Merge Pull Request #2954 from E3SM-Project/scream/mahf708/fix-rrtmgp-k-cuda

Automatically Merged using E3SM Pull Request AutoTester
PR Title: fix errors and warning in rrtmgp-k
PR Author: mahf708
PR LABELS: radiation, AT: AUTOMERGE, CUDA, pm-gpu

Author: E3SM-Bot

components/eamxx/src/physics/rrtmgp/eamxx_rrtmgp_process_interface.cpp

@@ -1171,10 +1171,24 @@ void RRTMGPRadiation::run_impl (const double dt) {
               aero_tau_lw_k(i,k,b) = d_aero_tau_lw(icol,b,k);
             });
           } else {
-            Kokkos::deep_copy(aero_tau_sw_k, 0);
-            Kokkos::deep_copy(aero_ssa_sw_k, 0);
-            Kokkos::deep_copy(aero_g_sw_k  , 0);
-            Kokkos::deep_copy(aero_tau_lw_k, 0);
+            // cuda complains (in warning only) about these being not allowed...
+            // Kokkos::deep_copy(aero_tau_sw_k, 0);
+            // Kokkos::deep_copy(aero_ssa_sw_k, 0);
+            // Kokkos::deep_copy(aero_g_sw_k  , 0);
+            // Kokkos::deep_copy(aero_tau_lw_k, 0);
+            // So, do the manual labor instead:
+            Kokkos::parallel_for(Kokkos::TeamVectorRange(team, nswbands*nlay), [&] (const int&idx) {
+              auto b = idx / nlay;
+              auto k = idx % nlay;
+              aero_tau_sw_k(i,k,b) = 0.0;
+              aero_ssa_sw_k(i,k,b) = 0.0;
+              aero_g_sw_k  (i,k,b) = 0.0;
+            });
+            Kokkos::parallel_for(Kokkos::TeamVectorRange(team, nlwbands*nlay), [&] (const int&idx) {
+              auto b = idx / nlay;
+              auto k = idx % nlay;
+              aero_tau_lw_k(i,k,b) = 0.0;
+            });
           }
 #endif
         });

components/eamxx/src/physics/rrtmgp/rrtmgp_utils.hpp

@@ -112,7 +112,7 @@ bool check_range(T x, Real xmin, Real xmax, std::string msg, std::ostream& out=s
 }
 #endif
 #ifdef RRTMGP_ENABLE_KOKKOS
-template <class T, typename std::enable_if<T::rank == 1>::type* = nullptr>
+template <class T, typename std::enable_if<T::rank == 1>::type* dummy = nullptr>
 bool check_range_k(T x, typename T::const_value_type xmin, typename T::const_value_type xmax,
                    std::string msg, std::ostream& out=std::cout) {
   bool pass = true;
@@ -137,7 +137,7 @@ bool check_range_k(T x, typename T::const_value_type xmin, typename T::const_val
   return pass;
 }
 
-template <class T, typename std::enable_if<T::rank == 2>::type* = nullptr>
+template <class T, typename std::enable_if<T::rank == 2>::type* dummy = nullptr>
 bool check_range_k(T x, typename T::const_value_type xmin, typename T::const_value_type xmax,
                    std::string msg, std::ostream& out=std::cout) {
   bool pass = true;
@@ -166,7 +166,7 @@ bool check_range_k(T x, typename T::const_value_type xmin, typename T::const_val
   return pass;
 }
 
-template <class T, typename std::enable_if<T::rank == 3>::type* = nullptr>
+template <class T, typename std::enable_if<T::rank == 3>::type* dummy = nullptr>
 bool check_range_k(T x, typename T::const_value_type xmin, typename T::const_value_type xmax,
                    std::string msg, std::ostream& out=std::cout) {
   bool pass = true;

components/eamxx/src/physics/rrtmgp/scream_rrtmgp_interface.hpp

@@ -1190,14 +1190,14 @@ static void mixing_ratio_to_cloud_mass(
  * property look-up tables, but could be used to limit other
  * fields as well.
  */
-template<typename InT, typename T, typename OutT, typename std::enable_if<OutT::rank == 1>::type* = nullptr>
+template<typename InT, typename T, typename OutT, typename std::enable_if<OutT::rank == 1>::type* dummy = nullptr>
 static void limit_to_bounds_k(InT const &arr_in, T const lower, T const upper, OutT &arr_out) {
   Kokkos::parallel_for(arr_out.size(), KOKKOS_LAMBDA(int i) {
     arr_out(i) = std::min(std::max(arr_in(i), lower), upper);
   });
 }
 
-template<typename InT, typename T, typename OutT, typename std::enable_if<OutT::rank == 2>::type* = nullptr>
+template<typename InT, typename T, typename OutT, typename std::enable_if<OutT::rank == 2>::type* dummy = nullptr>
 static void limit_to_bounds_k(InT const &arr_in, T const lower, T const upper, OutT &arr_out) {
   Kokkos::parallel_for(MDRP::template get<2>({arr_out.extent(0), arr_out.extent(1)}), KOKKOS_LAMBDA(int i, int j) {
     arr_out(i, j) = std::min(std::max(arr_in(i, j), lower), upper);