy)/sizeof(Spack);
+  const int wsm_size = WSM::get_total_bytes_needed(nk_pack_p1, 55, policy)/sizeof(Spack);
   s_mem += wsm_size;
 
   size_t used_mem = (reinterpret_cast<Real*>(s_mem) - buffer_manager.get_memory())*sizeof(Real);
@@ -398,6 +398,9 @@ void P3Microphysics::initialize_impl (const RunType /* run_type */)
   temporaries.flux_qit                = m_buffer.flux_qit;
   temporaries.v_qr                    = m_buffer.v_qr;
   temporaries.v_nr                    = m_buffer.v_nr;
+  temporaries.latent_heat_vapor       = m_buffer.latent_heat_vapor;
+  temporaries.latent_heat_sublim      = m_buffer.latent_heat_sublim;
+  temporaries.latent_heat_fusion      = m_buffer.latent_heat_fusion;
 #endif
 
   // -- Set values for the post-amble structure
@@ -427,7 +430,7 @@ void P3Microphysics::initialize_impl (const RunType /* run_type */)
 
   // Setup WSM for internal local variables
   const auto policy = ekat::ExeSpaceUtils<KT::ExeSpace>::get_default_team_policy(m_num_cols, nk_pack);
-  workspace_mgr.setup(m_buffer.wsm_data, nk_pack_p1, 52, policy);
+  workspace_mgr.setup(m_buffer.wsm_data, nk_pack_p1, 55, policy);
 }
 
 // =========================================================================================

components/eamxx/src/physics/p3/eamxx_p3_process_interface.hpp

@@ -362,7 +362,7 @@ class P3Microphysics : public AtmosphereProcess
     static constexpr int num_1d_scalar = 2; //no 2d vars now, but keeping 1d struct for future expansion
     // 2d view packed, size (ncol, nlev_packs)
 #ifdef SCREAM_P3_SMALL_KERNELS
-    static constexpr int num_2d_vector = 64;
+    static constexpr int num_2d_vector = 67;
 #else
     static constexpr int num_2d_vector = 8;
 #endif
@@ -393,7 +393,7 @@ class P3Microphysics : public AtmosphereProcess
       diag_diam_qi, pratot, prctot, qtend_ignore, ntend_ignore,
       mu_c, lamc, qr_evap_tend, v_qc, v_nc, flux_qx, flux_nx,
       v_qit, v_nit, flux_nit, flux_bir, flux_qir, flux_qit,
-      v_qr, v_nr;
+      v_qr, v_nr, latent_heat_vapor, latent_heat_sublim, latent_heat_fusion;
 #endif
 
     suview_2d col_location;

components/eamxx/src/physics/p3/impl/p3_main_impl.hpp

@@ -40,10 +40,16 @@ ::p3_main_init(
   const uview_1d<Spack>& T_atm,
   const uview_1d<Spack>& qv,
   const uview_1d<Spack>& inv_dz,
+  const uview_1d<Spack>& latent_heat_vapor,
+  const uview_1d<Spack>& latent_heat_sublim,
+  const uview_1d<Spack>& latent_heat_fusion,
   Scalar& precip_liq_surf,
   Scalar& precip_ice_surf,
   view_1d_ptr_array<Spack, 36>& zero_init)
 {
+  constexpr Scalar latvap = C::LatVap;
+  constexpr Scalar latice = C::LatIce;
+
   precip_liq_surf = 0;
   precip_ice_surf = 0;
 
@@ -63,6 +69,10 @@ ::p3_main_init(
     T_atm(k)                 = th_atm(k) * exner(k);
     qv(k)                = max(qv(k), 0);
     inv_dz(k)            = 1 / dz(k);
+    // TODO: use constants througout P3 for latent_heat instead of views
+    latent_heat_vapor(k) = latvap;
+    latent_heat_sublim(k) = latvap+latice;
+    latent_heat_fusion(k) = latice;
 
     for (size_t j = 0; j < zero_init.size(); ++j) {
       (*zero_init[j])(k) = 0;
@@ -89,10 +99,6 @@ ::p3_main_internal(
   using ExeSpace = typename KT::ExeSpace;
   using ScratchViewType = Kokkos::View<bool*, typename ExeSpace::scratch_memory_space>;
 
