Replace TeamThreadRange by TeamVectorRange in local_deep_copy

[janciesko]

Replaces TeamThreadRange by TeamVectorRange in local_deep_copy. This potentially improves concurrency.
Implements issue: #4054

[masterleinad]

This looks reasonable. Have you tried to measure performance improvements?

[janciesko]

Good point, let me see if this scales with vec size > 1 now.

[janciesko]

#include <Kokkos_Core.hpp>
using view_t = Kokkos::View<double*>;
using teampolicy_t = Kokkos::TeamPolicy<>;

int main(int argc, char *argv[]) {
    Kokkos::ScopeGuard kokkos(argc, argv);
    int N = argc > 1 ? atoi(argv[1]) : 100000000;
    int tsize = argc > 2 ? atoi(argv[2]) : 1;
    int vecsize = argc > 3 ? atoi(argv[3]) : 1; 
    {
        view_t view = view_t("View", N);        
        teampolicy_t team_policy = teampolicy_t(1,tsize,vecsize);
        Kokkos::Timer timer;
        Kokkos::parallel_for(team_policy,
          KOKKOS_LAMBDA (const teampolicy_t::member_type & team)
          {
                Kokkos::Experimental::local_deep_copy(team, view, 1.234);   
          });

        Kokkos::fence();
        double t = timer.seconds();
        printf("%f\n", t);
    }
}

is giving:

[janciesko]

Retest this please

[dalg24]

(resolved conflicts introduced by merge of #4520)

[janciesko]

Retest this please

[janciesko]

Retest this please