add Dmitry's solution

main.cpp

@@ -173,6 +173,38 @@ void scenario_4(double* array, int nblocks, int nlev, int nproma, bool print=tru
 
 }
 
+void scenario_4b(double* array, int nblocks, int nlev, int nproma, bool print = true) {
+  if (print)
+    std::cout << "scenario 4b (Dmitry's solution): view(array, nproma, nlev, nblocks); d_view(jc, jk, jb) ----- "
+              << std::endl;
+
+  Kokkos::View<double***, Kokkos::HostSpace, Kokkos::MemoryUnmanaged> view(array, nproma, nlev, nblocks);
+
+  using space_t   = Kokkos::DefaultExecutionSpace::memory_space;
+  auto d_view_tmp = Kokkos::create_mirror_view_and_copy(space_t(), view);
+  Kokkos::View<double***, space_t> d_view("d_view", nproma, nlev, nblocks);
+  Kokkos::deep_copy(d_view_tmp, view);
+  Kokkos::deep_copy(d_view, d_view_tmp);
+
+  timer.reset();
+  for (int jb = 0; jb < nblocks; ++jb)
+    Kokkos::parallel_for(
+        "", Kokkos::RangePolicy<>(0, nproma), KOKKOS_LAMBDA(const int jc) {
+          for (int jk = 0; jk < nlev; ++jk) {
+            int p              = jc * nlev * nblocks + jk * nblocks + jb;
+            d_view(jc, jk, jb) = p;
+
+            //  printf("%f ", d_view(jb, jk, jc));
+          }
+        });
+
+  Kokkos::fence();
+  if (print) printf("Time = %f ms\n\n", timer.seconds() * 1000);
+  Kokkos::deep_copy(d_view_tmp, d_view);
+  Kokkos::deep_copy(view, d_view_tmp);
+  validate(array, nblocks, nlev, nproma);
+}
+
 #if defined(KOKKOS_ENABLE_CUDA) or defined(KOKKOS_ENABLE_HIP)
     #define gpu 1
     using Layout = Kokkos::LayoutLeft;
@@ -346,6 +378,7 @@ int main() {
     scenario_2b(array, nblocks, nlev, nproma);
     scenario_3(array, nblocks, nlev, nproma);
     scenario_4(array, nblocks, nlev, nproma);
+    scenario_4b(array, nblocks, nlev, nproma);
     scenario_5(array, nblocks, nlev, nproma);
     scenario_6(array, nblocks, nlev, nproma);