-  view_2d<Spack> latent_heat_sublim("latent_heat_sublim", nj, nk), latent_heat_vapor("latent_heat_vapor", nj, nk), latent_heat_fusion("latent_heat_fusion", nj, nk);
-
-  get_latent_heat(nj, nk, latent_heat_vapor, latent_heat_sublim, latent_heat_fusion);
-
   const Int nk_pack = ekat::npack<Spack>(nk);
   const auto scratch_size = ScratchViewType::shmem_size(2);
   const auto policy = ekat::ExeSpaceUtils<ExeSpace>::get_default_team_policy(nj, nk_pack).set_scratch_size(0, Kokkos::PerTeam(scratch_size));
@@ -141,9 +147,12 @@ ::p3_main_internal(
       qtend_ignore, ntend_ignore,
 
       // Variables still used in F90 but removed from C++ interface
-      mu_c, lamc, qr_evap_tend;
+      mu_c, lamc, qr_evap_tend,
+
+      // TODO: use contants instead of WSM vars for these values
+      latent_heat_vapor, latent_heat_sublim, latent_heat_fusion;
 
-    workspace.template take_many_and_reset<44>(
+    workspace.template take_many_and_reset<47>(
       {
         "mu_r", "T_atm", "lamr", "logn0r", "nu", "cdist", "cdist1", "cdistr",
         "inv_cld_frac_i", "inv_cld_frac_l", "inv_cld_frac_r", "qc_incld", "qr_incld", "qi_incld", "qm_incld",
@@ -152,7 +161,7 @@ ::p3_main_internal(
         "rhofacr", "rhofaci", "acn", "qv_sat_l", "qv_sat_i", "sup", "qv_supersat_i",
         "tmparr1", "exner", "diag_equiv_reflectivity", "diag_vm_qi", "diag_diam_qi",
         "pratot", "prctot", "qtend_ignore", "ntend_ignore",
-        "mu_c", "lamc", "qr_evap_tend"
+        "mu_c", "lamc", "qr_evap_tend", "latent_heat_vapor", "latent_heat_sublim", "latent_heat_fusion"
       },
       {
         &mu_r, &T_atm, &lamr, &logn0r, &nu, &cdist, &cdist1, &cdistr,
@@ -162,7 +171,7 @@ ::p3_main_internal(
         &rhofacr, &rhofaci, &acn, &qv_sat_l, &qv_sat_i, &sup, &qv_supersat_i,
         &tmparr1, &exner, &diag_equiv_reflectivity, &diag_vm_qi, &diag_diam_qi,
         &pratot, &prctot, &qtend_ignore, &ntend_ignore,
-        &mu_c, &lamc, &qr_evap_tend
+        &mu_c, &lamc, &qr_evap_tend, &latent_heat_vapor, &latent_heat_sublim, &latent_heat_fusion
       });
 
     // Get single-column subviews of all inputs, shouldn't need any i-indexing
@@ -201,9 +210,6 @@ ::p3_main_internal(
     const auto oliq_ice_exchange   = ekat::subview(history_only.liq_ice_exchange, i);
     const auto ovap_liq_exchange   = ekat::subview(history_only.vap_liq_exchange, i);
     const auto ovap_ice_exchange   = ekat::subview(history_only.vap_ice_exchange, i);
-    const auto olatent_heat_vapor  = ekat::subview(latent_heat_vapor, i);
-    const auto olatent_heat_sublim = ekat::subview(latent_heat_sublim, i);
-    const auto olatent_heat_fusion = ekat::subview(latent_heat_fusion, i);
     const auto oqv_prev            = ekat::subview(diagnostic_inputs.qv_prev, i);
     const auto ot_prev             = ekat::subview(diagnostic_inputs.t_prev, i);
 
