From 9deb8f8ef2c292b611253a998da1ec64e2ce7872 Mon Sep 17 00:00:00 2001
From: Luca Bertagna <lbertag@sandia.gov>
Date: Wed, 12 Oct 2022 14:30:04 -0600
Subject: [PATCH 1/3] Fix CPP precision settings in kokkos stuff unit tests

---
 tests/kokkos/CMakeLists.txt | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tests/kokkos/CMakeLists.txt b/tests/kokkos/CMakeLists.txt
index f6641b82..013283a3 100644
--- a/tests/kokkos/CMakeLists.txt
+++ b/tests/kokkos/CMakeLists.txt
@@ -5,7 +5,7 @@ if (EKAT_TEST_DOUBLE_PRECISION)
   EkatCreateUnitTest(kokkos_utils${DP_POSTFIX} kokkos_utils_tests.cpp
     LIBS ekat
     PRINT_OMP_AFFINITY
-    COMPILER_DEFS EKAT_TEST_SINGLE_PRECISION
+    COMPILER_DEFS EKAT_TEST_DOUBLE_PRECISION
     THREADS 1 ${EKAT_TEST_MAX_THREADS} ${EKAT_TEST_THREAD_INC}
   )
 
@@ -31,7 +31,7 @@ if (EKAT_TEST_SINGLE_PRECISION)
   EkatCreateUnitTest(wsm${SP_POSTFIX} workspace_tests.cpp
     LIBS ekat
     PRINT_OMP_AFFINITY
-    COMPILER_DEFS EKAT_TEST_DOUBLE_PRECISION
+    COMPILER_DEFS EKAT_TEST_SINGLE_PRECISION
     THREADS 1 ${EKAT_TEST_MAX_THREADS} ${EKAT_TEST_THREAD_INC}
   )
 endif ()

From 38176bd795efefdd49c476be75592e58a48d08e2 Mon Sep 17 00:00:00 2001
From: Luca Bertagna <lbertag@sandia.gov>
Date: Wed, 12 Oct 2022 17:34:12 -0600
Subject: [PATCH 2/3] Fix view_reduction in case the result var comes from
 shared mem

---
 src/ekat/kokkos/ekat_kokkos_utils.hpp | 20 ++++++++++++++++----
 1 file changed, 16 insertions(+), 4 deletions(-)

diff --git a/src/ekat/kokkos/ekat_kokkos_utils.hpp b/src/ekat/kokkos/ekat_kokkos_utils.hpp
index c394a761..784c1f40 100644
--- a/src/ekat/kokkos/ekat_kokkos_utils.hpp
+++ b/src/ekat/kokkos/ekat_kokkos_utils.hpp
@@ -117,6 +117,17 @@ void view_reduction (const TeamMember& team,
   using PackType = typename std::remove_reference<decltype(input(0))>::type;
   constexpr int vector_size = PackType::n;
 
+  // We need to use a temporary, since we don't know whether result refers to a thread-local
+  // variable (e.g., automatic variable) or to shared-memory (e.g., an entry of a view).
+  // Hence, perform calculations on a local var, then copy back into the output result.
+  ValueType temp = result;
+
+  // Note: this team barrier is needed in some extreme case. Without it, it *could* happen that,
+  //       if result is a ref to shared mem (e.g., an entry of a view) rather than thread-local,
+  //       one team member might reach the end of the fcn (hence, updating result) *before*
+  //       another thread might have the chance to init temp.
+  team.team_barrier();
+
   // Perform a packed reduction over scalar indices
   const bool has_garbage_begin = begin%vector_size != 0;
   const bool has_garbage_end   = end%vector_size != 0;
@@ -130,7 +141,7 @@ void view_reduction (const TeamMember& team,
     const int first_indx = begin%vector_size;
     Kokkos::single(Kokkos::PerThread(team),[&] {
       for (int j=first_indx; j<vector_size; ++j) {
-        result += temp_input[j];
+        temp += temp_input[j];
       }
     });
   }
@@ -143,7 +154,7 @@ void view_reduction (const TeamMember& team,
                                        [&](const int k, ValueType& local_sum) {
           // Sum over pack entries and add to local_sum
           ekat::reduce_sum<Serialize>(input(k),local_sum);
-      }, result);
+      }, temp);
     } else {
       PackType packed_result(0);
       impl::parallel_reduce<Serialize>(team, pack_loop_begin, pack_loop_end,
@@ -152,7 +163,7 @@ void view_reduction (const TeamMember& team,
           local_packed_sum += input(k);
       }, packed_result);
 
-      result += ekat::reduce_sum<Serialize>(packed_result);
+      temp += ekat::reduce_sum<Serialize>(packed_result);
     }
   }
 
@@ -165,10 +176,11 @@ void view_reduction (const TeamMember& team,
     ConstExceptGnu int last_indx = end%vector_size;
     Kokkos::single(Kokkos::PerThread(team),[&] {
       for (int j=0; j<last_indx; ++j) {
-        result += temp_input[j];
+        temp += temp_input[j];
       }
     });
   }
+  result = temp;
 }
 } //namespace impl
 

From 4c8d6495b80c6a8fd6e1b8c940a28b9d3f5f894a Mon Sep 17 00:00:00 2001
From: Luca Bertagna <lbertag@sandia.gov>
Date: Wed, 12 Oct 2022 17:36:20 -0600
Subject: [PATCH 3/3] Use pack size to 2^n while testing view_reduction

---
 tests/kokkos/kokkos_utils_tests.cpp | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/tests/kokkos/kokkos_utils_tests.cpp b/tests/kokkos/kokkos_utils_tests.cpp
index 5b2061bf..14cb003c 100644
--- a/tests/kokkos/kokkos_utils_tests.cpp
+++ b/tests/kokkos/kokkos_utils_tests.cpp
@@ -364,10 +364,10 @@ TEST_CASE("view_reduction", "[kokkos_utils]")
   test_view_reduction<Real,false,false,3,4> ();
 
   // Sum subset of entries, non-zero starting value
-  test_view_reduction<Real, true,true,16,3> (1.0/3.0,2,11);
-  test_view_reduction<Real,false,true,16,3> (1.0/3.0,2,11);
-  test_view_reduction<Real, true,false,16,3> (1.0/3.0,2,11);
-  test_view_reduction<Real,false,false,16,3> (1.0/3.0,2,11);
+  test_view_reduction<Real, true,true,18,4> (1.0/3.0,2,11);
+  test_view_reduction<Real,false,true,18,4> (1.0/3.0,2,11);
+  test_view_reduction<Real, true,false,18,4> (1.0/3.0,2,11);
+  test_view_reduction<Real,false,false,18,4> (1.0/3.0,2,11);
 
 }