Merge pull request #1970 from CliMA/ck/fill_bench

Fix fill benchmark n-reads-writes

Author: charleskawczynski

test/DataLayouts/benchmark_fill.jl

@@ -24,7 +24,7 @@ function benchmarkfill!(bm, device, data, val, name)
     trial = @benchmark ClimaComms.@cuda_sync $device fill!($data, $val)
     t_min = minimum(trial.times) * 1e-9 # to seconds
     nreps = length(trial.times)
-    n_reads_writes = DataLayouts.ncomponents(data) * 2
+    n_reads_writes = DataLayouts.ncomponents(data)
     push_info(
         bm;
         kernel_time_s = t_min,

test/Spaces/distributed_cuda/ddss2.jl

@@ -108,7 +108,7 @@ pid, nprocs = ClimaComms.init(context)
     end
 #! format: on
     p = @allocated Spaces.weighted_dss!(y0, dss_buffer)
-    iamroot && @test p ≤ 8832
+    iamroot && @test p ≤ 9088
 
     #testing weighted dss on a vector field
     init_vectorstate(local_geometry, p) = Geometry.Covariant12Vector(1.0, -1.0)