@@ -228,12 +234,13 @@ ::p3_main_internal(
       ocld_frac_i, ocld_frac_l, ocld_frac_r, oinv_exner, oth, odz, diag_equiv_reflectivity,
       ze_ice, ze_rain, odiag_eff_radius_qc, odiag_eff_radius_qi, odiag_eff_radius_qr,
       inv_cld_frac_i, inv_cld_frac_l, inv_cld_frac_r, exner, T_atm, oqv, inv_dz,
+      latent_heat_vapor, latent_heat_sublim, latent_heat_fusion,
       diagnostic_outputs.precip_liq_surf(i), diagnostic_outputs.precip_ice_surf(i), zero_init);
 
     p3_main_part1(
       team, nk, infrastructure.predictNc, infrastructure.prescribedCCN, infrastructure.dt,
       opres, odpres, odz, onc_nuceat_tend, onccn_prescribed, oinv_exner, exner, inv_cld_frac_l, inv_cld_frac_i,
-      inv_cld_frac_r, olatent_heat_vapor, olatent_heat_sublim, olatent_heat_fusion,
+      inv_cld_frac_r, latent_heat_vapor, latent_heat_sublim, latent_heat_fusion,
       T_atm, rho, inv_rho, qv_sat_l, qv_sat_i, qv_supersat_i, rhofacr,
       rhofaci, acn, oqv, oth, oqc, onc, oqr, onr, oqi, oni, oqm,
       obm, qc_incld, qr_incld, qi_incld, qm_incld, nc_incld, nr_incld,
@@ -252,8 +259,8 @@ ::p3_main_internal(
       lookup_tables.dnu_table_vals, lookup_tables.ice_table_vals, lookup_tables.collect_table_vals, lookup_tables.revap_table_vals, opres, odpres, odz, onc_nuceat_tend, oinv_exner,
       exner, inv_cld_frac_l, inv_cld_frac_i, inv_cld_frac_r, oni_activated, oinv_qc_relvar, ocld_frac_i,
       ocld_frac_l, ocld_frac_r, oqv_prev, ot_prev, T_atm, rho, inv_rho, qv_sat_l, qv_sat_i, qv_supersat_i, rhofacr, rhofaci, acn,
-      oqv, oth, oqc, onc, oqr, onr, oqi, oni, oqm, obm, olatent_heat_vapor,
-      olatent_heat_sublim, olatent_heat_fusion, qc_incld, qr_incld, qi_incld, qm_incld, nc_incld,
+      oqv, oth, oqc, onc, oqr, onr, oqi, oni, oqm, obm, latent_heat_vapor,
+      latent_heat_sublim, latent_heat_fusion, qc_incld, qr_incld, qi_incld, qm_incld, nc_incld,
       nr_incld, ni_incld, bm_incld, mu_c, nu, lamc, cdist, cdist1, cdistr,
       mu_r, lamr, logn0r, oqv2qi_depos_tend, oprecip_total_tend, onevapr, qr_evap_tend,
       ovap_liq_exchange, ovap_ice_exchange, oliq_ice_exchange,
@@ -296,7 +303,7 @@ ::p3_main_internal(
 
     // homogeneous freezing of cloud and rain
     homogeneous_freezing(
-      T_atm, oinv_exner, olatent_heat_fusion, team, nk, ktop, kbot, kdir, oqc, onc, oqr, onr, oqi,
+      T_atm, oinv_exner, latent_heat_fusion, team, nk, ktop, kbot, kdir, oqc, onc, oqr, onr, oqi,
       oni, oqm, obm, oth);
 
     //
@@ -306,7 +313,7 @@ ::p3_main_internal(
     p3_main_part3(
       team, nk_pack, runtime_options.max_total_ni, lookup_tables.dnu_table_vals, lookup_tables.ice_table_vals, oinv_exner, ocld_frac_l, ocld_frac_r, ocld_frac_i,
       rho, inv_rho, rhofaci, oqv, oth, oqc, onc, oqr, onr, oqi, oni,
-      oqm, obm, olatent_heat_vapor, olatent_heat_sublim, mu_c, nu, lamc, mu_r, lamr,
+      oqm, obm, latent_heat_vapor, latent_heat_sublim, mu_c, nu, lamc, mu_r, lamr,
       ovap_liq_exchange, ze_rain, ze_ice, diag_vm_qi, odiag_eff_radius_qi, diag_diam_qi,
       orho_qi, diag_equiv_reflectivity, odiag_eff_radius_qc, odiag_eff_radius_qr, p3constants);
 

components/eamxx/src/physics/p3/p3_functions.hpp

@@ -282,6 +282,9 @@ struct Functions
     view_2d<Spack> flux_qir, flux_qit;
     // rain sedimentation
     view_2d<Spack> v_qr, v_nr;
+    // latent heat
+    // TODO: Remove and just use constants
+    view_2d<Spack> latent_heat_vapor, latent_heat_sublim, latent_heat_fusion;
   };
 #endif
 
@@ -899,9 +902,6 @@ struct Functions
                                     Smask& log_wetgrowth, Spack& qr2qi_collect_tend, Spack& qc2qi_collect_tend, Spack& qc_growth_rate,
 				    Spack& nr_ice_shed_tend, Spack& qc2qr_ice_shed_tend, const Smask& context = Smask(true));
 
-  // Note: not a kernel function
-  static void get_latent_heat(const Int& nj, const Int& nk, view_2d<Spack>& v, view_2d<Spack>& s, view_2d<Spack>& f);
-
   KOKKOS_FUNCTION
   static void check_values(const uview_1d<const Spack>& qv, const uview_1d<const Spack>& temp, const Int& ktop, const Int& kbot,
                            const Int& timestepcount, const bool& force_abort, const Int& source_ind, const MemberType& team,
@@ -949,6 +949,9 @@ struct Functions
     const uview_1d<Spack>& T_atm,
     const uview_1d<Spack>& qv,
     const uview_1d<Spack>& inv_dz,
+    const uview_1d<Spack>& latent_heat_vapor,
+    const uview_1d<Spack>& latent_heat_sublim,
+    const uview_1d<Spack>& latent_heat_fusion,
     Scalar& precip_liq_surf,
     Scalar& precip_ice_surf,
     view_1d_ptr_array<Spack, 36>& zero_init);
@@ -964,7 +967,9 @@ struct Functions
     const uview_2d<Spack>& diag_eff_radius_qi, const uview_2d<Spack>& diag_eff_radius_qr, const uview_2d<Spack>& inv_cld_frac_i,
     const uview_2d<Spack>& inv_cld_frac_l, const uview_2d<Spack>& inv_cld_frac_r,
     const uview_2d<Spack>& exner, const uview_2d<Spack>& T_atm, const uview_2d<Spack>& qv,
-    const uview_2d<Spack>& inv_dz, const uview_1d<Scalar>& precip_liq_surf, const uview_1d<Scalar>& precip_ice_surf,
+    const uview_2d<Spack>& inv_dz,
+    const uview_2d<Spack>& latent_heat_vapor, const uview_2d<Spack>& latent_heat_sublim, const uview_2d<Spack>& latent_heat_fusion,
+    const uview_1d<Scalar>& precip_liq_surf, const uview_1d<Scalar>& precip_ice_surf,
     const uview_2d<Spack>& mu_r, const uview_2d<Spack>& lamr, const uview_2d<Spack>& logn0r, const uview_2d<Spack>& nu,
     const uview_2d<Spack>& cdist, const uview_2d<Spack>& cdist1, const uview_2d<Spack>& cdistr,
     const uview_2d<Spack>& qc_incld, const uview_2d<Spack>& qr_incld, const uview_2d<Spack>& qi_incld,
@@ -1438,7 +1443,6 @@ void init_tables_from_f90_c(Real* vn_table_vals_data, Real* vm_table_vals_data,
 # include "p3_ice_melting_impl.hpp"
 # include "p3_calc_liq_relaxation_timescale_impl.hpp"
 # include "p3_ice_cldliq_wet_growth_impl.hpp"
-# include "p3_get_latent_heat_impl.hpp"
 # include "p3_check_values_impl.hpp"
 # include "p3_incloud_mixingratios_impl.hpp"
 # include "p3_subgrid_variance_scaling_impl.hpp"